UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays on the economics of aging and housing Kei, Wendy Wai Yee 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_september_kei_wendy.pdf [ 1.73MB ]
Metadata
JSON: 24-1.0307301.json
JSON-LD: 24-1.0307301-ld.json
RDF/XML (Pretty): 24-1.0307301-rdf.xml
RDF/JSON: 24-1.0307301-rdf.json
Turtle: 24-1.0307301-turtle.txt
N-Triples: 24-1.0307301-rdf-ntriples.txt
Original Record: 24-1.0307301-source.json
Full Text
24-1.0307301-fulltext.txt
Citation
24-1.0307301.ris

Full Text

Essays on the economics of aging andhousingbyWendy Wai Yee KeiB.Com. (Hon.), The University of British Columbia, 2007M.Sc.B., The University of British Columbia, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Economics)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)August 2016c©Wendy Wai Yee Kei, 2016AbstractThis dissertation uses three different studies to explore the impact of household finance on elderly house-holds’ labour market activities. Chapter 1 provides an introduction. Chapter 2 uses the eligibility require-ments of the Canadian Old Age Security (OAS) program as a novel approach to estimate a causal effectof public pensions on elderly immigrants’ labour supply decisions. I also evaluate the extent to whichthese individuals may adjust their labour supply behaviour prior to the receipt of public pension entitle-ments. The findings in this chapter illustrate that seniors only respond to the OAS benefits with a decreasein labour force participation rates. A combination of estimates implies that elderly immigrants may exhibitbehavioural response in anticipation for OAS benefits. Chapter 3 investigates the effect of immigration onmobility decisions of native-born near-retirees. The research findings in this chapter push this area of litera-ture forward by suggesting an alternative perspective for explaining native out-migration. The heterogeneityin mobility preferences across dwelling tenure groups is an important result because it may explain whyCard (2001) fails to find any significant effect from immigration on aggregate native relocation decisions.Finally, Chapter 4 explores how the recent house price shock in the U.S. affected the labour supply decisionsof near-retirees. This is the first study to use the national lending conditions for residential mortgage seriesas part of an instrumental variable strategy to explore this context. The final chapter sheds light by showingthat housing exerts insignificant impact on the near-retirees’ work and retirement decisions.iiPrefaceThis dissertation is original, unpublished and independent work by the author, Wendy Wai Yee Kei.iiiTable of contentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Public pensions and elderly immigrants’ labour supply decisions . . . . . . . . . . . . . . . . 32.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 The economics of immigrants’ labour supply decisions and social assistance . . . . . . . . 52.3 Application of the Canadian Old Age Security benefits to the static labour supply model . . 72.4 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.5 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.8 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.9 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 Immigration and housing for the near-retirees . . . . . . . . . . . . . . . . . . . . . . . . . . 593.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.2 Contexts and contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623.3 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633.4 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.5 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693.6 Main findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.8 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89iv3.9 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 954 Housing supply elasticity and the elderly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1174.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1174.2 Background information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1194.3 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1214.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1234.5 Main findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1374.7 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1394.8 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1555 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168AppendicesA Additional tables and figures for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176B Additional tables for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202vList of tables2.1 Share of permanent residents per selected occupational groups . . . . . . . . . . . . . . . . 392.2 Effect of residency requirements on OAS/GIS take-up . . . . . . . . . . . . . . . . . . . . . 402.3 Effect of OAS/GIS on family members’ labour supply decisions . . . . . . . . . . . . . . . 412.4 Effect of OAS/GIS on labour supply decisions of elderly immigrants using different band-widths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432.5 Effect of OAS/GIS on extensive margins of labour supply - 2006 Census versus 2011 Na-tional Household Survey (NHS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442.6 Effect of OAS/GIS on extensive margins of labour supply - 2006 Census versus Survey ofLabour and Income Dynamics (SLID) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462.7 Effect of OAS/GIS on extensive margins of labour supply - A falsification test . . . . . . . . 472.8 Effect of OAS/GIS on extensive margins of labour supply - Social Security Agreement . . . 482.9 Effect of OAS/GIS on extensive margins of labour supply - Anticipation effect . . . . . . . . 512.10 Effect of OAS/GIS on extensive margins of labour supply - High housing asset versus lowhousing asset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542.11 Effect of OAS/GIS on intensive margins of labour supply - Census data . . . . . . . . . . . 552.12 Effect of OAS/GIS on intensive margins of labour supply - GSS data . . . . . . . . . . . . . 573.1 Immigration’s impact on housing market - 2006 Census data . . . . . . . . . . . . . . . . . 953.2 Immigration’s impact on housing market - 2011 National Household Survey data . . . . . . 973.3 Which ethnic groups drive housing value and rental cost growth? . . . . . . . . . . . . . . 993.4 Immigration inflows and older native mobility . . . . . . . . . . . . . . . . . . . . . . . . 1003.5 Immigration inflows and younger native mobility . . . . . . . . . . . . . . . . . . . . . . . 1023.6 Immigration inflows and native mobility - 2011 National Household Survey data . . . . . . . 1043.7 Immigration inflows and mobility decisions for higher income natives who have alreadybeen out of the labour force for at least two years . . . . . . . . . . . . . . . . . . . . . . . 1063.8 Which ethnic groups drive native out-migration? . . . . . . . . . . . . . . . . . . . . . . . . 1083.9 Immigration inflows and native wage growth . . . . . . . . . . . . . . . . . . . . . . . . . . 1123.10 Linking empirical results to theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . 1143.11 Weighted average predicted probabilities of out-migration . . . . . . . . . . . . . . . . . . . 1153.12 Synthetic panel analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1164.1 A test of the exclusion restriction requirement . . . . . . . . . . . . . . . . . . . . . . . . . 1554.2 Difference-in-difference estimations - AHS data . . . . . . . . . . . . . . . . . . . . . . . . 156vi4.3 Difference-in-difference estimations - RAND HRS data . . . . . . . . . . . . . . . . . . . . 1574.4 Triple difference estimations - AHS data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1594.5 Triple difference estimations - RAND HRS data . . . . . . . . . . . . . . . . . . . . . . . . 1604.6 IV estimations for labour supply decisions - AHS data . . . . . . . . . . . . . . . . . . . . 1624.7 IV estimations for labour supply decisions - RAND HRS data . . . . . . . . . . . . . . . . 1634.8 IV estimations for wealth variables - RAND HRS data . . . . . . . . . . . . . . . . . . . . 1654.9 Wealth distributions for ages 55-64 - RAND HRS data . . . . . . . . . . . . . . . . . . . . 166A.1 Effect of OAS/GIS on family members’ labour supply decisions - Difference-in-difference(DD) and triple difference (DDD) estimations . . . . . . . . . . . . . . . . . . . . . . . . . 180A.2 Effect of OAS/GIS on family members’ labour supply decisions - Probit . . . . . . . . . . . 182A.3 Effect of OAS/GIS on family members’ labour supply decisions - Accounting for CanadaPension Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184A.4 Effect of OAS/GIS on family members’ labour supply decisions - Accounting for CanadaPension Plan and for anticipation effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186A.5 Effect of OAS/GIS on family members’ labour supply decisions - Without demographiccontrols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188A.6 Effect of OAS/GIS on family members’ labour supply decisions - With education cohortdummies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190A.7 Effect of OAS/GIS on family members’ labour supply decisions - Imposing legislation totest for anticipation effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191A.8 Effect of OAS/GIS on family members’ labour supply decisions - IV analog of expression(2.1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193B.1 A comparison of housing value and rental cost growth measures . . . . . . . . . . . . . . . 203B.2 Immigration inflows and older native mobility - Probit and logit results . . . . . . . . . . . 204B.3 Immigration inflows and younger native mobility - Probit and logit results . . . . . . . . . . 205B.4 Mobility regressions - Mean values for dependent and independent variables . . . . . . . . 206B.5 Dwelling tenure transition matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208B.6 Historical ethnic distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209B.7 First-stage regression results for “Table 1: Immigration’s impact on housing market” . . . . 210B.8 First-stage regression results for “Table 4: Immigration inflows and older native households’mobility decisions” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214B.9 First-stage regression results for “Table 5: Immigration inflows and younger native house-holds’ mobility decisions” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218B.10 Immigration’s impact on housing market (median values) - 2006 Census data . . . . . . . . 222viiList of figures2.1 Relationship between the number of years since immigrated to Canada and the OAS take-uprate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.2 Static labour supply model for households without working-age children . . . . . . . . . . . 312.3 Static labour supply model for households with working-age children . . . . . . . . . . . . 322.4 Relationship between the number of years since immigrated to Canada and demographiccontrols - Immigrants of ages 65 and over . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.5 Relationship between the number of years since immigrated to Canada and demographiccontrols - Immigrants of ages 25-54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.6 Share of respondents with at least a Bachelor’s degree, for years 1986-2006 . . . . . . . . . 352.7 Number of parent/grandparent immigrants and density, by forcing variable . . . . . . . . . . 362.8 Intensive margins of labour supply - Immigrants of ages 65 and over versus those of ages25-54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372.9 Extensive margins of labour supply - Immigrants of ages 65 and over versus those of ages25-54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383.1 Near-retirees’ labour force participation decisions and immigration inflows . . . . . . . . . 893.2 Labour force participation decisions for all workers and immigration inflows . . . . . . . . . 903.3 Immigration shares and international airport locations - Vancouver CMA . . . . . . . . . . 913.4 Immigration shares and international airport locations - Toronto CMA . . . . . . . . . . . . 923.5 Immigration shares and international airport locations - Winnipeg CMA . . . . . . . . . . . 933.6 Share of near-retirees with HELOC holdings versus immigration shares . . . . . . . . . . . 944.1 U.S. labour force participation rate, debt-to-disposable income ratio, and the aggregatehouse price index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1394.2 Value of primary residence and total net housing wealth for the inelastic and elastic housingsupply regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1404.3 Housing and non-housing debt for the inelastic and elastic housing supply regions . . . . . . 1414.4 Household characteristics - Inelastic versus elastic housing supply regions . . . . . . . . . . 1424.5 Dwelling characteristics - Inelastic versus elastic housing supply regions . . . . . . . . . . . 1434.6 Near-retirees’ labour supply decisions - AHS data . . . . . . . . . . . . . . . . . . . . . . . 1454.7 Near-retirees’ labour supply decisions in levels - RAND HRS data . . . . . . . . . . . . . . 1464.8 Changes in near-retirees’ labour supply decisions - RAND HRS data . . . . . . . . . . . . . 148viii4.9 Changes in labour supply decisions for household heads, split by dwelling tenure status -RAND HRS data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1504.10 Macroeconomic variables - Inelastic versus elastic housing supply regions . . . . . . . . . . 1524.11 U.S. house price index and national lending conditions - Inelastic versus elastic housingsupply regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154A.1 Second-stage results - Without imputation processes based on legislation . . . . . . . . . . . 177A.2 Second-stage results - Imposed imputation processes based on legislation . . . . . . . . . . 178A.3 Immigration share by place of origins and by forcing variable . . . . . . . . . . . . . . . . . 179ixAcknowledgementsCompleting a Ph.D. in Economics degree is an important milestone in my life. I am indebted to many peoplewho have helped me at various stages of this program.First, I would like to thank my research supervisor Dr. Kevin Milligan for his invaluable guidance,patience, and support throughout my Ph.D. program. Kevin is a smart researcher and provides very helpfuland unbiased advice on my career and research work. Without his encouragement, I would not have carvedout my niche in my current area of research. I also thank my dissertation committee members, Dr. NicoleFortin and Dr. Giovanni Gallipoli, for their time and support since the start of the dissertation process. Ibenefited from several discussions with them on empirical strategies, on data knowledge, and on researchideas and implications. I am grateful to have a great thesis supervisory committee. In addition, I thankDr. Hugh Neary for providing me an opportunity to teach as a Sessional Lecturer at UBC and for hisinsightful advice on my teaching skills. This unforgettable teaching experience complements well withdifferent aspects of my research work. Furthermore, I thank Dr. Cheryl Fu and Mr. Lee Grenon at the UBCResearch Data Centre for providing the data and assistance at different stages of my research.I appreciate all of the comments that I received from participants at various UBC seminars, at the Cana-dian Economics Association conferences, and at the Bank of Canada. I would like to thank Dr. ThomasLemieux and Dr. Patrick Francois for hosting Economics 640, which provided an excellent platform forbrainstorming new research ideas. I also thank the following UBC faculty members (in alphabetical order)for their help at different stages of the Ph.D. program: Dr. Werner Antweiler, Dr. Ruth Freedman, Dr. DavidGillen, Dr. David Green, Dr. Keith Head, Dr. Alfred Kong, Dr. Amartya Lahiri, Dr. Vadim Marmer, Dr.Gerald McIntyre, Dr. Steven Minns, Dr. Henry Siu, and Dr. Yaniv Yedid-Levi.Most importantly, I thank my parents for their unconditional love and support. Pursuing a Ph.D. degreeis definitely not an easy task and cannot be completed in a short amount of time. I am very lucky to have thebest parents, who always provide wise advice and are always willing to listen. I would not have completedthis academic marathon without their continued encouragement.xChapter 1IntroductionThis dissertation uses three different studies to explore the impact of household finance on elderly house-holds’ labour market activities. Chapter 2 adapts an alternative strategy to explore how the eligibility re-quirements of the Canadian public pension system affect elderly immigrants’ labour supply decisions. Ialso evaluate the extent to which these individuals may adjust their labour supply behaviour a few yearsprior to the receipt of public pension entitlements. In Chapter 3, I introduce net housing wealth effects intothe prevailing model to explain native out-migration decisions in response to an immigration shock. Thisdeviates from the existing academic framework, which suggests that distaste for immigrants of low socioe-conomic status and of visible minority groups is the primary reason behind native outflow (Sá, 2014; Saizand Wachter, 2011). Finally, Chapter 4 investigates how the recent house price shock in the U.S. affectedthe labour supply decisions of near-retirees. This is the first study to use the national lending conditions forresidential mortgage series as part of an instrumental variable strategy to look into this context.In Chapter 2, I use the eligibility requirements of the Canadian Old Age Security (OAS) program as anovel approach to estimate a causal effect of public pensions on elderly immigrants’ labour supply decisions.In contrast to previous studies, this study can directly compare the labour market responses from publicpension entitlements of newly-arrived immigrants and of early-arrived immigrants. The senior sub-groupresponds to the OAS benefits with a decrease in labour force participation rates. The OAS benefits do notexert any large influence on work intensity. A combination of estimates suggests that elderly immigrantsmay exhibit anticipatory behaviour in response to public pension entitlements.Chapter 3 examines the impact immigration exerts on mobility decisions of Canadian near-retirees.Some of the media as well as policymakers have associated immigration with house price appreciation andlabour out-migration. However, existing academic literature has attributed distaste towards immigrationas the primary driver behind native flight. The elderly group represents an interesting economic case forexamining the linkages between housing, mobility, and immigration because this subpopulation tends tobe asset-rich, and the older household’s labour supply and housing decisions may be more responsive tooverall economic conditions. To examine this mechanism, I unbundle the analysis by dwelling tenure typesand use an instrumental variable strategy that is based on historical ethnic distributions. A combinationof results points to the possibility that in addition to a taste channel, housing affordability and householdfinance could influence mobility decisions. In addition, synthetic cohort analysis shows that in responseto an immigration shock, the elderly homeowners who stay in the same neighbourhood do not exhibit anyform of housing asset downsizing (i.e. sell high and then buy low). Therefore, there is insufficient evidenceto conclude that the near-retirees extract housing equity by relocating. This study makes several importantcontributions. To the best of my knowledge, this is the first paper to separate the analysis by dwelling1tenure types and by age groups when investigating the impact immigration exerts on older native mobilitydecisions. The research findings in this study push this area of literature forward by suggesting an alternativeperspective for explaining native out-migration. The heterogeneity in mobility preferences across dwellingtenure groups is an important result because it may explain why Card (2001) fails to find any significanteffect from immigration on aggregate native relocation decisions.Finally, in Chapter 4, this study evaluates how the recent house price shock affected the labour supplydecisions of U.S. older households. The common impression is that housing market fluctuations and house-hold wealth are closely linked, and any changes to household wealth should alter labour supply behaviour.I first utilize a quasi-experimental approach by comparing the labour supply responses of near-retirees whoare more and less exposed to housing market variations. I then use the instrumental variable approach byapplying the interaction of the housing supply elasticity and the national lending conditions as an instru-ment for housing value growth to examine this linkage. Overall, this study cannot find evidence of a stronglinkage between housing and labour markets due to large standard errors. The results from Chapters 3 and4 are aligned with Skinner’s (1996), which suggests that only a small fraction of elderly households wouldtap into their housing equity.2Chapter 2Public pensions and elderly immigrants’labour supply decisions2.1 IntroductionTo date, most of the literature that examines the relationship between social assistance and labour supplybehaviour of immigrants has focused on the working-age population. Relatively few papers have exploredthe same context for the elderly immigrants. The elderly immigrant is an important research focus. First,Baker et al. (2009) mention that the income-based poverty rate tends to be driven by senior immigrants, sincenewly-arrived migrants tend to show low employment rate and are more likely to rely on social assistanceprograms. However, 77% of the Canadian elderly immigrants are homeowners, whereas only 67% of thenon-immigrants own a house1. This implies that many older migrants are asset-rich and income-poor. Veall(2013, 2014) suggest that low-income taxfilers could avoid GIS clawbacks by skillful tax planning. There-fore, their labour supply decisions could either be more vulnerable to economic conditions or be strategic inorder to qualify for public pension benefits.One of the key policy concerns is to find ways to encourage efficient work decisions by seniors beyondretirement age in order to reduce the burden on public finances. High clawback rates that accompany di-rect transfers discourage any labour market participation among recipients (see for example, Saez (2002)).However, not all immigrants receive full public pension benefits, since these transfers are based on the yearsof residence. This implies that the labour supply decisions of early-arrived immigrants who qualify for thepublic pension benefits could differ from those of newly-arrived immigrants who do not meet the eligibilityrequirements. Therefore, this paper exploits the discontinuity from the Canadian Old Age Security (OAS)and the Guaranteed Income Supplement (GIS) residency requirement as a quasi-experiment to examine themagnitude of the extensive and the intensive responses of labour supply to public pension policies for el-derly immigrant households. Specifically, I investigate whether labour market responses for the recipient,the spouse, and the working-age children exhibit any heterogeneities across different family compositiontypes.In contrast to previous studies, this paper uses a novel approach to estimate a causal effect of publicpensions on family labour supply decisions of elderly immigrants. I also evaluate the extent to which theseindividuals may exhibit anticipatory behaviour prior to receipt of public pension entitlements. Existingacademic literature suggests that anticipation effect may exist in response to public pension benefits, at1These numbers are based on the Statistics Canada’s Survey of Labour and Income Dynamics data for years 2002 onwards. Notethat the dwelling-related variables contain coding errors for years 1999 to 2001.3least for the Canadian context. For example, Veall (2013, 2014) mention that the dominant strategy fornear-retirees is to withdraw RRSPs immediately before reaching the GIS-eligible age and such withdrawalscan precede the actual GIS reduction by as much as 18 months. These two papers also show using theLongitudinal Administrative Database that significant RRSP withdrawals happen during ages 60 to 64. ForCanada, the spouse can qualify for Spousal Allowance if he/she is between ages 60 to 64 and Baker (2002)illustrates that individuals in eligible couples respond to the Spousal Allowance through a reduction in labourforce participation rate. Furthermore, Finnie et al. (2016) note that changes in marital status could influencethe take-up rate of GIS benefits. Given these findings, estimating for anticipation effects is necessary in thispaper in order to better understand whether elderly individuals make efficient labour supply choices as theybecome eligible for OAS/GIS benefits. This also provides an indirect answer as to why the income-basedpoverty rates are high for this sub-population group.My paper is more closely related to Danzer (2013) and to Baker (2002), which also exploit pensioneligibility rules to examine the impact of public pensions on labour supply decisions. My study differsfrom theirs in several important ways. First, I examine labour supply effects of elderly immigrant families,where relatively few papers have investigated this context. This group tends to show low incomes andhigh homeownership rates. I build off Danzer (2013) and Baker’s (2002) work by evaluating the extentto which these families may exhibit anticipatory behaviour in response to public pension entitlements. InDanzer’s (2013) work, he uses an unanticipated and exogenous increase of the legal minimum pension inUkraine as a quasi-experiment to investigate the pure income effect on elderly labour supply. In this case,his estimation does not allow respondents to adjust their labour supply behaviour in anticipation for anyincreases in old-age pension benefits. I relax this restriction by applying the Canadian pension system intothis study. Furthermore, I compare the labour supply behaviour of low-income households with high housingasset holdings and those with low housing asset holdings to further address the question of anticipatorylabour supply behaviour. This is the first study to examine this issue.This study improves the estimation in Borjas (2011). My paper uses a regression discontinuity design todirectly compare the labour supply behaviour of newly-arrived immigrants and of early-arrived counterparts.This cannot be computed in Borjas (2011). Baker et al. (2009) and Borjas (2011) note that this type ofestimation has not been done in existing literature due to small sample size and data limitations. Borjas(2011) mentions that the 1980 and 1990 U.S. Census data only provide five-year intervals for the yearof immigration. Therefore, Borjas (2011) cannot precisely compare the work and retirement decisions ofimmigrants who just qualify for public pension entitlements versus those who don’t. My paper can carryout this estimation using the 2006 Census dataset with sufficient size and accuracy because the income datain the 2006 Census are linked to tax information. Moreover, the Census data also provide the actual yearsof arrival to Canada with detailed place of birth information.Unlike previous literature, I examine the labour market attachment of different family members, includ-ing the working-age children’s in response to the parent’s receipt of OAS/GIS benefits. Blundell et al. (2016)suggest that family labour supply plays a particularly important role in response to a permanent wage shock.This study assesses whether these findings hold under the elderly immigrant family context. Moreover, theregression discontinuity design in this paper provides a more transparent indication of the changes in the4elderly immigrants’ labour supply decisions when they transition from having no or limited amounts of re-tirement benefits to having a positive, large amounts of public pension benefits (mainly from GIS). Unlikethe U.S., Canadian immigrants do not need to work at least 40 quarters in order to qualify for the Old AgeSecurity. Therefore, the data sample is less likely to be confounded by recollection bias on employmentperiods.My results indicate that elderly immigrants aged 65 and over respond to the Canadian Old Age Securitypension benefits with a decrease in labour force participation rates. The effect of OAS/GIS benefits isheterogeneous across family types. The household maintainer and the working-age children’s labour marketattachment depends on whether Spousal Allowance is available in the family. The OAS benefits do not seemto exert any major influence to work intensity for all family members.A combination of datasets suggests that elderly immigrants exhibit anticipation effect towards OAS/GISbenefits. Relative to the working-age population, time use data indicates that seniors who remain in thelabour force first reduce working hours by replacing market production with home production. Both theregression discontinuity design and the difference-in-difference frameworks do not find any evidence ofstrong linkages between public pension eligibility and work intensity once I include the households withlow labour effort into the estimations. Referring back to the static labour supply framework, the minisculechange in work intensity for these families suggests that the original utility curve would have already beenlocated near the kink for not working. This implies that working hours would have already been reducedsome time prior to the OAS/GIS eligibility date.2.2 The economics of immigrants’ labour supply decisions and socialassistanceRecent literature has suggested that various forms of social assistance programs (including public pensionbenefits) influence labour supply decisions. These findings are not country-specific. For example, Börsch-Supan (2000) shows using German data and the option value analysis that an actuarial fair retirement systemwill lower retirement before age 60 by more than a third. For the U.S., both Liebman et al. (2009) and Vere(2011) find that Social Security benefits do influence labour supply responses. By exploiting the 1977amendments to the Social Security Act, where seniors born after January 1, 1917 are subject to lowerSocial Security benefits, Vere’s (2011) results suggest that reductions in benefits induce recipients to worklonger hours during retirement even in their 70s and 80s. The effect from the reduction in benefits isstronger for singles, spouses of beneficiaries, and the less-educated individuals. Liebman et al. (2009) usethe discontinuities in five provisions of the Social Security benefit rules and find that individuals are morelikely to retire when the effective marginal Social Security tax is high. Furthermore, Borjas (2003) finds thatimmigrants tend to increase labour supply in order to increase the probability of being covered by employer-sponsored health insurance. For Canada, Baker (2002) finds that the introduction of the Spousal Allowance,which is intended to be for wives of younger age, is associated with a six to seven percentage point relativedecrease in labour force participation among males in eligible couples. Finally, for developing and emergingmarkets like Ukraine, Danzer (2013) demonstrates that labour supply reductions are disproportionally large5for the less-educated, because the opportunity costs of foregone earnings from immediate retirement is muchsmaller for the least-educated.Previous work has also illustrated that immigrants’ working hours are lower than natives’. Immigrants’labour supply decisions could be distorted by credit constraints and by human capital accumulations (Blau etal., 2003; Baker and Benjamin, 1997; Worswick, 1999). Similarly, the impact of social assistance programson working-age natives and on immigrants is also different (Borjas and Hilton, 1996). Yet, the evidenceon whether immigrants rely more on social assistance programs is mixed. Generally, existing literatureshows that less-skilled immigrants are heavy users of social assistance and welfare programs. For example,Borjas (1999) and Bratsberg et al. (2014) find that less-skilled immigrants cluster in high-benefit locationsand immigrants from low-income source countries tend to show rising participation in disability insuranceprograms. Similarly, Dustmann et al. (2010) show that A8 immigrants in U.K. (immigrants from the Centraland Eastern European countries that joined the European Union in 2004) are less likely to use welfareservices and receive state benefits because these in-migrants tend to be younger and better educated andalso tend to have fewer children than natives. Using a more general model specification, Crossley et al.(2001) also do not find immigrants of any cohorts at any number of years-since-migration to rely more onsocial assistance programs relative to Canadian-born men. However, Swedish data show that immigrantstend to stay on welfare and are less likely to remain employed than natives (Hansen and Lofstrom, 2003,2009). Baker et al. (2009) also mention that recent elderly immigrants who do not qualify for public pensionbenefits in Canada show low employment rate and rely on social assistance programs.To date, relatively few papers have specifically explored the impact of retirement benefits on laboursupply decisions of elderly immigrants. Two recent papers have examined this context. First, Borjas (2011)uses the U.S. Census data from the 1960-2000 to compare how the eligibility requirements for Social Se-curity benefits affect the immigrant and native men’s decisions to exit the labour force as they approachretirement. Immigrants are required to work at least 10 years before they can qualify for Social Securitybenefits, which motivate the immigrants who arrived in their 50s and 60s to work longer relative to thenatives. Consistent with this conjecture, Borjas’s (2011) results do show that the probability of employmentfor older immigrant men drops by seven to eleven percentage points once they qualify for Social Securitybenefits. Second, Kaushal (2010) applies the change in the 1996 Personal Responsibility and Work Oppor-tunity Reconciliation Act (PRWORA), which banned the receipt of Supplemental Security Income (SSI)for the majority of elderly immigrants, to examine the effect on immigrants’ employment, retirement, andfamily incomes. This policy change is associated with an increase in employment and in delayed retirementfor elderly male immigrants, but not for female immigrants.Yet, little is known about the effect of public pensions on the joint family labour supply decisions ofelderly immigrants and whether these individuals alter their behaviour a few years prior to the receipt ofthese benefits. This study adds to this area of literature by examining the causal impact of public pensionson the labour supply decisions of the household maintainer, the spouse, and the working-age children.62.3 Application of the Canadian Old Age Security benefits to the staticlabour supply modelThis section first presents a description of the Canadian Old Age Security (OAS) program. I then apply theCanadian OAS program to the static labour supply model to predict the elderly immigrants’ labour marketattachment from this public pension entitlement.2.3.1 The Canadian Old Age Security programThe Canadian Old Age Security (OAS) program is the Government of Canada’s largest pension program,which is a monthly benefit paid to seniors who have either a Canadian citizenship or permanent residencestatus. Unlike the U.S., Canadian applicants do not need to work at least 40 quarters in order to qualify forthis public pension entitlement. For those who live in Canada, they are eligible for the OAS pension if theyare at least 65 years of age and have resided in Canada for at least ten years after age 18. Conversely, forseniors who live outside of Canada, they must present proof that they previously lived in Canada for at least20 years after age 18 in order to qualify for this public pension entitlement. If neither of the above scenariosapply, the applicant could still meet the residence requirement under the terms of an international socialsecurity agreement. A social security agreement (SSA) allows the periods of contributions and/or residencein Canada and in an agreement country to be added together to meet the residence requirements for the OASbenefits. In other words, an individual can still apply for OAS if they live in Canada for less than 10 yearsas long as they have previously worked in and/or lived in a country that has an SSA with Canada. Startingin April 2013, the government implemented a process to automatically enroll a subset of seniors who areeligible to receive the Old Age Security pension. For those without automatic enrolment, they will need toapply in writing to receive OAS benefits.The residency period in Canada determines the amount of Old Age Security pension that the elderlyperson receives. Generally, the full OAS pension amount is payable if the individual has resided in Canadafor at least 40 years since the age of 18.2 The person can also qualify for the OAS benefit if he/she wasborn on or before July 1, 1952 and: (1) on July 1, 1977, he/she resided in Canada; or (2) on July 1, 1977,he/she did not reside in Canada but after turning 18, he/she resided in Canada for a period of time prior toJuly 1, 1977; or (3) on July 1, 1977, he/she possessed a valid Canadian immigration visa. In the two lattercases, the respondent must have resided in Canada continuously for the ten years immediately before theapproval of the OAS pension. Otherwise, if the person does not satisfy this ten-year period restriction, theapplicant could have made up for partial absences by having periods of prior residence in Canada that wereequal to at least three times the period of absence during the ten-year period. For example, three years ofresidence in Canada between ages 19 and 22 could offset the one year absence between ages 63 and 64.Alternatively, the applicant must reside in Canada for at least one year immediately before the approval ofthe OAS pension.On the other hand, the OAS pension amount is prorated by the number of years of residence if the2For the October to December 2015 period, the maximum OAS pension amount was $569.95 (source:http://www.esdc.gc.ca/en/cpp/oas/payments.page)7respondent has a minimum of ten years of residence, but has lived for less than forty years in Canada. Forexample, the senior receives 10/40th of the full OAS pension if he/she has lived in Canada for ten years afterage 18, and is entitled to the same pension amount for life (prior to any inflation adjustments). The OASbenefits are inflation-adjusted on a quarterly-basis.Within the Old Age Security program, there are three other types of benefits: (1) Guaranteed IncomeSupplement (GIS); (2) Spousal Allowance (Allowance); and (3) Allowance for the Survivor. The GIS is anincome-tested program that is based on previous year’s family income excluding OAS and GIS benefits.3The family annual income must be lower than the maximum annual threshold. The senior can only applyfor this income-tested benefit if the individual is eligible for the OAS. The GIS benefits are clawed back,where the benefits are reduced by $1 for every $2 of income for single-person families; and by $1 for every$4 for married couples. The GIS clawback only applies to combined yearly income of couples, and does notinclude income of working-age children. Starting on July 1, 2008, the GIS earnings exemption was raisedfrom $500 to $3,500. This means that for a single pensioner earning $3,500 or more, the clawback startsfrom $3,501 to the maximum annual threshold amount. Unlike the OAS, the GIS benefit amounts are notprorated by the years of residence. Therefore, immigrants who have landed for 10-19 years or those who sat-isfy the residence requirement through social security agreements can benefit from the full GIS amount eventhough the OAS benefit is a partial amount. Between years 2001 and 2006, the legislation allowed spon-sored immigrants from SSA countries to be eligible for income-tested benefits when they became citizens.However, the legislation was amended in 2007, where seniors from social security agreement countries whobecome Canadian citizens while under sponsorship cannot receive GIS until they have resided in Canada forten years.The Spousal Allowance is a benefit available to the spouses or common-law partners of GuaranteedIncome Supplement recipients.4 The spouse or the common-law partner must be between ages 60 and 64,be a Canadian citizen or a legal resident, have resided in Canada for at least ten years since age 18, and thefamily income must be less than the maximum allowable annual threshold. The spouse could still qualify forpartial allowance benefit if he/she has previously worked in a country that has a social security agreementwith Canada. Conversely, the Allowance for the Survivor is a benefit available to Canadian citizens orlegal residents of ages 60-64 whose spouse or common-law partner is deceased, who have met the 10-yearresidency requirement, and whose annual income is less than the maximum annual threshold.5 Clawbackprovisions apply to both the Spousal Allowance and the Allowance for the Survivor, and the clawback ratesvary across different income ranges.Effective in July 2013, the applicant could defer the receipt of Old Age Security benefits by up to fiveyears in exchange for a higher monthly amount. However, the increase only applies to the OAS pensions, andnot to the Guaranteed Income Supplement, the Spousal Allowance, or the Allowance for the Survivor. If the3For the October to December 2015 period, the maximum monthly GIS amount was $772.83 for single-person household; and $512.44 if the spouse/common-law partner receives the full OAS pension (source:http://www.esdc.gc.ca/en/cpp/oas/payments.page).4For the October to December 2015 period, the maximum Allowance amount was $1,082.39 (source:http://www.esdc.gc.ca/en/cpp/oas/payments.page)5For the October to December 2015 period, the maximum amount for the Allowance for the Survivor benefit was $1,211.79(source: http://www.esdc.gc.ca/en/cpp/oas/payments.page)8senior chooses to defer receipt of the OAS pension, he/she cannot receive the GIS and his/her spouse cannotreceive the Spousal Allowance. Furthermore, although the Old Age Security pension is not an income-testedbenefit, the applicant must pay a recovery tax on OAS benefits if the individual’s net annual world incomeis more than $71,592 in 2014 and/or if he/she lives in a country where the non-resident tax on Canadianpensions is 25 percent or more.Figure 2.1 shows the OAS take-up rate and the combined OAS and GIS mean amounts for years 2001,2006, and 2011, using data retrieved from the Canadian Censuses and the 2011 National Household Survey(NHS).6 The top panel shows the raw data for the two series, which shows a discrete jump in the OAS/GIStake-up rate and the public pension benefit amounts at around year 10-11 of immigration.7 I exploit thisvariation by comparing the labour supply decisions of the elderly who landed for just more than 10 years inCanada versus those with less than 10 years since arrival. The raw data series in Figure 2.1 show an upwardspike in the OAS take-up rate at the fourth year of arrival for years 2001 and 2006 data, which coincides withthe 2001 legislation that allowed sponsored immigrants from SSA countries to be eligible for GIS when theybecame citizens. As noted in the Data Description section below, Statistics Canada excluded the respondentswho reported a positive Old Age Security amount and fewer than eleven years of arrival to Canada for the2011 National Household Survey dataset. To be consistent with Statistics Canada’s imputation proceduresfor the 2011 NHS dataset, the bottom panel also shows the case where I apply the same process to the 2001and 2006 data.Generally, the raw data shows that the residence requirement for OAS/GIS pension benefits can stronglypredict public pension entitlements using years 2006 and 2011 data. However, the validity of the 2001 datacould be confounded by the start of the above-noted legislation regarding sponsored immigration, as wellas by potential misakes from self-reported income. The Data Description section highlights that most of therespondents’ income data are linked to tax information starting from year 2006 Census. This implies thatdata accuracy for year 2006 Census is higher than year 2001’s. In light of these potential issues with usingthe 2001 data, the rest of the analysis is based primarily on year 2006 data. I use 2011 NHS data as part ofthe robustness checks.2.3.2 Theoretical frameworkFollowing Baker (2002), I apply a simple neoclassical model to predict the effects of Old Age Security(OAS) and Guaranteed Income Supplement (GIS) benefits on family labour supply decisions. Figures 2.2and 2.3 present a static labour supply model incorporating the features of the OAS and GIS benefits forhouseholds without working-age children and for those with working-age children, respectively. The verticalaxis gives the total annual after-tax family income and the horizontal axis illustrates the total family non-6A few potential reasons could explain why the OAS take-up rate does not reach 100% in Figure 2.1. First, respondents couldperceive the OAS/GIS application to be a costly process and may refuse to apply for this public pension benefits (see for exampleVeall (2008)). Second, OAS/GIS is not paid to individuals who have left Canada for more than six months. For the 2006 Census,residences for senior citizens were self-enumerated using Forms 2A and 2B, and they are not considered as institutionalized.Therefore, the 10% gap may also account for the non-responses from seniors who had difficulty filling out the Census forms.7Note that the Census and the 2011 National Household Survey are conducted in mid-year, and neither datasets provides theactual date of immigration. Therefore, just taking the year of the dataset and subtracting by the year of immigration may cause atleast a half-year lag. In this case, the figures show the discontinuity point to be around year 11 instead of year 10.9market production time spent per day. The daily maximum allocation to non-market production is 24 hours,which is denoted by “T=24”. Non-market production time includes leisure and home production activities.For both cases, the thin grey line shows the after-tax income for immigrant families with less than ten yearsof residence in Canada, which means that they do not qualify for the OAS/GIS benefits. The thick greyand black lines reflect the after-tax income for households without working-age children and those withworking-age children who qualify for partial amounts of the public pension benefits in Canada, respectively.With no work, new immigrants (i.e. those who lived in Canada for less than 10 years) receive the amountAA’. Without OAS/GIS benefits, newly-arrived in-migrants work at the wage rate denoted by the segmentA’G.8 On the other hand, earlier immigrants (i.e. those who lived in Canada for more than 10 years) receivethe amount AC or AC’, which includes OAS and GIS benefits and other sources of non-employment income.For simplicity, I exclude the GIS exemption from the static labour supply model.9For families without working-age children, which is illustrated by the thick grey line of Figure 2.2, GISbenefits are clawed back between points C and D. Benefits are reduced by $1 for every $2 of income forperson living alone; and by $1 for every $4 for married couples. At point D, the GIS benefit is exhaustedand the household increases consumption by working more. Segment EF shows that the OAS recipient willneed to repay part of or the entire OAS pension if he/she has a net world income of approximately $71,000or more. On the other hand, for families with working-age children, the clawback rate for GIS benefitswould implicitly be smaller. This is denoted by the black line in Figure 2.3. As noted, the GIS clawbackis applicable to combined yearly income of couples. Therefore, any source of earnings from working-agechildren would not be subject to clawback in the GIS computation. This implies that the slope of segmentC’D’ (black line) must be larger than that of segment CD (thick grey line). Therefore, point D’ is locatedto the right of point D to illustrate the possibility that family members work fewer hours with the presenceof working-age children. Families with working-age children also face the OAS clawback at segment EF inFigure 2.3.The impact of OAS/GIS benefits on labour supply decisions varies across income groups. I includetwo different preference sets into one static labour supply diagram to illustrate this heterogeneity. The firstpreference set is for low-income households, which is denoted by the utility curves U0 and U1. The optimalallocations for these two utility curves are X0 and X1, respectively. In the absence of OAS/GIS benefits,a low-income household with the utility preference of U0 will work and will prefer the bundle X0. Withthe OAS/GIS benefits, the same type of household will choose not to work because the utility curve U1is now higher than U0. For families with and without working-age children, OAS/GIS benefits encouragelow-income recipients to exit the labour force.The second preference set is for middle-income recipients. I use utility curves U2, U3, and U4 toillustrate the point that the effect of OAS/GIS benefits on middle income households is ambiguous. X2,X3, and X4 reflect the optimal allocations for utility curves U2, U3, and U4, respectively. Based on thestatic labour supply model, the decision to withdraw from the labour force depends on the curvature of theutility function. Regardless of family composition types, by income effect, the middle income household8 I follow a similar assumption as Baker’s (2002), where the relative wage rate reflects the main income earner’s comparativeadvantage in market production.9The overall result does not change with and without the GIS exemption in the model.10will reduce working hours by moving from bundle X2 (without social assistance) to bundle X3 (with socialassistance). The same type of household will also choose to reduce working hours at segment DE (or D’E)due to both income and substitution effects, if the utility for bundle X3 exceeds that for bundle X4. However,if the utility preference for X3 is lower than the utility at point C (the point for exiting the labour force), thenthe middle-income group will retire. In order for the major income earner(s) in the household to exit thelabour market, the model predicts that family members reduce working hours sequentially. In other words,the total family market production time must fall in response to OAS/GIS benefits, but the rate at whichmarket production time declines should be faster for families with working-age children.Finally, this type of social assistance has no impact on the high-income group, (denoted by segmentFG) because the public pension benefit for this type of household is zero. The subsequent sections willuse econometric techniques to explore the exact magnitude of the impact from OAS/GIS onto the elderlyimmigrants’ and their family members’ labour supply decisions.2.4 Data descriptionThis paper uses a combination of datasets produced by Statistics Canada to investigate a causal effect ofpublic pension benefits on family labour supply decisions of elderly immigrants.10 First, I apply the 2006Census data as part of the main estimation. The voluntary nature of the 2011 National Household Survey(NHS) dataset raises concerns over the validity of the immigration and place of birth numbers.11 Therefore,I estimate the regression model using 2011 NHS data as part of robustness checks. Starting from the 2006Census, most of the respondents’ income data are linked to tax information. This change improves theincome data’s accuracy. Both datasets contain variables on family composition, income sources, actualyears of immigration, country of residence from five years ago, detailed place of birth information, propertyvalue, and labour force status.However, the Census does not identify the main income earner in the family. Therefore, I assume thereference person in the Census corresponds to the household maintainer. As part of the supplementaryanalysis, I use the longitudinal version of the Survey of Labour and Income Dynamics dataset to furtheridentify how the OAS/GIS benefits affect the main income earner’s and his/her spouse’s labour supplydecisions. The SLID data only provides labour force status information for employed individuals of ages16-69. Therefore, for this part of the supplementary analysis, I use wage data, which includes the wholeuniverse of respondents, to create a binary variable for the labour force participation decision.12To gather more evidence on the effect of OAS/GIS benefits on the intensive margins of labour supply,this study also uses the 1998, 2005, and 2010 General Social Survey (GSS) datasets to compare time use10The research and analysis in this chapter are based on data from Statistics Canada and the opinions expressed do not representthe views of Statistics Canada.11See for example, Grant (2015).12I also use the variable “person’s major activity at end of reference year” to construct a binary variable for labour force partic-ipation when using the SLID data. The results are similar to those where I apply wages to construct the same measure. However,the major activity variable faces problems in the collection of data for year 1999 SLID. Respondents aged 70 and older were notasked about their major activity that year. To compensate for this problem, an imputation program was developed, based on datafor years 1998 and 2000. The number of “retired” responses is high in 1999 as compared to other years. In light of this problem,this study uses wage data to construct the 0/1 labour force participation variable.11of elderly immigrants and of working-age immigrants.13 The GSS dataset contains information on the yearof arrival to Canada in 5 or 10-year groups, as well as detailed time use and demographic information.However, small sample size is one of the major drawbacks to using the GSS data, especially for immigrantfamilies. To address this problem, I include all respondents into the empirical strategy in order to preserveas much information as possible. I also compare the estimates generated by the GSS against those producedby the Census to test robustness of results.One of the shortcomings with the Census and the NHS data is that both datasets do not report the actualyears of residence in Canada. As noted above, the eligibility requirement for OAS depends on the years ofresidence. Yet, the years of residence could be different from the number of years since immigration. Forexample, years of residence would be a non-zero number for a person who lives in Canada under studentvisa; but the number of years since immigration would be zero in this case because the individual is a non-permanent resident. Given data limitations, I use the number of years since immigration to proxy for theyears of residence in Canada. The estimation with the number of years since immigration is expected toprovide a lower bound answer since this measure carries a lag.Although the percentage of immigrants with less than ten years of landing who reported a non-zeroOAS/GIS amount is roughly a quarter for year 2006 data, a few stylized facts suggest that the estimation withthe years of residence is not expected to be substantially different from that with the number of years sinceimmigration.14 In Section 2.6, I show that the labour supply decisions are very similar whether I include orexclude the individuals who reported receiving OAS/GIS benefits for less than 10 years of landing.15 Basedon the OAS numbers from the Employment and Social Development Canada (ESDC), roughly 6.3% of theseniors of ages 65 and above with less than ten years of immigration would qualify for the OAS throughsocial security agreements.16 Furthermore, the longitudinal version of the Survey of Labour and IncomeDynamics dataset also shows that a very small proportion of respondents with less than ten years of landingwould have received the OAS benefits.Yet, there are three potential factors that could explain the discrepancy between the overall OAS take-uprate and the OAS take-up rate through social security agreements (6.3%). First, imputation procedures couldbe one possibility. Statistics Canada confirmed that only a small proportion of respondents with less thanten years of landing would have reported an OAS amount in the raw data (prior to any imputations), andthe difference in results between using their imputed data and the raw data is roughly 10%. Second, for the2006 Census, senior citizens in residences were self-enumerated using Forms 2A and 2B, and they are notconsidered as institutionalized. The gap could account for some of these seniors who made a mistake in13This paper uses the aggregate time use categories compiled by Statistics Canada for years 1998, 2005, and 2010. I do notinclude data from the 1986 and 1992 GSS because the categories for these two years are different.14According to the 2006 Census PUMF file, roughly 23% of the respondents with less than ten years since immigration receiveOAS/GIS benefits (which would also include allowances).15See also Figures A.1 and A.2 in the Supplementary Appendix.16This value gives an upper bound answer and is computed as follows using the data provided by the Employment and SocialDevelopment Canada: I take the number of OAS recipients applied through social security agreements (SSA) and divided by thetotal number of OAS immigrant recipients of ages 65 and above for year 2006. I assume that the number of OAS applied throughSSA, which is 80,770 in year 2006, applies to all immigrants who landed for less than ten years. For individuals of ages 65 andover, the share of immigrants over total population is 29.9%. Therefore, the number of OAS immigrant recipients is approximatelyequal to 30% of the total number of OAS recipients in Canada (4,261,262), which is 1,278,379. Therefore, the percentage of seniorsof ages 65 who qualify through SSA is roughly 80,770 / 1,278,379 or 6.3%.12recalling the year of immigration information. Finally, the years that an immigrant have spent as a studentor as as temporary foreign worker would also count towards the 10-year benefit eligibility rule. Using the2006 Census, I find that roughly one-third of the elderly immigrants who landed for less than five years inCanada have previously lived in Canada five years ago.Another shortcoming with using the Census and National Household Survey data is that StatisticsCanada did not impose consistent imputation procedures to both datasets for determining the respondentswith a non-missing OAS/GIS amount in the field. For the NHS data, Statistics Canada moved the OASamounts to the “other government transfer” category for any observations with less than eleven years oflanding to Canada. However, this imputation procedure was not done to the naturalized Canadian citizensfor the 2001 and 2006 Censuses. Given this shortcoming, as part of the robustness checks, I first exclude therespondents with a non-missing OAS amount and with less than eleven years of immigration from the 2006Census for the regression discontinuity analyses. I also conduct a similar set of regression analyses usingthe tax-linked data from the longitudinal version of the Survey of Labour and Income Dynamics (SLID) asa second robustness check. Statistics Canada confirmed that they did not impose any imputation proceduresto tax-linked respondents in the SLID data. However, small sample is a major concern with the SLID; andthus, I estimate the full sample using this dataset for most of the analyses. Despite these data limitations, todate, these datasets are the best sources available for studying the linkages between public pension benefits,labour supply responses, and living arrangements.2.5 Empirical strategyThe main empirical approach exploits the discontinuity in the Canadian Old Age Security (OAS) benefits atthe tenth year of residence, by using a regression discontinuity (RD) design. One of the advantages of usingthe RD design over the difference-in-difference (DD) method is that the labour supply comparisons can beplaced in more closely similar markets. In this study, I can directly compare the labour supply decisions ofthe elderly population (ages 65 and over) who landed for less than 10 years versus those who arrived formore than 10 years as of year 2006. The RD strategy is more transparent and the results are also less likelyto be confounded by the selection of discontinuity points. Conversely, the DD estimates are sensitive to thechoice of the control groups (see for example Appendix Table A.1).17 Finding an appropriate control groupthat has the same pre-existing trends under the difference-in-difference framework can be challenging inthis paper’s context. As discussed below, for certain labour supply variables, the trends for the working-agepopulation (ages 25-54) are different from those for the elderly population (ages 65 and over). This impliesthat the common trends assumption could be violated under the DD method. In addition, the comparisonacross years can be tricky since Statistics Canada imposed inconsistent imputation procedures for differentyears of the Census. As such, the estimates from the difference-in-difference technique could be confoundedby other external factors that cannot be easily eliminated from the regressions.Furthermore, one of the questions of interest in this paper is whether individuals show any anticipationeffect in response to public pension entitlements. Technically, for a valid RD design, there should be no17See also Lemieux and Milligan (2008).13jumps in the dependent variable at non-discontinuity points (see for example, Imbens and Lemieux (2007)).However, estimating for anticipation effects is necessary in this paper from a policy standpoint, in order tobetter understand whether elderly immigrants are making efficient labour supply choices. This also providesan indirect answer as to why the income-based poverty rates are high for this sub-population group. The RDsetup is more suitable in this context to quantify the anticipation effect resulting from the OAS/GIS benefitsbecause the treatment variable in this framework gives a discrete answer. Conversely, the difference-in-difference estimation is less direct, as the treatment variable only reflects the impact coming from the changein eligibility for a range of points on years of residence.I start the econometric analysis by examining the impact of OAS/GIS on the extensive margins of laboursupply. As noted in the Data Description section, I set the number of years since immigration to be thesame as the year of residence in this paper. Furthermore, this study assumes that the reference person in thecensus family taken from the 2006 Census corresponds to the household maintainer. Therefore, this paperuses the terms “reference person”, “household maintainer”, and “main respondent” interchangeably. I firstextract the raw data series using the 2006 Census dataset, which also include the applicants with less than tenyears of residence in Canada and who are able to meet the residence requirement through a social securityagreement (SSA). I then run various robustness checks to address the potential concern that immigrants whoare from a country with an SSA with Canada could be self-selecting themselves into Canada to benefit fromthe OAS/GIS benefits. These robustness checks include incorporating an SSA dummy to account for thisvariation as well as eliminating the respondents with less than ten years of arrival to Canada and reported apositive OAS benefit amount. To estimate the intensity of anticipation effect, I apply two different strategies.First, I use a donut hole by excluding the respondents with 9 to 12 years of immigration from the analysis.In another estimation, I move the threshold point to earlier dates to explicitly test for anticipation effects.In addition, I unbundle the estimation by different family composition types in order to estimate theeffect from public pension entitlements on the main respondent, on the spouse, and on the working-age chil-dren. The 2006 Census does not provide information on the primary and secondary income earners withina family. Therefore, I use the longitudinal version of the Survey of Labour and Income Dynamics (SLID)dataset to find a causal relationship between OAS/GIS and labour force exit rate for the main income earnerand for his/her spouse for years 1993-2010. The voluntary nature of the 2011 National Household Survey(NHS) raises concerns over the validity of the immigration and place of origin numbers.18 Therefore, I com-pare the results obtained using 2006 Census against those from the 2011 NHS for broad family compositiontypes as part of robustness checks.This paper then investigates the effect of public pension entitlements on the intensive margins of laboursupply by applying the 2006 Census data and the time use information obtained from the General SocialSurvey (GSS). For the 2006 Census data, I regress the hours worked per week and the weeks worked peryear variables, as well as the binary variable for zero hours spent on unpaid housework onto the point ofdiscontinuity for various types of family structures. Given small sample size problems, for the GSS data, Iutilize a difference-in-difference approach to compare the number of minutes spent on market production,home production, and leisure-related activities for immigrants of ages 65 and over and the number of minutes18See for example, Grant (2015).14spent on the same set of activities for immigrants of ages 25-54.2.5.1 Extensive margins of labour supplyThis paper begins the econometric analysis with a regression discontinuity (RD) design, where I exploit thediscontinuity in the OAS/GIS benefits at the tenth year of residence to explore the effect of public pensionentitlements on the extensive margins of labour supply. I estimate three different sets of regression modelsusing the 2006 Census data as part of my main analysis. The first set examines the impact of the OAS/GISbenefits on the main respondent’s labour supply outcomes; the second set explores the effect from this policyon the spouse’s; and finally, the last set focuses on the working-age children’s work decisions.19 I defineworking-age children as those of ages 25 and above, who are no longer considered as dependents while livingtogether with their parents. This paper examines the labour market attachments for the following familycompositions: single-person families; married and common law couples with working age children; andmarried and common law couples without children. The goal of the latter two estimations is to investigatewhether family members would exhibit weak labour market attachment with the presence of working-agechildren. This links back to the static labour supply model, which shows that the presence of working-agechildren could help accelerate the family labour force exit rate. Any sources of earnings from the child arenot subject to clawback in the GIS computation, and this provides a bigger cushion for senior couples towork less. I also compute the case where the spouse’s age is less than 60 and greater than 60 in order todetermine whether Spousal Allowance may be driving labour supply behaviour. The main purpose of havingthese estimations is to determine whether family members exhibit any form of joint labour supply decisions.This study estimates the following main regression model for respondent i using Ordinary Least Squares(OLS):Yirt = β0+β1Xirt +β2 provr +δ1{t ≥ 11}︸ ︷︷ ︸=T REATirt+γ11{t ≥ 11} · (t−11)+ γ21{t < 11} · (t−11)+ εirt (2.1)where r indexes the province of residence and t indexes the number of years since immigrated to Canada.Respondent i refers to the main respondent, the main respondent’s spouse, or the respondent’s working-agechildren.20 I include province fixed effects provr in order to account for other provincial-based supplementalprograms that may drive labour supply behaviours as the household maintainer turns age 65. εirt could becorrelated across the number of years since immigration. Therefore, I address this problem by clusteringthe standard error with number of years since immigrated to Canada. Yirt is a measure of labour supplybehaviour for respondent i in province r with t years of immigration to Canada. I use four measures tocapture labour market attachment (Yirt) at the extensive margins of labour supply: (1) employment rate; (2)labour force participation rate; (3) an indicator for having an annual wage of less than $500; and (4) an19The estimations in this paper only include spouses and working-age children who are also immigrants. The results are similarwhen I include non-immigrant spouses and children into the computations. These results are available upon request.20I use OLS for these estimations in order to be consistent across all of the strategies in this chapter. The estimates using OLSare similar to those produced by the probit model. See Appendix Table A.2 for the probit results.15indicator for having an annual wage of less than $3,500.21 For measures (3) and (4), the threshold pointsrefer to the exemption levels for GIS benefits pre- and post-2008, respectively.I plot various household demographic control variables by the forcing variable to detect whether anypossible factors unrelated to the OAS/GIS benefits could be confounding the estimates (see Figures 2.4 and2.5). First, Figure 2.4 shows a discrete jump at around year 7 of immigration for the share of respondentswith private pensions. This gap could possibly reflect retirees’ adjustment to RRSP and/or RRIF holdingsin anticipation for OAS/GIS benefits. Early-arrived immigrants could also have greater access to better paidjobs than newly-arrived immigrants, which could be a potential explanation behind the sharp run-up in theshare of individuals with private pension holdings after the eleventh year since immigrated to Canada.22For both senior and working-age respondents, Figures 2.4 and 2.5 show that most of the control vari-ables exhibit a smooth pattern around the point of discontinuity, except for the share of respondents with aBachelor’s degree. As shown in Figure 2.6, the discontinuity for the education plot at around year 11 is onlyapplicable to year 2006. The same pattern can be observed in the 2001 series at around years 3-5 of immi-gration, but not for the other series. Therefore, the gap in the education series at around the discontinuitypoint for the 2006 data most likely reflects cohort effects. These trends also coincide with the time pe-riod when the Canadian government made changes to the selection system for skilled immigrants, businessimmigrants, and self-employed immigrants, with a greater emphasis on higher levels of education, officiallanguage ability, decision-making skills, and motivation and initiative (Canada, 1994; Canada, 1995). Con-sistent with the timing of the reform, Table 2.1 shows a sharp jump in the share of immigrants (permanentresidents) in natural sciences, engineering, and mathematics occupations between years 1994 and 1996.In light of these potential estimation concerns, I include several demographic control variables in Xirtto correct for these possible confounding factors and to increase the estimates’ precision.23 The Xirt vectoralso addresses the potential concern that immigrants who are eligible for OAS/GIS benefits could differ fromthose who do not. In particular, immigrants could be changing their characteristics over time. The vectorof observable characteristics for individual i is denoted by Xirt , which includes dummy variables controllingfor age, gender, disability, education, ethnicity, language, dwelling tenure status, and other private pensionsources. The inclusion of the private pension dummy accounts for the possibility that immigrants may beadjusting their pension holdings in anticipation for public pension entitlements. The education dummies andthe language indicator variable absorb the variation that results from the change in the immigration selectionsystem, with a tilt towards higher education levels and official language abilities. Generally, the results withand without the X vector are similar (see Appendix Table A.5 for the results without the X vector). As asupplementary analysis, I also present results with an additional control variable – the interaction of higheducation dummy and cohort groups – in order to further test whether the gap in the education control21The Canadian Census does not provide any variables specifically on retirement.22As an extra test, I also include a dummy variable for Canada Pension Plan (CPP) into the estimation. The results are similarwith and without the CPP dummy variable, which suggest that CPP exerts a minimal impact on labour supply choices. This findingalso holds when the discontinuity is set at year 9 (i.e. two years prior to the OAS/GIS eligibility date). These findings are consistentwith the findings from Baker and Benjamin (1999), who do not find any huge impact on labour supply choices through the earlyretirement provisions of the CPP. See Appendix Tables A.3 and A.4 for the results with the CPP dummy variable.23Bui et al. (2014) also apply a similar technique to an education context. They find a discontinuity in one of their covariates andcorrect for this problem by providing results both with and without controls.16variable could drive the estimations (see Appendix Table A.6). Overall, the gap in the education variableand in the private pension holdings does not seem to be driving the results in the dependent variable.δ is the main coefficient of interest. This gives the effect of the household maintainer’s receipt of theCanadian Old Age Security benefit at ages 65 and over starting at the tenth year of immigration. The Censusand the 2011 National Household Survey are conducted in mid-year, and both datasets do not provide theactual date of immigration. As such, just taking the year of the dataset and subtracting by the year ofimmigration may cause at least a half-year lag in the regression discontinuity design. A potential problemby setting the threshold point at year 10 instead of year 11 is that some of the individuals who are ineligiblefor the OAS/GIS benefits may be included into the treatment group.24 Furthermore, the income data inthe Census are based on year 2005 tax year and the application process for the OAS/GIS benefits is notautomatic. Therefore, to address these issues, for the main estimation with the 2006 Census data, I set thethreshold point to be at year 11 instead of at year 10 for the RD estimations.25 In other words,T REATirt =0 if t < 11 and age of maintainer ≥ 651 if t ≥ 11 and age of maintainer ≥ 65I restrict the analysis to immigrants who landed for 5 to 20 years. As noted in Section 2.3.1, individualswho do not live in Canada at the time of OAS application will need to show proof of residence for 20 yearssince age 18. This implies that the type of immigrants who have lived for 20 years or more could be differentfrom those who have lived for 20 years or less. On the other hand, I exclude the respondents who landedfor less than 5 years as immigrants take 4-5 years to fully settle in one location.26 Moreover, some of theimmigrants may immediately apply for non-residents and return to their home country once they obtainCanadian citizenship at the fourth year of immigration. This also suggests that the labour supply decisionsof new immigrants may not be fully representative of those who landed for more than 5 years. As part ofthe robustness checks, this paper re-estimates expression (2.1) using different time windows to determinewhether the results are sensitive to the choice of the bandwidth. I use the full sample for this sensitivity testin order to preserve as many observations as possible. I assume that the findings from this robustness checkare applicable to all family composition types.In addition, I re-estimate expression (2.1) using the 2011 National Household Survey (NHS) data toexamine whether the results are time-specific. As noted in the Data Description section, although the NHSdata are linked to income tax information, the voluntary nature of this dataset leads to low response ratesfrom certain types of respondents. Statistics Canada also did not implement consistent imputation proce-dures across the Census years. In light of the shortcomings as discussed in the Data Description section, Ifirst exclude the respondents with a non-missing OAS amount and with less than eleven years since immi-gration from the 2006 Census.27 Then, I compare the results produced by the modified 2006 data against24For example, individuals who arrived in Canada in December 1996 would be coded as having 10 years since immigrationinstead of 9.5. On the other hand, individuals who landed in December 1995 would be coded as having 11 years since immigrationinstead of 10.25The results are similar if I set the threshold point at year 10 instead of year 11 for the full sample. However, the results areweaker for finer family breakdowns.26See for example, Haan (2012).27Figure 2.1 shows the imputation procedure for 2011 NHS data goes up to the eleventh year since immigrated to Canada.17those generated by the 2011 National Household Survey (NHS) data for broad family compositions in orderto increase precision and to preserve as many observations as possible.As an extra check, I also compare the results produced by the 2006 Census data against those generatedby the longitudinal version of the Survey of Labour and Income Dynamics (SLID) dataset using expression(2.1). This supplementary analysis is necessary for three reasons. First, the SLID data contains a variablefor the actual number of years since immigrated to Canada, which eliminates the half-year lag problemin constructing the discontinuity point when using the 2006 Census. Second, Statistics Canada does notimpose any imputation procedures to tax-linked SLID data. Therefore, unlike the first comparison with theNHS data, I can keep all of the tax-linked observations including those who reported a non-zero OAS/GISvalue with less than ten years of immigration to Canada with the SLID data. This helps preserve as muchinformation as possible. Finally, the SLID data contains an indicator variable that identifies the main incomeearner in the census family, which helps distinguish the effect of the OAS/GIS benefits from the primaryand secondary earners. This cannot be done using the Census data because it only provides information onthe “reference person” in the census family. Yet, the “reference person” does not necessarily correspond tothe main income earner.For the supplementary analyses with the SLID data, the X vector is different from the one when usingthe Census data. First, I include year effects. Second, the SLID does not provide any information onlanguage usage and on visible minority (ethnicity) groups. Also, the non-response rate for the dwellingtenure variable was high for reference years 1999 and 2000 due to an error in collection. Statistics Canadaalso replaced many “don’t know” responses with the value for the household from a preceding or subsequentyear (if from the same dwelling) for year 2001. The screening questions for the disability status variablewere significantly modified starting in year 1999. In light of these data issues, I exclude language, visibleminority group, disability, and homeownership control variables from the X vector for this extra analysis.The estimation using the SLID data is from years 1993 to 2010.One of the key identification assumptions that underlies the regression discontinuity design is that f (t)is a smooth continuous function. This means that changes in the OAS/GIS benefits are the only source ofdiscontinuity in outcomes around year 10 of landing and for household maintainers of over age 64. I runvarious falsification tests to verify this. First, I examine the labour supply responses of main respondents ofages 25-54, both graphically and econometrically. I then restrict the sample to households with a spouse ofless than age 60 and a household maintainer of ages 25-54 and estimate both maintainer and spouse’s laboursupply behaviour. I select ages 25-54 for the placebo tests because this age group reflects the working-age population and the labour supply decisions of this group of individuals should be less likely to beconfounded by pension benefits. This falsification test should thus provide a cleaner comparison of labourmarket attachments between the treatment and control groups. If δ only captures the effect from the receiptof OAS/GIS benefits, then δ should converge to zero or be insignificant for any of these falsification tests.Yet, a few additional factors could threaten the identification strategy. For example, some applicantscould lie on the application form regarding their year of residence. If this were possible, then people withnine years of residence and with a higher propensity to receive social assistance could claim to have lived forTherefore, I set the threshold point to be at year 12 instead of year 11 for this comparison.18ten years in Canada. Since the OAS partial benefits depend on the number of years of residence at the timeof application, a person who landed for ten years could also falsify the year of immigration with a highervalue in order to receive more benefits. This problem is unlikely to occur. The Income Security and SocialDevelopment Branch confirms in writing that their Integrity Services Branch undertakes strict measures,including collaborating with other government departments, to detect client error, fraud and abuse; and theappropriate actions would be taken immediately if any anomalies or discrepancies were identified.Moreover, the choice to defer the public pension benefits generates a non-random selection in the mainanalysis sample. The partial OAS benefit amounts are pro-rated by the number of years resided in Canada atthe time of application, and the recipient receives the same monthly amount (prior to adjusting for inflation)for life. Therefore, this policy creates an incentive for immigrants to choose the timing of receipt. Thisproblem is also not likely to affect the validity of the estimation results. As noted above, the deferral programstarted in July 2013, which does not directly affect the time period of analysis in this paper. Moreover, therecipient would not be eligible for the GIS if he/she chooses to defer receipt of the OAS pension. The lossin GIS benefits is greater than the gain from waiting for a bigger OAS amount. Therefore, it is unlikely thatthe elderly immigrants will pursue this action.A related concern is that respondents may misreport the year of immigration in the Census data. Asnoted above, for the 2006 Census, senior citizens in residences were self-enumerated using Forms 2A and2B. This implies that some of these seniors could provide an incorrect answer to the year of immigrationquestion. However, this possibility is expected to be small. First, the Census form is not linked to anyofficial OAS/GIS applications and the respondents do not gain from lying on the Census form. Second, the2006 Census is linked to tax information, which could implicitly encourage respondents to be careful onthe Census questionnaire. In addition, Figure 2.7 shows that the density of the forcing variable is smoothand continuous around the point of discontinuity. The escalation in the density of the forcing variable thathappens around years 8 to 14 of immigration does not seem to be related to any “manipulation” effects or toany coding mistakes. This trend coincides with the increase in the number of parent/grandparent immigrantsduring the early to mid 1990s, which matches with the period when a mass number of Chinese immigrantsarrived in Canada due to the transfer of sovereignty over Hong Kong.28Furthermore, the estimation could face self-selection problems if immigrants choose to migrate toCanada based on their eligibility for public pension entitlements. For the Canadian context, this threatdepends on whether the immigrant’s place of origin is one of the 55 countries that has a social securityagreement (SSA) with Canada. This means that some immigrants could apply for Old Age Security pensionbenefits with less than ten years of residence in Canada. This could also cause a non-random selection inthe main estimation sample. Naturalization decisions may also confound the OAS take-up rate. ESDC sug-gests that at the time of the 2001 Census, the legislation allowed sponsored immigrants from social securityagreement countries to be eligible for income-tested benefits when they became citizens. Therefore, someelderly immigrants could apply for and start to receive the OAS pension between three and four years ofresidence in Canada. The data provided by ESDC shows that at most only 10% of the elderly immigrants28Using the 2006 Census data, Figure A.3 shows that only the immigration shares for Chinese immigrants rose between years1992-1998.19would receive OAS pension benefits via the social security agreement. Therefore, it is expected that theimplementation of SSA would cause minimal impact on non-random selection. As part of the robustnesschecks, I include a new dummy variable into the estimation to account for the effect from SSA, which equalsone if the respondent’s place of birth is from a country with social security agreements with Canada and zerootherwise.29 If non-random selection from SSA were to be negligible, then the impact from this new dummyvariable should pose a miniscule influence on the estimation results. In addition, I run another regressionwhere I limit the analysis to those who never receive OAS for the first ten years of landing to verify whetherSSA plays any role.Finally, anticipation effects could also affect the results. Immigrants could gradually adjust their laboursupply as they approach year 10 of immigration. In particular, the GIS is means-tested, which may encour-age households to adjust their labour supply decisions a few years prior to the OAS/GIS eligibility date inorder to maximize the amount of public pension entitlements. To evaluate the extent of this effect, I conductthree different estimations using the 2006 Census data.In the first test, I widen the threshold, by setting the treatment to equal 1 if the number of years sinceimmigration is greater than 9 instead of greater than 11. I set the new threshold point to be at year 9 to test foranticipation effect because as noted above, an RRSP/RRIF withdrawal can precede the actual GIS reductionby as much as 18 months. This implies that respondents could start re-adjusting their work decisions as earlyas two years prior to the receipt of OAS/GIS benefits. If an anticipation effect were to hold, then δ wouldshow a significant effect.30 Next, I create a donut hole by excluding the respondents with 9 to 12 years ofimmigration. The construction of the donut hole compares the labour supply decisions of immigrants wholanded for 5-8 years versus those who landed for 13-20 years. Similar to the first estimation, if respondentsexhibit any anticipation effect, the results from the donut hole estimation should exhibit a larger coefficientin absolute magnitude than those with a discrete discontinuity at year 11. Finally, I compare the outcomeresponses of households with low housing assets and low income against those with high housing asset andlow income in order to investigate whether asset-rich families also exhibit anticipation effects. I use 0.5 ofmedian income (low income measure) for identifying low income households and use a similar technique todefine households with low and high housing asset values. The Census does not provide any information onnon-housing wealth. Therefore, I assume that families with high housing asset value are asset-rich.2.5.2 Intensive margins of labour supply and time useThis study uses a combination of Census and Statistics Canada’s General Social Survey (GSS) data toestimate the effect of OAS/GIS pension benefits onto intensive margins of labour supply. I present theresults for all respondents and for employed individuals only. The findings for employed workers are aproxy for families that are situated in the middle of the static labour supply model.31 I use the results29The Census does not report the place of residence 10 years ago. It only provides the place of birth and the country of residencefive years ago.30One of the concerns with this estimation is whether the coefficients reflect self-selection bias due to social security agreementsor anticipation effects. To differentiate these two effects, I repeat this computation by eliminating the OAS recipients with less than10 years of immigration in order to extract out self-selection bias. The results are similar between the two cases, which implies thatthe coefficients reflect anticipation effect instead of self-selection bias. See Appendix Table A.7 for details.31See Figure 2.2 for details.20for all respondents, which include the immigrants who have already exited the labour force at the time ofsurvey, to proxy for households that are near the kink point in the same graphical framework. The sampleof all observations also accounts for the individuals who are only in or out of the labour force for partialyears. I run two different estimations to gather more information on elderly immigrants’ time use. First, thispaper applies the 2006 Census data to expression (2.1), where I replace the dependent variable Yirt with thefollowing three measures: (1) number of hours worked per week; (2) a binary variable for zero hours spenton unpaid housework; and (3) number of weeks during which persons worked for pay. Similar to the laboursupply estimations, I unbundle the time use analysis by various family composition types to investigate theextent to which elderly immigrant families exhibit joint labour supply behaviour.32The theoretical model predicts that lower income households will choose to reduce market productiontime once they are eligible for OAS/GIS benefits. Therefore, a question of interest is whether these individ-uals substitute market production hours with home production and/or with leisure, and which categories ofleisure increase once the individual qualifies for public pension entitlements. To answer this question, thispaper uses the GSS data as a supplementary analysis to better understand how elderly immigrants allocatetheir time in response to public pension entitlements.33 Given small sample size problem with the GSS,it is technically infeasible to construct a regression discontinuity design for individuals aged 65 and overand with 0-20 years of arrival to Canada with 72 observations.34 Although imperfect, this paper utilizes thefollowing difference-in-difference estimation to explore elderly immigrants’ time use:T IME jit = αj +X ′itβj +θDDELIGit ·AGE65PLit + γ1ELIGit + γ2AGE65PLit + ε jit (2.2)where T IME jit is the number of minutes spent on activity j per day for respondent i at time t. Xit containsyear and provincial effects; indicator variables for male, for speaking bilingual language at home, for home-ownership, and for disability; as well as dummy variables for education levels. ELIGit equals one if therespondent has arrived for more than ten years to Canada and zero otherwise. AGE65PLit equals one if theindividual’s age is 65 and over; and zero if the respondent’s age is between 25 and 54 (i.e. working-agepopulation).The key coefficient of interest is θDD, which captures the difference-in-difference measure and approxi-mates the average treatment effect on the treated (ATT). In order for θDD to converge to the ATT, the commontrends assumption is a necessary condition for identification. The validity of θDD requires the underlyingtrends in the time use variable to be the same for respondents of ages 65 and over (treatment group) andfor individuals of ages 25-54 (control group). However, it is graphically infeasible to plot the labour marketvariables by each year of immigration using the GSS data to check for common trends assumption. In lightof this issue, I rely on the Census data to graphically illustrate this condition, where I assume the Censusdata is representative and is comparable to the GSS Time Use data. Figure 2.8 shows the measures for the32I cannot perform similar computations using the SLID data due to small sample problems because the SLID only providesinformation on hours worked for employed individuals of ages 16-69.33This additional estimation follows from Aguiar and Hurst’s (2005) work on the retirement-consumption puzzle, and Aguiar etal.’s (2013) research on time use during the Great Recession.34I use the PUMF version of the GSS dataset for this part of the analysis. The number of observations is the same between thePUMF and the restricted versions of the dataset, and thus, there is no additional gain from using the restricted version of the GSS.21intensive margins of labour supply – (1) number of hours worked per week; (2) number of weeks duringwhich person worked for pay; and (3) share of workers with part-time jobs – which captures the marketproduction time. To summarize, Figure 2.8 suggests that the common trends assumption may hold for theworking individuals for certain labour supply variables. I assume that the trend before year 11 of immi-gration captures the path that the treatment group (immigrants of ages 65 and over) would have undergoneif it had not been treated. For employed respondents, all panels of Figure 2.8 show that the trends for thetreatment group (ages 65 and over) prior to the threshold point are somewhat aligned to the trends for thecontrol group (ages 25-54) for market production. This implies that similar patterns would have held forleisure and/or home production, as these components form the residual time. However, the last panel inFigure 2.8 reveals that the trends for the treatment and control groups prior to the point of discontinuity aredifferent if the sample includes all respondents (i.e. including those not in labour force).Under usual circumstances, the difference-in-difference method would not be the best choice to comparethe time usage for these two sub-groups. Although the GSS Time Use and the Census data might notprovide a comprehensive analysis of time usage within the family, these two supplementary estimationsprovide some insights as to where in the static labour supply model the elderly immigrants locate. This is animportant starting point in understanding whether these individuals respond to the OAS/GIS benefits withanticipation.2.6 ResultsOverall, the eligibility requirement for OAS/GIS pension benefits can strongly predict public pension en-titlements. Table 2.2 presents the regression analogs of Figure 2.1. The dependent variable equals one ifthe respondent receives OAS/GIS benefits and zero otherwise. The T REAT variable from expression (2.1)is the main explanatory variable of interest, which denotes the point of discontinuity based on residencyrequirements. Each column reports results from a separate regression. Columns (1) and (2) give the resultswith a linear spline. Columns (3) and (4) provide the estimations with a quadratic spline. The regressionincludes demographic controls for the even-numbered columns. For example, column (1) and row (1) sug-gests that OAS/GIS take-up rate increases by 47% for each additional eligible respondent. Table 2.2 showsthat the first-stage regressions are insensitive to the addition of demographic controls, which implies thatpublic pension entitlements are not confounded by specific household characteristics. The general resultsare similar with and without a quadratic spline.To summarize, elderly immigrants tend to be responsive to public pension entitlements for the extensivemargins of labour supply. However, there is not enough evidence to conclude that this senior sub-groupresponds to the OAS/GIS benefits at the intensive margins of labour supply. The time use data indicatesthe possibility of anticipation effect, since the total additional time spent on leisure and home productionactivities only changes slightly relative to the working-age population.Sections 2.6.1 and 2.6.2 present the results for the extensive margins of labour supply and the intensivemargins of labour supply, respectively. Note that all regression models are estimated by OLS. The advantageof using the OLS over the IV is that the OLS estimations provide a more explicit answer to the research22question of interest, especially when the threshold point moves away from the point of discontinuity to testfor anticipation effects.2.6.1 Extensive margins of labour supplyElderly immigrant families respond to public pension policies at the point of discontinuity for the extensivemargins of labour supply. Figure 2.9 illustrates graphically the relationship between the extensive marginsof labour supply and the number of years since immigrated to Canada for individuals of ages 65 and overand for those of ages 25-54. The figure uses the restricted version of the 2006 Census data to present fourdifferent labour market attachment measures, which include the employment rate, the share of respondentsnot in labour force, the share of respondents with less than $500 annual wage, and the share of individualswith less than $3,500 annual wage. As mentioned in the Empirical Strategy section, the elderly immigrantsmake up the treatment group. As such, the plot for the working-age population represents the results ofthe falsification tests. The graphs clearly show that only the senior population is responsive to the OldAge Security entitlement at the point of discontinuity, with a sharp decline in the employment rate, a steepincrease in the labour force exit rate, and an abrupt rise in the share of respondents with less than $500 and$3,500 annual wage for the aging group. On the other hand, labour supply responses of the working-agepopulation do not exhibit any discontinuity around the threshold point.35Table 2.3 presents the main regression results for the parameter on T REAT from expression (2.1), usingthe 2006 Census data. Each row reports the results for different family compositions. Rows (2), (3), (6),(7), and (10) provide the cases where the spouse is ineligible for Spousal Allowance, whereas rows (4),(5), (8), (9), and (11) exhibit the contrary cases. The first panel provides the labour supply responsesfor household maintainers; the second panel lists the findings for spouses; and the last panel presents thefindings for working-age children. Consistent with the results from Figure 2.9, the effect of OAS/GIS publicpension benefits applies to the extensive margins of labour supply. The impact of public pensions on laboursupply decisions is heterogeneous across family types. For example, column (1), row (1) shows that theemployment rate will decline by 4 percentage points (12 percent) when the household maintainer of single-person families reaches age 65 and has arrived in Canada more than ten years ago. In contrast, rows (2) to(5) of column (1) illustrate that employment rate will drop by approximately 2-42 percentage points (4-137percent) as the household maintainer of multi-person households qualifies for OAS/GIS benefits.36The results from Table 2.3 provide a few additional insights. First, the estimates for the householdmaintainer are similar across the four labour market attachment measures for single-person households,which imply that movements in or out of unemployment do not play any role. However, this finding doesnot seem to hold for multi-person families. One possible explanation for this is that the timing of the labourforce status and the income questions does not match in the Census. Specifically, the employment rate and35Overall, the findings are robust to the inclusion of recipients who may qualify for OAS/GIS through social security agree-ments. See Figures A.1 and A.2 in the Supplementary Appendix for a comparison of the labour supply responses with and withoutrecipients who received OAS/GIS payments and with less than ten years of immigration to Canada.36The results are similar using the probit model. See Appendix A for details. The numbers in bracket provide the averagetreatment on the treated computed using the fuzzy regression discontinuity (RD) approach, which is based on a two-stage leastsquares regression with the residence requirement as the instrument. Appendix Table A.8 provides the coefficients for the fuzzyregression discontinuity (RD) approach.23the labour force exit dependent variables are derived from the “Labour Market Activities: Labour ForceActivity” question, which refers to the labour market activity in the week prior to Census Day (May 16,2006). On the other hand, the wage-based dependent variable refers to income earned in the 2005 calendaryear. The discrepancy in results for the multi-person families indicates that certain family members did notwork the full year.In other words, household members may exhibit joint labour supply decisions in response to publicpension entitlements. The presence of working-age children appears to speed up the overall labour forceexit rate of elderly couples. This matches the predictions of the static labor supply model. For familieswith spouses who are ineligible for Spousal Allowance and with working-age children, row (2) shows thatthe labour force participation rate declines by nearly 40 percentage points for the household maintainer. Inaddition, the chance of receiving a wage of less than $500 increases by roughly 25 percentage points for themain respondent. On the other hand, there is insufficient evidence to suggest that the spouse adjusts his/herbehaviour as the household maintainer qualifies for public pensions. Yet, within this family structure, theworking-age child reduces the probability of earning low income by roughly 15 percentage points as oneof the parents becomes eligible for OAS/GIS benefits. These estimates point to the possibility that thechildren’s labour supply may be substituting for the parents’ work reductions in order to maximize publicpension entitlements for the maintainer.In contrast, for the same type of spouse with no kids, row (3) shows that the drop in labour forceparticipation rate for the household maintainer is approximately 20 percentage points. This magnitude isroughly half of that for the households with working-age kids. Unlike the previous case, the spouse showsa reduction in the employment rate of 17 percentage points, but the same trend cannot be observed for theother labour market attachment variables. Therefore, it is unclear that the spouse fully adjusts his/her laboursupply behaviour in response to public pension entitlements. Nonetheless, the reduced probability of labourforce exit for the main respondent implies that the presence of working-age kids provides a cushion for themaintainer to earn low wages.When the spouse is eligible for Spousal Allowance, the decline in the employment rate and the increasein the labour force exit rate for the household maintainer and the spouse are roughly similar between thefamilies with and without working-age kids. The drop in the employment rate is also much less in thesetwo types of family structures. The spouse shows insignificant effects on the labour force participation ratein response to the receipt of OAS/GIS benefits. On the other hand, row (11) from Table 2.3 shows that thelabour force participation rate will decline by nearly 7 percentage points for the working-age children as bothparents qualify for OAS/GIS benefits and the Spousal Allowance. Unlike the previous case, the working-agechildren do not show any increase in probability of earning high income. Instead, the coefficients for earninglow wages are small. By first glance, the OAS/GIS benefits seem to discourage the working-age childrenfrom working, which is a puzzling and an unintended consequence. The timing of the Census questionscould be one potential factor in explaining the discrepancy between the results produced by the labour forcestatus question and by the income-based question. Moreover, this result somewhat matches the predictionsfrom the static labour supply model, where the entire family reduces market production time as the marriedcouples qualify for public pension entitlements. To summarize, labour supply smoothing does not seem to24exist in families where both spouses qualify for public pension benefits.I conduct several robustness checks to test the validity of the estimation results. First, I use differentbandwidth windows to test the sensitivity of the results. Table 2.4 shows the regression results using the fullsample for different time windows, ranging from 5-20 years of immigration to 5-16 years of immigration(i.e. roughly 6 years away from the point of discontinuity). I also test the case with bandwidth of 2-20 years,which accounts for 9 years away from the point of discontinuity. The estimates using the wide bandwidthare nearly double those from the main estimations. On the other hand, the optimal bandwidth that minimizesthe asymptotic mean square error of the joint estimator gives the range of 8-14 years of immigration. Table2.4 shows that the results using the income-based dependent variables are roughly halved of those from themain estimations. Yet, the effect of OAS/GIS on the employment rate and labour force exit rate seems tobe negligible with small bandwidth. As noted above, the labour force status-based variables are based onCensus day, which creates a half-year lag in the estimation. Therefore, the drop in the employment rate isexpected to have occurred at year 10 rather than at year 11 of immigration. As such, it is reasonable thatthe employment rate and the labour force exit rate variables show insignificant effects when the point ofdiscontinuity is set at year 11. Generally, the results are aligned with expectations.In Tables 2.5 and 2.6, I compare the results produced by the 2006 Census data with those from the 2011National Household Survey and from the longitudinal Survey of Labour and Income Dynamics datasets,respectively. The overall message is similar between the datasets. The magnitude of the results generated bythe 2011 NHS tends to be smaller than those computed using the 2006 Census data, whereas those producedby the SLID are larger than the Census’s. Nonetheless, the magnitude of the coefficients for the full samplespecification is similar to Baker’s (2002) and Borjas’s (2011). Given small sample problems, the estimationsfor the main income earner and for the spouse using the SLID data yields insignificant coefficients. Despitethis, the magnitude of the results suggests that only the main income earner chooses to reduce labour effortin response to OAS/GIS benefits.Table 2.7 presents the regression analog for the falsification tests. The first panel reports the results forall respondents of ages 25-54. The second panel shows the findings for respondents of ages 25-54 in single-person families. The third and fourth panels illustrate the coefficients for household maintainers aged 25-54and with a spouse of less than 60 years of age in multi-person families, respectively. In all cases, most of thecoefficients are either statistically insignificant or are small. While there is a positive correlation betweenthe OAS eligibility and the dependent variable for earning a wage amount of less than $500 in single-personhouseholds, this correlation is not robust across different measures of labour market attachment. Thesefalsification test results are consistent with the graphical illustrations from Figure 2.9, which further confirmthat changes in the OAS/GIS benefits are the only source of discontinuity in outcomes around the point ofdiscontinuity for individuals of ages 65 and over.To test whether immigrants could be self-selecting to migrate to Canada for OAS/GIS benefits, Table2.8 reports regression results that account for social security agreements. Columns (1) to (4) show the casewhere I incorporate the SSA dummy variable, and columns (5) to (8) illustrate the scenario where I excludethe households with OAS/GIS benefits for less than 11 years of immigration to Canada. The coefficientsfor the main explanatory variable T REAT under the case with the SSA dummy variable and for the set with25excluded observations are similar to those presented in Table 2.3. This implies that non-random selectiondoes not seem to play any influential role in the estimations.Table 2.9 reveals whether any anticipation effect exists. Columns (1) to (8) report the estimates of theparameter on T REAT . Columns (1) to (4) provide the estimates with the point of discontinuity locatedat year 9 of immigration to Canada; and columns (5) to (8) illustrate the results for the donut hole using2006 Census data. The estimates in the first panel include all observations. The Census data suggeststhat elderly immigrants of ages 65 and over may exhibit anticipation effects in response to public pensionentitlements. The results are significant for the case when the point of discontinuity is moved back by twoyears, where respondents show a labour force exit rate of roughly 6-7 percentage points. The estimates fromthe donut hole show a decline in labour force participation rate of 7-10 percentage points. These numbersare larger than the case with a discrete point of discontinuity at year 11 of immigration, which suggests thatrespondents may be smoothing their behaviour in response to OAS/GIS benefits.The Census data also confirms that anticipation effects do exist for some family structures. Single-person households do exhibit different behaviour starting at year 9 of immigration. Row (2) shows that thelabour force exit rate decreases by 6 percentage points when the respondent enters year 9 of immigration.The donut hole also demonstrates similar findings. For families with spouse greater than age 60 and withno kids, both respondent and spouse seems to exit the labour force at year 9 of immigration, if the labourmarket attachment variable is based on annual wages. The results with the donut hole also shows somewhatsimilar findings. On the other hand, for families with spouses who do not qualify for Spousal Allowances,the household maintainer reduces earnings prior to the OAS/GIS eligibility date. This holds regardless ofthe presence of working-age children in the family. Conversely, the spouse seems to reduce labour effortonly if the working-age child is present. The working-age child shows a greater reduction in the proba-bility of earning low wages only for families with a parent who does not qualify for Spousal Allowances.These results point to the possibility that the parents may be adjusting labour supply behaviour prior to theOAS/GIS eligibility date, in order to maximize the amount of public pension entitlements through reportinga lower household income on tax returns.Finally, Table 2.10 compares the labour supply responses of low income households with high and lowamounts of housing assets. Generally, the decline in the labour force participation rate is stronger for familieswith low income and low housing asset values. However, the magnitude of the exit rate for households withhigh housing asset holdings is roughly 2 to 7 percentage points when the discontinuity point is at year 11.For both household types, the exit rate doubles as the point of discontinuity is pushed back by two years.These results suggest that certain asset-rich households may exhibit anticipation effect a few years prior tothe receipt of OAS/GIS benefits.2.6.2 Intensive margins of labour supply and time useThe regression results reveal that the receipt of the OAS/GIS benefits does not affect the intensive marginsof labour supply, which is consistent with the graphical results.37 Table 2.11 presents the estimates of theparameter on T REAT for the intensive margins of labour supply using the 2006 Census data. Using hours37See Figure 2.8.26worked per week as the dependent variable, Columns (1) and (2) list the coefficients for the regressionswith only working seniors and with all respondents as the sample, respectively. Column (3) shows theresults where the binary variable for zero hours of unpaid housework is the dependent variable. Column (4)displays the estimates for the number of weeks worked as the outcome variable. Each of the rows gives thefamily composition types.The first panel provides the responses for the main respondent. For seniors who are working, only thesingle-person households seem to reduce labour effort by 15 hours per week, which translates to roughlytwo days off per week. However, the results are insignificant, with a potential small sample problem.As discussed, one of the issues with the Census is that the labour force status questions only focus at aspecific point in time. Therefore, the individual could be in the labour force for part of the year. To accountfor this possibility, I also report the results for all observations under columns (2) and (3), which would alsoinclude the individuals who were not in the labour force at the time of the Census week. On net, the resultstend to show a reduction in hours worked per week by 2-4 hours. The time spent on housework does notseem to respond to public pension entitlements, except for single-person families. To further address thepotential concern that the time use variable may only capture the partial year of labour market activities, Ialso present the estimates for the number of weeks during which elderly respondents worked for the full year.Generally, the effect of public pension entitlement on the number of weeks worked per year is insignificant.Overall, the receipt of the OAS/GIS benefits does not significantly affect the intensive margins of laboursupply for the main respondent. Consistent with the static labour supply model, the main respondent seemsto gradually decline working hours prior to withdrawal from the labour market. Other family members,such as the spouse and the working-age children, do not seem to exhibit joint labour supply decisions byadjusting market production time, as the household maintainer qualifies for OAS/GIS benefits.Table 2.12 presents estimates of the main coefficient θDD from expression (2.2) using the General SocialSurvey (GSS) data.38 The unit for the estimates is minutes per day. Given the small sample, this paper showsthe results for the elderly respondents who arrived to Canada for 0-20 years in column (1); and for all yearsof immigration in columns (2) and (3). For most of the categories, the results are similar between the twotypes of immigrants (columns (1) and (2)), when we include all observations into the analysis. This impliesthat the findings are generally robust to the choice of the year of arrival to Canada. Based on this finding, Ionly report the regression results for employed immigrants of all years of landing, where I assume that theaverage time usage of all immigrants is representative.39For market production activities (“paid work” and “activities related to paid work”), θDD is insignificantwhen I include respondents who have already been out of the labour force. Nonetheless, in column (1) ofrow (1), Table 2.12 shows that the magnitude of the estimate for the “paid work” category is 73 minutesper day. This value translates to a reduction of roughly 7 hours of paid work per week as compared to theworking-age population. This estimate is consistent with the results from Table 2.11, which shows that onaverage elderly respondents tend to reduce work effort by up to 2 working days in response to public pensionentitlements. On the other hand, relative to the control group, column (3) shows that employed respondents38See Appendix A for a detailed description of the time use categories.39The GSS data (years 1998, 2005, and 2010 combined) only contains 7 observations for employed immigrants of ages 65 andover with 0-20 years of landing; and 149 observations for employed immigrants of the same age range for all years of landing.27tend to respond to OAS/GIS benefits by decreasing the number of minutes worked per day by 350 minutes,which translates to taking 1 week off. This seems to be consistent with the static labour supply model, whichpredicts that overall family market production time would gradually decline to the kink point.Given that elderly respondents do respond to OAS/GIS benefits through decreases in their labour forceparticipation rates, the ultimate question of interest is where do these immigrants devote the extra non-marketproduction time. Although incomplete, the GSS data suggests that relative to the working-age population(control group), the elderly immigrants (both in labour force and retirees) who landed for 0-20 years inCanada tend to devote roughly an extra 45 minutes per day to meals eaten at non-restaurant locations.However, this sub-population group does not respond to public pension entitlements through an increasein other home production activities. Conversely, the employed individuals replace market production withhome production, such as housekeeping, shopping, and other household work.Immigrants who continue working beyond retirement age seem to increase their time on leisure activitiessuch as reading and watching TV in response to OAS/GIS benefits. However, the effect of OAS/GIS onleisure time is mixed once I include non-working individuals into the sample. On one hand, the elderlyimmigrants show an increase of 30 minutes per day on education-related activity and 25 minutes per dayon certain active leisure activities (i.e. hobbies, games, pleasure drives). On the other hand, relative to theworking-age population, this sub-population group shows a reduction of approximately 1 hour in socializingtime at home. Yet, for activities such as night sleep/essential sleep, θDD is significant only under column (2)because these categories show large standard errors for more restricted samples. Despite this, relative to thecontrol group, the estimates under column (2) for all years of immigration suggest that elderly immigrants(both in labour force and retired) generally spend about an extra 15 minutes per day on child care; at leastan extra hour per day on night sleep/essential sleep; and at least 30 minutes more per day on other activeleisure activities.These findings suggest that the elderly immigrants may have responded to OAS/GIS benefits with an-ticipation. Those who are employed first reduce working hours by replacing market production with homeproduction. Referring back to the static labour supply model, this implies that working families first re-adjust to a location near the kink point. By including individuals who have already exited the labour force,relative to the working-age population, the total additional time spent on leisure and on non-restaurant meal-related activities changes slightly before and after the receipt of OAS/GIS benefits. The Census data alsoreveals that other family members do not respond to public pension entitlements by adjusting their marketand home production time. Therefore, for households with low labour efforts, the original utility curve isexpected to have already located near the kink point for not working (i.e. point C), and households wouldmake an upward (near) parallel move to the utility curve with OAS+GIS benefits. In order for this finding tohold theoretically, working hours would have been reduced some time prior to the OAS/GIS eligibility date.Therefore, a combination of results points to the possibility of anticipation effects.282.7 ConclusionThis paper uses the eligibility requirements of the Canadian Old Age Security program to estimate a purecausal effect of public pensions on elderly immigrants’ labour supply decisions. Specifically, I investigatewhether labour market responses for the recipient, the spouse, and the working-age children exhibit anyheterogeneities across different family composition types and housing wealth. I also apply time use data toprovide a more comprehensive analysis on work intensity decisions. This paper improves Borjas’s (2011)estimation by using a regression discontinuity design to directly compare the labour market responses frompublic pension entitlements of newly-arrived immigrants and of early-arrived immigrants. This estimationcannot be computed in the previous literature. This study also evaluates the extent to which these immigrantfamilies may exhibit anticipatory behaviour in response to OAS benefits.My results reveal that elderly immigrants aged 65 and over respond to the OAS benefits with a decreasein labour force participation rates. The effect of OAS/GIS benefits is heterogeneous across family types. Thespouse’s and the working-age children’s labour supply decisions depend on whether Spousal Allowance isavailable in the family.The results of this study raise a key concern of whether elderly immigrants are making efficient workchoices. The implementation of OAS/GIS benefits seems to encourage weak labour market attachment afew years prior to the eligibility date. Several pieces of evidence point to the possibility of an anticipationeffect. For example, elderly immigrants including those with high asset holdings tend to show a decline inlabour force participation rates even when the point of discontinuity is moved back by 2 years. Relative tothe working-age population, time use data also reveals that employed individuals replace market productionwith home production. Both the regression discontinuity design and the difference-in-difference frameworksdo not find any evidence of strong linkages between public pension eligibility and work intensity once Iinclude the households with low labour effort into the estimations.My results, which show a strong impact only on the extensive margins of labour supply, are consistentwith expectations. First, these estimates are consistent with Danzer’s (2013), which also does not find anystrong labour supply effects at the intensive margins for an Ukrainian context. He argues that reducingworking hours is possible only for a few service occupations and/or for low-skilled workers. Baker andBenjamin (1999) also illustrate using the introduction of early retirement provisions to Canada’s publicpension plan that the reform only led to an increase in pension receipt, but had a minimal effect on laboursupply behaviour. They suggest that men who took advantage of the early retirement provisions tend toexhibit weak labour market attachment. Furthermore, my findings imply that the predictions from the staticlabour supply model are valid. The model suggests that family members as a whole will first react to publicpension entitlements by gradually declining labour hours; and the main income earner will be the last toleave the labour market. Referring back to the theoretical framework, the small reaction in labour hoursto the OAS for a majority of main recipients points to the possibility that the household’s utility curve islocated near the kink point for not working. This implies that immigrant families may have already reducedlabour effort in anticipation for public pension entitlements.292.8 FiguresFigure 2.1: Relationship between the number of years since immigrated to Canada and the OAS take-up ratePlots constructed using raw data [1]Plots constructed using imputed data [2]Notes:[1] At the time of the 2001 Census, the legislation allowed sponsored immigrants from social security agreementcountries to be eligible for income-tested benefits when they became citizens. Therefore, both the 2001 and 2006Censuses show an upward spike in the OAS take-up rates between the third and fourth year of residence in Canada.[2] Statistics Canada imposed an imputation procedure to the 2011 National Household Survey data, by moving theOAS amounts to the “other government transfer” category for any observations with less than eleven years of landingto Canada. However, this imputation procedure was not done to the naturalized Canadian citizens for the 2001 and2006 Censuses. For a better comparison, I imposed the legislation that respondents can only receive OAS benefitsafter year 10 of immigration to the 2001 and 2006 Census datasets. The Census and the National Household Surveyare conducted in mid-year, and both datasets do not provide the actual date of immigration. Therefore, just takingthe year of the dataset and subtracting by the year of immigration may cause at least a half-year lag in the regressiondiscontinuity (RD) design. In this case, I set the threshold point to be at year 11 instead of at year 10 for the RDestimations.Sources: Statistics Canada’s 2001 and 2006 Censuses and 2011 National Household Survey; and author’s calculations30Figure 2.2: Static labour supply model for households without working-age children31Figure 2.3: Static labour supply model for households with working-age children32Figure 2.4: Relationship between the number of years since immigrated to Canada and demographic controls- Immigrants of ages 65 and over35404550In percent5 10 15 20Number of years since immigrationShare of male respondents15202530In percent5 10 15 20Number of years since immigrationShare of disabled respondents657075In percent5 10 15 20Number of years since immigrationHomeownership rate1015202530In percent5 10 15 20Number of years since immigrationShare of respondents who have private pensions15202530In percent5 10 15 20Number of years since immigrationShare of respondents with at least a Bachelor's degree40455055606570In percent5 10 15 20Number of years since immigrationShare of respondents who are bilingualSources: Statistics Canada’s 2006 Census; and author’s calculations33Figure 2.5: Relationship between the number of years since immigrated to Canada and demographic controls- Immigrants of ages 25-5435404550In percent5 10 15 20Number of years since immigrationShare of male respondents468In percent5 10 15 20Number of years since immigrationShare of disabled respondents50556065707580In percent5 10 15 20Number of years since immigrationHomeownership rate0.100.200.300.400.50In percent5 10 15 20Number of years since immigrationShare of respondents who have private pensions3040506070In percent5 10 15 20Number of years since immigrationShare of respondents with at least a Bachelor's degree9092949698100In percent5 10 15 20Number of years since immigrationShare of respondents who are bilingualSources: Statistics Canada’s 2006 Census; and author’s calculations34Figure 2.6: Share of respondents with at least a Bachelor’s degree, for years 1986-2006101520253035In percent0 5 10 15 20Number of years since immigration1986 1991 19962001 2006Share of respondents with at least a Bachelor's degreeSources: Statistics Canada 1986 - 2006 Censuses and author’s calculations35Figure 2.7: Number of parent/grandparent immigrants and density, by forcing variable36Figure 2.8: Intensive margins of labour supply - Immigrants of ages 65 and over versus those of ages 25-54101112131415In percent203040506070In percent5 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)Share of workers with part-time jobs35404550354045505 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)during which persons worked for pay [1]Average number of weeks per year303540455020253035405 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)during which persons worked for payAverage number of hours per week51015202530351234565 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)for all respondentsAverage number of hours worked per weekNote:[1] This variable is intended for employed individuals and refers to the number of weeks in 2005 during which therespondent worked for pay or in self-employment.Sources: Statistics Canada’s 2006 Census; and author’s calculations37Figure 2.9: Extensive margins of labour supply - Immigrants of ages 65 and over versus those of ages 25-547075808590In percent05101520In percent5 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)Employment Rate051015202530In percent707580859095100In percent5 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)Share of respondents not in labour force051015202530In percent707580859095100In percent5 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)Share of respondents with wages < $5005101520253035In percent707580859095100In percent5 10 15 20Number of years since immigrationAge 65 plus (left axis)Age 25-54 (right axis)Share of respondents with wages < $3500Sources: Statistics Canada’s 2006 Census; and author’s calculations382.9 TablesTable 2.1: Share of permanent residents per selected occupational groups1992 1993 1994 1995 1996Managerial, Administrative 3.1% 3.3% 4.8% 5.6% 7.2%Natural Sciences, Engineering and Mathematics 4.5% 7.1% 10.9% 14.1% 17.0%Teaching 1.5% 1.7% 1.7% 1.6% 1.5%Medicine and Health 2.2% 2.4% 2.7% 2.6% 2.3%Clerical 4.7% 5.0% 6.2% 6.8% 6.8%Sales 2.1% 2.2% 2.5% 2.4% 2.4%Services 6.6% 6.6% 4.6% 5.1% 4.3%Farming, Horticultural and Animal-Husbandry 1.2% 1.6% 1.1% 0.7% 0.8%Fabricating, Assembling and Repairing 4.6% 4.7% 4.2% 4.0% 3.9%Construction 2.1% 1.6% 1.3% 1.2% 0.9%Machining and Related Occupations 1.1% 1.0% 0.9% 0.8% 0.7%Sources: Citizenship and Immigration Statistics 1996, Service Canada; and author’s calculations39Table 2.2: Effect of residency requirements on OAS/GIS take-upDependent variable: Binary variable for receiving OAS/GIS benefitsIndependent variable: (1) (2) (3) (4)Full sample:(1) T REAT : discontinuity at year 11 0.469***(0.030)0.471***(0.030)0.482***(0.030)0.481***(0.029)Number of observations 26,640 26,635 26,640 26,635Living alone:(2) T REAT : discontinuity at year 11 0.359***(0.049)0.353***(0.043)0.250***(0.038)0.256***(0.033)Number of observations 2,950 2,945 2,950 2,945With spouse < age 60, with kids ≥ age 25(3) T REAT : discontinuity at year 11 0.263(0.151)0.305**(0.137)0.394***(0.131)0.448***(0.120)Number of observations 385 385 385 385With spouse < age 60, with no kids(4) T REAT : discontinuity at year 11 0.429***(0.042)0.447***(0.041)0.324***(0.059)0.326***(0.046)Number of observations 525 525 525 525With spouse ≥ age 60, with kids ≥ age 25(5) T REAT : discontinuity at year 11 0.564***(0.045)0.566***(0.044)0.749***(0.069)0.744***(0.063)Number of observations 1,655 1,655 1,655 1,655With spouse ≥ age 60, with no kids(5) T REAT : discontinuity at year 11 0.526***(0.032)0.524***(0.031)0.534***(0.037)0.532***(0.036)Number of observations 5,390 5,390 5,390 5,390Linear spline? Yes Yes No NoQuadratic spline? No No Yes YesControls? No Yes No YesThis table shows the regression results for year 2006 using expression (2.1). All regression models contain provincedummies. The demographic controls include dummy variables controlling for age, gender, disability, education, eth-nicity, dwelling tenure status, language, and other private pension sources. These models are weighted by individualCensus weights. Standard errors are clustered by the number of years since immigration and are in parentheses. ***Significant at 1%; ** significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.40Table 2.3: Effect of OAS/GIS on family members’ labour supply decisionsThis table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.Dependent variables Indicator foremploymentIndicator fornot in labourforceIndicator forannual wage< $500Indicator forannual wage< $3,500Family types (1) (2) (3) (4)Household maintainer’s labour supply decision in response to OAS/GIS benefits(1) Live alone -0.042*(0.022)0.044**(0.019)0.041(0.036)0.045(0.035)Number of observations 2,945 2,945 2,945 2,945(2) Multi-person family:Spouse < age 60 + kids ≥ age 25-0.417**(0.195)0.395*(0.188)0.246**(0.112)0.247***(0.070)Number of observations 385 385 385 385(3) Multi-person family:Spouse < age 60 + No kids-0.163***(0.041)0.219***(0.065)0.262***(0.046)0.238***(0.041)Number of observations 525 525 525 525(4) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.057(0.077)0.062(0.067)0.092***(0.028)0.069(0.045)Number of observations 1,655 1,655 1,655 1,655(5) Multi-person family:Spouse ≥ age 60 + No kids-0.022(0.016)0.032*(0.016)0.099***(0.014)0.096***(0.014)Number of observations 5,390 5,390 5,390 5,390Spouse’s labour supply decision in response to household maintainer’s receipt of OAS/GIS benefits(6) Multi-person family:Spouse < age 60 + kids ≥ age 25-0.040(0.077)0.044(0.110)0.133(0.095)0.100(0.081)Number of observations 385 385 385 385(7) Multi-person family:Spouse < age 60 + No kids-0.171**(0.067)0.136(0.082)-0.055(0.087)-0.138(0.086)Number of observations 525 525 525 525(8) Multi-person family:Spouse ≥ age 60 + kids ≥ age 250.014(0.053)-0.029(0.052)0.085**(0.034)0.074**(0.031)Number of observations 1,655 1,655 1,655 1,655(9) Multi-person family:Spouse ≥ age 60 + No kids0.033*(0.016)-0.018(0.014)0.047***(0.013)0.047***(0.014)Number of observations 5,390 5,390 5,390 5,390(continued on next page...)41Table 2.3: Effect of OAS/GIS on family members’ labour supply decisions (continued)This table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.Dependent variables Indicator foremploymentIndicator fornot in labourforceIndicator forannual wage< $500Indicator forannual wage< $3,500Family types (1) (2) (3) (4)Working-age child’s labour supply decisions in response to household maintainer’s receiptof OAS/GIS benefits(10) Multi-person family:Spouse < age 60 + kids ≥ age 250.085(0.093)0.053(0.074)-0.132**(0.060)-0.147*(0.082)Number of observations 645 645 645 645(11) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.072(0.042)0.075*(0.036)-0.016(0.040)-0.005(0.056)Number of observations 2,255 2,255 2,255 2,255This table shows the OLS regression results for year 2006 using expression (2.1). All regression models containprovince dummies and demographic controls. Each row denotes the type of family composition. The demographiccontrols include dummy variables controlling for age, gender, disability, education, ethnicity, language, dwellingtenure status, and other private pension sources. These models are weighted by individual Census weights. Standarderrors are clustered by the number of years since immigration and are in parentheses. *** Significant at 1%; **significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.42Table 2.4: Effect of OAS/GIS on labour supply decisions of elderly immigrants using different bandwidthsFull sample: All respondents of ages 65 and overDependent variables Indicator foremploymentIndicator fornot in labourforceIndicator forannual wages< $500Indicator forannual wages< $3,500Independent variables: (1) (2) (3) (4)Time window: 5-20 years since immigration(1) T REAT : discontinuity at year 11 -0.026*(0.015)0.031**(0.011)0.059***(0.014)0.051***(0.016)Number of observations 26.635 26,635 26,635 26,635Time window: 5-18 years since immigration(3) T REAT : discontinuity at year 11 -0.024***(0.015)0.029**(0.012)0.059***(0.015)0.050**(0.017)Number of observations 23,825 23,825 23,825 23,825Time window: 5-16 years since immigration(4) T REAT : discontinuity at year 11 -0.024(0.016)0.030**(0.012)0.058***(0.016)0.048**(0.018)Number of observations 20,250 20,250 20,250 20,250Time window: 2-20 years since immigration(4) T REAT : discontinuity at year 11 -0.054**(0.021)0.064***(0.021)0.089***(0.021)0.079***(0.022)Number of observations 29,725 29,725 29,725 29,725Time window: 8-14 years since immigration(4) T REAT : discontinuity at year 11 0.006(0.009)0.007(0.008)0.032*(0.014)0.025(0.017)Number of observations 12,755 12,755 12,755 12,755This table shows the OLS regression results for year 2006, for all respondents of ages 65 and over. See expression(2.1) for the construction of the regression discontinuity design. All regression models contain province dummiesand demographic controls. The demographic controls include dummy variables controlling for age, gender, disability,education, ethnicity, language, dwelling tenure status, and other private pension sources. These models are weightedby individual Census weights. Standard errors are clustered by the number of years since immigration and are inparentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.43Table 2.5: Effect of OAS/GIS on extensive margins of labour supply - 2006 Census versus 2011 NationalHousehold Survey (NHS)This table reports the coefficients for the main explanatory variable T REAT , with thediscontinuity at year 12.Dependent variables: Indicator foremploymentIndicator fornot in labourforceIndicator forannual wages <$500Indicator forannual wages <$3,500(1) (2) (3) (4)Full sample (all respondents aged 65 and over)(1) 2006 Census data: -0.041*(0.023)0.042**(0.018)0.087***(0.019)0.086***(0.023)Number of observations 23,320 23,320 23,320 23,320(2) 2011 NHS data: -0.018(0.012)0.020(0.015)0.050***(0.013)0.041***(0.013)Number of observations 23,905 23,905 23,905 23,905Single-person households(3) 2006 Census data: -0.089***(0.027)0.075**(0.028)0.068(0.043)0.062*(0.035)Number of observations 2,535 2,535 2,535 2,535(4) 2011 NHS data: 0.030(0.030)-0.025(0.023)0.022(0.031)-0.006(0.030)Number of observations 2,525 2,525 2,525 2,525Spouse ≥ age 60(5) 2006 Census data: -0.050(0.033)0.058*(0.028)0.128***(0.009)0.134***(0.019)Number of observations 6,315 6,315 6,315 6,315(6) 2011 NHS data: -0.030(0.028)0.033(0.032)0.094***(0.028)0.084**(0.032)Number of observations 7,045 7,045 7,045 7,045(continued on next page...)44Table 2.5: Effect of OAS/GIS on extensive margins of labour supply - 2006 Census versus 2011 NationalHousehold Survey (NHS) (continued)This table reports the coefficients for the main explanatory variable T REAT , with thediscontinuity at year 12.Dependent variables: Indicator foremploymentIndicator fornot in labourforceIndicator forannual wages <$500Indicator forannual wages <$3,500(1) (2) (3) (4)Spouse < age 60(7) 2006 Census data: -0.225***(0.073)0.199***(0.057)0.105(0.081)0.147*(0.081)Number of observations 1,140 1,140 1,140 1,140(8) 2011 NHS data: -0.080(0.050)0.121**(0.055)0.063(0.093)0.051(0.115)Number of observations 860 860 860 860This table shows the OLS regression results using expression (2.1) for respondents of ages 65 and over. In order for theregression results produced by the 2006 Census to be comparable to those generated by the 2011 National HouseholdSurvey (NHS), I exclude the respondents who reported receiving OAS/GIS payments and with less than 11 years ofimmigration from the 2006 Census. To account for the half-year lag in the Census and in the NHS data, I set the pointof discontinuity at year 12. All regression models contain province dummies and demographic controls. See text fora description of the demographic controls. These models are weighted by person-level weights. Standard errors areclustered by the number of years since immigration and are in parentheses. *** Significant at 1%; ** significant at5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census and 2011 National Household Survey; and author’s calculations.45Table 2.6: Effect of OAS/GIS on extensive margins of labour supply - 2006 Census versus Survey of Labourand Income Dynamics (SLID)Dependent variable: Indicator for earning a wage amount of less than $500 [1]Datasets (1) 2006 Census data[2](2) 2006 Census data[2](3) 1993-2010 SLIDdata [3]Independent variables: T REAT Discontinuity atyear 11Discontinuity atyear 10Discontinuity atyear 10(1) (2) (3)All respondents (ages 65 and above) 0.059***(0.014)0.077***(0.016)0.088*(0.044)Number of observations 26,635 26,635 1,878Main income earner inmulti-person households(ages 65 and above)N/A N/A 0.046(0.030)Number of observations N/A N/A 1,034Spouse (all ages) inmulti-person householdsN/A N/A -0.075(0.048)Number of observations N/A N/A 630This table shows the OLS regression results using expression (2.1). All regression models contain province dummiesand demographic controls. These models are weighted by individual weights. Standard errors are clustered by thenumber of years since immigration and are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at10%.[1] The SLID dataset only provides labour force status for respondents of ages 16-69. Therefore, I define respondentswho exited the labour market as those who reported a wage amount of less than $500.[2] This regression contains the following demographic control variables: age, gender, disability, education, ethnicity,language, dwelling tenure status, and other private pension sources.[3] This regression contains the following demographic control variables: age, gender, education, and other privatepension sources.Sources: Statistics Canada’s 2006 Census and Survey of Labour and Income Dynamics; and author’s calculations.46Table 2.7: Effect of OAS/GIS on extensive margins of labour supply - A falsification testDependent variables Indicator foremploymentIndicator fornot in labourforceIndicator forannual wages <$500Indicator forannual wages <$3,500Independent variables: (1) (2) (3) (4)Full sample: labour supply decisions of respondents of ages 25-54(1) T REAT : discontinuity at year 11 -0.001(0.004)0.000(0.004)-0.008**(0.003)-0.008***(0.002)Number of observations 243,050 243,050 243,050 243,050Single-person families: Labour supply decisions of respondents of ages 25-54(2) T REAT : discontinuity at year 11 0.014(0.024)-0.015(0.017)-0.041***(0.017)-0.027(0.023)Number of observations 13,960 13,960 13,960 13,960Multi-person families with a spouse of ages < 60:Labour supply decisions of household maintainers of ages 25-54(3) T REAT : discontinuity at year 11 -0.009(0.009)0.009(0.005)-0.006(0.008)-0.009(0.010)Number of observations 76,215 76,215 76,215 76,215Multi-person families with a household maintainer of ages 25-54:Labour supply decisions of spouses of ages < 60(4) T REAT : discontinuity at year 11 0.019*(0.010)-0.011(0.010)-0.020*(0.011)-0.018*(0.010)Number of observations 76,215 76,215 76,215 76,215This table shows the regression results for year 2006 using expression (2.1). All regression models contain provincedummies and demographic controls. The demographic controls include dummy variables controlling for age, gender,disability, education, ethnicity, language, dwelling tenure status, and other private pension sources. These models areweighted by individual Census weights. Standard errors are clustered by the number of years since immigration andare in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.47Table 2.8: Effect of OAS/GIS on extensive margins of labour supply - Social Security AgreementThis table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.With SSA dummy Exclude respondents with OAS & with lessthan 11 years of immigrationDependent variables Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)Household maintainer’s labour supply decision in response to OAS/GIS benefits.(1) Live alone -0.042*(0.022)0.044**(0.019)0.041(0.036)0.045**(0.035)-0.069**(0.036)0.065*(0.036)0.050(0.053)0.062(0.049)Number of observations 2,945 2,945 2,945 2,945 2,680 2,680 2,680 2,680(2) Multi-person family:Spouse < age 60 + kids ≥ age 25-0.413*(0.196)0.392*(0.189)0.234*(0.112)0.242***(0.070)-0.434**(0.197)0.400**(0.182)0.246**(0.105)0.262***(0.066)Number of observations 385 385 385 385 370 370 370 370(3) Multi-person family:Spouse < age 60 + No kids-0.166***(0.043)0.222***(0.063)0.263***(0.047)0.240***(0.043)-0.286***(0.047)0.349***(0.085)0.347***(0.064)0.343***(0.055)Number of observations 525 525 525 525 495 495 495 495(4) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.058(0.076)0.062(0.066)0.093***(0.027)0.070(0.044)-0.076**(0.095)0.083(0.076)0.118**(0.054)0.110(0.065)Number of observations 1,655 1,655 1,655 1,655 1,555 1,555 1,555 1,555(5) Multi-person family:Spouse ≥ age 60 + No kids-0.022(0.016)0.032*(0.016)0.099***(0.014)0.096***(0.014)-0.032**(0.012)0.048***(0.013)0.138***(0.014)0.132***(0.013)Number of observations 5,390 5,390 5,390 5,390 5,095 5,095 5,095 5,095(continued on next page...)48Table 2.8: Effect of OAS/GIS on extensive margins of labour supply - Social Security Agreement (continued)This table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.With SSA dummy Exclude respondents with OAS & with lessthan 11 years of immigrationDependent variables Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)Spouse’s labour supply decision in response to household maintainer’s receipt of OAS/GIS benefits(6) Multi-person family:Spouse < age 60 + kids ≥ age 25-0.052(0.074)0.051(0.109)0.130(0.096)0.097(0.081)-0.078(0.086)0.074(0.117)0.163(0.093)0.128(0.081)Number of observations 385 385 385 385 370 370 370 370(7) Multi-person family:Spouse < age 60 + No kids-0.170**(0.063)0.135**(0.081)-0.056(0.087)-0.138(0.087)-0.155**(0.068)0.146(0.084)-0.060(0.092)-0.183*(0.091)Number of observations 525 525 525 525 495 495 495 495(8) Multi-person family:Spouse ≥ age 60 + kids ≥ age 250.013(0.053)-0.028(0.051)0.087**(0.033)0.075**(0.030)0.002(0.048)-0.019(0.050)0.093**(0.042)0.081**(0.036)Number of observations 1,655 1,655 1,655 1,655 1,555 1,555 1,555 1,555(9) Multi-person family:Spouse ≥ age 60 + No kids0.033*(0.016)-0.018(0.013)0.047***(0.013)0.047***(0.014)0.029(0.022)-0.009(0.017)0.079***(0.015)0.082***(0.017)Number of observations 5,390 5,390 5,390 5,390 5,095 5,095 5,095 5,095(continued on next page...)49Table 2.8: Effect of OAS/GIS on extensive margins of labour supply - Social Security Agreement (continued)This table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.With SSA dummy Exclude respondents with OAS & with lessthan 11 years of immigrationDependent variables Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)Working-age child’s labour supply decisions in response to household maintainer’s receipt of OAS/GIS benefits(10) Multi-person family:Spouse < age 60 + kids ≥ age 250.103(0.093)0.039(0.074)-0.144**(0.066)-0.164*(0.085)-0.053(0.080)0.131*(0.068)-0.092(0.063)-0.115(0.074)Number of observations 645 645 645 645 620 620 620 620(11) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.073(0.042)0.075*(0.036)-0.015(0.040)-0.005(0.055)-0.057(0.044)0.067*(0.033)0.000(0.053)0.009(0.078)Number of observations 2,255 2,255 2,255 2,255 2,120 2,120 2,120 2,120This table shows the OLS regression results for year 2006 using expression (2.1). All regression models contain province dummies and demographic controls.Each row denotes the type of family composition. The demographic controls include dummy variables controlling for age, gender, disability, education, ethnicity,language, dwelling tenure status, and other private pension sources. These models are weighted by individual Census weights. Standard errors are clustered by thenumber of years since immigration and are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.50Table 2.9: Effect of OAS/GIS on extensive margins of labour supply - Anticipation effectMain explanatory variable:Discontinuity at year 9Main explanatory variable:Donut hole at years 9-12Dependent variables: Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)All respondents of ages 65+(1) Full sample -0.062***(0.015)0.061***(0.016)0.074***(0.021)0.065**(0.023)-0.077***(0.025)0.081**(0.027)0.101**(0.035)0.102**(0.034)Number of observations 26,635 26,635 26,635 26,635 21,925 21,925 21,925 21,925Household maintainer’s labour supply decision in response to OAS/GIS benefits.(2) Live alone 0.032(0.019)-0.064***(0.020)0.030(0.027)0.022(0.025)0.006(0.021)-0.035***(0.009)0.085***(0.020)0.073***(0.014)Number of observations 2,945 2,945 2,945 2,945 2,475 2,475 2,475 2,475(3) Multi-person family:Spouse < age 60 + kids ≥ age 25-0.238(0.182)0.252(0.165)0.294***(0.104)0.193**(0.078)-0.113(0.341)0.200(0.269)-0.042(0.148)0.051(0.113)Number of observations 385 385 385 385 330 330 330 330(4) Multi-person family:Spouse < age 60 + No kids-0.105(0.095)0.153(0.102)0.231***(0.072)0.174**(0.079)-0.168(0.209)0.125(0.253)0.172(0.144)0.199(0.144)Number of observations 525 525 525 525 435 435 435 435(5) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.030(0.048)0.022(0.062)0.051(0.037)0.057(0.046)-0.101(0.073)0.127(0.119)0.078(0.061)0.028(0.069)Number of observations 1,655 1,655 1,655 1,655 1,380 1,380 1,380 1,380(6) Multi-person family:Spouse ≥ age 60 + No kids-0.114***(0.023)0.117***(0.026)0.092***(0.032)0.108***(0.028)-0.055(0.048)0.060(0.050)0.091*(0.048)0.138**(0.046)Number of observations 5,390 5,390 5,390 5,390 4,440 4,440 4,440 4,440(continued on next page...)51Table 2.9: Effect of OAS/GIS on extensive margins of labour supply - Anticipation effect (continued)Main explanatory variable:Discontinuity at year 9Main explanatory variable:Donut hole at years 9-12Dependent variables: Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)Spouse’s labour supply decision in response to household maintainer’s receipt of OAS/GIS benefits(7) Multi-person family:Spouse < age 60 + kids ≥ age 250.017(0.067)0.066(0.126)0.099(0.078)0.138*(0.066)-0.179(0.158)0.451**(0.204)0.309***(0.091)0.317***(0.086)Number of observations 385 385 385 385 330 330 330 330(8) Multi-person family:Spouse < age 60 + No kids-0.099(0.113)0.112(0.122)-0.099(0.121)-0.043(0.118)-0.033(0.298)-0.027(0.291)0.147(0.245)0.201(0.213)Number of observations 525 525 525 525 435 435 435 435(9) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.146(0.117)0.099(0.103)0.044(0.048)0.019(0.044)-0.372***(0.035)0.313***(0.044)0.183***(0.038)0.151***(0.041)Number of observations 1,655 1,655 1,655 1,655 1,380 1,380 1,380 1,380(10) Multi-person family:Spouse ≥ age 60 + No kids-0.026(0.045)0.033(0.039)0.075***(0.024)0.093***(0.025)0.044(0.025)-0.026(0.028)0.053(0.041)0.046(0.045)Number of observations 5,390 5,390 5,390 5,390 4,440 4,440 4,440 4,440(continued on next page...)52Table 2.9: Effect of OAS/GIS on extensive margins of labour supply - Anticipation effect (continued)Main explanatory variable:Discontinuity at year 9Main explanatory variable:Donut hole at years 9-12Dependent variables: Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Indicatorfor em-ploymentIndicatorfor not inlabourforceIndicatorfor annualwage <$500Indicatorfor annualwage <$3,500Family types (1) (2) (3) (4) (5) (6) (7) (8)Working-age child’s labour supply decisions in response to household maintainer’s receipt of OAS/GIS benefits(11) Multi-person family:Spouse < age 60 + kids ≥ age 250.068(0.080)0.116(0.103)-0.095(0.110)-0.023(0.090)0.243(0.235)-0.261**(0.092)-0.249*(0.125)-0.187**(0.083)Number of observations 645 645 645 645 550 550 550 550(12) Multi-person family:Spouse ≥ age 60 + kids ≥ age 25-0.120*(0.057)0.097**(0.033)0.062(0.038)0.012(0.037)0.050(0.069)0.084(0.052)-0.013(0.087)-0.022(0.092)Number of observations 2,255 2,255 2,255 2,255 1,845 1,845 1,845 1,845This table shows the OLS regression results for year 2006. All regression models contain province dummies and demographic controls. Each row denotes the typeof family composition. The demographic controls include dummy variables controlling for age, gender, disability, education, ethnicity, language, dwelling tenurestatus, and other private pension sources. These models are weighted by individual Census weights. Standard errors are clustered by the number of years sinceimmigration and are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: Statistics Canada’s 2006 Census; and author’s calculations.53Table 2.10: Effect of OAS/GIS on extensive margins of labour supply - High housing asset versus low housing assetDependent variables: Indicator foremploymentIndicator for not inlabour forceIndicator for annualwages < $500Indicator for annualwages < $3,500Independent variables: (1) (2) (3) (4) (5) (6) (7) (8)(1) T REAT : discontinuity at year 11 -0.021(0.012)-0.129(0.123)0.022*(0.012)0.140(0.094)0.074***(0.021)0.218**(0.097)0.061*(0.024)0.237*(0.118)Low asset? [1] No Yes No Yes No Yes No YesNumber of observations 6,880 305 6,880 305 6,880 305 6,880 305(2) T REAT : discontinuity at year 9 -0.068*(0.036)-0.296***(0.068)0.073(0.043)0.259***(0.060)0.133***(0.026)0.305***(0.046)0.132***(0.022)0.383***(0.030)Low asset? [1] No Yes No Yes No Yes No YesNumber of observations 6,880 305 6,880 305 6,880 305 6,880 305This table shows the regression results for year 2006 using expression (2.1). The estimation includes all household maintainers of ages 65 and above. All regressionmodels contain province dummies and demographic controls. The demographic controls include dummy variables controlling for age, gender, disability, education,ethnicity, language, dwelling tenure status, and other private pension sources. These models are weighted by individual Census weights. Standard errors areclustered by the number of years since immigration and are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.[1] I use 0.5 of median housing value to identify the households with low housing asset.Sources: Statistics Canada’s 2006 Census; and author’s calculations.54Table 2.11: Effect of OAS/GIS on intensive margins of labour supply - Census dataThis table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.Dependent variables: Hours workedfor pay,employed onlyHours workedfor pay, allrespondentsIndicator for 0hours on unpaidhousework [1]Number ofweeks workedfor payFamily types (1) (2) (3) (4)Household maintainer’s labour supply decision in response to OAS/GIS benefits.(1) Live alone -15.052(11.526)-2.543***(0.708)-0.054**(0.021)1.620(2.924)Number of observations 185 2,945 2,945 180(2) Multi-person family:Spouse < age 60 + kids ≥ age 250.212(12.641)-14.363*(6.819)-0.129(0.117)8.205(8.281)Number of observations 85 385 385 80(3) Multi-person family:Spouse < age 60 + No kids3.676(6.041)-3.994*(1.967)0.030(0.082)3.279(6.033)Number of observations 130 525 525 125(4) Multi-person family:Spouse ≥ age 60 + kids ≥ age 254.482(6.254)-1.312(2.125)0.031(0.030)6.182(6.294)Number of observations 180 1,655 1,655 170(5) Multi-person family:Spouse ≥ age 60 + No kids-0.143(5.778)-0.857(0.962)0.030(0.024)0.383(2.522)Number of observations 495 5,390 5,390 470Spouse’s labour supply decision in response to household maintainer’s receipt of OAS/GIS benefits(6) Multi-person family:Spouse < age 60 + kids ≥ age 25-3.558(9.712)-2.458(3.530)0.029(0.076)9.767(11.953)Number of observations 125 385 385 125(7) Multi-person family:Spouse < age 60 + No kids0.325(6.153)-5.137(3.275)-0.021(0.036)-3.043(4.972)Number of observations 220 525 525 210(8) Multi-person family:Spouse ≥ age 60 + kids ≥ age 254.351(7.543)0.970(1.800)0.040(0.046)-6.269(4.094)Number of observations 185 1,655 1,655 175(9) Multi-person family:Spouse ≥ age 60 + No kids-6.864(4.118)0.242(0.495)-0.003(0.015)-1.067(2.235)Number of observations 425 5,390 5,390 400(continued on next page...)55Table 2.11: Effect of OAS/GIS on intensive margins of labour supply - Census data (continued)This table reports the coefficients for the main explanatory variable T REAT , with the discontinuity at year 11.Dependent variables: Hours workedfor pay,employed onlyHours workedfor pay, allrespondentsIndicator for 0hours on unpaidhousework [1]Number ofweeks workedfor payFamily types (1) (2) (3) (4)Working-age child’s labour supply decisions in response to household maintainer’s receipt ofOAS/GIS benefits(10) Multi-person family:Spouse < age 60 + kids ≥ age 25-2.136(3.242)1.796(1.981)-0.045(0.064)-3.771*(1.942)Number of observations 455 645 645 430(11) Multi-person family:Spouse ≥ age 60 + kids ≥ age 252.337(1.831)-1.077(1.280)0.005(0.017)0.380(2.097)Number of observations 1,705 2,255 2,255 1,660This table shows the OLS regression results for year 2006 using expression (2.1). All regression models containprovince dummies and demographic controls. Each row denotes the type of family composition. The demographiccontrols include dummy variables controlling for age, gender, disability, education, ethnicity, language, dwellingtenure status, and other private pension sources. These models are weighted by individual Census weights. Standarderrors are clustered by the number of years since immigration and are in parentheses. *** Significant at 1%; **significant at 5%; * significant at 10%.[1]: Includes all respondents.Sources: Statistics Canada’s 2006 Census; and author’s calculations.56Table 2.12: Effect of OAS/GIS on intensive margins of labour supply - GSS dataThis table reports the coefficient θDD.Samples: 0-20 years ofimmigrationAll years ofimmigrationAll years ofimmigrationAll respondents All respondents Employedindividuals onlyDependent variables:Activities (total duration in minutes)(1) (2) (3)(1) Paid work -73.040**(36.365)-33.469(23.826)-349.950***(69.392)(2) Activities related to paid work 0.302(3.590)0.568(2.138)-0.506(2.271)(3) Cooking and washing up 5.326(30.350)-5.673(28.996)10.456(7.488)(4) Housekeeping -2.180(20.181)6.903(18.476)47.827***(14.089)(5) Maintenance and repair 7.188(9.448)5.499(3.959)5.754(9.333)(6) Other household work -19.716(17.383)-1.211(15.442)24.419***(7.906)(7) Shopping for goods and services -12.378(26.824)6.864(23.676)47.678***(15.272)(8) Child care 15.224*(8.748)12.574**(5.942)-4.354(9.934)(9) Civic and voluntary activity 1.502(10.877)8.262(5.626)16.614(11.105)(10) Education-related activity 27.022**(12.070)30.417***(8.268)14.429**(6.808)(11) Meals (excluding restaurants) 44.627***(13.632)34.251***(11.446)17.724*(9.983)(12) Other personal activities -9.702(34.341)-29.492(27.996)-33.134(26.942)(13) Restaurant meals 18.984(15.534)3.764(4.719)9.241(6.819)(14) Socializing in homes -70.019*(39.895)-69.240*(35.881)-10.308(16.045)(15) Other socializing activities -10.922(9.477)-2.264(9.166)8.000(7.974)(continued on next page...)57Table 2.12: Effect of OAS/GIS on intensive margins of labour supply - GSS data (continued)This table reports the coefficient θDD.Samples: 0-20 years ofimmigrationAll years ofimmigrationAll years ofimmigrationAll respondents All respondents Employedindividuals onlyDependent variables:Activities (total duration in minutes)(1) (2) (3)(16) Watching TV 33.921(36.362)4.718(17.119)48.476*(26.644)(17) Reading books, newspaper -30.746(20.164)-15.755(15.576)28.502***(8.103)(18) Other passive leisure 12.847(8.890)6.456(4.061)1.537(3.073)(19) Sports, movies & other 8.565(5.757)1.489(2.868)-3.977(6.567)(20) Active sports -17.527(16.055)-21.485(14.101)1.210(11.120)(21) Other active leisure 11.830(16.100)25.929**(12.296)59.441*(30.436)(22) Night sleep / essential sleep 91.346(66.239)61.113***(22.988)150.741(105.701)Number of observations 1,658 5,177 2,990This table shows the OLS regression results for years 1998, 2005, and 2010, where the number of minutes spent peractivity is the dependent variable. See expression (2.2) for the construction of these regressions. The main coefficientof interest is θDD, which compares the time use of immigrants of ages 65 and over who landed for more than 10 yearsversus those who don’t in year 2006, and then compare the result to the same difference for immigrants of ages 25-54. All regression models contain province and year dummies. The demographic controls include dummy variablescontrolling for gender, disability, education, language, and dwelling tenure status. These models are weighted byperson-level weights. Standard errors are in parentheses and are corrected for heteroskedasticity. *** Significant at1%; ** significant at 5%; * significant at 10%. See Appendix A for the components that make up the above-listed timeuse categories.Sources: Statistics Canada’s 1998, 2005, and 2010 General Social Survey; and author’s calculations.58Chapter 3Immigration and housing for thenear-retirees3.1 IntroductionIn recent years, the major Canadian cities have experienced unprecedented housing cost growth. Some of themedia as well as policymakers have associated the Canadian housing boom with the increasing number ofrich immigrants who arrived with foreign money under the now-terminated Federal Immigrant Investor andEntrepreneurs Program and the still-running Quebec Immigrant Investor Program.40 Economic reports byfinancial institutions have highlighted the chance of major labour out-migration as housing in metropolitanareas becomes less affordable.41 Academic research, such as Saiz and Wachter (2011), has instead suggestedthat the distaste for immigrants of low socioeconomic status and of visible minority groups is associated withnative flight.42To date, most academic research on this topic has focused on general populations and has not accountedfor other possibilities to explain native out-migration. None of the past studies on native mobility hasseparated out the estimation by dwelling tenure types and by age groups. By first impression, the renters’relocation decisions are expected to be different from the homeowners’. The amount of housing wealth afamily owns could be a potential determinant in predicting native out-migration decisions in response to animmigration shock. On one hand, renters’ mobility preferences would be more directly related to housingaffordability as they do not own any housing asset. On the other hand, for families with large amounts ofhousing wealth, capital gains made from selling the current property in the original place of residence couldmotivate them to relocate to less expensive communities. Yet, native homeowners could instead choose tostay in the same dwelling by extracting the extra housing equity obtained from an immigration shock throughborrowing against their housing collateral. Therefore, the net effect of immigration on native out-migrationis ambiguous.In this case, the near-retirees represent an interesting economic case for examining the interactionsbetween the labour and housing markets. This subpopulation tends to be asset-rich and more vulnerable tonegative health shocks. Between the years 1997-2009, approximately 75 percent of the near-retirees owned adwelling.43 Relative to the whole working age population, Canadian near-retirees’ labour force participationrate seems to be much more sensitive to immigration shocks (see Figures 3.1 and 3.2). Although the average40See for example, Conservative Party of Canada (2015), Gold (2015), Lee (2015), and Young (2015).41See Vancity (2015) for details.42See also Accetturo et al. (2014) for Italy, Akbari and Aydede (2012) for Canada, and Sá (2014) for the U.K.43This number is derived from Statistics Canada’s Survey of Financial Security and Survey of Household Spending datasets.59mortgage debt for primary residence is lower for the elderly households, their holdings on other mortgagesand on lines of credit are greater than that for the younger families.44 In fact, the proportion of home equityloans held by near-retirees has risen significantly since the late 1990s and tends to be positively correlatedwith the share of immigrant settlement. These findings point to the possibility that older households mayturn to mortgage refinancing or home equity lines of credit instead of selling off their primary residences toextract housing equity that arises from immigration shocks. Therefore, incorporating the near-retirees intothe immigration and housing context provides additional insights to how older native households use theirhousing assets and indicates alternative causes behind native out-migration. This is the first paper to explorethis mechanism.This study compares the short-term impact of immigration on housing cost growth and on mobilitydecisions for the rental and owner-occupied dwelling markets using data from the 1986-2006 CanadianCensuses and from the Statistics Canada’s 2011 National Household Survey. I extend Saiz’s (2007) and Sá’s(2014) theoretical framework and apply various econometric techniques to explore this research question.In this study, I also compare the relocation preferences of near-retirees (ages 55-64) and of the working-age population (ages 25-54) to explore whether mobility proclivities vary across the age groups. I usethe near-retirees to proxy the homeowners with low amounts of residential mortgage debt (i.e. sum of allmortgages) and with high amounts of home equity lines of credit; and the working-age population to proxythe homeowners with high amounts of outstanding mortgage balance and with low amounts of home equitylines of credit. Furthermore, this paper uses an ordered logistic regression with moving distance categoriesas the dependent variable to determine where native households move to in response to an immigrationshock. Finally, I conduct a synthetic panel analysis to examine whether the elderly homeowners who stayin the same municipality exhibit any housing asset downsizing in response to an immigration shock. Thesecomparisons help identify whether net housing wealth could influence overall mobility decisions.Theoretically, I extend Saiz’s (2007) and Sá’s (2014) model to include net housing wealth into thehomeowner’s budget constraint. Both the original model and my extended model suggest that immigrationleads to a rise in all types of housing cost growth as long as immigration exerts a positive effect on nativewages and the city experiences inelastic housing supply. In Sá’s (2014) model, native flight is positivelyassociated with an increase in housing cost growth, native’s distaste towards immigration, and a negativeeffect from immigration on native wages. However, my extended model shows that net housing wealthalso drives native homeowners’ mobility decisions. This implies that renters and homeowners’ relocationpreferences would differ only if the homeowners hold a low amount of residential mortgage debt. Therefore,the net effect remains an empirical question.To examine this mechanism empirically at the municipality level within metropolitan areas, I use an44The 1999 and 2005 Statistics Canada Survey of Financial Security datasets present the following average debt holding figures,which show that younger families tend to hold higher amounts of mortgage debt on primary residence relative to near-retirees.The average value of mortgage debt on primary residence was $81,210.52 in 1999 and $93,324.41 in 2005 for respondents of ages55-64; and $103,226.00 in 1999 and $129,613.70 in 2005 for respondents of ages 25-54. The mean amount for other mortgagedebt holdings was $114,210.60 in 1999 and $166,202.20 in 2005 for respondents of ages 55-64; and $108,777.80 in 1999 and$165,939.30 in 2005 for respondents of ages 25-54. The average value of lines of credit was $21,591.70 in 1999 and $27,901.57 in2005 for respondents of ages 55-64; and $16,207.54 in 1999 and $23,374.09 in 2005 for respondents of ages 25-54. These holdingsexclude the respondents who reported zero debt amounts.60instrumental variable (IV) strategy that is based on historical ethnic distributions (enclave approach). ThisIV technique predicts the immigration flow value based on the initial geographical settlement patterns ofimmigrants, which exploits the fact that immigrants tend to move to locations with strong social networks.As part of the robustness checks, I include two additional instrumental variables to test the validity of theresults through two different over-identification tests. The first over-identification set includes the enclaveIV and the Gravity IV. The Gravity IV follows from Saiz and Wachter’s (2011) approach, which assumesthat neighbourhoods that are geographically close to existing immigrant enclaves have a higher probabilityof becoming immigrant areas in the future. The second over-identification set includes the enclave IV andthe Airport IV. The Airport IV makes use of the pattern that immigrant density tends to be high in censussubdivisions that are near the international airport.To summarize my findings, both new and established immigrants lead to rises in both average rental costand property value in all markets. However, housing cost growth is slower in smaller markets and/or lessimmigrant-dense municipalities. These findings coincide with the model predictions that show locationswith inelastic housing supply experience the largest increase in housing cost in response to an immigra-tion shock. A combination of results points to the possibility that in addition to a taste channel, housingaffordability and household finance could influence mobility decisions. Various stylized facts suggest thata growing number of older American and Canadian households has accessed home equity borrowing. In-creasing holdings of home equity borrowing could possibly explain the overall slow rate of out-migration byelderly native homeowners. In addition, synthetic cohort analysis shows that in response to an immigrationshock, the elderly homeowners who stay in the same neighbourhood do not exhibit any form of housingasset downsizing (i.e. sell high and then buy low). Therefore, there is insufficient evidence to conclude thatnear-retirees extract housing equity by relocating.This study makes several important contributions. To the best of my knowledge, this is the first paper toseparate the analysis by dwelling tenure types and by age groups when investigating the impacts immigra-tion exerts on older native mobility decisions. The research findings in this paper push this area of literatureforward by suggesting an alternative perspective for explaining native out-migration. The mobility regres-sions illustrate that relative to renters, homeowners are more likely to show distaste towards immigration.This implies that distaste may not completely address the reasons behind native out-migration, at least forthe Canadian context. The heterogeneity in mobility preferences across dwelling tenure groups is an impor-tant finding because it may explain why Card (2001) fails to find any significant effect from immigration onaggregate native relocation decisions.Moreover, this paper focuses on short-term effects from immigration, which deviates from existingresearch. U.S. literature, such as Card and DiNardo (2000), Saiz (2007), and Saiz and Wachter (2011), hasused the decennial census to explore the linkages between immigration, housing, and native flight, whichtends to miss the more immediate impacts. Conversely, European papers, such as Accetturo et al. (2014) andSá (2014), have used annual data for their analyses. However, annual data may not fully capture the gradualimpact immigration exerts on the housing market, since international migrants take roughly four years totransition from being a renter to homeowner and not all immigrants stay in the province to which they wereinitially destined (Haan, 2012; Pandey and Townsend, 2011). Therefore, this study uses the quinquennial61Canadian Census and 2011 National Household Survey data to better identify the shorter-term effects arisingfrom immigration.Section 3.2 reviews the literature that examines the effect of immigration on housing and on nativemobility. In Section 3.3, I present the theoretical framework and in Section 3.4, I discuss the data. I describethe econometric analyses in Section 3.5 and discuss the main results in Section 3.6. Finally, Section 3.7concludes.3.2 Contexts and contributionsPrevious research has explored the effect of immigration on housing costs. For the U.S. housing market,Ottaviano and Peri (2006) and Saiz (2003, 2007) find a strong, positive association between immigration andhousing costs. A further investigation on local residential dynamics in Saiz and Wachter (2011) shows thathousing values grow slowly in neighborhoods with an increasing immigrant density, with stronger impact inneighborhoods where the population was initially, predominantly occupied by wealthy caucasions. Some ofthe reasons behind the negative impact include (1) natives moving out in response to the immigration flow;and (2) immigrants tend to be of low socioeconomic status. On the other hand, Card (2001) and Card andDiNardo (2000) do not find any significant relationship between immigration and native mobility. There isinsufficient evidence to conclude that immigration made the native-born population worse.For other countries, Gonzalez and Ortega (2013) show that immigration between the years 2000-2010led to an increase in house prices and the number of housing units in Spain. However, Aldén et al. (2015) andSá’s (2014) findings for Sweden and for the U.K., respectively, are in line with Saiz and Wachter’s (2011).Both European papers illustrate that natives with high educational attainment and high wages are morelikely to out-migrate in response to an immigration shock. Specifically, low-skilled immigrants exert thestrongest negative impact on U.K. housing prices (Sá, 2014). For Italy, Accetturo et al. (2014) and Mocettiand Porello’s (2010) results are opposite to Sá’s (2014). Accetturo et al. (2014) find that native flight is morepronounced in poorer communities. Mocetti and Porello (2010) show that immigration displaces low-skillednative workers and encourages inflows of highly-educated natives. Stillman and Maré (2007) also find skillcomplementarities in New Zealand, where on net, immigration does not seem to displace the native-born.Therefore, the overall effect of immigration on housing and native mobility remains unclear.For Canada, relatively little research has examined the relationship between housing, native mobility,and immigration. More recent work includes Akbari and Aydede (2012), Latif (2015), and Li (2014). Latif(2015) and Li (2014) find a positive relationship between housing cost and immigration. On the other hand,Akbari and Aydede (2012) show a small effect of immigration on Canadian house prices. They argue thatnative out-migration may exert downward pressure on house prices. However, their estimation techniquealso raises some questions. These include potential endogeneity problems with using a mobility variable aspart of the control, collinearity issues with both the unemployment rate and labour force participation ratein the same regression model, and omitted variables relating to amenities. Other research also shows thatnew immigrants generally impose greater demand for housing. Mendez et al. (2006) and Haan (2012) findmost immigrant groups transitioned from being a renter to homeowners within four years time. Yet, roughly62a quarter of the newcomer tenants face financial stress and overcrowding problems.To summarize, existing work has not painted a very clear picture of the effect of immigration on thehousing market and on native relocation preferences. Specifically, none of these studies on native mobilityseparates out the estimation by dwelling tenure types, where native homeowners and renters are expected torespond differently to an immigration shock. Therefore, home equity extraction-related activities could haveconfounded the impact of immigration on native homeowners’ mobility intentions. Previous work couldinaccurately over-attribute amenity effects as the primary cause for native out-migration. For example, Sá(2014), Saiz and Wachter (2011), and Accetturo et al. (2014) suggest that reduced desirability of the citydue to growing immigrant density leads to native flight.Moreover, to date, most research on immigration has focused on general populations. Borjas (2008) isan exception, where he finds immigration to exert a depressing effect on the elderly native workers’ wagesand to lead to increases in retirement. However, Borjas (2008) does not account for the possibility that thenewcomers may drive housing costs, thereby could influence labour mobility decisions and wages. Theseinteractions have implications on the linkages between the labour and housing markets, which have beenoverlooked in the existing literature.3.3 Theoretical modelI extend Saiz’s (2007) and Sá’s (2014) theoretical models by introducing housing wealth effects to examinethe linkages between immigration, native mobility, and housing cost. This extension is necessary because theamount of housing wealth a family owns could be a key factor in predicting native out-migration decisions.For households with large amounts of housing wealth, capital gains made from selling the current propertyin the original place of residence could motivate them to relocate to less expensive communities. Yet, thesehouseholds could also choose to stay in the same dwelling by extracting the extra housing equity obtainedfrom an immigration shock through mortgage refinancing and/or via home equity lines of credit. On thecontrary, native renters’ relocation decisions are not confounded by housing wealth effects and may chooseto out-migrate if housing in the original place of residence becomes unaffordable. Therefore, mobilitypreferences in response to an immigration shock should be different across dwelling tenure groups.I begin with the model for renters, by considering Saiz (2007) and Sá’s (2014) frameworks without anyhousing wealth effect. I then extend their models for homeowners by including a housing wealth effect inthe budget constraint. By applying this extension, I show that increases in net housing equity do speed upthe rate of native out-migration.3.3.1 Model for rentersI start by considering Saiz (2007) and Sá’s (2014) frameworks with two types of workers in city c: natives(N) and immigrants (I). Unlike Sá (2014), I assume for simplicity that all native workers are homogeneous.Immigrants prefer city c and the supply of immigrants I is assumed to be exogenous.The preference for native renter i is:63Uic = Vic︸︷︷︸=Aic−γN+hαx(1−α)−δ I (3.1)Vic is the value of local amenities in city c for individual i; h is the amount of housing units consumed bythe renter; α is the elasticity of the demand for housing; x is the amount of non-housing consumption; andδ I captures the preference of natives for immigration. Natives show distaste towards immigrants if δ > 0.Following Saiz (2007) and Sá (2014), I set Vic = Aic− γN, where the preferences for local amenities willdecline at the rate of γ as more natives N are in city c.Similarly, the preference for immigrant renter i is:Uic = Vic︸︷︷︸=Aic−γI+hαi x(1−α)i (3.2)With R as the rental cost, w as the individual labour income, and y as the individual non-labour income,renters maximize the utility function subject to the following budget constraint:Rhi+ xi = wi+ yi (3.3)The utility maximization problem provides the following demand functions for housing consumption:hi = αwi+ yiR(3.4)The total housing demand for city c equals the sum of the housing consumption for the native andimmigrant households:HD =αR[W N +W I +Y N +Y I](3.5)Let W i and Y i be the aggregate labour and non-labour income, respectively, for agent of type i. Iassume that W N 6=W I because the skill sets for immigrants and natives are different. Specifically, wages fornative-born can be expressed as W N = Wˆ −ρI, where ρ > 0 means that immigrants are substitutes to nativeworkers.Therefore, the total housing demand after taking logarithms is:ln(HD) = ln(α)− ln(R)+ ln(N · [Wˆ −ρI]+ I ·W I +NY N + IY I) (3.6)I follow Saiz (2007) and Sá (2014) and adopt the following specification for housing supply in city c:ln(Hs) = β0+β1ln(R) (3.7)where R is the rental cost; β0 captures the construction cost; and Hs is the number of rental units available incity c. Each city has a distinct housing supply due to geographical and regulatory constraints and the term,64β1 characterizes this elasticity.45 A low β1 term implies that city c faces geographical and/or land regulationconstraints, which create challenges to supplying more rental units. Therefore, rental cost will rise with fewvacancies. On the other hand, a city faces high rental vacancy rate if β1 is large (elastic).In equilibrium, housing supply equals housing demand. Therefore, expressions (3.6) and (3.7) providethe following expression for rental cost:ln(R) =(11+β1)· [ln(α)−β0+ ln(N · [Wˆ −ρI]+ I ·W I +NY N + IY I)] (3.8)Therefore, the rental cost growth as a function of immigration can be expressed as:dRdI= R ·(11+β1)·(1N · [W N +Y N ]+ I · [W I +Y I])·[dNdI· (W N +Y N)+(W I +Y I)−Nρ](3.9)The key point of expression (3.9) is that rental cost will rise in response to an immigration shock if cityc experiences: (1) a positive inflow of natives (i.e. dNdI > 0) and (2) a positive effect from immigration onnative wages or labour complementarity (i.e. ρ < 0). Furthermore, an elastic housing supply (i.e. high β1)will slow the impact of immigration on rental cost growth. Therefore, it is expected that major markets withinelastic housing supplies, such as Vancouver and Toronto metropolitan areas, will experience the biggestjump in housing cost growth in response to immigration shocks.In order to find dNdI , I assume that the marginal native renter is indifferent between staying and leavingcity c. The overall utility outside of city c is U¯ . Therefore, by combining expressions (3.1) and (3.4), theindirect utility function can be set as follows for the marginal native renter:Aic− γN+αα(1−α)(1−α)(W N +Y N)R−α −δ I = U¯ (3.10)By re-arranging expression (3.10), the number of native renters in city c is:N =1γ·[A−U¯ +αα(1−α)(1−α)(W N +Y N)R−α −δ I](3.11)Therefore, the mobility response of native renters in response to an immigration shock is:dNdI=−1γ[αα(1−α)(1−α)R−αρ+α(1+α)(1−α)(1−α)(W N +Y N)R−1−α · dRdI+δ](3.12)Expression (3.12) illustrates the critical point that native renters will out-migrate to another location inresponse to an immigration shock if rental costs ( dRdI ) rise as an influx of immigrants enters city c, in additionto reasons relating to discontent towards immigrants. The rate of out-migration will slow if total incomeis low. However, the effect of labour substitution (ρ) on native mobility is ambiguous. On one hand, theeffect of immigration on rental cost growth is negative if ρ > 0, which would lower the rate of native out-migration. On the other hand, based on expression (3.12), dNdI < 0 if ρ > 0, holding everything else constant.45See for example, Saiz (2010)65Therefore, the net effect remains an empirical question.3.3.2 Model for homeownersThe frameworks presented in Saiz (2007) and Sá (2014) do not consider the possibility that housing wealthcould influence native out-migration decisions. To account for this dynamic, I extend the basic frameworkfor homeowners by introducing a new budget constraint, where I assume that households choose to first sellthe current property and then acquire a new house:ψPhi+(µ+φb)Pκhi+ xi = wi+ yi+(1−φs)Phi−mPhi (3.13)On the left-hand side of the new budget constraint (expression (3.13)), the term ψPhi represents themaintenance cost for the current residence hi. The term (µ +φb)Pκhi denotes the total initial expenditurearising from purchasing a new home of size κhi. The newly purchased unit is a fraction κ of the currentproperty hi, where κ can hold any rational values. For example, if the homeowner decides to downsize thehousing asset by selling a single-detached home and moving to an apartment unit, then 0 < κ < 1. On theother hand, κ > 1 if the household purchases a bigger house. The initial expenditure consists of two parts:(1) µPκhi covers the downpayment; and (2) φbPκhi is the proportional transaction cost associated withpurchasing the new home. Relative to the renter’s model, the right-hand side budget constraint in this caseincludes an extra term, (1− φs)Phi−mPhi. This covers the net revenue from selling the current property,where φsPhi is the proportional transaction cost and mPhi is the outstanding mortgage balance at the time ofselling the house.The utility maximization problem yields the following housing demand equation for the native home-owners’ current residence:hi = αwi+ yiP[ψ+µ+φb+m− (1−φs)] (3.14)Unlike the renter’s expression, in addition to income (wi + yi), the homeowners’ housing demand forthe current residence also relies on the term, P[ψ + µ +φb +m− (1−φs)]. For simplicity, I set ζ = [ψ +µ+φb+m− (1−φs)], where Pζ equals the sum of the unit cost from maintaining and purchasing a house,plus mortgages, less the revenue from selling the current property. In other words, −Pζ represents the nethousing wealth component. The main point of expression (3.14) is that homeowners will reduce the amountof current housing consumption if net housing wealth is high (i.e. Pζ →−∞). This implies that homeownerswill sell off their existing property if they could extract housing equity, which means that h < 0 and Pζ < 0.For the homeowners’ model, the total housing demand is:ln(HD) = ln(α)− ln(P)− ln(ζ )+ ln(N · [Wˆ −ρI]+ I ·W I +N ·Y N + I ·Y I) (3.15)The total housing supply is:ln(Hs) = β0+β1ln(P), (3.16)66In this case, P is the value of dwelling; β0 captures the construction cost; β1 reflects the housing supplyelasticity in the owned-dwelling market; and Hs is the number of owned-dwelling units available in city c.Setting expressions (3.15) and (3.16) equal and totally differentiating gives:dPdI= P ·(11+β1)·(1N · [W N +Y N ]+ I · [W I +Y I])·[dNdI· (W N +Y N)+(W I +Y I)−Nρ](3.17)Expression (3.17) shows that the factors contributing to the growth of property value in response toan immigration shock is the same as those pushing rental cost growth. However, Expression (3.18) illus-trates that native homeowners’ mobility decisions are different from native renters’, where homeowners’preferences now depend on net housing wealth (−Pζ ):dNdI=−1γ[αα(1−α)(1−α)(Pζ )−αρ+α(1+α)(1−α)(1−α)(W N +Y N)(Pζ )−1−α · dPdI+δ](3.18)By taking the partial derivative of dNdI with respect to Pζ :∂ dNdI∂Pζ=1γ[α1+α(1−α)(1−α)(Pζ )−α−1ρ+α(1+α)(1−α)(1−α)(1+α)(W N +Y N)(Pζ )−2−α · dPdI](3.19)As noted above, Pζ < 0 if the homeowner can profit from an extra housing windfall due to an immigra-tion shock (i.e. sell high and then buy low). Therefore, holding all other variables constant, expression (3.19)suggests that ∂dNdI∂Pζ > 0. This means that the rate of out-migration by native homeowners will accelerate ifthe amount of housing equity grows. Yet, Pζ is positively-associated with mortgage debt m. This meansthat homeowners with large amounts of mortgage debt will lower their intention to out-migrate. There-fore, expression (3.19) implies that mortgage refinancing and/or home equity lines of credit could act as acounter-force towards native relocation.To summarize, the theoretical model results show that native homeowners’ mobility proclivities couldbe different from native renters’ due to net housing wealth impacts. Therefore, it is essential empirically tounbundle the analysis by dwelling tenure groups in order to accurately narrow down the reason behind nativeflight. This approach implicitly assumes that the selection to be in or out of the owner-occupied dwellingmarket is exogenous. This assumption may not necessarily hold if the switch in dwelling tenure statusis driven by overall local market conditions. In subsequent sections, I describe an instrumental variablestrategy that I apply to address this potential selection bias.3.4 Data descriptionThis paper uses the restricted version of the 1986 - 2006 Canadian Census datasets and the 2011 StatisticsCanada’s National Household Survey (NHS) to examine the linkages between immigration, housing, and67native mobility at the municipality level.46 By setting the analysis at this geographical level, the results areless likely to be confounded by job search and wage bargaining costs, since native workers could keep thesame job within the same metropolitan area (commuting zone). Statistics Canada applies the term, “censussubdivision (CSD)” to define a municipality, which is the finest geographical level available for analyzinghouseholds’ mobility decisions between the Census years. The census subdivision geographic definitionschange quinquennially. I utilize a combination of information retrieved from the Postal Code ConversionFile (PCCF) datasets; from Statistics Canada’s Standard Geographical Classification Concordance Tables;from Commission de toponymie du Québec; and from land area information obtained from the Universityof Toronto’s Census Analyzer database to form consistent CSD categories across the years. Unlike Saiz andWachter (2011), who only concentrate on neighborhoods with more than five percent change in foreign-bornpopulation, I focus on all census subdivisions within a census metropolitan area for my analysis. The targetpopulation for the native mobility estimations includes all households living in private dwellings.To estimate immigration’s impact on housing decisions, the immigration rate variable is computed asthe difference in the weighted count of immigrants between the consecutive Census years normalized bythe current population. For the enclave instrumental variable approach (described in detail below), I useplace of birth information taken from the 1986 Census to construct the historical ethnic distributions.47Except for the province of Alberta, housing cost data taken from the Canadian Census and the NationalHousehold Survey are generally comparable to those published by Teranet-National Bank and by CanadaMortgage and Housing Corporation (CMHC).48,49 For the housing cost regressions, which are estimated atthe municipality level, this paper normalizes the housing rent and housing value variables by the number ofrooms to determine whether immigration exerts differential impact on the Canadian housing market.50 Thisstudy uses the provincial Consumer Price Index for all items to convert the housing cost numbers to realterms.51 Furthermore, I create binary and categorical variables for out-migration based on the household’s46The research and analysis in this chapter are based on data from Statistics Canada and the opinions expressed do not representthe views of Statistics Canada. The public use microdata file version (PUMF version) of the Census and the NHS do not provideany information at the municipality level. Starting from the 1986 Census, the value of dwelling and the gross rent variables do notinclude reserve dwellings. Therefore, to ensure that the housing-related variables are comparable across time, I exclude the 1981Census from this study (see Statistics Canada (2010) for more details).47This study uses the “Immigrant Status” and the “Year of Immigration” variables to construct the immigration rates; and the“Country of Birth” variable from the year 1986 Census to construct the ethnic distributions.48See the Online Supplementary Appendix. The Teranet-National Bank House Price Index and CMHC’s average rental valuesmay under-estimate Alberta’s housing cost growth, as the province experienced strong interprovincial and international migrationflows starting in the 2000’s. In particular, international migration flows are concentrated in Calgary and Edmonton, with low rentalvacancies and housing inventories around the 2005-2007 period (Government of Alberta, 2006, 2007, 2009). The Teranet-NationalBank House Price Index (HPI) is estimated using the repeated sales methodology, which tracks properties with at least two sales.However, the estimation process for this HPI does not account for the physical characteristics of the property (i.e. renovations),non-arms-length transactions, and high turnover frequency; and cannot account for new residential constructions and housing startsin response to a large influx of in-migrants (Teranet-National Bank, 2015). The average rent data are based on the CMHC’sRental Market Survey, which only targets privately initiated structures with at least three rental units. This methodology overlooksrentals made through sub-leases and from single-detached residential structures, which are alternative leasing choices especially inresponse to low rental vacancies and housing inventories (Canada Mortgage and Housing Corporation, 2014).49I use the “Value of Dwelling” and “Gross Rent” variables to define housing costs. Statistics Canada describes the “Value ofDwelling” variable as the dollar amount expected by the owner if the dwelling were to be sold during the reference year; and the“Gross Rent” variable as the tenant’s total average monthly payment to secure shelter. The tenant’s total average monthly paymentcovers payments for electricity, oil, gas, coal, wood, or other fuels, water and other municipal services, and cash rent.50Census data do not provide any information on the square footage of the dwelling unit.51The estimation results are very similar if I use the provincial Consumer Price Index for shelter to deflate the housing cost68place of residence information for five years ago and for the contemporaneous period in order to estimatethe native mobility regressions at the household level.52 For constructing the instrumental variables for therobustness checks, I use the “geosphere” package from the program R and the geocodes retrieved from thePostal Code Conversion File datasets to compute the air distances between the centres of the origin anddestination municipalities.3.5 MethodologyThis study uses various econometric techniques to explore the linkages between immigration, housing cost,and native mobility within metropolitan areas. I define metropolitan areas as locations with a population ofat least 10,000. For the Canadian context, in this study, I include both census agglomerations and censusmetropolitan areas into the analysis, and I use the acronym “MA” to broadly cover all metropolitan areas.Existing literature has attributed distaste towards immigration as the primary force behind native flight.As illustrated by the extended theoretical framework, other reasons, such as total household income, housingaffordability, and labour substitution, could drive relocation decisions. The model predicts that housing costgrowth is positively related to native renters and homeowners’ mobility preferences. Therefore, I start theeconometric analysis by comparing the impact immigration exerts on housing value growth and gross rentgrowth within metropolitan areas. I also unbundle the immigration rate term by place of birth categories todetermine which ethnic group drives up housing cost growth.This paper then investigates native out-migration by separating the estimation by dwelling tenure groups.I focus on two main econometric specifications for the mobility regressions. For the first specification, Iregress the binary variable for change in municipality of residence onto the immigration rate. For the sec-ond mobility specification, I replace the binary variable for mobility with a categorical variable for movingdistance to investigate whether natives prefer to move to adjacent census subdivisions with lower hous-ing costs. Unlike the housing cost regressions, the two native out-migration models are indexed to thehousehold-municipality-time level. This chapter also conducts additional estimations to gather more evi-dence on the potential contributors behind native flight. These estimations include restricting the sample tohigh income households to strip out the effects from financial and spatial constraints; and unbundling theimmigration rate term by source country groups to explore whether native households show any distastetowards immigration. Furthermore, this study estimates the relationship between native wage growth andimmigration to explore whether labour substitution drives native flight. Finally, I construct a synthetic panelfor household maintainers born between the years 1924 and 1953 to examine whether households who re-ported staying in the same municipality with a different dwelling downsized their house in response to animmigration shock.53 I index the synthetic cohort estimation to be at the municipality-year-cohort level.The main econometric analyses apply the historical ethnic distributions (enclave approach) to instrumentvariables.52I use the “Mobility 5: Census subdivision of residence 5 years ago” and the “Census subdivision of current residence” variablestaken from the Census and from the NHS.53The Health and Retirement Study, which is a longitudinal survey for the elderly in the United States, also focuses on the cohortsborn between years 1924 and 1953 (except for Versions M and onward, which start to add Mid Baby Boomers born between years1954 and 1959).69for the immigration rate terms. This technique predicts the immigration rate based on the initial geographicalsettlement patterns of immigrants, which exploits the fact that newcomers tend to reside in locations withstrong social networks. To test the robustness of the estimates, I run three checks. The first robustnesscheck involves including census subdivision dummies instead of metropolitan area dummies to capturelocal amenity effects. I then use two different over-identification specifications as the second and thirdrobustness checks to test the validity of results. The first IV set includes the enclave and the Gravity IV; andthe second IV set includes the enclave and the Airport IV. The Gravity IV follows from Saiz and Wachter’s(2011) approach, which assumes that neighbourhoods that are geographically close to existing immigrantenclaves have a higher probability of becoming immigrant areas in the future. The Airport IV makes useof the pattern that immigrant density tends to be strong in municipalities (census subdivisions) that are nearthe major international airports (see for example, Figures 3.3 to 3.5). I describe these three instrumentalvariables in detail in Section 3.5.4.For all regression specifications estimated at the municipality level, I present the results for: (1) first-tierimmigration markets, which include the census subdivisions within the Vancouver, Montreal, and Torontocensus metropolitan areas; and (2) all census subdivisions within metropolitan areas.54 This breakdownprovides a clearer indication on whether international migration flows only influence certain markets. I alsodecompose the immigration rate variable by the timing of arrival (i.e. by year since immigration), in order toexamine which type of in-migrants exerts the biggest impact on the Canadian housing and labour markets.The empirical analyses exclude the municipalities with more than 50% of the population living in Indianreserves and with a population less than 100.This paper sets years 1991-2006 as the main estimation period, where I use the Canadian Census datafor the main analysis. The voluntary nature of the 2011 National Household Survey raises concerns over thevalidity of the immigration and place of origin numbers.55 Therefore, I extend the regression model to year2011 using the National Household Survey as part of the robustness analysis.3.5.1 Econometric specification for housing costAs noted in Section 3.4, I use the percentage change in gross rent and the percentage change in housing valueas the dependent variable for two different sets of regression models. Both sets of findings provide insightfulperspectives about the rental and housing sale markets. The model with the gross rent numbers better reflectsactivities of local residents and newly-arrived immigrants who may not choose to purchase a dwelling unitimmediately upon arrival. On the other hand, the housing value specification gives some additional insightsinto how native households may be using their housing assets in response to an immigration shock.I use expression (3.20) to estimate the overall effect of immigration on housing cost growth; and Ex-pression (3.21) to compute the impact of immigration by different place of origins.54The classification of first-tier markets is taken from Citizenship and Immigration Canada (2005).55See for example, Grant (2015).70∆pc,t = β0+β1timet +β2cmac+β3Xc,t +θ IMM · ∆immigrantsIMMc,tpopulationc,t+ εc,t (3.20)∆pc,t = β0+β1timet +β2cmac+β3Xc,t +∑mθm · ∆immigrantsm,c,tpopulationc,t + εc,t (3.21)Subscripts m, c and t are for place of origin, CSD, and year, respectively. Superscript IMM denotesthe timing of immigration and can take two different sets of values for two separate regressions – (1) 0 -5 years; and (2) all years. ∆pc,t denotes the percentage change in housing cost normalized by the numberof rooms; timet contains year effects;∆immigrantsIMMc,tpopulationc,tis the change in immigration between the consecutiveCensus years (i.e. between time t− 5 and time t) for different years of landing, normalized by the currentpopulation; and εc,t is the error term. The first-difference controls for any time-invariant omitted variablesrelated to the quality of the municipality. The cmac vector contains MA dummies to capture the effectswithin metropolitan areas.56 Xc,t is a vector of CSD attributes. These CSD attributes include the employmentrate; the share of population with at least a Bachelor’s degree, in the construction industry, and in themanufacturing industry, which may affect future city development; the share of dwelling units built prior toyear 1980 and in year 1980, to account for local housing quality; and the share of single-detached homes,apartments with at least five storeys, and movable dwellings to capture housing supply.57 The regressionswith housing value growth are weighted by the number of owned-dwelling units, whereas the models withgross rent growth are weighted by the number of rental units. The coefficient of interest is θ IMM, whichprovides the causal impact immigrants of specific year of landing exert on housing cost growth in CSD c attime t. Since θ captures CSD and time variations, I cluster the standard errors at the census subdivision level.This follows from one of the suggested strategies in Bertrand et al. (2004) for a conventional difference-in-difference framework with state and time effects, where they adjust the standard errors by clustering at thestate-level.3.5.2 Econometric specification for native mobilityAccetturo et al. (2014), Sá (2014), and Saiz and Wachter (2011) note that neighborhoods of growing im-migrant settlement are becoming less attractive to natives, which contribute to native flight. However, thetheoretical framework also shows that other factors, such as household income, housing cost, and labour56I incorporate MA effects instead of CSD effects into the regression models for two reasons. First, certain public policies areconducted at the metropolitan area level. For example, Translink provides the transportation network for the Metro Vancouverarea. Second, I use MA effects in order to prevent running into variance singularity issues for the first-stage regressions. Resultsare similar if I use CSD effects instead of MA effects. Similarly, Saiz and Wachter (2011) apply metropolitan statistical area(MSA)-year effects for housing value estimations made at the census tract level.57The regression results remain very similar if land area information as of year 1986 are included into the regression models.71substitutability, could drive mobility decisions. In order to narrow down the reason for native out-migration,unlike the existing literature’s approach, I first separate the native mobility estimation by dwelling tenuretypes. Net housing wealth effects could be influencing homeowners’ relocation intentions if the coefficientsfor the immigration rate term are different between the homeowners’ and renters’ specifications.58This study also compares the mobility decisions of the younger and older native households. I use thenear-retirees to proxy the homeowners with low amounts of residential mortgage debt (i.e. sum of all mort-gages) and with high amounts of lines of credit; and the working-age population to proxy the homeownerswith high amounts of outstanding mortgage balance and with low amounts of lines of credit. Although Iincorporate a disability dummy into the estimation to control for activity limitation, other unobserved fac-tors specific to the near-retirees could influence their taste for relocation. For the older group, I restrict thesample to families with household maintainer within ages 60-69 at time t, which means that the decision tomove was made when the household head was near retirement five years ago (i.e. ages 55-64 at time t−5).The younger set is limited to households with the maintainer of ages 30-59 at time t.I first use expression (3.22) to investigate whether immigration causes native household i to out-migratefrom origin r at time t:59Mobilityirt = β0+β1timet +β2cmar +β3 ·ageirt +β4ageirt · timet+ψ1Xrt +ψ2Hirt +θ IMM · ∆immigrantsIMMrtpopulationrt+ εirt (3.22)Mobilityirt is a binary variable, which equals one if the respondent has relocated to another censussubdivision between times t−5 and t. I report the linear probability model results in Section 3.6, but probitand logit techniques also produce similar findings as OLS.60 In addition to incorporating the CSD attributesof the original city into the model, the mobility analysis adds an extra household attribute vector Hirt .61 Thisvector includes dummy variables for gender, disability status, education, number of children, householdincome, and marital status. The Census only provides the demographic variables at time t. Therefore, Iassume that these household characteristics are invariant between consecutive Census years. Moreover, timet58For the mobility regressions, I assume that households who out-migrate to another municipality will need to sell their primaryresidence. This is a reasonable assumption as roughly three-quarters of the Canadian near-retiree households did not own secondaryresidences or other forms of real estate investments between years 1999 and 2005. Statistics Canada’s Survey of Financial Securitydatasets show that the proportion of Canadians of ages 55-64 who owned secondary residences or other forms of real estate invest-ments is 26% in 1999 and 22% in 2005. Between years 1998-2010, the RAND Health and Retirement Study dataset also illustratessimilar findings for U.S. households of the same age group.59The year of immigration variable is based on the year landed immigrant status was first obtained in Canada. The constructionof the immigration rate variable by various year of arrivals should be precise. For simplicity, I only focus on the impact of externalmigrants on local residents. Internal migrants (i.e. movements to/from another province) could also influence mobility decisions,but this is beyond the scope of this paper.60See Appendix B for the probit and logit results.61I do not include crime data into the native mobility estimations due to data limitations. The Uniform Crime Reporting Incident-Based Survey provides Canadian crime data. However, the crime data are not available for all census subdivisions for all years. Saizand Wachter (2011) and Sá (2014) also show that crime does not seem to play any role in explaining native out-migration for theU.S. and the U.K., respectively. Furthermore, Zhang (2014) chooses to use larger geographical areas (census divisions) to examinethe relationship between crime and immigration to account for the fact that crimes tend to be committed by residents of nearbyneighbourhood or community. Therefore, I assume that the MA effects in the regression models can absorb the impact from crime.72contains year dummies for the contemporaneous period; cmar includes Metropolitan Area effects for theoriginal municipality; and ageirt are age dummies for the contemporaneous period.62 ageirt · timet controlsfor age composition changes (cohort effects) across time. The standard errors continue to be clustered at thecensus subdivision level, and the regressions are weighted using household weights. The main explanatoryvariable of interest is θ IMM, which provides the probability that households choose to relocate in responseto an immigration shock between time t−5 and time t.63 Table B.4 in Appendix B presents the mean valuesof the dependent and the main independent variables.The Census and the National Household Survey datasets do not provide housing tenure status informa-tion for the previous residence. Therefore, for the mobility regressions, I assume that the native residentswho self-reported as homeowners (renters) in the current period were also homeowners (renters) five yearsago. This is a reasonable assumption because the longitudinal version of the Survey of Labour and IncomeDynamics (SLID) dataset shows that, for all age groups, a large percentage of households keeps the samedwelling tenure types for at least five years.64 Canada Mortgage and Housing Corporation (CMHC) alsosuggests that elderly homeowners are more likely to stay as homeowners even if they choose to downsizetheir house before age 65 (Canada Mortgage and Housing Corporation, 2012).This paper conducts several additional estimations in order to gather more evidence on the driving forcesbehind native out-migration. First, I run two additional estimations using expression (3.22). The first set ofestimations focuses on natives who self-reported as out of the labour force for at least two years and self-reported their income to be greater than $70,000, in order to strip out any impacts from employment-relatedand/or household finance-related factors.65 As illustrated by the theoretical model (expressions (3.12) and(3.18)), both household income and labour market trends can influence native relocation preferences. Anempirical challenge with the original mobility estimation is that renters may not be directly comparable tohomeowners because renters tend to show lower household income. Therefore, the sub-group in this firstadditional set of regression models provides a proxy for households who are less likely to be financially andspatially-constrained, which helps address the concern that renters and homeowners are not fully compara-ble.Next, similar to the housing cost regressions, I unbundle the immigration flow term by the place of birthcategories in order to evaluate whether native households show any distaste towards immigration. This setof estimations examines whether δ > 0 in the theoretical model. If distaste towards immigration were tobe the primary force driving native out-migration, then the coefficients for the immigration rate of visibleminority groups should be positive for both dwelling tenure groups.In addition, I follow Sá’s (2014) wage regression model specification to explore whether labour substi-tution exists between natives and immigrants in Canada. Given that renters and homeowners may exhibit62These estimations capture the changes in households’ location preferences over a five-year period. Therefore, by construction,the time invariant omitted variables specific to the households’ original place of residence have already been controlled. Also, theregression should produce the same results if I use the year and age dummies from five years ago instead of the contemporaneousperiod’s (i.e. subtract five for all year and age values to shift the time period by five years ago).63The regression results give similar findings if I lag the immigration rate term by one Census period (i.e. back by five years).64See Appendix B for the dwelling tenure transition matrix.65The Canadian Old Age Security starts to claw back the pension benefits when the respondent’s gross income exceeds $71,592in 2014. Therefore, I use $70,000 as the benchmark for defining higher income families. Regression results are similar if I focuson households with income greater than $100,000.73different mobility intentions, I split the estimation by dwelling tenure types to investigate if immigrationexerts differential impact on the average wage growth of the two sub-population groups. Specifically, I useexpression (3.23) to investigate whether ρ > 0 in the theoretical model:∆wdc,t = β0+β1timet +β2regionc+θIMM · ∆immigrantsIMMc,tpopulationc,t+ εc,t (3.23)Subscripts c and t denote CSD and time, respectively; and superscript d denotes the dwelling tenuregroup. ∆wc,t is the percentage change in average wage for native workers between the consecutive Censusyears. regionc denotes regional effects. I run two specifications, where one of them includes MA effectsand another includes CSD effects. The coefficient of interest θ IMM would be negative if labour substitutionwere to exist (i.e. ρ > 0 in the theoretical model).Finally, I estimate expression (3.24) using an ordered logistic model in order to evaluate the intensity ofout-migration:distirt = β0+β1timet +β2cmar +β3 ·ageirt +β4ageirt · timet+ψ1Xrt +ψ2Hirt +θ IMM · ∆immigrantsIMMrtpopulationrt+ εirt (3.24)dist contains four categories:• = 1 if households stayed in the same census subdivision, but moved to another dwelling.• =2 if households moved to a neighbouring census subdivision within the same metropolitan area.• =3 if households out-migrated to another metropolitan area within the same province.• =4 if households made an inter-provincial move.I present the weighted average predicted probabilities in order to determine where native households aremore likely to out-migrate to in response to an immigration shock, taking into account household character-istics and CSD attributes. The predicted probabilities are computed as follows, where distirt = ζB+ εirt :p j = Pr(distirt = j) = Pr (k j−1 < ζB+ εirt ≤ k j)=11+ exp(−k j +ζB) −11+ exp(−k j−1+ζB) (3.25)To summarize, I benchmark the empirical results against the theoretical model as a starting point tonarrow down the possible driving forces behind native relocation decisions and to explore whether nethousing wealth could be playing any role. This is an important innovation to the recent literature, which hasprimarily focused on the taste channel.743.5.3 Synthetic panel analysisAnother question of interest is how elderly native homeowners who stay in the same municipality use theirhousing asset in immigrant-dense locations. Asset downsizing, such as moving from a single-detached hometo an apartment, could be one possibility. As illustrated in the theoretical model, large mortgage debt balanceis one of the possible reasons holding back native homeowners’ out-migration intentions at times of risinghousing value. Given data limitations, I cannot directly estimate the impact of mortgage debt on nativeout-migration decisions using the Canadian Census data. Therefore, I use the results from the syntheticcohort analysis to indirectly determine whether native homeowners who are reluctant to move out to anothermunicipality because of attachment cost may instead choose to extract housing equity by selling their housein the current CSD at high price and purchase another residence at a lower value.Since the Census does not provide the value of the previous residence, I use expression (3.26) to con-struct a synthetic panel analysis for household maintainers who were born between the years 1924 and 1953and who reported a change in residence within the same census subdivision to examine this mechanism.66∆hpc,t,h = β0+β1yeart +β2cohorth+β3cmac+β4H¯c,t,h+ψ1Xc,t +θ IMM · ∆immigrantsIMMctpopulationct+νc,t,h (3.26)Subscript h denotes cohort. ∆hpc,t,h denotes the percentage change in mean property value for CSD c,cohort h at time t. The vector cohorth contains the dummies for the following cohort groups – (1) 1924-1930; (2) 1931-1941; (3) 1942-1947; and (4) 1948-1953. H¯ contains the mean household characteristicsfor specific cohort group, municipality, and year. The household characteristics vector includes the follow-ing variables: age, age-squared, education (no certificate; high school; some college or trade certificate;and university), household composition (married with children; married without children; single with chil-dren; single without children), household income (<30000; 30000-100000; >100000), male indicator, anddisability flag. Similar to all other regression specifications, X depicts the CSD traits.The main explanatory variable of interest is θ IMM, which shows the relationship between immigrationflow and the percentage change in the native elderly’s mean property value over time. A negative valuefor θ IMM signals the possibility that elderly native households may choose to downsize their residence inresponse to a large influx of in-migrants.3.5.4 Instrumental variable for immigration flowThe empirical concern in expressions (3.20), (3.22), and (3.26) is that certain metropolitan areas may besubject to differential external policies that are demand-driven (such as zoning regulations); and these poli-cies could be correlated with the immigration flow variable, ∆immigrationctpopulationct and thus could drive housing costand native mobility. Location preferences, as well as households’ selection to be in or out of the housing66The Health and Retirement Study, which is a longitudinal survey for the elderly in the United States, also focuses on the cohortsborn between years 1924 and 1953 (except for Version M, which starts to add Mid Baby Boomers born between years 1954 and1959).75rental markets, are generally unobserved but are correlated with the immigration flow term. Immigrants maychoose locations that are more prosperous and/or more affordable, which could cause upward bias in the re-lationship between immigration, housing cost, and native out-migration. All of these endogeneity concernspoint towards finding an exogenous measure to instrument for immigration flow.The enclaves IV, which follows a shift-share approach, is a natural candidate for instrumenting theimmigration flow variables in expressions (3.20), (3.22), and (3.26). Previous authors have computed theenclaves IV using expression (3.27) by interacting the initial share of immigrants from country m who landedin city c with the total national number of immigrants from country m at time t, summed over the country oforigin (m):Enclave IVc,t =M∑m=1( immigrantsm,c,1986immigrantsm,1986)·∆immigrantsm,Canada,t (3.27)where,• immigrantsm,c,t0immigrantsm,t0 is the initial share of immigrants from country m who first landed in CSD c at time t0. Iset the initial time period to be year 1986, which is when the Census starts to use consistent categoriesfor the dwelling-related variables.• ∆immigrantsm,Canada,t is the total number of immigrants from country m to Canada at time t.Under the conventional design, this technique predicts the immigration flow value based on the initialgeographical settlement patterns of immigrants, which exploits the fact that new immigrants with imperfectinformation tend to move to locations with strong social networks. Findings from the Canadian Census dataand the Longitudinal Survey of Immigrants to Canada dataset also confirm this settlement pattern, where im-migrants have relied on geographical closeness of friendship to improve employment opportunities (Qadeeret al., 2010; Xue, 2008). Therefore, the predicted immigration flow values based on the enclaves approachshould be strongly correlated with the actual numbers for the Canadian data. Many earlier economics pa-pers have used similar strategies to instrument for immigration flows under various research contexts andhave generally shown strong first stage regression results using this technique.67 This IV makes an implicitassumption that the initial immigrant settlement pattern is uncorrelated with the future prosperity of themunicipality.Although most of my results go through under the standard enclave IV design, this study refines theexisting IV strategy by using Expression (3.28):Refined Enclave IVc,t =M∑m=1(immigrants0−5 yearsm,c,1986immigrants0−5 yearsm,1986)·∆immigrants0−5 yearsm,Canada,t (3.28)where,67See for example, Saiz (2007) and Ottaviano and Peri (2006) for American housing cost; Cortés (2008) for prices of non-tradedgoods and services; Cortés and Tessada (2011) for time use decisions of high-skilled women; and Peri and Sparber (2009) for taskspecialization and wages.76• immigrants0−5 yearsm,c,t0immigrants0−5 yearsm,t0is the initial share of newly-arrived immigrants from country m who first landed inCSD c at time t0. I set the initial time period to be year 1986, which is when the Census starts touse consistent categories for the dwelling-related variables. I define the newly-arrived immigrants asthose with 0-5 years of landing.• ∆immigrants0−5 yearsm,Canada,t is the total number of new immigrants from country m to Canada with 0-5 yearsof landing at time t.The refined enclave IV takes the stock and the flow of newly-arrived immigrants to impute the immi-gration rate at the CSD-time level. This refinement can more precisely capture inflow dynamics, wherenew immigrants tend to settle in locations that are more accommodative to newcomers of their own ethnicorigin.68 In fact, the robust F-statistics for the first-stage regression is stronger with the refined version thanwith the standard enclave IV. This suggests that the refinement is necessary to more accurately predict newimmigrants’ settlement patterns.The exclusion restriction requirement for the refined enclave IV depends on two assumptions. First, thetotal flow of immigrants from country m to Canada(∆immigrants0−5 yearsm,Canada,t)is uncorrelated with the errorterm. Generally, almost all of the census subdivisions represent a small share of national total number ofimmigrants in Canada. This implies that the national immigration term should not be correlated with anyCSD-specific productivity or amenity effects. Immigration policies are usually implemented at the nationalor provincial level, and not at the CSD-level. Therefore, the national term should not reflect any CSD-specific factors. Furthermore, the Canadian immigration flow trends could be driven by exogenous politicalfactors from the source country, such as the transfer of sovereignty over Hong Kong, and these politicaldecisions should be independent from any unobserved CSD-trends.The second assumption for the exclusion restriction requirement relies on the initial share of newly-arrived immigrants(immigrants0−5 yearsm,c,t0immigrants0−5 yearsm,t0)not directly influencing the current housing market and native mobil-ity at the CSD level. However, one of the empirical challenges is that the geographical settlement patternsof immigrants in 1986 could influence the current amenities within a CSD, which would be priced into thecurrent and future housing market. Qadeer et al. (2010) also show that in Toronto, enclaves tend to expandto suburban areas with high homeownership rates and with growing number of residential constructions.These demand-driven expansions could alter current zoning regulations in CSD c and therefore, may havespill-over effects on current and future housing costs and native mobility preferences for the same munici-pality.Using data from the 1986 Canadian Census, I find that ethnic groups are generally not concentrated inone specific metropolitan area. Specifically, none of the groups have over 65% of their migrants settled inthe major immigration hubs, such as Toronto, Vancouver, or Montreal census metropolitan areas.69 Thisimplies that no particular settlement should exert dominant influence on any non-quantifiable CSD-specific68See for example, Saunders (2015).69See Appendix B for details, which shows the percentage of immigrant population from place of origin m who moved intometropolitan area c. For example, the first cell shows that 9.13% of the U.S.-born migrants moved into Montreal. This tablesuggests that ethnic groups are generally not concentrated in one specific metropolitan area.77attributes, providing confidence that the enclave instrumental variable approach satisfies the exclusion re-striction requirement. First-differencing the regression models also helps control for any time invariantfactors that may be correlated with the initial geographical settlement patterns of immigrants.Transforming the instrumental variable to a “Bartik-like” construction should technically solve this em-pirical problem. The Bartik construction similarly follows a “shift-share” approach, but unlike the enclavemeasure, it excludes the own city’s variation from the national flow term. This means that the predictedimmigration rate based on this modified technique is equal to the initial share of immigrants from countrym who first landed in city c multiplied by the total change in national immigration at time t, excluding theown city’s. The exclusion of the own-city’s immigration flow from the national term means that the varia-tion in the predicted value should be uncorrelated to any current fluctuations specific to the own-city overtime. Therefore, any unobserved components that influence housing cost growth and/or native mobility inmunicipality c should not be correlated with this IV. Autor and Duggan (2003) and Charles et al. (2013)use similar approaches for disability and manufacturing contexts, respectively. Expression (3.29) shows theBartik IV, where “−c” means that city c is excluded from the national measure:∆ ˆimmigrantsIMMc,t =M∑m=1(immigrantsm,c,t0immigrantsm,t0)·∆immigrantsIMMm,Canada−c,t (3.29)Although the Bartik IV is a cleaner approach, the cost of using this method rises dramatically when theanalysis is conducted at more disaggregated geographical levels. The cost is most severe for small citiesor towns with volatile immigration flow values. The Bartik predictions provide a much larger number ofoutliers relative to the enclave approach, which weakens the first stage relationship. Unlike municipalitiesin the U.S., numerous Canadian census subdivisions have small population numbers. Therefore, the enclaveIV is still a more suitable approach for this study.70Instead of using the Bartik IV, I run a total of three robustness checks for the instrumental variablestrategy to verify the validity of the estimation results. The first robustness check involves including CSDeffects instead of MA effects into the regression models. The CSD effects should account for any amenityeffects specific to the local markets. However, the cost of including the more disaggregated fixed effects isthat these indicators may absorb all of the variations coming from the enclave instrument.To account for this estimation concern, I introduce two additional instrumental variables for the secondand third robustness checks in order to construct two different over-identification specifications. Using anover-identification specification, I can apply the J-test (Hansen test) to explicitly check the moment conditionfor the exclusion restriction requirement. Under the null hypothesis for this test, the IV set would satisfy theexclusion restriction requirement. This test is not feasible under the exactly identified case.The first overidentification strategy combines the enclave IV with the Gravity IV. This first IV additionfollows from Saiz and Wachter’s (2011) approach and I use the Gravity IV as a benchmark to the existing70Overall, the regression results with the Bartik IV are generally consistent with those produced using the enclave IV, givingconfidence that the instrumental variable strategy satisfies the exclusion restriction requirement. The robust F-statistics for the first-stage regressions are generally lower under the Bartik specifications, but still exceed the minimum threshold of 16.38 based on theStock and Yogo’s (2005) benchmark in almost all cases. However, the regression results with the Bartik IV show larger standarderrors.78literature. This gravity pull instrument assumes that neighbourhoods that are geographically close to existingimmigrant enclaves have a higher probability of becoming immigrant areas in the future. In Saiz and Wachter(2011), a neighbourhood is defined at the census tract level. They argue that the impact of proximity toan immigrant enclave is heterogeneous across locations, which could help satisfy the exclusion restrictionrequirement. However, the validity of the Gravity IV also requires the inclusion of various amenity-relatedcontrol variables into the regression model in order to account for factors that could attract immigrants tolocations with characteristics that were becoming relatively less valuable to natives.Due to data limitations, in this study, I set a neighbourhood to be at the census subdivision level. Thisis the geographical level that the Canadian Census provides information on household’s mobility decisions.Expression (3.30) shows the construction of the Gravity IV in this paper for CSD c at time t.Gravity IVc,t = J∑j=1c6= j(immigrants0−5yearsj,1986population j,1986)·(landarea j,1986distance2c, j,1986) · immigrants0−5yearsCanada,tpopulationCanada,t(3.30)The first term of the instrument is a weighted average of immigrant densities in the neighbouring CSDsfor year 1986, where the weights are based on the ratio of the land area of the neighbouring CSD j tothe distance between CSD c and j. This construction is slightly different from Saiz and Wachter’s (2011)in two ways. First, they utilize the share of total number of immigrants in a census tract in the previouscensus period. Applying the same reasoning as that for refining the enclave IV, I modify the gravity factor(first term of the instrument) to be based on the number of newly-arrived immigrants with 0-5 years oflanding in year 1986 in order to better capture immigration flow dynamics. Second, instead of using theimmigration share from the previous Census period for the gravity factor, I set the share to be at year 1986in order to minimize the potential concern that more recent immigrant settlement patterns may be spatiallycorrelated with amenities in a specific location. As noted above, the validity of the gravity instrument relieson including housing quality attributes into the regression model to control for amenity effects. In Saiz andWachter’s (2011) work, these attributes are derived from the American Housing Survey. Although I includevarious dwelling-related characteristics into the estimation as controls, it is still empirically challenging toinclude a set of amenity factors that are as comprehensive as Saiz and Wachter’s (2011) into this study sinceno such dataset exists in the Canadian context. By replacing the gravity factor term with the fixed sharein 1986, I interact the gravity factor by the national share of new immigrants (second term of the IV) toobtain CSD-time variations for the instrument. The national share of new immigrants is expected to passthe exclusion restriction requirement by the same argument as for the enclave IV.As noted above, the Gravity IV could run into the same estimation problem as the enclave IV if in-sufficient number of amenity-based controls are included into the regression model. I use a different over-identification strategy to address this potential concern, which combines the enclave IV with the Airport IV.Expression (3.31) presents this instrument. The Airport IV makes use of the pattern that immigrant densitytends to be strong in municipalities (census subdivisions) that are near the major international airports. Forexample, Richmond, Mississauga, and Winnipeg, which are the locations of the international airports for79Vancouver, Toronto, and Winnipeg metropolitan areas, respectively, have the largest shares of immigrantswithin the census metropolitan areas (see for example, Figures 3.3 to 3.5). The Airport IV imputes theimmigration share at the CSD-year level by interacting the national immigration stock each year and theinverse of the distance between the CSD and the nearest international airport location.71Airport IVc,t =1distance2c,airport,1986· immigrants0−5yearsCanada,tpopulationCanada,t(3.31)The advantage of using the Airport IV is that immigrant settlement patterns are not expected to be cor-related with the ethnic enclave effect. As such, the mechanism behind the Airport IV is expected to bedifferent from the enclave IV. For example, “astronaut families” could be one possible reason behind thistrend, where the “Astronaut Dad” or “Astronaut Mom” may need to travel frequently back to the homecountry for work-related purposes. In this case, these families may choose to reside near the airport for con-venience. A potential concern with this instrument is that airport could still be correlated with productivityshocks. On one hand, airports are perceived to be economic generators for local markets, and immigrants’location choices may implicitly reflect self-selection into markets that are economically prosperous. On theother hand, municipalities that are near the major international airports or of where the airport is situateddo not necessarily have the highest productivity. For example, municipalities such as Vancouver city centreand Burnaby are near the location of the Vancouver International Airport (YVR), and Richmond is hometo YVR. Yet, the unemployment rate is highest in these locations in recent years. A similar pattern is alsoobserved for Thunder Bay. Although the airport is located at the core of the Thunder Bay metropolitan area,the 2006 Census data shows this area to be one of the locations with the lowest employment rate in Canada.Despite the fact that each of the instrument by itself could be flawed, the p-values for the over-identificationtest (J-test) are greater than 0.10 in most of the specifications. In other words, there is insufficient evidencein most cases to reject the null hypothesis that the moment condition is valid. The overall results are alsoconsistent between the exactly identified and the over-identified cases, which imply that the instrument setsare likely to satisfy the exclusion restriction requirement. Almost all forms of regression display the robustF-statistics to be greater than 10 based on Stock et al.’s (2002) benchmark. For the exactly-identified cases,the robust F-statistics exceed the Stock and Yogo’s (2005) critical values of 16.38.72 This suggests that theenclave IV satisfies the relevance condition for all specifications under the Canadian context. Section 3.6shows the robust F-statistics for the first-stage regressions and the p-values for the over-identification tests.733.6 Main findingsTo begin the analysis of results, I present the findings for the housing cost regressions, which provide anindication of whether immigration exerts significant impact on the housing market and on which market(s).71Consistent with the construction of the trade gravity model, I use the squared distance for both the Airport IV and the GravityIV’s denominator.72The threshold point of 16.38 is based on a rejection rate of at most 10% at the 5% Wald test significance level.73See Appendix B for the first-stage coefficients.80I then report the findings for the native mobility regressions to investigate the extent to which immigrationdrives native relocation. The native mobility results for the near-retirees motivate the need for the syntheticpanel analysis to determine whether the households who stay in the same municipality are more likely todownsize their housing assets in response to an immigration shock. I use the enclave IV and MA effects aspart of the main estimations. The cases with CSD effects, with the Gravity IV, and the Airport IV are a partof robustness checks.3.6.1 Housing cost regressionsTable 3.1 presents the coefficients (θ IMM) for the housing cost regressions, using the percentage change inthe average value of the dwelling and the average gross rent as the dependent variables.74 The first partillustrates the results for property value growth and the second part (on the next page) shows the findingsfor rental cost growth. The immigration rate is the main explanatory variable of interest. For the endoge-nous regressor, rows (1) and (2) use the inflow rate of newly-arrived immigrants (i.e. those who landed for0-5 years) and rows (3) and (4) apply the total immigration rate. The “First-tier markets” columns providethe average effects for the census subdivisions within Vancouver, Montreal, and Toronto census metropoli-tian areas. The remaining columns present the results for all census subdivisions (CSD) within the censusmetropolitan areas.For the OLS and the main IV estimations, I present the findings for two different sets of control variablesto examine the sensitivity of the results. The first set follows from Saiz and Wachter (2011), where I includethe change in employment rate, and the first-differenced and lagged values of socioeconomic and housingvariables as control variables. The second set is similar to Sá’s (2014), where I only include the lagged valuesof employment rate, socioeconomic factors, and housing-related variables into the estimation. Columns (1)- (4) show these results. Generally, the results are insensitive to the choice of the control variables.For the robustness checks, I focus on the case with the change in employment rate, and first-differencedand lagged values of socioeconomic and housing-related variables as controls. Column (5) shows the casewith CSD effects only; column (6) illustrates the models with both enclave and gravity instruments; andcolumn (7) presents the results with both enclave and Airport IV. Except for the owned-occupied dwellingmarket in the Vancouver, Toronto, and Montreal census metropolitan areas, which shows a positive insignif-icant coefficient for the main explanatory variable under the case with CSD effects, the regression resultsare similar across all of the specifications.To summarize, only the immigrant-dense locations (first-tier markets) show strong, positive relation-ships between property values and immigration. Both the OLS and the IV models show the expected sign.Without accounting for endogeneity, the OLS results overestimate the impact made by new immigrants andunderestimate the effect from more established immigrants. Looking at column (1), the OLS estimationssuggest that a 1% increase in the share of immigrants who landed for 0-5 years in the first-tier marketscontributes to a 1.90% increase in the housing value, whereas the total immigration rate only leads to aone-to-one increase for the same dependent variable. On the other hand, the IV model illustrates that new74Results are similar if the median values are used instead. See Appendix B for the coefficient estimates with the median valueof dwelling and the median gross rent as the dependent variables.81immigrants only exert slightly more positive impact on the owned-dwelling markets relative to the estab-lished immigrants. Conversely, immigration and property values exhibit a much weaker relationship whenI include smaller markets and/or less immigrant dense locations in the analysis. Although the OLS resultscontinue to show new immigrants exerting a strong positive influence on property values for the smallermarkets, the IV models do not show any significant linkages between these two factors. This is in line withthe theoretical model, which shows that property value growth in response to an immigration shock tends tobe strong in local markets with inelastic housing supply.Conversely, average rental costs are responsive to all forms of immigration in all markets, with theimpact strongest in the first-tier markets. Again, the OLS over-estimates the effects coming from newimmigrants and under-estimates those from older in-migrants. After accounting for endogeneity, IV resultsgenerally show newer immigrants to exert more than one-to-one increase in average rental costs for allmarkets. In the first-tier markets, newcomers cause nearly a doubling of rental cost. The estimation thatincludes CSD effects illustrates that new immigrants contribute to a more than doubled increase for thesame dependent variable in the immigrant-dense locations. Again, similar to the property value regressionresults, the smaller municipalities show a weaker relationship between immigration and average rental cost.This is consistent with the theoretical model that shows housing cost growth to slow in markets with elastichousing supply.I also extend the analysis to include the 2011 National Household Survey (NHS) data (see Table 3.2).As noted above, researchers have expressed concerns over the validity of the immigration and place of birthnumbers in the NHS. In my estimation, the year dummies should absorb any effect specific to the NHS.Overall, the findings are robust to the addition of this extra set of data. The OLS specification continues toshow the newer immigrants exert the most impact on property values and rental costs, whereas the IV resultsillustrate that both the new and established immigrants contribute nearly equal influence onto the housingmarket. Unlike the main estimation results, the smaller markets now show positive and significant effect ofimmigration on property value growth if the changes in socioeconomic characteristics were to be includedinto the estimation. In this case, including the NHS 2011 data gives slightly larger coefficients than themain estimation’s. These results coincide with the current popular idea that recent international migrantsare driving up housing markets not only in the major Canadian cities, but also in smaller markets such asHamilton.In terms of the timing of immigration, my results are somewhat consistent with Akbari and Aydede(2012) and Haan’s (2012). Haan (2012) shows that new immigrants tend to face affordability constraintsupon arrival and take at least four years to transition from renters to homeowners. Akbari and Aydede(2012) illustrate that only immigrants who landed for more than ten years have an impact on the Canadianhousing market, with the maximum rise in house price to be in the range of 0.10-0.12%. Therefore, it isreasonable to find in the regression results that more established immigrants also exert a strong impact on thehousing market, especially for the smaller census subdivisions. However, my results deviate from Akbariand Aydede (2012) and Saiz and Wachter (2011), where both of these papers show immigration to slow thehousing market.Table 3.3 shows the OLS regression results for housing cost by unbundling the immigration flow term82by the country of origin. Consistent with common belief, the results suggest that an increase in the numberof Asian immigrants especially those from Hong Kong is correlated with both housing value and gross rent.On the other hand, non visible minority groups as well as Asians who are not from Hong Kong appear to bepositively-associated with gross rent. Saiz and Wachter (2011) note that local house price growth tends to beslower in neighbourhoods with high density of immigrants that belong to the visible minority groups, whichimplies that native households segregate themselves by ethnicity. My results appear to be different fromSaiz and Wachter’s (2011), where visible minority groups are actually driving up housing value. However,given that the magnitudes and the direction of the coefficients are not uniform across the place of origins, itis possible that a subset of native households may show distaste towards certain types of immigrants. Thefollowing sub-section investigates this context in more detail.3.6.2 Mobility regressionsThe housing cost regressions generally show immigration to exert a positive influence on both the housingsale and rental markets. These results imply that native households may out-migrate to another censussubdivision due to rising housing cost from an immigration shock. The main research question of interestis whether relocation preferences are different across the dwelling tenure and age groups due to net housingwealth accumulations.Tables 3.4 and 3.5 show the native out-migration decisions of homeowners and renters for ages 55-64 and ages 25-54 at the time of migration, respectively. For both tables, the first section corresponds tohomeowners and the second section (on the next page) is for renters, with the binary variable for change inmunicipality over a five-year period as the dependent variable. Again rows (1) and (2) present the findingsfor new immigrants who landed for 0-5 years; and rows (3) and (4) use the total immigration rate as themain explanatory variable. The layout for Tables 3.4 and 3.5 is the same as the ones for the housing costregressions. The regression results for the robustness checks are generally consistent with the main estima-tions’ with the enclave IV as the only instrument. This evidence provides confidence that the IV regressionresults are valid and satisfy both the exclusion restriction and the relevance requirements.Among the elderly native homeowners, the overall effect of immigration on native mobility is relativelymodest. The OLS results show a significant positive relationship between these two factors for all markettypes and for all types of immigration. For example, looking at column (1) and row (3) of Table 3.4, a 100%increase in the total immigration rate is associated with a 24% chance of relocating to another municipalityfor native homeowners who live in the Vancouver, Toronto, and Montreal metropolitan areas. After account-ing for endogeneity, near-retiree homeowners are approximately 35% more likely to out-migrate for every100% increase in the total share of immigrants in the first-tier markets. The effect of immigration on nativerelocation decisions is strongest for immigrant-dense locations. Yet, the IV results show that both the newerand more established immigrants exert roughly equal impact on the elderly native homeowners’ mobilitydecisions. This seems to be consistent with the IV regression results for the housing value specifications.Conversely, Table 3.5 shows that younger native homeowners have stronger intentions to out-migrate inresponse to an immigration shock. The IV specification shows that a 100% increase in the total immigrationrate for the first-tier markets leads to nearly 50% chance that these younger households will relocate to83another municipality. Similar to the elderly case, immigration exerts differential impact on native mobilityacross geographical areas where the impact from immigration is strongest on the first-tier markets.For all regression forms and for both age groups, renters exhibit an insignificant relationship betweennative out-migration and immigration. The magnitude of the coefficients tends to be much smaller than isthe case for homeowners. At most, the elderly shows approximately a 10% chance of out-migration and theyounger families exhibit about a 5% chance. This implies that renters have no intentions of moving out oftheir current place of residence in response to an immigration shock. This provides additional insight to whyimmigration pushes up average rental cost in all markets. The mobility regression results with the NHS dataare generally consistent with the main estimation’s (see Table 3.6). After accounting for endogeneity, theyounger native homeowners continue to show greater intentions to relocate relative to the elderly families.Table 3.7 presents the results for native household maintainers who self-reported as out of the labourforce for at least two years and self-reported their income to be greater than $70,000, in order to strip outthe possible confounding effects arising from employment and/or from household finance.75 These effectsare expected to limit relocation decisions, with the impact to be stronger for financially-constrained renters.The empirical concern is that the negative and insignificant relationship between native out-migration andimmigration for the renters’ regression model may not be reflecting their acceptance to the newcomers.Instead, the negative linkage may point to the possibility that renters experience high perceived costs ofmoving, which discourage them from relocating. Therefore, by focusing on the higher income householdswho are less likely to be financially and spatially-constrained, this additional test should provide a clearerindication of the causal relationships between immigration and mobility decisions for homeowners and forrenters.Table 3.7 shows that the regression results with this subsample are only consistent with those producedusing the full sample for the elderly homeowners. The results are generally reversed for the renters and forthe younger homeowners, which imply that household finance may play some role for these sub-populationsin general. By stripping out the effect from employment and from financial constraints, the younger nativehomeowners show low intentions to out-migrate in immigrant-dense locations and only slightly higher pref-erence to relocate in the smaller markets. Conversely, for the first-tier markets, the elderly native rentersshow high intentions to move out while the younger native renters do not show the same intentions. How-ever, once the estimation incorporates the smaller municipalities, the effect of immigration on youngernative renters’ mobility preference remains ambiguous because it is sensitive to the type of control variablesincluded into the model.To narrow down whether the reversal in the renters’ results relate to distaste for cultural diversity, I un-bundle the immigration flow term by the country of origins for the mobility regressions (Table 3.8). Saiz andWachter (2011) note that local house price growth tends to be slower in neighbourhoods with high density ofimmigrants that belong to the visible minority groups, which implies that native households segregate them-selves by ethnicity. Younger native homeowners appear to show stronger distaste for immigrants. Generally,homeowners tend to out-migrate in response to the group pushing up property values, such as immigrantsfrom Hong Kong. However, the younger homeowners also show relocation intentions in response to immi-75Regression results are similar if I focus on households with income greater than $100,000.84grants from Africa. On the other hand, renters do not seem to segregate by ethnic groups.In addition, Table 3.9 suggests that labour substitution between native and immigrant workers doesnot seem to hold for the Canadian context. Except for the younger renters, which illustrates a significantpositive relationship between native wage growth and the immigration rate, all other groups generally showinsignificant coefficients for the immigration rate variable. This implies that, for most cases, ρ ≤ 0 in thetheoretical model and that native flight does not seem to be associated with labour substitution.Table 3.10 relates the empirical findings to the theoretical model. In addition to distaste towards immi-gration, housing value growth (affordability), and labour substitution, Table 3.10 suggests that other com-ponents could be driving elderly native homeowners’ relocation decisions. The empirical findings show thatthe taste channel and housing cost growth do drive homeowners to move out, whereas the overall rate ofout-migration is small. Home equity borrowing could be one possible factor in explaining such discrepancy,which may explain the reason for the modest rate of out-migration among the near-retirees’.Unfortunately, the Canadian Census data does not provide information on home equity borrowing.76 Todate, no datasets similar to the RAND Health and Retirement Study (HRS) exist for the Canadian context.Despite this shortcoming, various stylized facts suggest that an increasing number of older American andCanadian households has accessed home equity borrowing. First, Crawford and Faruqui (2012) show usingthe Canadian Financial Monitor dataset that the proportion of secured personal lines of credit debt heldby households of ages 55-64 has risen significantly between years 1999-2010. For the same age group,the Survey of Financial Security (SFS) datasets also illustrate similar findings, with the mean proportionof line of credit users to grow from 15% in 1999 to 28% in 2005; and the outstanding balance of averagelines of credit to increase by 30% for the same time period (i.e. from $14,650.99 in 1999 to $19,161.52 in2005).77 For the U.S., the proportion of near-retirees with home equity loans in inelastic housing supplymarkets escalated from 15% to 22% between the years 1998 and 2010.78 Consistent with this finding, Figure3.6 shows that the share of U.S. near-retirees with home equity loans tends to be high in immigrant-denselocations. These stylized facts point to the possibility that older Canadians may be turning to home equityborrowing to extract the extra housing windfall arising from growing immigrant settlement.The factors that drive the younger homeowners’ relocation decisions are less obvious. On one hand, theout-migration rate for high-income homeowners is low, especially in the first-tier markets. However, thedistaste for immigration and the property value growth are non-zero. Given that the younger homeownerstend to show larger mortgage debt holdings, this could be one possible factor in limiting out-migrationintentions in the first-tier markets for these higher income individuals. On the other hand, once the smallermunicipalities are incorporated into the computation, this sub-group shows positive out-migration rate. Theempirical findings cannot identify which factors drive net relocation decisions. This ambiguity also holdswhen lower income younger homeowners are included into the computation.76The Canadian Censuses do not provide any information on outstanding balances of mortgage credit and home equity lines ofcredit.77The SFS data only provides total personal lines of credit data, which also includes non-housing lines of credit. Crawford andFaruqui (2012) suggest that secured lines of credit are mostly secured by HELOC; therefore I assume that the secured debt seriesare driven mainly by mortgage debt and HELOC. These numbers are normalized by the square root of the household size and theaverage is taken only among the households with positive HELOC holdings.78These numbers were computed using the RAND Health and Retirement Study.85Conversely, financial constraints could be another factor in limiting the elderly native renters’ out-migration decisions. For the first-tier markets, housing affordability appears to be the primary driver behindnative flight for the elderly native renters who have stopped working and with relatively higher householdincome. In other words, only the households that are more financially and spatially-flexible show preferencefor out-migration in response to an influx of newcomers. On the other hand, despite that rental cost growthincreases in response to an immigration shock, near-retiree renters who continue to work and with relativelyless income do not show any preference for relocation. The effect of immigration on native wages doesnot play any significant role in mobility preferences, which means that employment effects do not seem tobe slowing down native flight. Yet, the residual effects seem to be slowing down the rate of out-migration.Combining these findings, household financial constraints could be one potential factor limiting relocationdecisions among the renters.On the contrary, both ethnic diversity and labour complementarity seem to deter the younger nativerenters from out-migrating to another census subdivisions. These two factors counteract the positive effecton relocation from rising rental cost growth. However, in order to sum up the effects, other residual factorsseem to be discouraging the higher income renters from relocating, whereas the opposite trend applieswhen the lower income renters are included into the model. Transportation networks could be one possibleexplanation behind this finding. Families that are more spatially and financially constrained may need tomove to locations that are more accessible to transit. In any case, a combination of these findings suggeststhat other factors aside from distaste affect relocation decisions.To examine the intensity of out-migration, I use the ordered logistic model to explore where nativerenters and homeowners move to and whether renters are more likely to stay in the same CSD relative tothe homeowners. Table 3.11 shows the ordered logistic results. The top panel shows the results for nativehomeowners of different age groups and housing markets; and the bottom panel presents the results fornative renters. Consistent with the results above, relative to native homeowners, native renters do tend tostay in the same census subdivision. For all markets, nearly three-quarters of the renters relocate within thesame municipality, which implies that financial constraints may be pushing them back from out-migrating.Conversely, about 40-50% of the native homeowners out-migrate to another census subdivision, where amajority of them move to neighbouring municipalities.To summarize, the research findings in this paper open up another question of whether a housing wealthchannel could explain mobility decisions in addition to a taste channel. Overall, a combination of resultssuggests that housing affordability and household finance could matter. Moreover, mobility preferenceseems to be heterogeneous across dwelling tenure groups, which may explain why Card (2001) fails to findany significant effect from immigration on aggregate native relocation decisions.3.6.3 Synthetic panel analysisGiven that roughly half of the near-retirees stay in the same municipality, another research question ofinterest is whether these elderly households choose to downsize their house over time in response to animmigration shock (i.e. sell high and then buy low). Table 3.12 presents the results for the synthetic panelanalysis for the cohorts born between years 1924 and 1953 to determine whether the percentage change86in property value for homeowners who changed residence declines relative to a one-percent increase inthe immigration flow rate. This estimation assumes that households are switching to dwellings of similarquality, which provides an upper bound answer to the research question of interest.The OLS specification shows that the newer immigrants in the first-tier markets push up housing valuesover time by more than a one-to-one relationship. This implies that the elderly native homeowners switchto more expensive dwellings in response to an immigration shock. However, after adjusting for location andselection biases, the IV results do not show any significant relationship between immigration and propertyvalues for the older generation.79 Overall, there is insufficient evidence to suggest that elderly householdsextract housing equity by relocating.3.7 ConclusionThis study compares the short-term impacts of immigration on housing cost growth and on mobility deci-sions for the rental and owner-occupied dwelling markets. I extend Saiz’s (2007) and Sá’s (2014) theoreticalframework and apply various econometric techniques to explore this research question. In this study, I alsocompare the relocation preferences of near-retirees (ages 55-64) and of the working-age population (ages25-54) to explore whether mobility proclivities vary across the age groups. I use the near-retirees to proxythe homeowners with low amounts of residential mortgage debt (i.e. sum of all mortgages) and with highamounts of lines of credit; and the working-age population to proxy the homeowners with high amountsof outstanding mortgage balance and with low amounts of lines of credit. Furthermore, this paper uses anordered logistic regression with moving distance categories as the dependent variable to determine wherenative households move to in response to an immigration shock. Finally, I conduct a synthetic panel analysisto examine whether the elderly homeowners who stay in the same municipality exhibit any housing assetdownsizing in response to an immigration shock. These comparisons help identify whether net housingwealth could influence overall mobility decisions.I use the historical ethnic distributions (enclave approach) to construct the instrumental variable, whichexploits the fact that immigrants tend to move to locations with strong social networks. All forms of empiri-cal analysis show that in the short-run, immigration only influences property values in the major metropoli-tan areas, such as Vancouver, Montreal, and Toronto. This result is in line with the predictions from mytheoretical framework, which shows that housing cost growth tends to be high in markets with inelastichousing supplies. A combination of results points to the possibility that in addition to a taste channel, hous-ing affordability and household finance could influence mobility decisions. The heterogeneity in mobilitypreferences across dwelling tenure groups is an important finding because it may explain why Card (2001)fails to find any significant effect from immigration on aggregate native relocation decisions. The resultsfrom the synthetic panel do not provide sufficient evidence to conclude that near-retirees choose to extracthousing equity by relocating.This paper faces two major shortcomings. First, the Census data does not provide any information onproperty values, dwelling characteristics, and dwelling tenure statuses for the original place of residence.79The results still holds when I include the NHS 2011 data87Without this information, this paper cannot fully determine whether the elderly households switched tenuretypes in response to an immigration shock and cannot fully control for housing quality in the main regressionanalysis. Second, the Census data does not contain any information on mortgage refinancing and on homeequity lines of credit. Various stylized facts have shown that the proportion of home equity loans held byolder households has risen significantly since the 1990s and tends to be positively correlated with immigrantdensities. This study thus misses the linkages between immigration and home equity borrowing, where theelderly households could choose to use mortgage refinancing or home equity lines of credit instead of sellingoff their dwelling units to take advantage of the extra housing windfall arising from an immigration shock.Despite these shortcomings, the research findings in this paper push this area of literature forward bysuggesting an alternative perspective to explain native mobility decisions. The mobility regressions illustratethat relative to renters, homeowners are more likely to move. This implies that distaste may not completelyaddress the reasons behind native out-migration, at least for the Canadian context. Previous studies haveshown that Canadian immigrants tend to have higher levels of English fluency, cognitive ability, education,and income relative to natives than do U.S. immigrants; and the national-origin mix of the immigrant flowsis one explanation for this difference (Antecol et al., 2003; Borjas, 1993; and Kahn, 2004). A possible futurepath for this study is to extend this research question to a U.S. context in order to investigate whether thesame conclusion holds.This research is key from both public policy and financial stability standpoints. The findings from thisstudy indicate the extent to which elderly households extract housing equity in response to immigrationshocks and also provide additional insights to the possible consequences from implementing current immi-gration policy onto the high-priced Canadian housing market.883.8 FiguresFigure 3.1: Near-retirees’ labour force participation decisions and immigration inflows89Figure 3.2: Labour force participation decisions for all workers and immigration inflows90Figure 3.3: Immigration shares and international airport locations - Vancouver CMANote: The major international airport location is denoted by the red box. Dark yellow shade represents the census subdivisions with high immigrant densities.Sources: Statistics Canada’s 2006 Census geographic boundary files; University of Toronto Canadian Census Analyser (2006 Census / Profile of Census Subdivi-sions); and author’s calculations. These figures are generated by the following R packages: “maptools”, “rgeos”, and “ggmap”.91Figure 3.4: Immigration shares and international airport locations - Toronto CMANote: The major international airport location is denoted by the red box. Dark yellow shade represents the census subdivisions with high immigrant densities.Sources: Statistics Canada’s 2006 Census geographic boundary files; University of Toronto Canadian Census Analyser (2006 Census / Profile of Census Subdivi-sions); and author’s calculations. These figures are generated by the following R packages: “maptools”, “rgeos”, and “ggmap”.92Figure 3.5: Immigration shares and international airport locations - Winnipeg CMANote: The major international airport location is denoted by the red box. Dark yellow shade represents the census subdivisions with high immigrant densities.Sources: Statistics Canada’s 2006 Census geographic boundary files; University of Toronto Canadian Census Analyser (2006 Census / Profile of Census Subdivi-sions); and author’s calculations. These figures are generated by the following R packages: “maptools”, “rgeos”, and “ggmap”.93Figure 3.6: Share of near-retirees with HELOC holdings versus immigration sharesNote: The American Community Survey dataset does not contain any information on outstanding debt balances for all forms of home equity borrowing.Sources: U.S. Census Bureau’s 2005-2009 5-year American Community Survey; and author’s calculations943.9 TablesTable 3.1: Immigration’s impact on housing market - 2006 Census dataDependent variable: Percentage change in average value of dwellingOLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years1.900***(0.282)1.899***(0.354)1.578***(0.404)1.583***(0.588)1.135(0.776)1.573***(0.402)1.579***(0.404)F-test of excluded IV 28.807 48.208 10.948 15.172 22.264Hansen overidentification test[p-values for Hansen test]0.211[0.646]1.270[0.260]All markets(2) Immigration rate:0 - 5 years1.164***(0.259)1.142***(0.283)0.762(0.466)0.432(0.688)0.399(0.664)0.760(0.466)0.762(0.466)F-test of excluded IV 78.660 85.871 39.843 39.389 40.323Hansen overidentification test[p-values for Hansen test]0.697[0.404]1.881[0.170]First-tier markets [1](3) Immigration rate:All years1.003***(0.190)0.908***(0.246)1.387***(0.290)1.646**(0.661)1.037(0.700)1.266***(0.281)1.372***(0.291)F-test of excluded IV 24.651 68.178 10.130 14.151 12.177Hansen overidentification test[p-values for Hansen test]1.994[0.158]4.640[0.031]All markets(4) Immigration rate:All years0.332**(0.141)0.451***(0.167)0.786*(0.454)0.473(0.745)0.408(0.676)0.768*(0.449)0.790*(0.452)F-test of excluded IV 53.598 76.935 27.801 27.675 28.079Hansen overidentification test[p-values for Hansen test]1.090[0.297]0.956[0.328]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 460; for regressions with all markets = 1,915(continued on next page...)95Table 3.1: Immigration’s impact on housing market - 2006 Census data (continued)Dependent variable: Percentage change in gross rentOLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years1.493***(0.285)1.387***(0.272)1.862***(0.239)1.594***(0.254)2.196***(0.687)1.900***(0.248)1.861***(0.240)F-test of excluded IV 21.522 26.648 5.131 14.599 14.564Hansen overidentification test[p-values for Hansen test]7.748[0.005]0.045[0.832]All markets(2) Immigration rate:0 - 5 years0.971***(0.236)0.842***(0.208)1.193***(0.207)0.927***(0.184)1.371***(0.360)1.196***(0.208)1.193***(0.207)F-test of excluded IV 58.017 73.773 19.433 30.174 30.166Hansen overidentification test[p-values for Hansen test]1.292[0.256]0.103[0.748]First-tier markets [1](3) Immigration rate:All years1.269***(0.232)1.180***(0.287)1.652***(0.170)1.695***(0.292)1.666***(0.367)1.687***(0.173)1.650***(0.170)F-test of excluded IV 16.426 34.498 7.666 10.198 8.257Hansen overidentification test[p-values for Hansen test]0.852[0.356]0.523[0.470]All markets(4) Immigration rate:All years0.889***(0.217)0.819***(0.214)1.260***(0.199)1.048***(0.200)1.252***(0.332)1.262***(0.201)1.260***(0.199)F-test of excluded IV 35.856 66.161 20.931 19.422 20.464Hansen overidentification test[p-values for Hansen test]0.083[0.773]0.036[0.850]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 455; for regressions with all markets = 1,825This table shows the regression results for the sample period 1991-2006, using expression (3.20). Control set 1includes the change in employment rate, as well as the first-differenced and lagged values of socioeconomic andhousing variables. Control set 2 includes lagged values of employment rate, socioeconomic factors, and housing-related variables. See text for details regarding the control variables. These models are weighted by the number ofowned or rental dwellings per census subdivisions. Column “IV1” includes the enclave IV; column “IV2” includesthe enclave IV and the Gravity IV; and column “IV3” includes the enclave IV and the Airport IV. Standard errors arein parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitanareas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is basedon data licensed from Canada Post Corporation; and author’s calculations.96Table 3.2: Immigration’s impact on housing market - 2011 National Household Survey dataDependent variable: Percentage change in average value of dwelling [1]First-tier markets [2] All marketsOLS OLS IV IV OLS OLS IV IVIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8)(1) Immigration rate: 0 - 5 years 1.893***(0.237)1.902***(0.353)1.390***(0.260)1.5908***(0.5864)1.410***(0.310)1.056***(0.277)1.253***(0.443)0.373(0.671)F-test of excluded IV 48.303 48.375 97.853 85.514(2) Immigration rate: All years 0.856***(0.150)0.908***(0.246)1.363***(0.182)1.665**(0.664)0.542***(0.163)0.516***(0.151)1.325***(0.417)0.405(0.723)F-test of excluded IV 25.520 69.085 54.647 77.645Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 580 580 580 580 2,605 2,605 2,605 2,605(continued on next page...)97Table 3.2: Immigration’s impact on housing market - 2011 National Household Survey data (continued)Dependent variable: Percentage change in average gross rent [3]First-tier markets [2] All marketsOLS OLS IV IV OLS OLS IV IVIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8)(1) Immigration rate: 0 - 5 years 1.440***(0.198)1.399***(0.262)1.567***(0.164)1.605***(0.245)1.002***(0.187)0.814***(0.206)1.146***(0.137)0.887***(0.175)F-test of excluded IV 33.321 26.989 73.036 76.033(2) Immigration rate: All years 1.008***(0.270)1.188***(0.283)1.542***(0.204)1.727***(0.290)0.753***(0.200)0.793***(0.210)1.233***(0.135)0.999***(0.190)F-test of excluded IV 17.459 35.071 34.856 69.868Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 575 575 575 575 2,475 2,475 2,475 2,475This table shows the regression results for the sample period 1991-2011, using expression (3.20). All regression models contain year and MA effects. Controlset 1 includes the change in employment rate, as well as the first-differenced and lagged values of socioeconomic and housing variables. Control set 2 includeslagged values of employment rate, socioeconomic factors, and housing-related variables. The IV regressions use the enclave instrument. Standard errors are inparentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.[1]: These models are weighted by the number of owned dwellings per census subdivision.[2]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitan areas.[3]: These models are weighted by the number of rental units per census subdivision.Sources: Statistics Canada’s 1986 - 2006 Censuses; 2011 National Household Survey; and Postal CodeOM Conversion File (various dates), which is based on datalicensed from Canada Post Corporation; and author’s calculations.98Table 3.3: Which ethnic groups drive housing value and rental cost growth?Dependent variables: Percentage change in average value of dwelling [1] Percentage change in average gross rent [2]First-tier markets [3] All markets First-tier markets [3] All marketsIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8)(1) Immigration rate:Europe excluding U.K.0.953(0.799)1.737(1.080)0.272(0.630)1.277(0.925)1.975***(0.482)1.838***(0.619)0.792*(0.432)0.938*(0.495)(2) Immigration rate:Africa-0.486(2.085)-1.310(2.436)-0.683(3.523)-1.092(3.890)0.388(0.889)-1.498(1.251)-0.312(2.413)-1.314(2.548)(3) Immigration rate:Central and South America, includingCaribbean, Bermuda, and Jamaica2.425**(1.021)0.467(1.139)3.616**(1.547)0.835(1.489)0.066(0.735)0.303(1.189)0.844(0.829)0.050(0.994)(4) Immigration rate:Asia excluding Hong Kong0.132(0.350)0.306(0.615)-0.972**(0.404)-0.404(0.522)1.294***(0.412)1.106**(0.453)0.852**(0.332)0.904**(0.377)(5) Immigration rate:Hong Kong3.502***(0.786)2.762***(0.913)3.506***(1.021)2.953***(1.016)0.948*(0.520)1.265*(0.743)0.810(0.521)0.843(0.568)(6) Immigration rate:U.S., U.K., and the Oceania countries2.244*(1.239)2.992*(1.563)-0.007(0.396)0.051(0.734)3.342***(0.884)3.813***(0.795)2.135***(0.441)2.352***(0.467)Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 465 465 2,110 2,110 460 460 2,020 2,020This table shows the OLS results for expression (3.21), spanning the sample period of 1991-2006. All regression models contain year and MA effects. Control set1 includes the change in employment rate, as well as the first-differenced and lagged values of socioeconomic and housing variables. Control set 2 includes laggedvalues of employment rate, socioeconomic factors, and housing-related variables. Standard errors are in parentheses. *** Significant at 1%; ** significant at 5%; *significant at 10%.[1]: These models are weighted by the number of owned dwellings per census subdivision.[2]: These models are weighted by the number of rental units per census subdivision.[3]: First-tier markets include the census subdivisions within the Vancouver, Toronto, and Montreal census metropolitan areas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is based on data licensed from Canada PostCorporation; and author’s calculations.99Table 3.4: Immigration inflows and older native mobilityNative homeowners of ages 55-64 at time of migration:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent hasrelocated to another census subdivision between time t−5 and time t.OLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years0.330***(0.107)0.271***(0.098)0.370*(0.196)0.270*(0.151)0.221(0.147)0.372*(0.195)0.371*(0.196)F-test of excluded IV 42.118 67.879 20.839 22.652 28.281Hansen overidentification test[p-values for Hansen test]0.137[0.711]2.686[0.101]All markets(2) Immigration rate:0 - 5 years0.206**(0.084)0.174**(0.083)0.245**(0.117)0.179*(0.108)0.188**(0.089)0.246**(0.117)0.245**(0.117)F-test of excluded IV 95.994 98.519 68.480 48.098 48.024Hansen overidentification test[p-values for Hansen test]2.758[0.097]0.203[0.653]First-tier markets [1](3) Immigration rate:All years0.236***(0.084)0.199***(0.060)0.360*(0.204)0.308*(0.168)0.213(0.149)0.348*(0.188)0.366*(0.204)F-test of excluded IV 25.020 63.908 15.832 14.032 13.866Hansen overidentification test[p-values for Hansen test]0.256[0.613]2.375[0.123]All markets(4) Immigration rate:All years-0.006(0.065)-0.001(0.066)0.272**(0.135)0.217*(0.130)0.208*(0.106)0.281**(0.136)0.271**(0.135)F-test of excluded IV 54.994 63.202 35.352 27.859 29.169Hansen overidentification test[p-values for Hansen test]1.916[0.167]0.436[0.509]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 123,485Number of observations for regressions with all markets = 352,175(continued on next page...)100Table 3.4: Immigration inflows and older native mobility (continued)Native renters of ages 55-64 at time of migration:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent hasrelocated to another census subdivision between time t−5 and time t.OLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years0.014(0.117)-0.071(0.134)0.124(0.125)0.016(0.151)0.116(0.260)0.107(0.125)0.124(0.125)F-test of excluded IV 37.959 44.141 14.806 21.274 25.560Hansen overidentification test[p-values for Hansen test]2.206[0.137]0.006[0.939]All markets(2) Immigration rate:0 - 5 years0.054(0.097)-0.107(0.097)0.115(0.102)-0.080(0.104)0.069(0.128)0.118(0.102)0.114(0.102)F-test of excluded IV 83.662 117.548 42.075 42.477 41.847Hansen overidentification test[p-values for Hansen test]1.808[0.179]1.424[0.233]First-tier markets [1](3) Immigration rate:All years-0.130(0.107)-0.138(0.101)0.118(0.118)0.017(0.167)0.101(0.222)0.054(0.120)0.117(0.118)F-test of excluded IV 22.462 43.155 14.884 12.788 11.176Hansen overidentification test[p-values for Hansen test]2.379[0.123]0.020[0.888]All markets(4) Immigration rate:All years-0.028(0.094)-0.114(0.095)0.128(0.113)-0.094(0.124)0.070(0.128)0.142(0.111)0.122(0.114)F-test of excluded IV 42.937 75.658 33.038 26.451 26.868Hansen overidentification test[p-values for Hansen test]1.642[0.200]1.496[0.221]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 68,995Number of observations for regressions with all markets = 147,790This table shows the regression results for the sample period 1991-2006, using expression (3.22). Control set 1includes the change in employment rate, as well as the first-differenced and lagged values of socioeconomic andhousing variables. Control set 2 includes lagged values of employment rate, socioeconomic factors, and housing-related variables. These models are weighted by Census household weights. Column “IV1” includes the enclave IV;column “IV2” includes the enclave IV and the Gravity IV; and column “IV3” includes the enclave IV and the AirportIV. Standard errors are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%. [1]: First-tiermarkets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitan areas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is basedon data licensed from Canada Post Corporation; and author’s calculations.101Table 3.5: Immigration inflows and younger native mobilityNative homeowners of ages 25-54 at time of migration:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent hasrelocated to another census subdivision between time t−5 and time t.OLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years0.386***(0.101)0.478***(0.080)0.553***(0.160)0.582***(0.092)0.427***(0.129)0.530***(0.159)0.553***(0.160)F-test of excluded IV 41.680 65.262 23.022 22.904 30.141Hansen overidentification test[p-values for Hansen test]2.437[0.118]0.573[0.449]All markets(2) Immigration rate:0 - 5 years0.346***(0.076)0.388***(0.074)0.382***(0.096)0.397***(0.094)0.214**(0.107)0.383***(0.096)0.381***(0.096)F-test of excluded IV 86.181 84.914 66.887 43.153 43.057Hansen overidentification test[p-values for Hansen test]0.510[0.475]1.432[0.232]First-tier markets [1](3) Immigration rate:All years0.207***(0.078)0.229***(0.067)0.514***(0.157)0.620***(0.086)0.410***(0.136)0.383***(0.150)0.516***(0.158)F-test of excluded IV 29.189 81.243 18.308 16.360 15.295Hansen overidentification test[p-values for Hansen test]3.568[0.059]0.090[0.764]All markets(4) Immigration rate:All years0.052(0.082)0.080(0.092)0.403***(0.101)0.446***(0.100)0.231**(0.117)0.407***(0.100)0.396***(0.101)F-test of excluded IV 52.188 59.466 37.275 26.361 27.256Hansen overidentification test[p-values for Hansen test]0.153[0.696]1.631[0.202]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 658,950Number of observations for regressions with all markets = 1,803,360(continued on next page...)102Table 3.5: Immigration inflows and younger native mobility (continued)Native renters of ages 25-54 at time of migration:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent hasrelocated to another census subdivision between time t−5 and time t.OLS OLS IV1 IV1 IV1 IV2 IV3Independent variables: (1) (2) (3) (4) (5) (6) (7)First-tier markets [1](1) Immigration rate:0 - 5 years-0.165*(0.099)-0.198(0.120)-0.052(0.112)-0.037(0.162)0.171(0.187)-0.071(0.114)-0.049(0.112)F-test of excluded IV 27.240 32.446 8.847 17.406 17.797Hansen overidentification test[p-values for Hansen test]1.186[0.276]2.985[0.084]All markets(2) Immigration rate:0 - 5 years-0.030(0.081)-0.184**(0.083)0.050(0.095)-0.109(0.103)0.053(0.120)0.055(0.095)0.049(0.095)F-test of excluded IV 65.940 80.485 29.197 33.281 33.144Hansen overidentification test[p-values for Hansen test]2.447[0.118]1.461[0.227]First-tier markets [1](3) Immigration rate:All years-0.270***(0.092)-0.246***(0.084)-0.048(0.104)-0.039(0.173)0.137(0.142)-0.090(0.109)-0.052(0.105)F-test of excluded IV 19.638 46.773 11.918 11.547 9.735Hansen overidentification test[p-values for Hansen test]1.077[0.299]3.024[0.082]All markets(4) Immigration rate:All years-0.127***(0.032)-0.156***(0.045)0.053(0.101)-0.125(0.120)0.050(0.114)0.071(0.100)0.042(0.102)F-test of excluded IV 42.036 72.620 28.721 23.421 24.451Hansen overidentification test[p-values for Hansen test]2.374[0.123]1.471[0.225]MA effects Yes Yes Yes Yes No Yes YesCSD effects No No No No Yes No NoControl set 1 Yes No Yes No Yes Yes YesControl set 2 No Yes No Yes No No NoNumber of observations for regressions with first-tier markets = 386,630Number of observations for regressions with all markets = 842,425This table shows the regression results for the sample period 1991-2006, using expression (3.22). Control set 1includes the change in employment rate, as well as the first-differenced and lagged values of socioeconomic andhousing variables. Control set 2 includes lagged values of employment rate, socioeconomic factors, and housing-related variables. These models are weighted by Census household weights. Column “IV1” includes the enclave IV;column “IV2” includes the enclave IV and the Gravity IV; and column “IV3” includes the enclave IV and the AirportIV. Standard errors are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%. [1]: First-tiermarkets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitan areas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is basedon data licensed from Canada Post Corporation; and author’s calculations.103Table 3.6: Immigration inflows and native mobility - 2011 National Household Survey dataNative homeowners:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if therespondent has relocated to another census subdivision between time t−5 and time t.OLS OLS IV IVIndependent variables: (1) (2) (3) (4)Age 55-64: First-tier markets [1](1) Immigration rate:All years0.255***(0.078)0.201***(0.060)0.294(0.212)0.305*(0.168)F-test of excluded IV 26.855 64.778Number of observations 175,630 175,630 175,630 175,630Age 55-64: All markets(2) Immigration rate:All years0.151***(0.052)0.151***(0.056)0.257**(0.128)0.230*(0.127)F-test of excluded IV 61.607 70.243Number of observations 539,890 539,890 539,890 539,890Age 25-54: First-tier markets [1](3) Immigration rate:All years0.179**(0.072)0.227***(0.067)0.383**(0.177)0.604***(0.087)F-test of excluded IV 29.880 82.314Number of observations 884,285 884,285 884,285 884,285Age 25-54: All markets(4) Immigration rate:All years0.187***(0.054)0.248***(0.063)0.367***(0.088)0.466***(0.093)F-test of excluded IV 53.993 63.324Number of observations 2,557,640 2,557,640 2,557,640 2,557,640Control set 1 Yes No Yes NoControl set 2 No Yes No Yes(continued on next page...)104Table 3.6: Immigration inflows and native mobility - 2011 National Household Survey data (continued)Native renters:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if therespondent has relocated to another census subdivision between time t−5 and time t.OLS OLS IV IVIndependent variables: (1) (2) (3) (4)Age 55-64: First-tier markets [1](1) Immigration rate:All years-0.112(0.103)-0.133(0.097)0.154(0.136)-0.032(0.177)F-test of excluded IV 23.593 44.182Number of observations 90,015 90,015 90,015 90,015Age 55-64: All markets(2) Immigration rate:All years-0.085(0.088)-0.137(0.097)0.011(0.106)-0.151(0.137)F-test of excluded IV 45.628 88.300Number of observations 206,400 206,400 206,400 206,400Age 25-54: First-tier markets [1](3) Immigration rate:All years-0.275***(0.104)-0.248***(0.083)0.051(0.119)-0.074(0.181)F-test of excluded IV 19.165 47.716Number of observations 479,100 479,100 479,100 479,100Age 25-54: All markets(4) Immigration rate:All years-0.220**(0.085)-0.245***(0.087)-0.053(0.095)-0.200(0.134)F-test of excluded IV 40.801 78.769Number of observations 1,102,345 1,102,345 1,102,345 1,102,345Control set 1 Yes No Yes NoControl set 2 No Yes No YesThis table shows the mobility regression results for the sample period 1991-2011, using expression (3.22). All re-gression models contain year and MA effects. Control set 1 includes the change in employment rate, as well as thefirst-differenced and lagged values of socioeconomic and housing variables. Control set 2 includes lagged values ofemployment rate, socioeconomic factors, and housing-related variables. These models are weighted by Census house-hold weights. The IV regressions use the enclave instrument. Standard errors are in parentheses. *** Significant at1%; ** significant at 5%; * significant at 10%.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitanareas.Sources: Statistics Canada’s 1986 - 2006 Censuses; 2011 National Household Survey; and Postal CodeOM ConversionFile (various dates), which is based on data licensed from Canada Post Corporation; and author’s calculations.105Table 3.7: Immigration inflows and mobility decisions for higher income natives who have already been outof the labour force for at least two yearsNative homeowners:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if therespondent has relocated to another census subdivision between time t−5 and time t.OLS OLS IV IVIndependent variables: (1) (2) (3) (4)Age 55-64: First-tier markets [1](1) Immigration rate:All years0.188**(0.087)0.161**(0.066)0.351**(0.177)0.254(0.167)F-test of excluded IV 18.401 38.251Number of observations 21,765 21,765 21,765 21,765Age 55-64: All markets(2) Immigration rate:All years-0.019(0.055)-0.015(0.054)0.273***(0.106)0.190*(0.101)F-test of excluded IV 40.410 47.222Number of observations 63,865 63,865 63,865 63,865Age 25-54: First-tier markets [1](3) Immigration rate:All years-0.141(0.133)-0.059(0.099)0.051(0.157)0.195(0.131)F-test of excluded IV 23.754 59.334Number of observations 10,085 10,085 10,085 10,085Age 25-54: All markets(4) Immigration rate:All years0.046(0.087)0.044(0.081)0.239**(0.114)0.274***(0.102)F-test of excluded IV 42.342 47.700Number of observations 29,615 29,615 29,615 29,615Control set 1 Yes No Yes NoControl set 2 No Yes No Yes(continued on next page...)106Table 3.7: Immigration inflows and mobility decisions for higher income natives who have already been outof the labour force for at least two years (continued)Native renters:Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if therespondent has relocated to another census subdivision between time t−5 and time t.OLS OLS IV IVIndependent variables: (1) (2) (3) (4)Age 55-64: First-tier markets [1](1) Immigration rate:All years0.264(0.297)0.243(0.315)0.856**(0.392)0.600(0.521)F-test of excluded IV 23.170 44.019Number of observations 3,015 3,015 3,015 3,015Age 55-64: All markets(2) Immigration rate:All years-0.122(0.235)-0.211(0.218)0.257(0.266)-0.094(0.257)F-test of excluded IV 38.452 62.820Number of observations 6,480 6,480 6,480 6,480Age 25-54: First-tier markets [1](3) Immigration rate:All years-0.236(0.361)-0.215(0.381)0.209(0.605)-0.360(0.762)F-test of excluded IV 16.752 38.736Number of observations 2,050 2,050 2,050 2,050Age 25-54: All markets(4) Immigration rate:All years0.307(0.239)0.088(0.269)0.765*(0.395)-0.017(0.409)F-test of excluded IV 38.245 68.510Number of observations 4,655 4,655 4,655 4,655Control set 1 Yes No Yes NoControl set 2 No Yes No YesThis table shows the mobility regression results for the sample period 1991-2006, using expression (3.22). All re-gression models contain year and MA effects. Control set 1 includes the change in employment rate, as well as thefirst-differenced and lagged values of socioeconomic and housing variables. Control set 2 includes lagged values ofemployment rate, socioeconomic factors, and housing-related variables. These models are weighted by Census house-hold weights. The IV regressions use the enclave instrument. Standard errors are in parentheses. *** Significant at1%; ** significant at 5%; * significant at 10%.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitanareas.Sources: Statistics Canada’s 1986 - 2006 Censuses; and Postal CodeOM Conversion File (various dates), which isbased on data licensed from Canada Post Corporation; and author’s calculations.107Table 3.8: Which ethnic groups drive native out-migration?Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent has relocated to another census subdivisionbetween time t−5 and time t.Age 55-64 at the time of migrationNative homeownersFirst-tier markets [1] All marketsIndependent variables:All sample High-income [2] All sample High-income [2](1) (2) (3) (4) (5) (6) (7) (8)(1) Europe excluding U.K. 0.129(0.222)0.143(0.208)0.237(0.367)0.247(0.300)0.042(0.120)0.036(0.121)0.051(0.222)0.040(0.218)(2) Africa 0.566(0.542)0.773(0.654)0.383(1.197)1.002(1.337)0.578(0.475)0.629(0.480)1.078(0.825)1.162(0.870)(3) Central and South America -0.178(0.361)-0.616(0.372)0.769(0.613)-0.023(0.660)0.348(0.231)0.125(0.264)1.230***(0.450)0.750*(0.438)(4) Asia excluding Hong Kong 0.277**(0.121)0.287**(0.119)0.192(0.147)0.280*(0.155)0.002(0.118)0.053(0.118)-0.043(0.138)0.064(0.124)(5) Hong Kong 0.746***(0.255)0.507**(0.246)0.272(0.308)-0.045(0.323)0.863***(0.274)0.738***(0.267)0.357(0.265)0.144(0.244)(6) U.S., U.K., and the Oceania countries -0.752**(0.340)-0.583*(0.311)-1.388**(0.635)-1.346**(0.567)-0.250***(0.076)-0.231***(0.078)-0.292**(0.141)-0.258*(0.138)Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 124,605 124,605 21,765 21,765 382,685 382,685 63,865 63,865(continued on next page...)108Table 3.8: Which ethnic groups drive native out-migration? (continued)Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent has relocated to another census subdivisionbetween time t−5 and time t.Age 55-64 at the time of migrationNative rentersFirst-tier markets [1] All marketsIndependent variables:All sample High-income [2] All sample High-income [2](1) (2) (3) (4) (5) (6) (7) (8)(1) Europe excluding U.K. -0.771(0.548)-0.410(0.469)-2.265(1.406)-1.023(1.213)-0.424(0.379)-0.242(0.400)-1.657*(0.969)-1.305(0.980)(2) Africa 0.550(0.744)0.407(0.732)2.509(2.449)0.265(2.620)0.485(0.667)0.099(0.620)-1.119(2.146)-1.966(2.195)(3) Central and South America 1.069(0.818)-0.105(0.849)2.869(2.782)0.355(2.736)1.209**(0.546)0.235(0.523)1.027(1.893)0.459(1.791)(4) Asia excluding Hong Kong -0.022(0.226)-0.024(0.250)0.315(0.739)0.361(0.788)-0.274(0.193)-0.232(0.211)0.207(0.569)0.023(0.586)(5) Hong Kong -0.079(0.415)-0.251(0.402)0.828(1.696)0.530(1.559)0.335(0.383)0.091(0.375)0.732(1.427)0.628(1.364)(6) U.S., U.K., and the Oceania countries -2.574***(0.797)-0.824(0.705)-2.597(3.141)1.233(2.557)-0.377(0.344)-0.039(0.329)-1.849(1.449)-0.698(1.330)Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 69,850 69,850 3,015 3,015 158,120 158,120 6,480 6,480(continued on next page...)109Table 3.8: Which ethnic groups drive native out-migration? (continued)Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent has relocated to another census subdivisionbetween time t−5 and time t.Age 25-54 at the time of migrationNative homeownersFirst-tier markets [1] All marketsIndependent variables:All sample High-income [2] All sample High-income [2](1) (2) (3) (4) (5) (6) (7) (8)(1) Europe excluding U.K. -0.584*(0.310)-0.485(0.308)-0.654(0.471)-0.422(0.410)-0.235(0.174)-0.215(0.174)-0.193(0.335)-0.252(0.313)(2) Africa 1.221**(0.607)1.493**(0.718)2.062(1.397)2.004(1.429)1.377*(0.747)1.504**(0.732)1.743(1.271)1.651(1.303)(3) Central and South America 0.305(0.420)0.131(0.429)-0.625(0.676)-1.226*(0.653)0.867**(0.349)0.815**(0.384)0.340(0.585)0.232(0.606)(4) Asia excluding Hong Kong 0.193*(0.111)0.183*(0.107)-0.226(0.282)-0.092(0.283)-0.133(0.114)-0.082(0.110)-0.371*(0.208)-0.318(0.210)(5) Hong Kong 1.064***(0.264)1.016***(0.296)1.134*(0.634)1.113*(0.628)1.302***(0.306)1.319***(0.297)1.443***(0.550)1.374***(0.531)(6) U.S., U.K., and the Oceania countries -1.009**(0.446)-0.926**(0.405)-1.963**(0.831)-1.488*(0.816)-0.077(0.105)-0.092(0.105)-0.362(0.311)-0.320(0.323)Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 665,700 665,700 10,085 10,085 1,949,240 1,949,240 29,615 29,615(continued on next page...)110Table 3.8: Which ethnic groups drive native out-migration? (continued)Dependent variable: Binary variable for mobility (Mobilityirt), which equals one if the respondent has relocated to another census subdivisionbetween time t−5 and time t.Age 25-54 at the time of migrationNative rentersFirst-tier markets [1] All marketsIndependent variables:All sample High-income [2] All sample High-income [2](1) (2) (3) (4) (5) (6) (7) (8)(1) Europe excluding U.K. 0.033(0.546)0.116(0.552)2.216(2.004)1.864(1.985)0.628*(0.368)0.629(0.401)0.884(1.349)1.324(1.460)(2) Africa 0.013(0.732)-0.525(0.675)-4.661(3.462)-5.636*(3.387)-0.555(0.964)-1.147(0.852)-1.667(2.956)-3.119(3.096)(3) Central and South America 0.340(0.733)-0.415(0.820)2.301(2.159)2.507(2.438)0.775(0.602)0.184(0.617)2.439(1.621)1.914(1.794)(4) Asia excluding Hong Kong -0.190(0.225)-0.112(0.253)-0.633(0.864)-0.565(0.887)-0.567***(0.205)-0.469**(0.202)-0.133(0.703)-0.309(0.700)(5) Hong Kong -0.168(0.354)-0.221(0.348)0.379(1.445)0.449(1.275)0.520(0.350)0.355(0.320)0.702(1.014)0.426(0.934)(6) U.S., U.K., and the Oceania countries -3.650***(0.747)-2.249***(0.700)-4.513(3.693)-3.581(3.018)-0.637***(0.231)-0.587**(0.243)-0.335(1.559)-0.160(1.542)Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 392,715 392,715 2,050 2,050 904,835 904,835 4,660 4,660This table shows the OLS regressions for the sample period of 1991-2006. The main explanatory variable is the actual immigration rate segregated by the countryof origins. All regression models contain year and MA effects. Control set 1 includes the change in employment rate, as well as the first-differenced and laggedvalues of socioeconomic and housing variables. Control set 2 includes lagged values of employment rate, socioeconomic factors, and housing-related variables.These models are weighted by household weights. Standard errors are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitan areas. [2]: This category includes therespondents who reported a household income greater than $70,000 and who self-reported to be out of the labour force for at least two years.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is based on data licensed from Canada PostCorporation; and author’s calculations.111Table 3.9: Immigration inflows and native wage growthDependent variable: Percentage change in average native wage [1]First-tier markets [2] All marketsAge 55-64 Age 25-54 Age 55-64 Age 25-54OLS IV IV OLS IV IV OLS IV IV OLS IV IVIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Homeowners(1) Immigration rate:0 - 5 years-0.518(0.314)-0.659(0.416)-0.532(0.380)0.088(0.083)0.297(0.198)0.359*(0.190)-0.210(0.211)-0.743(0.473)-0.594(0.450)0.156***(0.058)0.081(0.183)0.113(0.182)F-test of excluded IV 84.750 62.404 84.750 62.404 97.259 77.473 97.319 77.618(2) Immigration rate:All years0.022(0.506)-0.613(0.399)-0.543(0.396)-0.110(0.092)0.276(0.189)0.366*(0.199)0.136(0.391)-0.792(0.520)-0.658(0.513)0.022(0.094)0.086(0.194)0.125(0.200)F-test of excluded IV 68.258 54.138 68.258 54.138 71.763 56.778 71.862 56.880MA effects Yes Yes No Yes Yes No Yes Yes No Yes Yes NoCSD effects No No Yes No No Yes No No Yes No No YesNumber of observations 455 455 455 455 455 455 1,980 1,980 1,980 1,990 1,990 1,990(continued on next page...)112Table 3.9: Immigration inflows and native wage growth (continued)Dependent variable: Percentage change in average native wage [1]First-tier markets [2] All marketsAge 55-64 Age 25-54 Age 55-64 Age 25-54OLS IV IV OLS IV IV OLS IV IV OLS IV IVIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Renters(1) Immigration rate:0 - 5 years-0.301(0.407)-1.182*(0.615)-0.623(0.488)0.438**(0.212)0.591***(0.172)0.651***(0.156)3.069(3.045)3.522(4.025)4.038(4.346)0.463***(0.169)0.389*(0.207)0.449**(0.183)F-test of excluded IV 43.290 31.921 44.080 33.193 75.508 59.989 77.737 63.232(2) Immigration rate:All years-0.628(0.396)-1.241*(0.641)-0.658(0.518)0.224(0.147)0.614***(0.179)0.689***(0.154)2.905(3.270)3.990(4.565)4.565(4.925)0.378**(0.149)0.441*(0.228)0.509**(0.200)F-test of excluded IV 40.554 28.696 41.621 29.789 66.793 52.927 68.838 55.385MA effects Yes Yes No Yes Yes No Yes Yes No Yes Yes NoCSD effects No No Yes No No Yes No No Yes No No YesNumber of observations 400 400 400 455 455 455 1,375 1,375 1,375 1,965 1,965 1,965This table shows the regression results for the sample period 1991-2006, using expression (3.23). All regression models contain year dummies. The models areweighted by the number of owned or rental dwellings per census subdivisions. The instrumental variable regression models apply the enclave IV. See text for detailsregarding the instrument. Standard errors are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%. [1]: Zero wages are not included inthe computation. [2]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitan areas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is based on data licensed from Canada PostCorporation; and author’s calculations.113Table 3.10: Linking empirical results to theoretical modelAge 55-64 Age 25-54All sampleHigh - incomeNot in labour forceAll sampleHigh - incomeNot in labour forceFirst-tiermarketsAllmarketsFirst-tiermarketsAllmarketsFirst-tiermarketsAllmarketsFirst-tiermarketsAllmarketsNative homeownersDistaste(δ > 0)↑↑ ↑↑ −− or ↓ ↑↑ ↑↑ ↑↑ ↑↑ ↑↑Housing cost growth( dPdI > 0 ordRdI > 0)↑↑ ↑ ↑↑ ↑ ↑↑ ↑ ↑↑ ↑Labour substitution(ρ > 0)−− −− N/A N/A −− −− N/A N/ANet housing wealth + residual effects ↓ ↓ ↓ ↓ ↑ or ↓ ↑ or ↓ ↓↓ ↑ or ↓Net out-migration intentions ↑ ↑ ↑ ↑ ↑↑ ↑↑ −− ↑Native rentersDistaste(δ > 0)↓↓ ↑ −− −− ↓↓ ↓↓ −− −−Housing cost growth( dPdI > 0 ordRdI > 0)↑↑ ↑↑ ↑↑ ↑↑ ↑↑ ↑↑ ↑↑ ↑↑Labour substitution(ρ > 0)↑ −− N/A N/A ↓↓ ↓↓ N/A N/AOther residual effects ↓ ↓↓ −− ↓↓ ↑↑ ↑↑ ↓↓ ↓↓Net out-migration intentions −− −− ↑↑ −− −− −− −− −−This table summarizes the findings from the mobility regression models for the sample period 1991-2006. “↑” denotes slight intentions to out-migrate, whereas “↓”represents small preference to stay in the same CSD. “↑↑” denotes large intentions to relocate and “↓↓” gives the opposite effect. “−−” means that the results weregenerally insignificant. The blue rows capture the residual effect that cannot be computed empirically. First-tier markets include the census subdivisions withinVancouver, Toronto, and Montreal census metropolitan areas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is based on data licensed from Canada PostCorporation; and author’s calculations.114Table 3.11: Weighted average predicted probabilities of out-migrationNative homeownersFirst-tier markets [1] All marketsAge groups at the time ofmigration:Age 55-64 Age 25-54 Age 55-64 Age 25-54Mobility categories: (1) (2) (3) (4) (5) (6) (7) (8)Native homeownersSame CSD, but differentdwelling42.15% 42.15% 50.81% 50.82% 56.64% 56.66% 63.26% 63.26%Different CSD, but sameMA31.79% 31.80% 34.07% 34.07% 20.22% 20.21% 20.29% 20.29%Different MA, sameprovince20.66% 20.67% 10.70% 10.70% 16.19% 16.19% 10.51% 10.51%Different province 5.39% 5.38% 4.42% 4.42% 6.95% 6.94% 5.94% 5.94%Number of observations 21,170 21,170 261,195 261,195 54,535 54,535 696,860 696,860Native rentersSame CSD, but differentdwelling69.57% 69.57% 67.17% 67.17% 75.25% 75.25% 69.17% 69.18%Different CSD, but sameMA20.48% 20.48% 21.19% 21.19% 12.50% 12.51% 12.98% 12.98%Different MA, sameprovince6.74% 6.74% 6.42% 6.42% 8.00% 8.00% 10.29% 10.29%Different province 3.21% 3.21% 5.22% 5.21% 4.25% 4.24% 7.55% 7.55%Number of observations 26,610 26,610 243,160 243,160 65,940 65,940 591,535 591,535Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesThis table shows the weighted average predicted probabilities of out-migration for native households, segregatedby mobility categories for the sample period of 1991-2006. These estimates follow from expression (3.25). Allregression models contain year and MA effects. Control set 1 includes the change in employment rate, as well asthe first-differenced and lagged values of socioeconomic and housing variables. Control set 2 includes lagged valuesof employment rate, socioeconomic factors, and housing-related variables. These models are weighted by Censushousehold weights.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitanareas.Sources: Statistics Canada’s 1986 - 2006 Censuses; and Postal CodeOM Conversion File (various dates), which isbased on data licensed from Canada Post Corporation; and author’s calculations.115Table 3.12: Synthetic panel analysisDependent variable: Percentage change in average value of dwellingFirst-tier markets [1] All marketsOLS OLS IV IV OLS OLS IV IVIndependent variables: (1) (2) (3) (4) (5) (6) (7) (8)(1) Immigration rate:0 - 5 years1.284***(0.435)1.893***(0.567)1.071(0.726)1.700*(0.920)0.698*(0.383)0.684*(0.405)0.724(0.543)0.076(0.754)F-test of excluded IV 45.758 70.903 103.078 95.068(2) Immigration rate:All years0.863**(0.358)0.866**(0.433)1.013(0.662)1.955*(1.077)0.511*(0.299)0.486*(0.291)0.780(0.570)0.088(0.867)F-test of excluded IV 41.469 85.608 66.180 62.158Control set 1 Yes No Yes No Yes No Yes NoControl set 2 No Yes No Yes No Yes No YesNumber of observations 1,275 1,275 1,275 1,275 4,715 4,715 4,715 4,715This table shows the results for the synthetic panel analysis for the sample period 1991-2006, using expression (3.26).All regression models contain year and MA effects. Control set 1 includes the change in employment rate, as well asthe first-differenced and lagged values of socioeconomic and housing variables. Control set 2 includes lagged valuesof employment rate, socioeconomic factors, and housing-related variables. These models are weighted by the numberof households who didn’t move at the CSD-cohort-year level. The IV regressions use the enclave instrument. Standarderrors are in parentheses. *** Significant at 1%; ** significant at 5%; * significant at 10%.[1]: First-tier markets include the census subdivisions within Vancouver, Toronto, and Montreal census metropolitanareas.Sources: Statistics Canada’s 1986 - 2006 Censuses and Postal CodeOM Conversion File (various dates), which is basedon data licensed from Canada Post Corporation; and author’s calculations.116Chapter 4Housing supply elasticity and the elderly4.1 IntroductionAs the household debt-to-income ratio skyrocketed during the housing boom period (1998-2006) for allof the western economies, policymakers and researchers have raised concerns over households’ financialpositions and over the labour supply decisions of individuals who are approaching retirement.80 In particular,policymakers, researchers, and the media have pointed to the idea that housing value and household wealthare closely linked, and any changes in household wealth should alter labour supply behaviour.81 The formerDeputy Governor of the Bank of Canada, Jean Boivin mentioned that the high household debt level is one ofthe pressure points for the aging group; and older individuals will need to save more and work more in orderto maintain current living standards (Boivin, 2012). The New York Times and the Federal Reserve Bankof San Francisco’s Economic Letter reported that more people are delaying retirement and the increase inlabour force participation rates for workers aged 55 and over is due to the sharp declines in housing equityneeded to cushion the recent financial devastation (Daly et al., 2009; New York Times, 2012). These speechesand reports suggest that housing, household wealth, and labour supply decisions are inter-related.Economic theory predicts that any increase in the present value of lifetime resources via a positive wealthshock should translate to an increase in consumption. The static first order condition from the householdoptimization problem suggests that additional utility derived from this extra consumption is matched withan additional pain from working more, implying that a positive wealth shock should motivate near-retireesto retire early. Yet, aggregate data does not show near-retirees to work less during the housing boom period.Figure 4.1 shows the aggregate U.S. labour force participation rate for individuals aged 55-64 to be increas-ing along with the debt-to-income ratio during the house price run-up period, which appears to contradicteconomic theory. Empirical analysis that focuses on relating house values and labour supply decisions ofnear-retirees is still a relatively young area of research. Goodstein (2007) is one of the first to examine howchanges in house prices affect retirement behaviour, which suggests that male labour force participation rateand household wealth are inversely-related. Conversely, Shan (2008) finds property tax, which is a func-tion of house value, to exert no effect on labour supply decisions of the near-elderly. Paradoxical findingsfrom aggregate data and mixed conclusions from empirical work provide motivation for using micro-data toexplore how house value affects the labour supply decisions of older households.This paper uses both descriptive and econometric analyses to investigate how the recent housing boomand bust affected the labour supply decisions of U.S. older households. This research deviates from the80See for example, Chart 21 of Bank of Canada’s Financial System Review (June 2009 edition).81See for example, Bernanke (2012); Bies (2005, 2006); Ferguson (2005); and Kartashova and Tomlin (2013).117estimations in Begley and Chan (2015), in Burge and Zhao (forthcoming), in Coile and Levine (2011), andin Farnham and Sevak (2015) when comparing the labour supply responses of individuals who are more andless exposed to a house price shock. These papers address endogeneity problems that surround the houseprice variable through: (1) using renters as a quasi-control group in order to difference out the change inretirement decisions due to unobserved variations in local amenities that were capitalized into local housingvalues; and/or (2) applying an auto-regressive model to estimate a measure of unanticipated change inhouse prices. However, both bullets (1) and (2) are not going to fully solve the endogeneity problems.First, renters’ labour supply decisions are not fully immune from local house price changes. Local rentsand housing values are expected to move hand-in-hand during economic cycles, which means that renterscould adjust their labour supply behaviour from an affordability perspective. Moreover, overall economicconditions could encourage homeowners to extract housing equity by selling off their dwellings and thenrenting a new home. In this case, retirement behaviour is expected to be confounded by this dwelling tenureswitch. Therefore, renters are not an appropriate comparison group for estimating the effect of housingwealth on labour supply decisions. For bullet (2), a potential concern is that unobserved changes in localamenities not only influence current housing value, but will also affect future housing prices. Therefore, thepredicted house price index that is based on past housing prices is still expected to be correlated with thecontemporaneous error term.To address these potential estimation concerns, I apply the land topology-based measure of housingsupply elasticities introduced by Saiz (2010) as an exogenous source of variation in house price growth,in order to compare the labour supply decisions of near-retirees who are more and less exposed to housingmarket conditions. I also interact the inverse of the regional housing supply elasticities with the U.S. nationallending conditions for residential mortgages as an instrumental variable strategy to estimate the impactfrom housing value growth on this sub-population’s labour supply behaviour. This instrument exploitsthe differential impact from the housing and the national lending conditions across regions and over time.The amount of easing or tightening in lending conditions is expected to be magnified in regions with lowhousing supply elasticities, as these regions are much more sensitive to house price fluctuations. This paperuses the following two major micro-datasets to investigate this context: (1) RAND Health and RetirementStudy (RAND HRS), produced by the RAND Center for the Study of Aging; and (2) U.S. Census Bureau’s(Department of Housing and Urban Development) American Housing Survey (AHS).This paper contributes to the existing literature in several respects. The empirical analysis provides noevidence that the recent housing market fluctuations affected the near-retirees’ retirement and work arrange-ments. Individuals also do not seem to respond asymmetrically to positive and negative wealth shocks. Thisresearch also sheds light on the popular claim by showing that housing value growth does not exert anysignificant influence on changes in net total household wealth. The household wealth distribution is skewed.The amount of net housing wealth for the median person is only half of that for the average person. Thisimplies that the effect of a house price shock on the median near-retiree is expected to be modest relative tothat for the average person.To the best of my knowledge, although central banks use the results from the credit condition surveysas part of the information set for monetary policy and financial stability assessments, this is the first paper118to apply the national lending conditions for residential mortgage series as part of an instrumental variablestrategy to measure the effect of a house price shock on household’s labour supply decisions.82 Moreover,I extend Coile and Levine’s (2006) work by expanding the set of households to those with homeownership.Coile and Levine (2006) argue that the stock market performance had minimal impact on the labour supplyof individuals aged 55-70 because less than half of the individuals had any major stock holdings. On theother hand, roughly 80% of Americans aged 55-64 have at least one house. Extending their argument further,the large proportion of homeownership implies that the housing boom and bust should have some impact onthe near-retirees. Begley and Chan (2015) also convey a similar message in their work. Yet, I show in thispaper that this argument does not hold: the proportion of households with homeownership does not conveyany information about the relationship between labour supply behaviour of the near-retirees and house pricegrowth.4.2 Background informationThe recent housing boom and bust has led to numerous works exploring the effect of house price on house-hold wealth. These studies generally suggest that the house price run-up and household wealth are inter-related, and such interactions then influence household consumption and employment figures. For example,the World Economic Outlook published by the International Monetary Fund shows that housing bust alongwith large run-up in gross household debt are associated with a greater reduction in household consumptionand a higher unemployment rate (Leigh et al., 2012). Other papers such as Baker (2015) and Christelis etal. (2014) also illustrate similar findings and show that liquidity constraints also drive the results.Other researchers have found that the housing boom contributed to a rise in home equity extraction.A majority of borrowed funds from this collateral channel was directed towards home renovations andconsumption (Bailliu et al., 2012; Kartashova and Tomlin, 2013; Mian and Sufi, 2011). Adelino et al.(2015) also find small businesses that are located in areas with larger house price appreciation hire moreworkers than large firms, because the housing boom facilitated these small firms to use borrowed fundsfrom the housing collateral channel for such expenditure. Moreover, using U.K. data, Campbell and Cocco(2007) show consumption for older households to be more sensitive to house prices.While numerous works have emphasized the effect of housing on household consumption and aggregateemployment activities, policymakers, researchers, and the media have also noted that the housing boom andbust also exerted impact on the labour supply decisions of the aging population. Economic theory predictsthat changes in net household wealth are associated with changes in labour supply. Yet, the results from thisarea of research are mixed.Several studies have utilized the Health and Retirement Study (HRS) dataset to investigate the linkagesbetween housing and elderly labour supply behaviour. Using house price growth and inheritance as instru-ments, Goodstein (2007) shows that a $20,000 increase in household wealth between years 1996 and 2004to be linked to a reduction of 1% in the labour force participation rate of men born between years 1931-1941. Ondrich and Falevich (2016) apply a proportional hazard model to illustrate that declines in housing82See for example, Faruqui et al. (2008).119wealth during the Great Recession lowered retirement probabilities of married males by as much as 14 to 17percent, but the defined benefit or contribution pensions could diminish some of these effects.In addition, a few papers have shown that female labour supply decisions are more responsive to changesin housing wealth relative to men. For example, Burge and Zhao (forthcoming) include both housing wealthand property tax liabilities in regression models and find that this result holds at the extensive margin oflabour participation. Begley and Chan (2015) demonstrate that borrowing and liquidity constraints couldinfluence labour supply decisions of certain sub-population groups. Specifically, women who experiencednegative house price shocks are 25% less likely to retire relative to those who experienced positive shocks,but the same effect does not apply to men. They also illustrate that retired homeowners are 50% more likelyto re-enter into the labour force or to increase hours worked in response to a negative housing shock. Fu etal. (2016) also find similar results using the 2011 China Household Finance Survey, where a 100,000 yuanincrease in housing wealth is associated with a decline in female’s labour force participation rate by 1.37percentage points and with an increase in the probability of becoming housewives of 1.49 percentage points.Yet, male’s labour force participation decisions are invariant to house price shocks.On the contrary, a handful of papers has also illustrated that the relationship between household wealthand retirement decisions is weak.83 For example, Farnham and Sevak (2015) show using the HRS data thatthe annual probability of retirement is statistically unaffected by housing capital gains. They find that a 10%increase in housing wealth is associated with the expected retirement age to be pushed back earlier by 4months. Using the HRS data, Shan (2008) illustrates that property tax, which is a function of house value,exert insignificant impact onto the elderly household’s retirement decisions. Moreover, other researchershave suggested that the financial crisis has a minimal influence on labour supply decisions of the elderly(Crawford, 2013; Gustman et al., 2011). In particular, stock market fluctuations and retirement behaviourare weakly correlated (Coile and Levine, 2006; Goda et al., 2012). Begley and Chan (2015) and Coile andLevine (2006) argue that the stock market should not affect the median person because fewer than half ofthe older households had any major stock investments. Begley and Chan (2015) suggest that homeownersmake up a large fraction of the older adult population and therefore it is more appropriate to explore theeffect of a wealth shock on labour market participation through a housing context. However, other papershave illustrated that individuals tend to be more responsive to local labour market fluctuations than to wealtheffects (Coile and Levine, 2011; Disney et al., 2015; Goda et al., 2011). Goda et al. (2012) and Goodstein(2007) suggest that the exclusion of risk aversion-related parameters along with measurement errors may bethe cause for having the estimated wealth effects to be biased towards zero.Overall, the papers on household finance and house value seem to be providing mixed conclusions. Onone hand, house value and household debt are heavily related to consumption and aggregate employment.On the other hand, the effect of housing and other forms of household wealth on individual labour supplydecisions is ambiguous. This raises the question of whether the current housing boom is a sideshow, froman individual household’s perspective.84 Skinner (1996) suggests that housing is not completely a sideshow83Using the 2012 Survey of Financial Security data, Amedah and Fougère (2016) showed at the 2016 Canadian EconomicsAssociation conference that this finding holds for the Canadian context.84Skinner (1996) states the following to describe what he means by “sideshow”: “Housing prices may decline by 47%, but ifyounger homeowners don’t save more in response to the price decline, and if retired homeowners don’t touch their housing equity,120due to precautionary savings motive. Under the precautionary savings motive, housing equity serves asinsurance for the household and few households would liquidate real estate assets in response to bad stateof the world. This provides some intuition behind why the median person is not affected by housing marketconditions. Furthermore, Costa (1998) and Munnell (2011) suggest that individuals may choose to workuntil the “official” retirement age in order to benefit from Social Security that starts at retirement age andfrom the employer’s health coverage that ends by retirement age. This implies that any forms of wealthshock would exert minimal impact onto the near-retirees’ labour supply decisions.Given these contradictory findings, it remains an open question of how the recent housing market varia-tions affected the labour supply decisions and the household wealth of older individuals. This is an importantresearch given the sizeable number of Baby Boomers entering retirement with improved health and longevityalong with house price booms that are currently happening in other countries around the world.4.3 Data descriptionThe American Housing Survey (AHS) and the RAND Health and Retirement Study (RAND HRS) are theprimary data sources for this paper.85 The national version of the AHS is a biennial survey on housingcharacteristics, such as housing quality and housing cost, that publishes in every odd-number year. TheRAND HRS is a national longitudinal household survey that contains health, demographic, income, asset,debt, employment, housing, and family structure-related data for households with at least one U.S. individualaged 51 and over. According to the RAND HRS codebook, “[t]he data include any individual interviewedat least once. This includes individuals who were age-eligible (born in eligible years) at the time of theirfirst interview, spouses that were not age-eligible at baseline, and spouses who married an age-eligiblerespondent between survey waves” (page 11). RAND HRS allocates zero weights to respondents who livein nursing homes. The study has 13 waves of data for years 1992, 1993, 1994, 1995, 1996, 1998, 2000, 2002,2004, 2006, 2008, 2010, and 2012. Overall, the AHS and RAND HRS data are nationally representative.For example, the value of primary residence series from both datasets follows closely to the aggregate U.S.FHFA House Price Index series.86The AHS and RAND HRS datasets complement each other. First, the AHS publishes data at the MSAlevel, which can more precisely compare the near-retirees’ labour supply decisions who are more and lessexposed to housing market fluctuations within metropolitan area levels. The AHS also provides a widevariety of housing quality measures, which can better quantify the amenities that may be priced into housingvalue, the explanatory variable of interest. However, the lack of information on households’ labour supplythen the price change will have little impact on overall welfare. In short, trends and fluctuations in house prices and housing equitywould be just a sideshow” (page 242).85The RAND HRS is developed at RAND with funding from the National Institute on Aging and the Social Security Adminis-tration (October 2015). This is the user-friendly version of the HRS data86Wealth-related variables are converted to year 2012 dollars for the RAND HRS dataset and to year 2013 dollars for the AHSdataset, using the Consumer Price Index series retrieved from the Bureau of Labour Statistics. The series identifier for the Con-sumer Price Index is CPIAUCSK. I excluded the records that provided incomplete and inconsistent answers for the wealth data.Incomplete answers include records that provided incomplete bracket answer, reported not knowing ownership, or provided nofinancial response. Inconsistent answers include respondents who reported zero value for primary residence (of which I assume nohomeownership), but reported positive outstanding mortgage and/or home equity borrowing balances.121activities is one of the major drawbacks to using the AHS data for this study. The dataset does not containany data on the intensive margins of labour supply. To estimate the effect on the extensive margins of laboursupply, I generate a dummy variable that equals one if the respondent switched from a non-zero to a zeroannual wage amount. I use this measure to proxy labour force exit.On the other hand, the RAND HRS dataset allows a more comprehensive analysis of labour supplydecisions. The RAND HRS contains various retirement-related measures. In this study, I apply the totalhours worked per week and the total weeks worked per year variables to estimate the effect of housing onthe intensive margins of labour supply.87 For the extensive margins, I rely on the labour force participationand the self-assessed retired variables from the RAND HRS. As part of the robustness check, I also generatea dummy that equals one if the annual wage switches from a non-zero to a zero value using the RAND HRSdata to see if the results are comparable to those generated by the AHS and to identify whether the resultscould be confounded by household’s mobility decisions. Furthermore, RAND HRS covers health insurance,pension, and bequest-related variables, and any of these factors could influence labour supply behaviour.These variables cannot be located in the AHS.Although RAND HRS seems to be a more preferable choice over the AHS, RAND HRS is also subjectto limitations. First, the data is available at the census division level, so the comparison of labour marketattachments between high and low elasticity regions is less precise. Second, the RAND HRS does notindicate household’s mobility status, so it is ambiguous whether the respondent stayed in the same residencethrough the life of the panel. Finally, this dataset does not contain any information on housing quality.In addition to the RAND HRS and the AHS, this study also uses a combination of data from Saiz (2010),from the Federal Reserve’s Senior Loan Officer Opinion Survey on Bank Lending Practices (SLOOS), andfrom the U.S. Bureau of Labor Statistics for the econometric analyses. Specifically, I use the housing supplyelasticities from Saiz (2010) to proxy the regions with high and low housing price fluctuations. Thesehousing supply elasticities are based on geographical and regulatory restrictions to new constructions. Inaddition to the difference-in-difference estimation, I also apply these elasticities to the instrumental variableapproach, by interacting the inverse of these elasticities with the SLOOS’s national lending condition forresidential mortgages.88 Currently, the SLOOS surveys up to 60 large domestically chartered commercialbanks and up to 24 large U.S. branches and agencies of foreign banks. The selection of respondents dependson the size of the financial institution, geographic coverage, and mutual independence (i.e. eliminate a bankif it is a subsidiary of another bank that is already in the panel). The lending conditions questions focus onthe changes in bank’s credit standards over the past three months. Prior to 2007Q2, the residential mortgageseries include prime mortgages, non-traditional mortgages, and subprime mortgages. The lending conditionseries is reported at an aggregate level, and is computed as the percentage of banks reporting tightenedcredit conditions minus the percentage reporting eased credit conditions. A positive value for this seriescorresponds to a net tightening of credit standards. The data frequencies for the RAND HRS, the AHS, and87I also remove records that responded working 168 hours per week (24 hours per day, 7 days per week; and misreportedretirement status. For example, individuals reported “retired” in the labour force question, but reported “not retired” in the retirementplan question or vice versa.88From the SLOOS, I use “net percentage of domestic respondents tightening standards for mortgage loans” series to capture thelending conditions for residential mortgages.122the SLOOS are different, where the RAND HRS and the AHS are biennial and SLOOS is quarterly. Giventhis, I first compute the annual average for the SLOOS series and then insert the annual SLOOS data intothe RAND HRS dataset, matched by the corresponding year.4.4 Empirical strategyThis study uses both difference-in-difference and instrumental variable approaches to explore the effect of ahouse price shock on labour supply decisions of near-retirees. I define near-retirees as individuals of ages 55-64. The impact from the housing market should be greater for agents who reside in the low elasticity regionsthan those who live in the high elasticity regions. Regions with low elasticities are generally geographicallyand/or regulatory constraint. For example, during the boom period, adding a new construction is expectedto be more challenging in inelastic regions, which should translate to high house prices and collateral values(Adelino et al., 2015; Mian and Sufi, 2011; Saiz, 2010). This implies that agents who reside in the inelasticregions should be much more sensitive to house price variations, which theoretically should translate togreater labour supply responses.Using the American Housing Survey (AHS) and the RAND Health and Retirement Study (RAND HRS)datasets, Figures 4.2 and 4.3 also provide evidence that housing supply elasticities and house prices arestrongly correlated. Relative to the high elasticity regions, Figure 4.2 illustrates that the low elasticity regionsexperienced much larger volatility in housing value during the boom (2002-2006) and the bust periods (afteryear 2007). On the other hand, the RAND HRS data does not reveal any concrete evidence that housingsupply elasticities are associated with fluctuations in non-housing components, such as consumer credit(Figure 4.3).89 These graphical results imply that the differences between the high and the low elasticityregions are driven primarily by housing-related factors, which are consistent with the findings from otherempirical work.90Given this trend, the empirical strategy uses the pre-existing regional housing supply elasticities toidentify the effect of the house price shock on near-retirees’ labour market attachment. The pre-existingregional housing supply elasticities are taken from Saiz (2010), which exploit the differences in geographicaland regulatory constraints across the regions. I start the analysis with a difference-in-difference method,where I compare the labour supply decision of individuals who reside in the low elasticity regions for thebefore and after periods, and then compare the result to the same time difference for those who reside inthe high elasticity region. This study then extends the estimation to an instrumental variable (IV) approachin order to account for the intensity of the treatment. The IV reflects the net percentage of tightening inlending standards for residential mortgages at the regional level and it exploits the differential impact fromthe housing and the lending markets across regions and over time.89The consumer credit data are derived from the RAND HRS dataset, which includes respondents of ages 51 and above. Thereader should be cautious that these graphical illustrations are applicable to the context of this paper, which covers individuals ofages 55-64.90For example, Mian and Sufi (2011).1234.4.1 Binary treatment approach: difference-in-difference estimationThis research applies the difference-in-difference method to compare the labour supply decisions of near-retirees who are more and less exposed to housing market fluctuations. The binary treatment is the mostbasic method. It provides transparency on the source of identification, as well as a direct answer to theresearch question of interest. Unlike the continuous treatment approach, this technique facilitates a discretecomparison across different time regimes. The treatment and control groups are based on the regionalhousing supply elasticities from Saiz (2010), where the low elasticity regions belong to the treatment group.I start the econometric analysis by measuring the impact of housing on the extensive margins of laboursupply, using the American Housing Survey (AHS) dataset. The main regression for respondent i at time tis:∆Yit = β0+∑rβ1rregionit +∑tβ2ttimet +β3Xit +θ1Y 0305it +θ2Y 0305it ·T REATit +θ3Y 0711it +θ4Y 0711it ·T REATit + εit (4.1)where respondent i refers to the household head (family respondent) or the head’s spouse. regionit containsthe MSA dummies and timet includes the year effects. I set years 1999-2001 as the “before” period anddefine years 2003-2011 as the “after” period. To identify whether the effect of house prices is asymmetricacross the boom and bust periods, I split the after period into two different variables. Therefore, Y 0305itis a dummy variable that equals one for years 2003-2005 to capture the housing boom period, and Y 0711itis a dummy variable that equals one for years 2007-2011 to capture the bust period. T REATit is a dummyvariable that equals one for individuals living in MSAs with low housing supply elasticities. The maincoefficients of interest are θ2 and θ4. The coefficient θ2 provides the difference-in-difference estimate forthe housing boom period, while θ4 provides the corresponding measure for the housing bust period. If theeffect from a house price shock were to be symmetric, the absolute magnitude of θ2 should be closely similarto that of θ4. The signs for θ2 and θ4 should be opposite from each other to reflect the effect from a positiveand a negative wealth shock, respectively.∆Yit is a labour market attachment measure, which equals one if the respondent transitioned from positiveearnings at time t−1 to zero earnings at time t. I use this indicator to proxy for individuals who exited thelabour force, since the AHS does not provide any information on job characteristics.Figure 4.4 illustrates that the trends for various household characteristics are similar between the treat-ment and control groups, implying that the θ ′s are not picking up effects from household characteristics.Despite this finding, I still incorporate a set of household demographic controls Xit into the regression mod-els. The vector Xit consists of individual demographic characteristics, which includes dummy variablesfor gender, marital status, age, birth cohort, ethnicity, education, and household income. Birth cohorts arebased on the classifications from the RAND HRS dataset.91 Using a similar technique to Kartashova and91RAND HRS defines the birth cohorts as follows: Cohort 1 refers to individuals born before 1924; Cohort 2 refers to individualsborn between years 1924-1930; Cohort 3 refers to individuals born between years 1931-1941; Cohort 4 refers to individuals born124Tomlin’s (2013), I include birth cohort dummies to account for potential heterogeneity across the differentgenerations. I cluster the standard errors at the household level.Labour and retirement outcomes could be different between households with and without homeowner-ship. Thus, I use the following triple difference specification to test whether households that own a houseduring the before period would exhibit different labour and retirement decisions:∆Yit = β0+∑rβ1rregionit +∑tβ2ttimet +β3Xit +γ1HOMEit + γ2HOMEit ·T REATit + γ3T REATit +λ1Y 0305it +λ2Y 0305it ·T REATit +λ3Y 0711it +λ4Y 0711it ·T REATit +θ1Y 0305it ·HOMEit +θ2Y 0305it ·HOMEit ·T REATit +θ3Y 0711it ·HOMEit +θ4Y 0711it ·HOMEit ·T REATit + εit (4.2)In this case, HOMEit is a dummy variable that equals one if the household reported homeownership inyear 1997, the year just prior to the start of the house price run-up. θ2 and θ4 provide the triple differencemeasure, where the changes in the dependent variable are measured relative to the change for the renters ineach MSA.Since the AHS does not provide a comprehensive set of labour market attachment measures, I utilize theRAND HRS dataset to gather more evidence on the effect of house prices on labour supply behaviour. I usethe following five measures taken from the RAND HRS dataset for ∆Yit for this additional analysis:• (1) Labour force exit: I set an indicator variable that equals one if the respondent was in labour forceat time t−1 and exited the labour force at time t.• (2) Self-assessed retirement: I set an indicator variable that equals one if the respondent was notretired or was partially retired at time t−1 and was retired at time t.• (3) Zero earnings: I set an indicator variable that equals one if the respondent earned positive salaryat time t−1 and did not earn any wage at time t. This is used to compare the results against the onescomputed using the AHS dataset.• (4) Percentage change in total hours worked per week.• (5) Percentage change in total weeks worked per year.Unlike the AHS, the RAND HRS is a biennial survey that is published in every even-number year. There-fore, I modify expressions (4.1) and (4.2) by replacing Y 0305it with Y 0206it ; and Y 0711it with Y 0812it . Forthe triple difference specifications, the homeownership dummy depends on the dwelling tenure status fromyears 1992-1996. Furthermore, the RAND HRS only provides data at the census division level. As such, thebetween years 1942-1947; Cohort 5 refers to the Early Baby Boomers born between years 1948-1953; and Cohort 6 refers to theMid Baby Boomers born between years 1954-1959.125regionit vector contains census division effects rather than MSA effects. In addition, with the RAND HRSdata, I also include dummy variables for disability, for bequest motive, and for health insurance to accountfor the possibility that individuals may adjust their retirement plans due to these factors.92For both binary treatment specifications, the error term is assumed to be uncorrelated with the regressors,which means that the household’s unobservable characteristics are independent of the treatment and alsoindependent of the observable characteristics. One of the concerns with this estimation is that householdsmay move from a high elasticity region to a low elasticity region in order to benefit from the house price run-up. Conversely, households may out-migrate from a low elasticity region to a high elasticity region duringthe bust period. Household’s mobility decisions may influence labour supply behaviour. Row (1), column(1) of Figure 4.5 shows that households in elastic regions do seem to move more than those in inelasticregions during the housing boom period. Yet, the reverse trend is not observed during the bust period. Assuch, household’s mobility decisions do not seem to be directly related to housing market fluctuations.The estimations using RAND HRS addresses this concern. Unlike the AHS, where the geographicvariation is at the MSA-level, the RAND HRS provides data at the census division level. Therefore, theestimation would capture the effect within a census division. Given the large size of each census division,it is unlikely that households would choose to move from one census division to another to profit from thehousing market. The high cost of mobility, such as relocation cost, cost of adapting to a new environment,job-search cost, to name a few, deters individuals from doing so at broad geographic levels. Therefore, ifthe results between the RAND HRS and the AHS are similar, this implies that household mobility decisionsare not a confounding factor to the difference-in-difference estimations.Differences in taste for work could also be embedded in the error term and may confound the estimates.Goodstein (2007) suggests that the assumption of setting the two groups of agents with the same taste forwork during the “before” period is unrealistic, and could be one of the drivers for deriving estimation resultsto be biased towards zero. Using the RAND HRS’s income risk aversion measure, I find that the shares ofindividuals with certain risk aversion characteristics are similar between the treatment and control groupsfor 1998-2006. Therefore, the assumption that the two types of individuals have similar work preferenceduring the “before” period is reasonable.93To further address this concern, I also examine how the overall housing quality measures differ betweenthe high and low elasticity regions. The quality of the house more or less reveals the household’s preferencefor lifestyle, which could potentially influence the individual’s taste for work. As Farnham and Sevak(2015) suggest, unobserved changes in local amenities could also be capitalized into local housing values.Any changes in local housing prices could influence retirement timing. Figure 4.5 shows that most of thehousing quality measures are similar between the two locations. This further suggests that taste and amenityeffects are not expected to directly influence labour supply decisions.Most importantly, the common trends assumption is a necessary condition for identification. The validityof θ2 and θ4 requires the underlying trends in the outcome variable to be the same for both the treatmentand control groups. Specifically, the counterfactual trend of the treatment group is the same as the observed92I set the bequest motive dummy equals to one if the respondent reported having a non-zero self-assessed probability of leavinga bequest of over $100,000.93The income risk aversion variable in the RAND HRS dataset is only available for years 1998-2006.126trend of the comparison group.Figures 4.6 to 4.8 illustrate the labour market attachment measures for the household head and his/herspouse, using the AHS and the RAND HRS data. These figures provide evidence that the common trendsassumption holds in most cases. To show that this condition holds, I first assume that the trend before thepeak of the housing boom period (before year 2001) captures the path that the treatment group (low elasticityregion) would have undergone if it had not been treated. Generally, the trends for the treatment group priorto year 2001 closely mimic the trends for the control group. Since the trends prior to the peak of the housingboom period are near identical between the two groups, these figures confirm that the high elasticity regionis a good comparison group for the difference-in-difference method. These observations suggest that thekey assumption required for identification holds.These figures also suggest that the house price shock has minimal impact on the labour and retirementdecisions of near-retirees. The figure shows miniscule differences between the treatment and control linesfor the boom and bust periods. Splitting the sample by dwelling tenure status also does not illustrate thathousing exerted any significant influence onto the labour and retirement decisions of the near-retirees (seeFigure 4.9). On one hand, this seems to be consistent with the findings in Coile and Levine (2006) and inShan (2008). On the other hand, this finding seems counterintuitive to the argument raised by Coile andLevine (2006). Extending the argument made by Coile and Levine (2006) to my findings, the median near-retiree should be affected by housing market variations, as approximately 80% of Americans aged 55-64own at least one house. Yet, the descriptive analysis based on the RAND HRS dataset shows the contrary.Finally, another major concern with this estimation is whether external policies could confound the es-timation. This may be a possible explanation behind why a small number of labour market attachmentvariables do not fully follow common trends in the before period. For example, legal or fiscal policy in-troduced for a specific region could distort the comparison between the low and high elasticity regions.Therefore, this study examines the changes in retirement decisions against the changes in house price, inorder to first-difference out the time-invariant idiosyncratic characteristics.Figure 4.10 shows the comparison trend plots for various measures to determine if the low and highelasticity regions are subject to differential impacts that are not related to housing. If no other regional dif-ferences were present, then the low and high elasticity series would exhibit parallel lines. This also impliesthat the control group (high elasticity region) is a good comparison group for the difference-in-differenceapproach. If structural breaks were observed in any one of these series, then certain policy changes or dis-tinct regional differences unrelated to housing may bias the overall result. I plot the unemployment rate,the employment rate, and the share of population in the construction industry to proxy for regional labourmarket activities; gross domestic product per capita series to proxy for income and productivity; annual realestate tax to proxy for fiscal policies; and the debt delinquency series to proxy for regional credit conditions.Figure 4.10 suggests that the estimation results are not likely to be affected by permanent income shocksand by fiscal policies. Generally, measures such as GDP per capita and annual real estate tax show relativelyparallel lines for the inelastic and elastic housing supply regions, suggesting that the difference in macroeco-nomic environments is constant over time. This gap can be eliminated through first differencing. Moreover,the employment rate and the unemployment rate for both regions appear to move hand-in-hand.127However, the same set of figures does show that mortgage delinquency rates and the share of populationin the construction industry seem to move with the housing market. The auto loan delinquency rates tend tobe quite similar between the treatment and control groups. On the other hand, the credit card delinquencyrates show a divergence starting in year 2009, which coincides with the start of the financial crisis period.Nonetheless, the credit card delinquency rates do not seem to be fully affected by the housing market, sincethe rates do not show differential trends prior to the financial turbulence period. Mian and Sufi (2011) alsouse similar arguments to show that the housing supply elasticity measure is exogenous to house prices. Theyfind no difference in income, payroll, and employment growth across the regions; house price appreciationhas no impact on credit card debt; and the results are insensitive to control variables such as individualincome, credit score, age and sex, which could drive changes in credit demand. The plots in Figure 4.10 aremore or less consistent with Mian and Sufi’s (2011) results.Thus, to further investigate the robustness of the coefficients, I present several estimations with varioussets of control variables as a form of sensitivity test. In the first set, I incorporate year dummies and regionaldummies to control for the effects from overall macroeconomic fluctuations through time and to accountfor location-specific factors, respectively. In another set of regressions, I also include the lagged regionalunemployment rate and the share of the population in construction industry to control for local labour marketactivities, as the labour market trends in the low elasticity regions could follow more closely to the housingmarket fluctuations.94 To account for permanent income shocks, I incorporate the lagged GDP per capitagrowth as an additional control variable.The debt delinquency rates, especially for mortgages, are another set of important controls for numerousreasons. First, subprime mortgage borrowing nearly tripled during the housing boom period due to looseunderwriting standards, and the increased rate of mortgage delinquency generally affected borrowers withadjustable-rate mortgages (Bernanke, 2007; Kroszner, 2007; Mian and Sufi, 2009). Brown et al. (2015) findthat both non-housing and housing debt for subprime borrowers move together with house price fluctuations.This implies that the effect from the subprime mortgage crisis is expected to also drive up the delinquencyrates for non-housing debt. As illustrated in Figure 4.10, the trends of the delinquency rates are differentbetween the inelastic and elastic housing supply regions during the slowdown period. In addition, as Adelinoet al. (2015) suggests, the number of small business employment depends on the collateral lending channel.States hit hardest by job cuts coincided with the locations with highest rates of foreclosures (Kroszner,2007). Markets with poor labour market prospects are expected to severely undermine an individual’s abilityto repay their debt, which would further exacerbate the number of defaults. To account for these potentialconfounding factors, I also incorporate the lagged regional level delinquency rates (mortgage, credit card,and auto loans) as part of another set of control variables.95 Generally, I do not find substantial changes tothe magnitudes of my estimates across these specifications.94The labour data are retrieved from the Bureau of Labor Statistics. These data are not available for all MSAs. Therefore, I usethe state-level labour market data and assume that these data can proxy for the MSA-level labour market trends.95The Federal Reserve Bank of New York only publishes state-level debt delinquency rates starting from year 1999. For theestimations at the MSA level, I assume that the rates at the state-level can proxy for those at the MSA-level. The lagged values startat year 2001.1284.4.2 Continuous treatment approach: instrumental variable (IV) estimationA shortcoming with the binary treatment approach is that this estimation only indicates whether the housingboom exerts any significant impact on the labour supply decisions of near-retirees aged 55-64, but thisquasi-experiment does not make use of the magnitude of the impact. The binary treatment approach alsoimplicitly assumes that the characteristics of all the regions within the low or within the high elasticityregions are homogeneous.To account for the intensity of the treatment, I explore the relationship between changes in housing valueand changes in labour supply decisions of older individuals using the following regression specification:∆yirt = β0+∑rβ1rregionr +∑tβ2ttimet +β3∆hprt +β4Xirt + εirt (4.3)where ∆yirt denotes the outcome variable for respondent i in region r at time t. I use the same set ofoutcome variables as in the difference-in-difference methods. ∆hprt is the average percentage change inhouse values for region r at time t.96 regionr contains a set of region dummies; and timet includes a setof year effects. The Xirt vector contains the same set of control variables as in the difference-in-differencemethods. In addition, I include a dummy variable that equals one if the household owns a house prior toyear 1998 (i.e. prior to the boom period) to account for the potential difference in labour supply behaviourbetween homeowners and renters. εirt is the error term and I cluster the standard errors at the householdlevel.β3 is the key coefficient of interest, which reflects the causal influence of an exogenous increase inthe value of the primary residence on the probability of retirement, on the change in the number of hoursworked per week, or on the change in the number of weeks worked per year. An exogeneity assumption isrequired in order for β3 to be identified, and it requires the regressors and the error term to be uncorrelated.The empirical challenge is that ∆hprt could be correlated with the error term, since unobservable spillovereffects from various external policies could drive house value and labour supply behaviours. For example,changes in the federal funds rate could affect household’s cost of borrowing for various types of financialinstruments. Lower borrowing cost along with household’s greater desire to borrow could influence overallhousehold wealth. Households could use the additional financing to invest in real estates, which couldfurther drive up house value. Any changes in household wealth may motivate older individuals to alter workarrangements.For β3 to be valid, the ideal instrumental variable (IV) needs to satisfy both the exclusion restriction re-quirement and the relevance condition. The exclusion restriction means that the IV needs to be uncorrelatedwith the error term, and the relevance condition means that the instrument needs to be correlated with theendogenous regressor. Identification relies on the idea that instrumental variable only influences household’swealth and labour supply decisions via the value of primary residence.96The results are similar if I use the change in the individual’s self-reported value of primary residence as the main explanatoryvariable. The relationship between the instrumental variable and the individual value is weaker. As suggested by Disney et al. (2015)and Farnham and Sevak (2015), the use of regional-level house price data eliminates the concern that additions or improvementsmade to individual homes could cause self-reported house prices to be correlated with other omitted factors, such as individualunobserved heterogeneity effects that are embedded in the error term. I eliminated the values at the top 1 and bottom 1 percentilesin order to remove outliers.129To address the endogeneity issue, this study uses the interaction of the inverse of the regional housingsupply elasticities and the U.S. national lending conditions for residential mortgages, (Elasticityr)−1 x SLOOSt ,as the instrumental variable.97 The IV reflects the net percentage of tightening in lending standards for res-idential mortgages at the regional level and it exploits the differential impact from the housing and thenational lending booms and busts across region and over time. The interaction of the housing supply elastic-ity and the U.S. national lending conditions is a suitable instrument for the value of the primary residence,since the housing boom, for example, was manifested by the amount of quantitative easing in lending con-ditions. The grey lines in Figure 4.11 show that national lending conditions for both business and householddebt eased during the housing boom period, while the black lines in the same figure show house price forthe low elasticity regions to increase at a much larger magnitude relative to the high elasticity regions. Insum, Figure 4.11 suggests that the amount of easing in lending standards is magnified in the low elasticityregions, as these regions are much more sensitive to the house price fluctuations. Although my instrumentalvariable is similar to Adelino et al. (2015), Chetty and Szeidl (2010), and Mian and Sufi’s (2011), to the bestof my knowledge, this is the first paper to consider using the national lending conditions as a component ofthe instrument.98I use two-stage least squares (2SLS) to conduct the instrumental variable estimation. In this paper’s IVframework, the second stage regression is given by equation (4.3). The first stage regression is given by:∆hprt = α0+∑rα1rregionr +∑tα2ttimet +α3(Elasticityr)−1 x SLOOSt +α4Xirt +νirt (4.4)where νirt denotes the error term in the first stage regression.The relevance condition is one of the key assumptions for the instrumental variable approach. Having aweak instrument is another common problem in IV regressions that is related to this condition. The modelexhibits strong IVs if the robust F-statistics for the first stage regressions exceed 10 based on Stock et al.’s(2002) benchmark; and exceed 16.38 based on Stock and Yogo’s (2005) framework, where 16.38 is basedon a rejection rate of at most 10% at the 5% Wald test significance level. The model is still identifiedunder a weak instrument, but the relationship between the endogenous regressor and the instrument is weak.The presence of such a problem generally shows the following symptoms: (1) unstable and inconsistentestimates; and (2) size distortion under the standard Wald test because the asymptotic distribution of theestimator depends on a random variable. These symptoms mean that the 2SLS would provide the wronginference for β3. The following section shows that the robust F-statistics for all of the regression modelsexceed the two above-listed minimum thresholds.On the other hand, several potential issues arise when assessing the exclusion restriction requirement.Due to the evaluation problem, one cannot observe the unobservable characteristics, so the exclusion re-striction requirement cannot be directly tested. By the same argument as for the difference-in-differencemethod, one of the concerns is that other external policies may be introduced during the housing boom97Note that the a high elasticity value corresponds to regions with low house price fluctuations. Therefore, I use the inverse ofthe housing supply elasticity to capture the degree of inelasticity for the location.98Adelino et al. (2015) and Mian and Sufi (2011) use MSA-level housing supply elasticities as IV; and Chetty and Szeidl (2010)use the interaction of national average of house prices interacted with the state housing supply elasticity as IV in one of theirspecifications.130and/or bust periods that could affect certain geographic regions. Such effect could then affect household’slabour supply decisions. The Conforming Loan Limit (CLL) could be one example of such policies. FannieMae and Freddie Mac are required to purchase residential mortgages below a specific amount, known asthe Conforming Loan Limit, and loan limits for the high cost areas vary by geographic locations.99 In otherwords, if certain external policies affected households of certain geographic division, then the housing sup-ply elasticities (which are used as proxies for geographic variation) would be correlated with the error term,and the instrument thus would be invalid.Another concern is that households may alter their financial and labour supply decisions due to health,pensions, and bequest-related factors. Individuals who face negative health shock may require more financ-ing for medical expenses and they may work less as poor health may limit their work abilities. Coile andMilligan (2009) show that aging and health shocks are related to a household’s ownership of various assets,where households increase liquid assets and reduce illiquid assets as they age. As Munnell (2011) suggests,individuals may have an incentive to keep working until retirement age (age 65) in order to benefit fromemployer-covered health insurance program. So, health insurance could affect labour supply decisions. Bysimilar argument, individuals may change labour supply decisions and borrowing trends in response to theamount and type of pensions he/she is entitled to, as well as to a specific bequest or precautionary savingsmotive. Given these possibilities, the exclusion restriction would be violated if the instrumental variablewere to be correlated with any one (or more) of these error components.Furthermore, the instrument could fail if it is correlated with households’ ability to borrow. On onehand, by construction, the SLOOS should not convey any information at the individual level, which impliesthat the second part of the instrument should not be correlated with the error term. A bank’s decision tolend or not to lend to a specific individual depends on the client’s credit background. Moreover, using thefraction of mortgage applications denied by financial institutions as a proportion of the total loan applicationsin a county and in a year, Adelino et al. (2015) find that credit conditions actually tightened for the lowhousing supply elasticity regions during years 2002-2007 period. This is the opposite of the national lendingconditions’ trend, where credit conditions eased during the same period. In addition, Mian and Sufi (2009)note that mortgage credit growth in subprime ZIP codes is unlikely to be explained by local housing supplyelasticities, which suggest that the instrument should be uncorrelated with the error term. On the other hand,as noted above, the delinquency rates exhibit differential trends between the inelastic and elastic housingsupply locations, which could threaten the validity condition.Following the strategy in Mian and Sufi (2011), I regress a set of macroeconomic variables as wellas various health, pensions, and bequest-related variables retrieved from the RAND HRS dataset onto theinstrumental variable to investigate whether permanent income shock, household characteristics, labourmarket, and credit conditions could influence the results (see Table 4.1). Generally, there is not enoughevidence to suggest that the instrument is strongly correlated with factors relating to permanent incomeshock and fiscal policies. Nonetheless, I include the lagged real GDP per capita growth as a control variablein one of the specifications to test the sensitivity of the estimates. With the RAND HRS data, I also comparethe results with and without bequest and health insurance variables as controls.99See the Federal Housing Finance Agency and Fannie Mae websites for details.131However, Table 4.1 shows that the effect of the instrument is unstable across the labour market variables.On one hand, the IV does not appear to be linked to the employment rate and to the share of the populationin construction sector. On the other hand, a percentage change in the instrument is associated with a three-percentage point increase in the unemployment rate, which means that locations with tightened lendingconditions are positively-related to job losses.100 One of the estimation concerns is that labour market trendsmay move more sharply in the low elasticity regions relative to the high elasticity regions. The instrumentmay not have accounted for the differential regional labour market trends that also happened simultaneouslywith the housing boom or bust. As Adelino et al. (2015) suggest, the housing boom facilitated small firmsto borrow against housing collateral in order to increase employment. This implies that locations withrelatively low housing supply elasticity may hire more individuals. Wages are expected to adjust in responseto the economic boom and individuals may alter their labour supply decisions in light of these regionalfluctuations. Given these variations, the instrument may run into problems if the differences across theregions are not purely from housing, but are affected by regional labour markets. To address this problem, Iinclude the lagged regional unemployment rate and the lagged share of population in the construction sectoras another set of control variables.Table 4.1 also illustrates that the delinquency rates are positively related to the instrumental variable.The significant relationship between the instrument and the delinquency rates raises a concern of omittedvariable problems in the estimation. Yet, the interpretation of this result should be cautious because thedefault rates are expressed at aggregate levels. Brown et al. (2015) show that non-housing and housingdebt allocations are heterogeneous across age groups; and the older sub-population tends to resemble primehomeowners. Therefore, these aggregate delinquency rates may not fully capture the trends specific to thenear-retirees. To determine whether default rates could be driving individual labour supply decisions, inanother specification, I include the lagged mortgage, credit card, and auto loan delinquency rates as anadditional set of control variables. In the next section, I show that the results are consistent across thedifferent sets of control variables and that there is insufficient evidence to suggest that debt default rates orregional labour market trends drive the estimation results.4.5 Main findingsTo begin the analysis of results, I present the findings for the binary treatment approach, which providesan indication of whether a house price shock exerts any significant impact onto near-retirees’ labour supplydecisions. I first start the analysis with the difference-in-difference (DD) estimations, followed by the resultsfrom the triple difference (DDD) specifications. Then, I conclude the section with the instrumental variable(IV) estimations. The quasi-experiments motivate the need for the instrumental variable strategy, whichincorporates the intensity of the treatment into the model. The IV analysis also indicates whether house100The high number of foreclosures in states with large job cuts may partially explain the positive relationship between the IV andthe unemployment rate. Moreover, existing research in the U.S. finds a strong positive linkage between the maximum duration ofbenefits and the length of an individual’s spell of unemployment benefits, as well as the possibility that some job losers may turn todisability insurance benefit as an outside option (Autor and Duggan, 2003; Katz and Meyer, 1990). These findings may address thediscrepancy between the results with the employment rate and with the unemployment rate as the dependent variable in Table 4.1.132value has any significant influence on household wealth, which provides more information about the linkagesbetween the housing boom and labour supply decisions.For both empirical strategies, I first illustrate the findings from the American Housing Survey (AHS),where the dependent variable equals one if the respondent transitioned from positive to zero earnings. Withinthe AHS results, I illustrate the coefficients for a total of six specifications. The specification ranges fromcontaining only regional and year dummies to including a combination of housing quality, household demo-graphics, and macroeconomic controls. Next, I present the results using the RAND Health and RetirementStudy dataset. For these data tables, I show the coefficients for two preferred specifications. The firstpreferred specification includes regional dummies, lagged unemployment rate, lagged share of the popula-tion in construction industry, and a set of household demographic controls. The second preferred regressionmodel includes all of the controls from the former specification plus dummy variables for bequest and healthinsurance.4.5.1 Binary treatment approachTable 4.2 reports the difference-in-difference (DD) estimates of equation (4.1) for various sets of controlvariables, using the American Housing Survey dataset. The key coefficients are θ2 and θ4, which are the DDcoefficients for the variables Y 0305it ·T REATit and Y 0711it ·T REATit , respectively. The binary variable forswitching to zero earnings is the dependent variable for Table 4.2, and I use this measure as one of the prox-ies for the labour force exit rate. These coefficients provide additional insights to whether housing marketfluctuations contribute to changes in retirement decisions among near-retirees who would have otherwisecontinued working. Columns (1) - (4) present the DD results for years 1999-2011; and columns (5)-(6)present the estimates for years 2001-2011. In all cases, the results are insensitive to the inclusion of housingquality-related variables (column (4)); lagged GDP per capita growth (column (5)); and debt default rates(column (6)). These results imply that overall lifestyle does not influence labour supply decisions, whichpartially addresses Goodstein’s (2007) concern of whether tastes for work could be driving the results to in-significance. Permanent income shocks, as well as the financial stability of the location also do not influencethe coefficients.Generally, I find no evidence that a house price shock exerted significant influence on the labour supplydecisions of older household heads. The first panel presents the findings for the household heads of ages55-64. Row (1) of Table 4.2 gives the coefficients for the Y 0305it variable, which suggests that householdheads exhibit a three-percentage point increase in exit rate in the elastic MSAs between 1999-2001 and2003-2005 periods. For the same time period, row (2) implies that the exit rate would rise by another 5-7percentage points if the household head were to live in the inelastic housing supply MSAs. On the otherhand, both rows (3) and (4) illustrate that the exit rate would barely change between 2003-2005 and 2007-2011 period for both the elastic and inelastic housing supply regions. In any case, none of the coefficientsfor the household head are statistically significant. Consistent with Figures 4.6 to 4.8, the overall effect ofhousing on labour market attachment is negligible. There is not enough evidence to conclude that θ2 isdifferent from θ4. Therefore, the impact from a house price shock seems to be relatively symmetric betweenthe boom and bust periods.133The second panel provides the estimates for spouses of ages 55-64. Row (5) of Table 4.2 suggests thatspouses would increase the labour force exit rate by 6-9 percentage points in the elastic regions betweenthe 1999-2001 and 2003-2005 periods. Relative to the elastic MSAs, spouses in the inelastic MSAs wouldreduce the exit rate by 5 percentage points for the same time frame. However, these variables are statisti-cally insignificant. Rows (7) and (8) show the coefficients for the bust period relative to the boom period.Compared to the household head, the spouse’s labour supply behaviour is more sensitive to the house pricedepreciation. This could be due to labour supply smoothing across family members against a negative wealthshock. In the elastic MSAs, spouses tend to increase the exit rate by 10-17 percentage points between 2003-2005 and 2007-2011 periods. However, relative to elastic locations, those in the inelastic MSAs tend tolower the exit rate by roughly 10 percentage points. On net, spouses do not seem to adjust the labour forceexit rate in response to a house price shock. Similar to the household heads, it is generally inconclusivewhether θ2 is different from θ4 due to large standard errors. This implies that the effect from a house priceshock is not expected to be different between the boom and bust periods.Table 4.3 presents the results using the RAND HRS for various labour market attachment measures asthe dependent variables. The results using the RAND HRS are generally consistent with those from theAHS. Columns (1) and (6) of Table 4.3 provide the coefficients for the regressions with the same dependentvariable as Table 4.2 with the AHS data. The results from these two columns are comparable to column (3)of Table 4.2. To summarize, the RAND HRS data shows that both the spouse and the household head do notswitch from positive to zero earnings in response to a house price shock. As noted above, the RAND HRScovers broader geographic areas relative to the AHS. Therefore, this finding suggests that the estimates arenot confounded by mobility decisions, which could be embedded in the error term. The coefficients are alsorobust to the additions of bequest and health insurance-related control variables, which imply that labourmarket activities are not affected by other external non-housing factors.The other labour market attachment indicators from Table 4.3 also convey a similar story. Columns(2) and (3) illustrate the effect of the housing market on labour force participation rate and on retirementrate, respectively. Although the coefficients are generally insignificant, the magnitude of the estimates isminiscule. This continues to suggest that housing value exerts negligible effect on labour supply decisionsof both household heads and spouses. Columns (7) and (8) show the corresponding results with the bequestand health insurance control variables. Again, the results are invariant to the choice of the controls.Similarly, a house price shock is associated with a very small effect on the intensive margins of laboursupply. For example, columns (4) and (5) show the effect from housing on the percentage change in totalhours worked per week and on the percentage change in total weeks worked per year, respectively. Thehousehold heads in the elastic census divisions do not seem to adjust the intensive margins of labour supplyin response to housing market fluctuations. Relative to the elastic locations, the household heads seem toreduce hours worked by at most 4.4 percentage points during the housing bust period. However, the sametrend does not apply to the weeks worked per year. The coefficients for the spouses are insignificant andsmall. Therefore, spouses’ labour supply does not seem to respond to housing market fluctuations.Table 4.4 shows the triple difference estimates from the AHS data, using the indicator for transitioning tozero earnings as the dependent variable. Again, the top panel refers to the results for the household heads and134the bottom panel for the spouses. The effect of a house price shock on work and retirement decisions coulddiffer between homeowners and renters. The AHS data suggests that this effect could hold within an MSA.For the household heads, the coefficient on the Y 0305it ·HOMEit variable is approximately 25 percentagepoints. Relative to the renters, the household heads who own a dwelling prior to 1997 would increase theexit rate by 25 percentage points in the elastic MSAs between 1999-2001 and 2003-2005. The numbers aregenerally statistically significant at the conventional level. Relative to the elastic locations for the same timerange, the homeowners in the inelastic MSAs tend to reduce the exit rate by 36-38 percentage points morethan renters. The bust period also shows similar trend, which again demonstrates that the effect of a houseprice shock on work and retirement decisions is uniform across the periods. Summing up the effects, relativeto the renters, homeowners would reduce exit rate by 5 to 15 percentage points in the inelastic locations forboth the boom and bust periods. Conversely, the coefficients for the spouses are insignificant. It is thusinconclusive whether the spouse would exhibit differential labour market trends between homeowners andrenters across the MSAs.Table 4.5 presents the triple difference results using the RAND HRS data. Generally, at broader geo-graphic levels, homeowners no longer show different labour supply behaviour relative to the renters. Col-umn (1) of Table 4.5 is comparable to column (3) of Table 4.4, which shows that the coefficients for boththe household heads and the spouses are insignificant and are small. The results are also invariant to theadditions of bequest and health insurance-related variables (see column (6)). This implies that mobilitydecisions could influence the comparisons between homeowners and renters. Therefore, in this case, theresults from the RAND HRS would more precisely capture the effect of housing on retirement choices.Similarly, the estimates for the other labour market attachment measures also do not show housing value toexert any differential impact on homeowners.To summarize, the difference-in-difference results show that labour supply decisions are uniform acrossthe housing boom and bust periods. There is insufficient evidence to conclude that housing wealth andretirement choices are strongly linked. Most of the results are robust to the various sets of control variables,which suggest that labour market attachment is not driven by other macroeconomic effects and/or preferenceshocks.4.5.2 Continuous treatment approach:Table 4.6 presents the instrumental variable estimation using the AHS data. The dependent variable is abinary variable for transitioning to zero earnings. The first two rows of each panel present the first stageregression results, which are given by Equation (4.4). As described in the Empirical Strategy section, theinstrument, (Elasticityr)−1 x SLOOSt reflects the additional exposure to the housing and lending booms (orbusts). The coefficient on the explanatory variable, (Elasticityr)−1 x SLOOSt for the first stage regressionreflects roughly the percentage increase in house value for each additional unit of lending exposure. Forexample, the coefficient “-0.442” in column (1) of row (1) means that house price at the MSA level willdecline by 0.442% for each additional percent of net tightening in lending conditions adjusted by the region’shousing supply. All of the first stage regressions show the IV to be strongly correlated with the percentagechange in house value, with the robust F-statistics to be over 16.38. By both Stock et al.’s (2002) and Stock135and Yogo’s (2005) benchmarks, these F-statistics suggest that the IV regression models are not subject toweak instrument problems.Each column in Table 4.6 illustrates the results for a specific set of control variables. The second part ofTable 4.6 presents the Ordinary Least Squares (OLS) and Instrumental Variable (IV) estimates for Equation(4.3) within MSA levels. Rows (3) and (8) show the OLS results for β3 for the household head and thespouse, respectively. The estimates for β3 reflect the change in the labour outcome variables for everypercentage increase in house value. If the regression models were estimated using the OLS, the β3 coefficientsuggests that a 1 percentage point increase in house value translates to roughly a three- to four-percentagepoint increase in the labour force exit rate for both the household head and the spouse. After accounting forendogeneity, the magnitude of the exit rate doubles for the household head and rises by more than 10 timesfor the spouse. However, the AHS data shows that the estimates for the household head are insignificant,and the coefficients are marginally significant for the spouse. These IV results are generally consistent withthe DD findings.Table 4.7 presents the effect of house price growth on other labour supply measures within the censusdivision level using RAND HRS data. Column (1) of Table 4.7 is comparable to column (3) of Table 4.6. Theresults at broader geographic level more or less convey similar story as those estimated using finer locationbreakdowns. The IV estimates show that household heads continue to show an insignificant relationshipbetween the probability of switching to zero earnings and housing value growth. Conversely, spouses exhibita positive and strong linkage between these two factors. The other regression models demonstrate thathousing value growth exerts a negligible effect on both extensive and intensive margins of labour supply.This finding applies to both family members.The labour supply results appear to contradict the findings from Goodstein (2007) and from Begleyand Chan (2015), which show that changes in house price should lead to changes in retirement decisions.However, a careful look at the IV results in Table 4.8 reveals that the findings on retirement decisionsare in line with expectations. The common impression is that the value of primary residence should bepositively correlated with household wealth, and any change in household wealth should alter labour supplybehaviour. Table 4.8 shows the linkages between the percentage change in housing value and various typesof wealth measures for the near-retirees. Column (1) shows the relationship between net housing wealth(i.e. value of all residences less residential mortgage credit). For both OLS and IV estimations, net housingwealth growth and housing value growth move hand-in-hand. This implies that the effect from real assetvalue growth dominates over the impact from mortgage debt growth. Column (2) of Table 4.8 suggests thathousing and financial (non-housing) wealth are not strongly related. Finally, Column (3) of the same tabledemonstrates that the linkages between housing value and total net household wealth is weak.This research brings light to the popular claim by showing that the percentage change in housing valueexerts no significant influence on the percentage change in net total household wealth for the near-retirees.The housing boom contributed to increases in net housing equity, but had no significant impact on net finan-cial wealth. Table 4.9 shows the wealth distributions for individuals of ages 55-64. The wealth distributionis skewed and the amount of net housing wealth for the median person is roughly half of that for the averageperson. This implies that the effect of a house price shock on the median near-retiree is expected to be136modest relative to that for the average person. Baker (2015) and Christelis et al. (2014) also suggest thatconsumption elasticities are more sensitive to liquid wealth holdings than to illiquid wealth holdings (hous-ing wealth). For example, Christelis et al. (2014) illustrate using the 2009 Internet Survey of the Health andRetirement Study data that the marginal propensities to consume with respect to housing and financial wealthare 1 and 3.3 percentage points, respectively. As such, given these findings, there is insufficient evidence toconclude a strong linkage between housing, net total household wealth, and labour supply decisions.4.6 ConclusionThis paper investigates the effect of house value onto labour supply decisions of U.S. older households.Policymakers, researchers, and the media have indicated that house values and household wealth are closelyrelated, and any changes in household wealth are expected to alter an individual’s labour supply behaviour.Given large standard errors, my empirical analysis does not provide sufficient evidence to conclude thathousing value and near-retirees’ work and retirement decisions are strongly-linked. Near-retirees also do notseem to respond asymmetrically to positive and negative wealth shocks. Using the interaction of housingsupply elasticity and national lending conditions for residential mortgages as an instrument, this researchsheds light on this popular claim by showing that the value of the primary residence does not exert anysignificant influence on net total household wealth and labour supply decisions. The wealth distribution isskewed. The amount of net housing wealth for the median person is roughly half of that for the averageperson. This implies that the effect of a house price shock on the median near-retiree is expected to bemodest relative to that for the average person. Furthermore, a house price shock is expected to be transitoryand thus the effect of housing wealth on work decisions is expected to be small.This paper faces a shortcoming with the estimation of the labour supply response. The RAND HRSdid not accurately compile the data for the risk aversion dummies, and all of the observations over age 65were coded as missing values from years 2002 to 2006. If the risk aversion dummies were not subject tocompliance issues, this variable could be used in the IV regressions. This could help answer the question ofwhether risk tolerance plays any role in explaining the weak relationship between housing and retirementdecisions.Despite this shortcoming, this paper makes several important contributions to this area of research. Tothe best of my knowledge, although central banks use the results from the credit condition surveys as partof the information set for monetary policy and financial stability assessments, this is the first paper to applythe national lending conditions for residential mortgage series as part of the instrumental variable strategyto measure the effect of a house price shock on household’s wealth and labour supply decisions. Moreover,this paper shows that the proportion of households with homeownership does not explain whether housingand labour supply are related. This contradicts Coile and Levine (2006), who argue that the insignificanceof their results is due to the small number of households with sizeable stock holdings.Future research should consider extending this type of work to a cross-country comparison analysis, todetermine if the lack of impact from house value onto labour supply and total net household wealth is onlyapplicable to the U.S.. Finding the reasons behind the lack of relationship between housing value and net137household wealth, and how the other non-housing (non-financial return-related) components may be drivingthe results could also be another future path for this area of research.1384.7 FiguresFigure 4.1: U.S. labour force participation rate, debt-to-disposable income ratio, and the aggregate houseprice index.Grey shading denotes housing bubble period. The household debt-to-disposable income measure is thesum of home mortgages and consumer credit extracted from Table B101, “Balance Sheet of Householdsand Nonprofit Organizations” of the U.S. Federal Reserve’s Z1 release. This household debt-to-disposableincome series is smaller than what is reported in the Bank of Canada’s Financial System Review (FSR)publications. The Bank of Canada’s U.S. household debt-to-disposable income series gives larger valuesbecause it includes the unincorporated business sector, in order for its series to be comparable with theCanadian data. Canadian household debt-to-disposable income series is from Statistics Canada, and Statis-tics Canada’s household sector includes unincorporated businesses (See Box 1 of December 2012 versionof the Bank of Canada’s Financial System Review).Sources: U.S. Bureau of Labor Statistics; U.S. Federal Housing Financing Agency; U.S. Federal Reserve’sFlow of Funds Account (Z1 Release); and author’s calculations. Last observation: year 2013139Figure 4.2: Value of primary residence and total net housing wealth for the inelastic and elastic housingsupply regionsValue of primary residence and total net housing wealth, using RAND Health and Retirement Study data:11.411.611.812.012.2In logs of 2012 dollars1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsValue of primary residence-20.0-10.00.010.020.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsValue of primary residence11.211.411.611.812.0In logs of 2012 dollars1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsNet housing wealth-20.0-10.00.010.020.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsNet housing wealthValue of primary residence, using American Housing Survey data4.24.44.64.85.05.2In logs of 2013 dollars1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsNormalized by unit square footageAverage housing value-20.0-10.00.010.020.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsNormalized by unit square footageAverage housing value growthSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study ofAging, with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA(October 2015); Saiz (2010); U.S. Bureau of Labor Statistics (for CPI measure); U.S. Census Bureau’s AmericanHousing Survey; and author’s calculations.140Figure 4.3: Housing and non-housing debt for the inelastic and elastic housing supply regionsHome equity loans and non-housing (consumer) debt, using RAND Health and Retirement Study data:8.18.28.38.48.58.6In logs of 2012 dollars1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsConsumer debt-20.0-10.00.010.020.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsConsumer debt growthHome equity loans, using American Housing Survey data2.02.22.42.62.8In logs of 2013 dollars1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsNormalized by unit square footageAverage value of home equity loans-15.0-10.0-5.00.05.010.015.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsNormalized by unit square footageAverage percentage change in home equity loansSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study ofAging, with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA(October 2015); Saiz (2010); U.S. Bureau of Labor Statistics (for CPI measure); U.S. Census Bureau’s AmericanHousing Survey; and author’s calculations.141Figure 4.4: Household characteristics - Inelastic versus elastic housing supply regions35.040.045.050.055.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionswith a bequest over $100,000Self-assessed probability of living0.010.020.030.040.050.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of male respondents15.020.025.030.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsvisible minority groupsShare of respondents belonging to10.020.030.040.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsa Bachelor's degreeShare of respondents with at least70.080.090.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of respondents with health insurance covered50.060.070.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of respondents who are marriedSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); and author’s calculations.142Figure 4.5: Dwelling characteristics - Inelastic versus elastic housing supply regions0.05.010.015.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of household heads who moved within 2 years30.040.050.060.0Age1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsMean building age0.02.04.06.08.010.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with wide open cracks0.05.010.015.020.025.030.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with evident of rodents in units0.02.04.06.08.010.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with no running water0.01.02.03.04.05.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with broken sewage(continued on next page...)143Figure 4.5: Dwelling characteristics - Inelastic versus elastic housing supply regions (continued)0.01.02.03.04.05.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with big peelings in paint0.010.020.030.040.050.060.070.080.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with no heat10.015.020.025.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of homes with water leakageSources: Saiz (2010); U.S. Census Bureau’s American Housing Survey; and author’s calculations.144Figure 4.6: Near-retirees’ labour supply decisions - AHS dataHousehold head Spouse20.025.030.035.040.0In percent1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of individuals with 0 earnings30.040.050.0In percent1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of individuals with 0 earnings0.05.010.015.020.025.0In percent1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of individuals who transitioned to 0 earnings5.010.015.020.025.0In percent1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of individuals who transitioned to 0 earningsSources: Saiz (2010); U.S. Census Bureau’s American Housing Survey; and author’s calculations.145Figure 4.7: Near-retirees’ labour supply decisions in levels - RAND HRS dataExtensive margins of labour supplyHousehold head Spouse50.060.070.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsLabour force participation rate60.070.080.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsLabour force participation rate10.020.030.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of individuals who completely retired18.020.022.024.026.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of individuals who completely retired40.042.044.046.048.050.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of individuals with zero earnings30.035.040.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsShare of individuals with zero earnings(continued on next page...)146Figure 4.7: Near-retirees’ labour supply decisions in levels - RAND HRS data (continued)Intensive margins of labour supplyHousehold head Spouse35.040.045.01998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTotal hours worked per week20.025.030.01998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTotal hours worked per week45.050.055.01998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsWeeks worked per year45.050.055.01998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsWeeks worked per yearSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); and author’s calculations.147Figure 4.8: Changes in near-retirees’ labour supply decisions - RAND HRS dataExtensive margins of labour supplyShare of household heads who... Share of spouses who...0.05.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsSwitched to labour force exit5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsSwitched to labour force exit0.05.010.015.020.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to completely retired5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to completely retired5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to zero earnings5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to zero earnings(continued on next page...)148Figure 4.8: Changes in near-retirees’ labour supply decisions - RAND HRS data (continued)Intensive margins of labour supplyHousehold head Spouse-10.0-8.0-6.0-4.0-2.00.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total hours worked per week-10.0-5.00.05.010.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total hours worked per week-5.0-4.0-3.0-2.0-1.00.01.02.03.04.05.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in weeks worked per year-5.0-3.0-1.01.03.05.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in weeks worked per yearSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); and author’s calculations.149Figure 4.9: Changes in labour supply decisions for household heads, split by dwelling tenure status - RANDHRS dataExtensive margins of labour supplyShare of homeowners who... Share of renters who...5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsSwitched to labour force exit5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsSwitched to labour force exit5.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to completely retired5.010.015.020.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to completely retired0.05.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to zero earnings0.05.010.015.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsTransitioned to zero earnings(continued on next page...)150Figure 4.9: Changes in labour supply decisions for household heads, split by dwelling tenure status - RANDHRS data (continued)Intensive margins of labour supplyHomeowners Renters-8.0-6.0-4.0-2.00.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total hours worked per week-20.0-15.0-10.0-5.00.05.010.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total hours worked per week-5.0-3.0-1.01.03.05.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total weeks worked per year-10.0-5.00.05.010.0In percent1998 2000 2002 2004 2006 2008 2010 2012YearElastic regions Inelastic regionsPercentage change in total weeks worked per yearSources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); and author’s calculations.151Figure 4.10: Macroeconomic variables - Inelastic versus elastic housing supply regions40.045.050.055.060.0In thousands of nominal dollars1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsGDP per capita1.52.02.53.03.54.0In thousands of nominal dollars1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsAnnual real estate tax50.055.060.065.070.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsEmployment rate4.06.08.010.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsUnemployment rate1.61.82.02.22.42.6In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsShare of population in construction industry(continued on next page...)152Figure 4.10: Macroeconomic variables - Inelastic versus elastic housing supply regions (continued)0.04.08.012.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsMortgage delinquency rate (over 90 days)5.010.015.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsCredit card delinquency rate (over 90 days)0.02.04.06.0In percent1997 1999 2001 2003 2005 2007 2009 2011YearElastic regions Inelastic regionsAuto loan delinquency rate (over 90 days)Sources: FRBNY Consumer Credit Panel (for delinquency rate series); Saiz (2010); U.S. Bureau of Economic Anal-ysis (for GDP per capita and employment rate series); U.S. Bureau of Labor Statistics (for construction share, un-employment rate, and CPI series); U.S. Census Bureau’s American Housing Survey (for real estate tax series); andauthor’s calculations.153Figure 4.11: U.S. house price index and national lending conditions - Inelastic versus elastic housing supplyregions[1]: Weighted by the number of housing units in each census division.[2]: Starting from 2007Q2, the U.S. Federal Reserve’s Senior Loan Officer Opinion Survey (SLOOS) does not reportthe net percentage of banks tightening standards for all mortgage loans. The SLOOS reports the changes in lendingstandards for prime, non-traditional, and subprime mortgages separately starting from 2007Q2. I use the shares ofprime, subprime, and government-insured mortgages that were reported by the Mortgage Bankers Association in itsDecember 6, 2007 press release to extrapolate the change in lending standard series for all mortgages for 2007Q2 -2013Q4. The shares of prime, subprime, and government-insured mortgages are 77.6%, 13.1%, and 9.3%, respec-tively. For the non-traditional and subprime mortgages series, I first average the lending conditions for these twocategories and then multiply by the sum of the subprime and government-insured mortgage shares. I then add thisseries to the product of prime mortgage share and changes in lending condition series for prime mortgages to retrievethe overall lending conditions series for all mortgages for 2007Q2 - 2013Q4.Sources: U.S. Census Bureau’s Population Division; U.S. Federal Housing Finance Agency; U.S. Federal Reserve;Mortgage Bankers Association; and author’s calculations. Last observation: year 20131544.8 TablesTable 4.1: A test of the exclusion restriction requirement(1) (2) (3) (4) (5) (6)Macroeconomic variables [1]Dependentvariables:% change inreal GDPper capita% change inaveragewage levels% change inaveragehouseholdincomeChange inemploymentrateChange in un-employmentrateChange in % ofpopulation inconstructionsector(1) IV -0.024 0.042 0.044 -0.008 0.030*** -0.008**(0.032) (0.070) (0.071) (0.010) (0.007) (0.003)Observations 791 791 791 780 794 794Tax, debt and delinquency ratesDependentvariables:% change inreal estatetax [1]% change inmortgagedebtoutstanding[2]% change inconsumerdebtoutstanding[2]Change inmortgagedelinquencyrate [3]Change incredit carddelinquencyrate [3]Change in autoloandelinquencyrate [3](2) IV 0.127 0.170 0.361 0.146*** 0.019** 0.100***(0.134) (0.171) (0.484) (0.020) (0.008) (0.019)Observations 791 6,432 3,758 681 681 681Household characteristics variables [2]Dependentvariables:Householdhead hashealthinsuranceSpouse hashealthinsuranceHouseholdhead plansto leavebequestSpouseplans toleavebequestHouseholdhead haspensionsSpouse haspensions(3) IV -0.004 0.086 -0.027 0.099 -0.025 0.227**(0.046) (0.060) (0.054) (0.072) (0.071) (0.095)Observations 31,410 15,856 31,558 16,701 16,863 10,554This table presents the results for regressing various macroeconomics and household characteristics variables ontothe instrumental variable, (Elasticityr)−1 x SLOOSt . All regressions contain dummy variables for year and regionaleffects. Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significantat 5%; * significant at 10%.Notes [1]: The sample period is 1999-2011. [2]: The sample period is 1998-2012 and covers respondents of ages55-64. [3]: The sample period is 2001-2011 because the delinquency rate series start in year 1999.Sources: FRBNY Consumer Credit Panel (for delinquency rate series); RAND Health and Retirement Study Data,Version O (for outstanding debt amounts, health insurance, bequest, and pension data); Saiz (2010); U.S. Bureauof Economic Analysis (for GDP per capita and employment rate series); U.S. Bureau of Labor Statistics (for con-struction share, unemployment rate, and CPI series); U.S. Census Bureau’s American Housing Survey; and author’scalculations.155Table 4.2: Difference-in-difference estimations - AHS dataDependent variable: Binary variable for transitioning to zero earningsVariables (1) (2) (3) (4) (5) (6)Household heads’ labour supply decisions(1) Y 0305it 0.031 0.030 0.026 0.026(0.035) (0.039) (0.039) (0.039)(2) Y 0305it ·T REATit 0.067 0.058 0.052 0.053(0.042) (0.042) (0.042) (0.042)(3) Y 0711it 0.026 0.002 0.012 0.012 -0.012 -0.026(0.030) (0.043) (0.044) (0.044) (0.033) (0.035)(4) Y 0711it ·T REATit 0.032 0.030 0.017 0.018 -0.038 -0.020(0.036) (0.035) (0.036) (0.037) (0.033) (0.035)Observations 2,293 2,293 2,293 2,293 2,021 2,021Spouse’s labour supply decisions(5) Y 0305it 0.063 0.090* 0.085* 0.085*(0.047) (0.050) (0.051) (0.051)(6) Y 0305it ·T REATit -0.054 -0.048 -0.048 -0.049(0.054) (0.054) (0.054) (0.054)(7) Y 0711it 0.105** 0.156*** 0.169*** 0.169*** 0.105** 0.118**(0.045) (0.059) (0.060) (0.060) (0.046) (0.049)(8) Y 0711it ·T REATit -0.096* -0.102** -0.108** -0.110** -0.089** -0.080*(0.051) (0.050) (0.051) (0.052) (0.045) (0.047)Observations 1,848 1,848 1,848 1,848 1,624 1,624SMSA FE YES YES YES YES YES YESDemographic controls? NO YES YES YES YES YESLagged unemployment rate? NO NO YES YES YES YESLagged % construction? NO NO YES YES YES YESLagged GDP per capita growth? NO NO NO NO YES NOLagged delinquency rates? NO NO NO NO NO YESHousing Quality? NO NO NO YES NO NOThis table presents the difference-in-difference (DD) regression results for household heads and spouses of ages 55-64.The sample period for columns (1)-(4) is 1999-2011; and for columns (5)-(6) is 2001-2011, because the delinquencyrates and the GDP per capita growth series start from year 1999. The regressions follow equation (4.1) in text.Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at5%; * significant at 10%.Sources: FRBNY Consumer Credit Panel (for delinquency rate series); Saiz (2010); U.S. Bureau of EconomicsAnalysis (for GDP per capita series); U.S. Bureau of Labor Statistics (for construction share and unemployment rateseries); U.S. Census Bureau’s American Housing Survey; and author’s calculations.156Table 4.3: Difference-in-difference estimations - RAND HRS dataExclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworked perweekChange inweeksworked peryearTransitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworked perweekChange inweeksworked peryear(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Household heads’ labour supply decisions(1) Y 0206it -0.008 -0.003 0.021** -0.005 0.003 -0.006 -0.002 0.021** -0.007 0.004(0.010) (0.009) (0.009) (0.019) (0.013) (0.010) (0.009) (0.009) (0.019) (0.013)(2) Y 0206it ·T REATit 0.012 0.013 -0.011 -0.029 0.002 0.011 0.013 -0.011 -0.028 0.002(0.011) (0.010) (0.010) (0.020) (0.013) (0.011) (0.010) (0.010) (0.020) (0.013)(3) Y 0812it -0.012 -0.017 0.015 0.000 0.000 -0.010 -0.016 0.014 -0.000 0.001(0.013) (0.012) (0.012) (0.023) (0.015) (0.013) (0.012) (0.012) (0.023) (0.015)(4) Y 0812it ·T REATit -0.008 -0.000 -0.015 -0.044** -0.003 -0.008 0.001 -0.014 -0.044** -0.003(0.011) (0.010) (0.010) (0.019) (0.012) (0.011) (0.010) (0.010) (0.019) (0.012)Observations 26,735 26,985 26,208 13,152 12,990 26,717 26,962 26,194 13,144 12,984(continued on next page...)157Table 4.3: Difference-in-difference estimations - RAND HRS data (continued)Exclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworked perweekChange inweeksworked peryearTransitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworked perweekChange inweeksworked peryear(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Spouse’s labour supply decisions(5) Y 0206it 0.019 -0.001 0.003 0.023 0.020 0.021* -0.001 0.003 0.022 0.021(0.012) (0.013) (0.014) (0.035) (0.023) (0.012) (0.013) (0.014) (0.035) (0.023)(6) Y 0206it ·T REATit -0.015 0.026* -0.001 -0.050 -0.013 -0.014 0.027** -0.001 -0.050 -0.014(0.014) (0.014) (0.015) (0.034) (0.023) (0.014) (0.014) (0.015) (0.034) (0.023)(7) Y 0812it 0.025 -0.014 -0.025 0.050 0.001 0.026 -0.014 -0.024 0.049 0.003(0.017) (0.017) (0.018) (0.039) (0.028) (0.017) (0.017) (0.018) (0.039) (0.027)(8) Y 0812it ·T REATit -0.028** 0.019 0.006 -0.011 -0.006 -0.026* 0.021 0.006 -0.012 -0.007(0.014) (0.013) (0.015) (0.030) (0.020) (0.014) (0.013) (0.015) (0.030) (0.020)Observations 13,684 13,294 11,557 4,984 4,955 13,680 13,290 11,555 4,984 4,955This table presents the difference-in-difference (DD) regression results for household heads and spouses of ages 55-64. The regressions follow from equation(4.1) in text. All regression models contain census division, demographic controls including a disability indicator, lagged unemployment rate, and lagged shareof population in construction industry. Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at 5%; *significant at 10%.Sources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging, with funding from the National Instituteon Aging and the Social Security Administration, Santa Monica, CA (October 2015); Saiz (2010); U.S. Bureau of Labor Statistics (for construction share andunemployment rate series); and author’s calculations.158Table 4.4: Triple difference estimations - AHS dataDependent variable: Binary variable for transitioning to zero earningsVariables (1) (2) (3) (4) (5) (6)Household heads’ labour supply decisions(1) Y 0305it ·HOMEit 0.225** 0.241** 0.234** 0.231**(0.114) (0.109) (0.110) (0.109)(2) Y 0305it ·T REATit ·HOMEit -0.359*** -0.382*** -0.382*** -0.380***(0.132) (0.129) (0.131) (0.131)(3) Y 0711it ·HOMEit 0.382** 0.355*** 0.346*** 0.347*** 0.386** 0.391**(0.154) (0.128) (0.126) (0.126) (0.184) (0.189)(4) Y 0711it ·T REATit ·HOMEit -0.427*** -0.393*** -0.393*** -0.394*** -0.325* -0.334*(0.162) (0.139) (0.137) (0.137) (0.194) (0.199)Observations 2,087 2,087 2,087 2,087 1,815 1,815Spouse’s labour supply decisions(5) Y 0305it ·HOMEit 0.061 0.114 0.073 0.075(0.052) (0.073) (0.076) (0.077)(6) Y 0305it ·T REATit ·HOMEit -0.037 -0.100 -0.073 -0.072(0.120) (0.125) (0.125) (0.126)(7) Y 0711it ·HOMEit -0.113 -0.180 -0.190 -0.182 -0.207 -0.390(0.154) (0.168) (0.166) (0.174) (0.233) (0.239)(8) Y 0711it ·T REATit ·HOMEit 0.159 0.217 0.204 0.195 0.195 0.384(0.183) (0.194) (0.191) (0.198) (0.253) (0.259)Observations 1,667 1,667 1,667 1,667 1,443 1,443SMSA FE YES YES YES YES YES YESDemographic controls? NO YES YES YES YES YESLagged unemployment rate? NO NO YES YES YES YESLagged % construction? NO NO YES YES YES YESLagged GDP per capita growth? NO NO NO NO YES NOLagged delinquency rates? NO NO NO NO NO YESHousing Quality? NO NO NO YES NO NOThis table presents the triple difference (DDD) regression results for household heads and spouses of ages 55-64. Thesample period for columns (1)-(4) is 1999-2011; and for columns (5)-(6) is 2001-2011, because the delinquency ratesand the GDP per capita growth series start in year 1999. The regressions follow equation (4.2) in text. Standard errorsare in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at 5%; * significantat 10%.Sources: FRBNY Consumer Credit Panel (for delinquency rates series); Saiz (2010); U.S. Bureau of EconomicAnalysis (for GDP per capita series); U.S. Bureau of Labor Statistics (for construction share and unemployment rateseries); U.S. Census Bureau’s American Housing Survey; and author’s calculations.159Table 4.5: Triple difference estimations - RAND HRS dataExclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworkedper weekChange inweeksworkedper yearTransitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworkedper weekChange inweeksworkedper year(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Household heads’ labour supply decisions(1) Y 0206it ·HOMEit -0.004 -0.039* -0.025 0.005 -0.042* -0.006 -0.041* -0.024 0.007 -0.043*(0.021) (0.023) (0.023) (0.043) (0.024) (0.021) (0.023) (0.023) (0.043) (0.024)(2) Y 0206it ·T REATit ·HOMEit -0.033 0.031 0.054* -0.004 0.044 -0.034 0.029 0.054* -0.006 0.045(0.027) (0.028) (0.029) (0.056) (0.029) (0.027) (0.028) (0.029) (0.056) (0.029)(3) Y 0812it ·HOMEit 0.038 -0.038 0.018 0.013 -0.047 0.038 -0.038 0.018 0.015 -0.046(0.038) (0.039) (0.033) (0.053) (0.033) (0.038) (0.039) (0.033) (0.053) (0.034)(4) Y 0812it ·T REATit ·HOMEit -0.050 -0.008 -0.040 0.137* 0.045 -0.047 -0.004 -0.040 0.131* 0.044(0.047) (0.048) (0.043) (0.074) (0.052) (0.047) (0.048) (0.043) (0.073) (0.052)Observations 14,043 14,136 13,791 6,208 6,182 14,035 14,126 13,784 6,204 6,179(continued on next page...)160Table 4.5: Triple difference estimations - RAND HRS data (continued)Exclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworkedper weekChange inweeksworkedper yearTransitionto zeroearningsLabourforceexitCompleteretire-mentChange inhoursworkedper weekChange inweeksworkedper year(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Spouse’s labour supply decisions(5) Y 0206it ·HOMEit 0.003 -0.010 -0.098* 0.098 -0.039 0.004 -0.010 -0.097* 0.103 -0.042(0.041) (0.043) (0.058) (0.071) (0.029) (0.041) (0.043) (0.057) (0.071) (0.029)(6) Y 0206it ·T REATit ·HOMEit -0.017 -0.023 0.073 -0.083 0.066 -0.021 -0.023 0.073 -0.090 0.071(0.052) (0.051) (0.067) (0.092) (0.043) (0.052) (0.051) (0.066) (0.092) (0.043)(7) Y 0812it ·HOMEit -0.077 -0.094 -0.143** 0.108 0.047 -0.073 -0.091 -0.149** 0.114 0.043(0.083) (0.066) (0.071) (0.094) (0.045) (0.084) (0.066) (0.072) (0.088) (0.047)(8) Y 0812it ·T REATit ·HOMEit 0.098 0.020 0.101 -0.057 -0.031 0.092 0.017 0.105 -0.056 -0.029(0.089) (0.087) (0.091) (0.151) (0.052) (0.090) (0.087) (0.092) (0.149) (0.055)Observations 6,601 6,376 5,407 1,992 1,985 6,600 6,375 5,406 1,992 1,985This table presents the triple difference (DDD) regression results for household heads and spouses of ages 55-64. The regressions follow from equation (4.2) in text.All regression models contain census division, demographic controls including a disability indicator, lagged unemployment rate, and lagged share of population inconstruction industry. Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging, with funding from the National Instituteon Aging and the Social Security Administration, Santa Monica, CA (October 2015); Saiz (2010); U.S. Bureau of Labor Statistics (for construction share andunemployment rate series); and author’s calculations.161Table 4.6: IV estimations for labour supply decisions - AHS dataDependent variable for second-stage regression: Binary variable for transitioning to zero earningsVariables (1) (2) (3) (4) (5) (6)Household heads’ labour supply decisionsFirst-stage regressions:(1) (Elasticityr)−1 ·SLOSt -0.442*** -0.434*** -0.513*** -0.511*** -0.535*** -0.387***(0.065) (0.065) (0.072) (0.072) (0.076) (0.090)(2) Robust F-statistics 46.21 44.27 50.28 49.80 50.10 18.56Second-stage regressions:(3) OLS: ∆HPrt 0.040 0.029 0.033 0.033 0.035 0.027(0.050) (0.050) (0.050) (0.050) (0.059) (0.062)(4) IV: ∆HPrt 0.011 -0.016 0.020 0.018 0.035 0.046(0.263) (0.269) (0.246) (0.248) (0.243) (0.365)(5) Observations 2,085 2,085 2,085 2,085 1,813 1,813Spouse’s labour supply decisionsFirst-stage regressions:(6) (Elasticityr)−1 ·SLOSt -0.397*** -0.399*** -0.485*** -0.483*** -0.501*** -0.381***(0.070) (0.069) (0.075) (0.075) (0.078) (0.094)(7) Robust F-statistics 32.22 33.01 41.85 41.34 41.02 16.60Second-stage regressions:(8) OLS: ∆HPrt 0.034 0.035 0.041 0.042 0.038 0.020(0.055) (0.055) (0.054) (0.054) (0.060) (0.063)(9) IV: ∆HPrt 0.596* 0.623* 0.468* 0.474* 0.417 0.659(0.333) (0.332) (0.281) (0.283) (0.280) (0.431)(10) Observations 1,848 1,848 1,848 1,848 1,624 1,624SMSA FE YES YES YES YES YES YESDemographic controls? NO YES YES YES YES YESLagged unemployment rate? NO NO YES YES YES YESLagged % construction? NO NO YES YES YES YESLagged GDP per capitagrowth?NO NO NO NO YES NOLagged delinquency rates? NO NO NO NO NO YESHousing Quality? NO NO NO YES NO NOThis table presents the instrumental variable (IV) regression results for household heads and spouses of ages 55-64.The sample period for columns (1)-(4) is 1999-2011; and for columns (5)-(6) is 2001-2011 because the delinquencyrate and the GDP per capita growth series start from year 1999. The regressions follow equations (4.3) and (4.4) intext. Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significantat 5%; * significant at 10%.Sources: FRBNY Consumer Credit Panel (for delinquency rate series); Saiz (2010); U.S. Bureau of Economic Analy-sis (for GDP per capita series); U.S. Bureau of Labor Statistics (for construction share and unemployment rate series);U.S. Census Bureau’s American Housing Survey; and author’s calculations.162Table 4.7: IV estimations for labour supply decisions - RAND HRS dataExclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforce exitCompleteretirementChange inhoursworkedper weekChange inweeksworkedper yearTransitionto zeroearningsLabourforce exitCompleteretirementChange inhoursworkedper weekChange inweeksworkedper year(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Household heads’ labour supply decisionsFirst-stage regressions:(1) (Elasticityr)−1 ·SLOSt -0.385***(0.009)-0.384***(0.009)-0.383***(0.009)-0.370***(0.013)-0.368***(0.012)-0.385***(0.009)-0.384***(0.009)-0.383***(0.009)-0.369***(0.013)-0.367***(0.012)(2) Robust F-statistics 1786.555 1820.387 1786.695 853.981 869.221 1792.371 1823.049 1789.337 854.577 870.699Second-stage regressions:(3) OLS: ∆HPrt 0.060 0.001 -0.046 -0.087 0.000 0.061 0.004 -0.043 -0.096 0.004(0.072) (0.072) (0.078) (0.159) (0.084) (0.071) (0.072) (0.079) (0.159) (0.084)(4) IV: ∆HPrt -0.003 0.187 0.022 0.653 -0.145 -0.008 0.180 0.022 0.630 -0.139(0.285) (0.224) (0.243) (0.497) (0.238) (0.285) (0.223) (0.243) (0.496) (0.237)(5) Observations 14,043 14,136 13,791 6,208 6,182 14,035 14,126 13,784 6,204 6,179(continued on next page...)163Table 4.7: IV estimations for labour supply decisions - RAND HRS data (continued)Exclude bequest and health insurance variables Include bequest and health insurance variablesDependent Variables: Transitionto zeroearningsLabourforce exitCompleteretirementChange inhoursworkedper weekChange inweeksworkedper yearTransitionto zeroearningsLabourforce exitCompleteretirementChange inhoursworkedper weekChange inweeksworkedper year(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Spouse’s labour supply decisionsFirst-stage regressions:(6) (Elasticityr)−1 ·SLOSt -0.371***(0.019)-0.365***(0.019)-0.364***(0.019)-0.366***(0.064)-0.363***(0.068)-0.372***(0.019)-0.366***(0.019)-0.365***(0.019)-0.362***(0.064)-0.358***(0.068)(7) Robust F-statistics 382.965 356.541 351.895 32.232 28.121 383.153 356.643 355.163 31.609 27.475Second-stage regressions:(8) OLS: ∆HPrt -0.171 0.068 0.106 -1.158*** -0.114 -0.179 0.069 0.109 -1.159*** -0.111(0.127) (0.120) (0.149) (0.346) (0.192) (0.125) (0.120) (0.149) (0.349) (0.190)(9) IV: ∆HPrt 0.798** 0.153 -0.498 -1.508 0.033 0.682* 0.119 -0.406 -1.438 0.000(0.374) (0.333) (0.495) (1.860) (0.555) (0.375) (0.334) (0.489) (1.923) (0.574)(10) Observations 6,601 6,376 5,407 1,992 1,985 6,600 6,375 5,406 1,992 1,985This table presents the instrumental variable regression results for household heads and spouses of ages 55-64. The regressions follow equations (4.3) and (4.4) intext. All regression models contain census division, demographic controls including a disability indicator, lagged unemployment rate, and lagged share of populationin construction industry. Standard errors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at 5%; * significant at 10%.Sources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging, with funding from the National Institute on Agingand the Social Security Administration, Santa Monica, CA (October 2015); Saiz (2010); U.S. Bureau of Labor Statistics (for construction share and unemployment rateseries); and author’s calculations.164Table 4.8: IV estimations for wealth variables - RAND HRS dataDependent Variables: Net housingwealth [1]Net financialwealth [2]Nethouseholdwealth [3](1) (2) (3)First-stage regressions:(1) (Elasticityr)−1 ·SLOSt -0.399*** -0.380*** -0.385***(0.011) (0.011) (0.010)(2) Robust F-statistics 1358.698 1160.468 1553.598Second-stage regressions:(3) OLS: ∆HPrt 0.613*** -0.134 0.441**(0.163) (0.528) (0.216)(4) IV: ∆HPrt 1.041** 0.752 0.102(0.490) (1.807) (0.797)(10) Observations 10,638 8,626 12,455This table presents the instrumental variable regression results for household heads of ages 55-64. The regressionsfollow equations (4.3) and (4.4) in text. All regression models contain census division, demographic controls includinga disability indicator, lagged unemployment rate, and lagged share of population in construction industry. Standarderrors are in parentheses and are clustered at the household level. *** Significant at 1%; ** significant at 5%; *significant at 10%.[1] The net value of housing wealth equals the value of the primary residence less mortgages and home equity loans.[2] The net value of financial wealth includes stocks, mutual funds, and investment funds; checking, savings, andmoney market accounts; CDs, government savings bonds and T-bills; bonds and bond funds; and other savings assets.[3] Total household wealth equals the sum of net value of non-housing wealth and housing wealth. The net value ofnon-housing wealth includes real estate (not primary residence), vehicles, businesses, IRA, and financial wealth, lessnon-mortgage debt.The computation includes individuals who reported a value of zero.Sources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); U.S. Bureau of Labor Statistics (for construction share, unemployment rate, and CPI series); andauthor’s calculations.165Table 4.9: Wealth distributions for ages 55-64 - RAND HRS dataIn thousands of 2012 dollars(1) (2) (3) (4) (5) (6)percentile10th 25th 50th 75th 90th AverageNet value of housing wealth 0.000 4.254 80.000 176.068 320.000 135.496Net value of non-housingwealth0.000 5.956 59.008 275.189 755.000 316.986Net value of financial wealth -6.467 0.000 9.116 78.961 281.044 121.921Total household wealth 0.013 36.247 166.514 475.383 1,034.345 452.483The net value of housing wealth equals the value of the primary residence less mortgages and home equity loans.The net value of non-housing wealth includes real estate (not primary residence), vehicles, businesses, IRA, andfinancial wealth, less non-mortgage debt.The net value of financial wealth includes stocks, mutual funds, and investment funds; checking, savings, and moneymarket accounts; CDs, government savings bonds and T-bills; bonds and bond funds; and other savings assets.Total household wealth equals the sum of net value of non-housing wealth and housing wealth.The computation includes individuals who reported a value of zero.Sources: RAND Health and Retirement Study Data, Version O. Produced by the RAND Center for the Study of Aging,with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (October2015); Saiz (2010); U.S. Bureau of Labor Statistics (for CPI series); and author’s calculations.166Chapter 5ConclusionThis dissertation contributes to the literature on the economics of aging and housing from various perspec-tives. Using the eligibility requirements of the Canadian Old Age Security (OAS) program to investigateelderly labour market attachments, Chapter 2 illustrates that seniors only respond to public pension en-titlements with a decrease in labour force participation rates. The effect from public pension benefits isheterogeneous across family types. A combination of estimates suggests that elderly immigrants show weaklabour market attachment a few years prior to the OAS eligibility date. In Chapter 3, the research find-ings push this area of literature forward by suggesting an alternative perspective for explaining native out-migration. The heterogeneity in mobility preferences across dwelling tenure groups is an important resultbecause it may explain why Card (2001) fails to find any significant effect from immigration on aggregatenative relocation decisions. The synthetic cohort analysis does not provide enough evidence to concludethat near-retirees extract housing equity by relocating. Finally, Chapter 4 sheds light to the popular claimby showing that housing does not exert any significant influence on elderly labour supply decisions. Theresults from Chapters 3 and 4 are aligned with Skinner’s (1996), which suggests that only a small fractionof elderly households would tap into their housing equity.To summarize, the findings in all three chapters seem to be linked by the Permanent Income Hypothesis.In Chapter 2, while the timing of the OAS benefits is anticipated, the benefit amount that is disbursed is un-known due to clawback provisions and inflationary effects. Although agents could respond to the OAS/GISbenefits by re-adjusting their labour supply behaviour prior to the receipt of this public pension entitlement,it is also reasonable to see that elderly immigrants reduce labour efforts as they become eligible for OAS/GISbenefits. For both Chapters 3 and 4, changes in housing wealth shock are expected to be transitory ratherthan permanent. As such, the impact from the housing market should not lead to significant changes in workand mobility decisions. Therefore, the results in Chapters 3 and 4 are aligned with expectations.167BibliographyAccetturo, A., Manaresi, F., Mocetti, S., & Olivieri, E. (2014). "Don’t stand so close to me: The urbanimpact of immigration". Regional Science and Urban Economics, 45(6), 45-56.Adelino, M., Schoar, A., & Severino, F. (2015). "House prices, collateral and self-employment". Journal ofFinancial Economics, 117, 288-306.Aguiar, M., & Hurst, E. (2005). "Consumption versus expenditure". Journal of Political Economy, 113(5),919-948.Aguiar, M., Hurst, E., & Karabarbounis, L. (2013). "Time use during the Great Recession". The AmericanEconomic Review, 103(5), 1664-1696.Akbari, A. H., & Aydede, Y. (2012). "Effects of immigration on house prices in Canada". Applied Eco-nomics, 44(13), 1645-1658.Aldén, L., Hammarstedt, M., & Neuman, E. (2015). "Ethnic segregation, tipping behavior, and nativeresidential mobility". International Migration Review, 49(1), 36-69.Amedah, S. A., & Fougère, M. (2016, June). "Impact of financial and housing wealth on retirementdecision". Presented at the Canadian Economics Association conference on June 3, 2016, Ottawa,Ontario. Retrieved 2016-06-03, from http://economics.ca/cgi/cms?task=abstract&item=2016:AS0032-1Antecol, H., Cobb-Clark, D. A., & Trejo, S. J. (2003). "Immigration policy and the skills of immigrants toAustralia, Canada, and the United States". The Journal of Human Resources, 38(1), 192-218.Autor, D. H., & Duggan, M. G. (2003). "The rise in the disability rolls and the decline in unemployment".The Quarterly Journal of Economics, 118(1), 157-205.Bailliu, J., Kartashova, K., & Meh, C. (2011-2012). "Household borrowing and spending in Canada". Bankof Canada Review (Winter 2011-2012), 16-29.Baker, M. (2002). "The retirement behaviour of married couples: Evidence from the Spouse’s Allowance".The Journal of Human Resources, 37(1), 1-34.Baker, M., & Benjamin, D. (1997). "The role of the family in immigrants’ labour-market activity: Anevaluation of alternative explanations". The American Economic Review, 87(4), 705-727.Baker, M., & Benjamin, D. (1999). "Early retirement provisions and the labor force behavior of older men:Evidence from Canada". Journal of Labor Economics, 17(4), 724-756.Baker, M., Benjamin, D., & Fan, E. (2009). "Public policy and the economic wellbeing of elderly immi-grants". CLSRN Working Paper No. 52.Baker, S. R. (2015, February). "Debt and the consumption response to household income shocks". SSRNWorking Paper No. 2541142.168Bank of Canada. (Various issues). Financial System Review. Retrieved 2013-09-22, from http://www.bankofcanada.ca/publications-research/periodicals/fsrBegley, J., & Chan, S. (2015, January). "The effect of housing wealth shocks on work and retire-ment decisions". Retrieved 2016-04-01, from https://www.researchgate.net/publication/280291521Bernanke, B. (2012, February 10). "Housing markets in transition". Speech to the 2012 National Associationof Home Builders International Builders’ Show, Orlando, Florida.Bernanke, B. S. (2007, June 5). "The housing market and subprime lending". Speech to the 2007 Interna-tional Monetary Conference, Cape Town, South Africa (via satellite).Bertrand, M., Duflo, E., & Mullainathan, S. (2004). "How much should we trust differences-in-differencesestimates?". The Quarterly Journal of Economics, 119(1), 249-275.Bies, S. (2005, January 18). "The economy and challenges in retirement savings". Speech to the FinancialExecutives International Baltimore Chapter, Baltimore, Maryland.Bies, S. (2006, June 14). "A supervisor’s perspective on mortgage markets and mortgage lending practices".Speech to the Mortgage Bankers Association Presidents Conference, Half Moon Bay, California.Blau, F. D., Kahn, L. M., Moriarty, J. Y., & Souza, A. P. (2003). "The role of the family in immigrants’labor-market activity: An evaluation of alternative explanations: Comment". The American EconomicReview, 63(1), 429-447.Blundell, R., Pistaferri, L., & Saporta-Eksten, I. (2016). "Consumption inequality and family labor supply".The American Economic Review(2), 387-435.Boivin, J. (2012, April 4). "Aging gracefully: Canada’s inevitable demographic shift". Speech to theEconomic Club of Canada, Toronto, Ontario.Borjas, G. J. (1993). "Immigration policy, national origin, and immigrant skills: A comparison of Canadaand the United States". In D. Card & R. B. Freeman (Eds.), Small Differences That Matter: LaborMarkets and Income Maintenance in Canada and the United States (p. 21-43). Chicago: Universityof Chicago Press.Borjas, G. J. (1999). "Immigration and welfare magnet". Journal of Labor Economics, 17(1), 607-637.Borjas, G. J. (2003). "Welfare reform, labor supply, and health insurance in the immigrant population".Journal of Health Economics, 22, 933-958.Borjas, G. J. (2008). "Immigration and labor market outcomes in the native elderly population". NBERRetirement Research Center Paper No. NB08-10.Borjas, G. J. (2011). "Social security eligibility and the labor supply of older immigrants". Industrial andLabor Relations Review, 64(3), 485-501.Borjas, G. J., & Hilton, L. (1996). "Immigration and the welfare state: Immigrant participation in means-tested entitlement programs". The Quarterly Journal of Economics, 111, 575-604.Bratsberg, B., Raaum, O., & Røed, K. (2014). "Immigrants, labour market performance and social insur-ance". The Economic Journal, 124, 644-683.Brown, M., Stein, S., & Zafar, B. (2015). "The impact of housing markets on consumer debt: Credit reportevidence from 1999 to 2012". Journal of Money, Credit and Banking, 47(1), 175-213.169Börsch-Supan, A. (2000). "Incentive effects of social security on labor force participation: Evidence inGermany and across Europe". Journal of Public Economics, 78, 25-49.Bui, S. A., Craig, S. G., & Imberman, S. A. (2014). "Is gifted education a bright idea? Assessing the impactof gifted and talented programs on students". American Economic Journal: Economic Policy, 6(3),30-62.Burge, G., & Zhao, L. (Forthcoming). "Housing wealth, property taxes, and labor supply among the elderly".Journal of Labor Economics.Campbell, J. Y., & Cocco, J. F. (2007). "How do house prices affect consumption? Evidence from microdata". Journal of Monetary Economics, 54(3), 591-621.Canada. (1994). "A Broader Vision: Immigration and Citizenship Plan 1995-2000, Annual Report toParliament". Ottawa: Supply and Services Canada.Canada. (1995). "A Broader Vision: Immigration Plan, 1996 Annual Report to Parliament". Ottawa: Supplyand Services Canada.Canada Mortgage and Housing Corporation. (2012). "Housing for older Canadians: The definitive guide tothe over-55 market". Ottawa, ON. Retrieved 2015-03-30, from http://www.cmhc-schl.gc.ca/Canada Mortgage and Housing Corporation. (2014). "Rental market statistics". Ottawa, ON. Retrieved2015-03-30, from http://www.cmhc-schl.gc.ca/Canada Mortgage and Housing Corporation. (Various years). Canadian Housing Observer [Data file].Ottawa, ON. Retrieved 2015-03-30, from http://www.cmhc-schl.gc.ca/Card, D. (2001). "Immigration inflow, native outflows, and the local market impacts of higher immigration".Journal of Labor Economics, 19(1), 22-64.Card, D., & DiNardo, J. (2000). "Do immigrant inflows lead to native outflow". The American EconomicsReview, 90(2), 360-367.Charles, K. K., Hurst, E., & Notowidigdo, M. J. (2013, April). "Manufacturing decline, housing booms,and non-employment". NBER Working Paper No. 18949.Chetty, R., & Szeidl, A. (2010, May). "The effect of housing on portfolio choice". NBER Working PaperNo. 15998.Christelis, D., Georgarakos, D., & Jappelli, T. (2014). "Wealth shocks, unemployment shocks and con-sumption in the wake of the Great Recession". CFS Working Paper Series, No. 471.Citizenship and Immigration Canada. (2005, April). "Recent immigrants in metropolitan areas: Canada - acomparative profile based on the 2001 Census". Ottawa, ON. Retrieved 2015-03-30, from http://www.cic.gc.ca/Coile, C. C., & Levine, P. B. (2006). "Bulls, bears, and retirement behaviour". Industrial and LaborRelations Review, 59(3), 408-429.Coile, C. C., & Levine, P. B. (2011). "The market crash and mass layoffs: How the current economic crisismay affect retirement". The B.E. Journal of Economic Analysis & Policy, 11(1), 1-40.Coile, C. C., & Milligan, K. (2009, June). "How household portfolios evolve after retirement: The effect ofaging and health shocks". The Review of Income and Wealth, 55(2), 226-248.Conservative Party of Canada. (2015, August 12). "Prime Minister Harper announces help for aspiring170homeowners". Retrieved 2015-08-15, from http://www.conservative.ca/Cortés, P. (2008). "The effect of low-skilled immigration on U.S. prices: Evidence from CPI data". Journalof Political Economy, 116(3), 381-422.Cortés, P., & Tessada, J. (2011). "Low-skilled immigration and the labor supply of highly skilled women".American Economic Journal: Applied Economics, 3, 88-123.Costa, D. L. (1998). "The evolution of retirement: Summary of a research project". The American EconomicReview, 88(2), 232-236. (Papers and proceedings of the hundred and tenth annual meeting of theAmerican Economic Association)Crawford, A., & Faruqui, U. (2012). "What explains trends in household debt in Canada". Bank of CanadaReview, Winter 2011-2012, 3-15.Crawford, R. (2013). "The effect of the financial crisis on the retirement plans of older workers in England".Economics Letters, 121, 156-159.Crossley, T. F., McDonald, J. T., & Worswick, C. (2001). "Immigrant benefit receipt revisited: Sensitivityto the choice of survey years and model specification". The Journal of Human Resources, 36(2),379-397.Daly, M., Jobijn, B., & Kwok, J. (2009, January 30). "Labor supply responses to changes in wealth andcredit". FRBSF Economic Letter No. 2009-05.Danzer, A. M. (2013). "Benefit generosity and the income effect on labour supply: Quasi-experimentalevidence". The Economic Journal, 123, 1059-1084.Disney, R., Ratcliffe, A., & Smith, S. (2015). "Booms, busts, and retirement timing". Economica, 82,399-419.Dustmann, C., Frattini, T., & Halls, C. (2010). "Assessing the fiscal costs and benefits of A8 migration tothe UK". Fiscal Studies, 31(1), 1-41.Farnham, M., & Sevak, P. (2015). "Housing wealth and retirement timing". CESifo Economic Studies,1-21.Faruqui, U., Gilbert, P., & Kei, W. (2008). "The Bank of Canada’s Senior Loan Officer Survey". Bank ofCanada Review (Autumn), 55-62.Ferguson, R. W., Jr. (2005, November 18). "Recent changes in household finances and home lending".Speech to the monthly meeting of the Epsilon Boule Fraternity for African-American Leaders, Wash-ington D.C..Finnie, R. E., Gray, D. M., & Zhang, Y. (2016, September). "A longitudinal analysis of GIS entries andexits". Canadian Public Policy, 42(3).Fu, S., Liao, Y., & Zhang, J. (2016, February). "Housing wealth effect and labor force participation:Evidence from China Household Finance Survey". SSRN Working Paper No. 2735293.Goda, G. S., Shoven, J. B., & Slavov, S. N. (2011). "What explain changes in retirement plans duringthe Great Recession". The American Economic Review, Papers and proceedings of the 123rd annualmeeting of the American Economic Association (May 2011), 101(3), 29-34.Goda, G. S., Shoven, J. B., & Slavov, S. N. (2012). "Does stock market performance influence retirementintentions". Journal of Human Resources, 47(4), 1055-1081.171Gold, K. (2015, May 29). "Affordable housing out of reach for Vancouver millenials". Retrieved 2015-07-04, from http://www.theglobeandmail.comGonzalez, L., & Ortega, F. (2013). "Immigration and housing booms: Evidence from Spain". Journal ofRegional Science, 53(1), 37-59.Goodstein, R. (2007, October). "The effect of wealth on labor force participation of older men". Pre-sented at the FDIC’s Center for Financial Research Seminar Series on January 24, 2008, Arling-ton, Virginia. Retrieved 2013-08-10, from www.fdic.gov/bank/analytical/cfr/2008/seminar_series_2008.htmlGovernment of Alberta. (2006, June). "Alberta’s housing market". Retrieved 2015-04-16, from http://www.finance.alberta.ca/Government of Alberta. (2007, November). "Alberta’s housing markets boom". Retrieved 2015-04-16,from http://www.finance.alberta.ca/Government of Alberta. (2009, September). "Demographic trends and Alberta’s housing market". Retrieved2015-04-16, from http://www.finance.alberta.ca/Grant, T. (2015, January 29). "Damage from cancelled census as bad as feared, researchers say". Retrieved2015-03-28, from http://www.theglobeandmail.comGustman, A. L., Steinmeier, T. L., & Tabatabai, N. (2011, October). "How did the recession of 2007-2009affect the wealth and retirement of the near retirement age population in the Health and RetirementStudy". NBER Working Paper No. 17547.Haan, M. (2012). "The housing experiences of new Canadians: Insights from the Longitudinal Surveyof Immigrants to Canada". Citizenship and Immigration Canada: Research and Evaluation WorkingPapers.Hansen, J., & Lofstrom, M. (2003). "Immigrant assimilation and welfare participation: Do immigrantsassimilate into or out of welfare". The Journal of Human Resources, 38, 74-98.Hansen, J., & Lofstrom, M. (2009). "The dynamics of immigrant welfare and labor market behavior".Journal of Population Economics, 22, 941-970.Imbens, G. W., & Lemieux, T. (2007). "Regression discontinuity designs: A guide to practice". Journal ofEconometrics. doi: 10.1016/j.jeconom.2007.05.001Kahn, L. M. (2004). "Immigration, skills and the labor markets: International evidence". Journal ofPopulation Economics, 17(3), 501-534.Kartashova, K., & Tomlin, B. (2013, January). "House prices, consumption, and the role of non-mortgagedebt". Bank of Canada Working Paper No. 2013-2.Katz, L. F., & Meyer, B. D. (1990). "The impact of the potential duration of unemployment benefits on theduration of unemployment". Journal of Public Economics, 41, 45-72.Kaushal, N. (2010). "Elderly immigrants’ labor supply responses to supplemental security income". Journalof Policy Analysis and Management, 29(1), 137-162.Kroszner, R. S. (2007, November 5). "The challenges facing subprime mortgage borrowers". Speech at theConsumer Bankers Association 2007 Fair Lending Conference, Washington D.C..Latif, E. (2015). "Immigration and housing rents in Canada: A panel data analysis". Economic Issues,17220(1), 91-108.Lee, J. (2015, June 5). "Vancouver mayor seeks to curb housing speculation". Retrieved 2015-07-04, fromhttp://www.vancouversun.com/Leigh, D., Igan, D., Simon, J., & Topalova, P. (2012, April). "Dealing with household debt". In WorldEconomic Outlook: Growth Resuming, Dangers Remain (chap. 3). International Monetary Funds.Lemieux, T., & Milligan, K. (2008). "Incentive effects of social assistance: A regression discontinuityapproach". Journal of Econometrics, 142, 807-828.Li, Q. (2014). "Ethnic diversity and neighborhood house prices". Regional Science and Urban Economics,48, 21-38.Liebman, J. B., Luttmer, E. F. P., & Seif, D. G. (2009). "Labor supply responses to marginal Social Securitybenefits: Evidence from discontinuities". Journal of Public Economics, 93(3), 1208-1223.Mendez, P., Hiebert, D., & Wyly, E. (2006). "Landing at home: Insights on immigration and metropolitanhousing markets from the Longitudinal Survey of Immigrants to Canada". Canadian Journal of UrbanResearch, 15(2), 82-104.Mian, A., & Sufi, A. (2009). "The consequences of mortgage credit expansion: Evidence from the U.S.mortgage default crisis". Quarterly Journal of Economics, 124, 1449-1496.Mian, A., & Sufi, A. (2011, August). "House prices, home equity-based borrowing, and the U.S. householdleverage crisis". American Economic Review, 101, 2132-2156.Mocetti, S., & Porello, C. (2010). "How does immigration affect native internal mobility? New evidencefrom Italy". Regional Science and Urban Economics, 40, 427-439.Munnell, A. H. (2011, August). "What is the average retirement age?". Center for Retirement Research atBoston College Issue Number 11-11.Ondrich, J., & Falevich, A. (2016). "The great recession, housing wealth, and the retirement decisions ofolder workers". Public Finance Review, 44(1), 109-131.Ottaviano, G. I. P., & Peri, G. (2006, November). "Wages, rents and prices: The effects of immigration onU.S. natives". Unpublished manuscript.Pandey, M., & Townsend, J. (2011). "Quantifying the effects of the Provincial Nominee Program". Cana-dian Public Policy, 37(4), 495-512.Peri, G., & Sparber, C. (2009). "Task specialization, immigration, and wages". American Economic Journal:Applied Economics, 1(3), 135-169.Qadeer, M., Agrawal, S. K., & Lovell, A. (2010). "Evolution of ethnic enclaves in the Toronto metropolitanarea". International Migration & Integration, 11, 315-339.RAND Center for the Study of Aging. (2015, October). RAND Health and Retirement Study data, VersionO [Data file and codebook]. Santa Monica, CA. Retrieved 2016-03-25, from http://hrsonline.isr.umich.edu/ (With funding from the National Institute on Aging and the Social SecurityAdministration)Sá, F. (2014). "Immigration and house prices in the UK". The Economic Journal, 1-32. doi: 10.1111/ecoj.12158Saez, E. (2002). "Optimal income transfer programs: Intensive versus extensive labor supply responses".173Quarterly Journal of Economics, 117(3), 1039-1073.Saiz, A. (2003). "Room in the kitchen for the melting pot: Immigration and rental prices". The Review ofEconomics and Statistics, 85(3), 502-521.Saiz, A. (2007). "Immigration and housing rents in American cities". Journal of Urban Economics, 61,345-371.Saiz, A. (2010, August). "The geographic determinants of housing supply". Quarterly Journal of Eco-nomics, 125(3), 1253-1296.Saiz, A., & Wachter, S. (2011). "Immigration and the neighborhood". American Economic Journal:Economic Policy, 3, 169-188.Saunders, D. (2015, June 27). "Why newcomers are beginning to bypass Canada’s big cities". Retrieved2015-07-10, from http://www.theglobeandmail.comShan, H. (2008). "Property taxes and elderly labor supply". Federal Reserve Board Finance and EconomicsDiscussion Series Working Paper No. 2008-51.Skinner, J. S. (1996). "Is housing wealth a sideshow?". In D. A. Wise (Ed.), Advances in the Economics ofAging (p. 241-272). University of Chicago Press (for NBER).Statistics Canada. (1986, 1991, 1996, 2001, 2006). Census, years 1986-2006 [Data file and codebook].Ottawa, ON. Retrieved 2014-05-01, from http://www.statcan.gc.ca/Statistics Canada. (2010, January). "2006 Census dictionary". Catalogue no. 92-566-X.Statistics Canada. (2011). National Household Survey [Data file and codebook]. Ottawa, ON. Retrieved2014-05-01, from http://www.statcan.gc.ca/Statistics Canada. (2014). "Standard geographical classification concordance tables". Ottawa, ON. Retrieved2015-10-16, from http://www.statcan.gc.ca/Statistics Canada. (Various years). Postal Code Conversion File (various years) [Data file and codebook,use with permission from Canada Post]. Ottawa, ON. Retrieved 2014-12-16, from http://www.statcan.gc.ca/Stillman, S., & Maré, D. C. (2007). "The impact of immigration on the geographic mobility of NewZealanders". CReAM Discussion Paper Series No. 14.Stock, J. H., Wright, J. H., & Yogo, M. (2002). "A survey of weak instruments and weak identification ingeneralized method of moments". Journal of Business and Economic Statistics, 20, 218-529.Stock, J. H., & Yogo, M. (2005). "Testing for weak instruments in linear IV regression". In D. W. K. An-drews & J. H. Stock (Eds.), Identification and inference for econometric models: Essays in honor ofThomas Rothenberg, ed. (p. 80-108). New York: Cambridge University Press.Teranet-National Bank. (2015). Teranet-National Bank House Price Index [Data file and Methodology].Ottawa, ON. Retrieved 2014-06-14, from http://www.housepriceindex.ca/The New York Times. (2012, September 15). "The Road to Retirement". Retrieved 2013-08-10, fromhttp://www.nytimes.comU.S. Census Bureau. (Various years). American Housing Survey, 1997-2013 [Data file and codebook].Retrieved 2016-03-28, from http://www.census.gov/housing/ahs/U.S. Federal Reserve System. (2016). Senior Loan Officer Opinion Survey on Bank Lending Practices174[Data file and documentation]. Retrieved 2016-03-28, from http://www.federalreserve.gov/boarddocs/snloansurvey/about.htmVancity. (2015, May). "Help wanted: salaries, affordability and the exodus of labour from Metro Vancou-ver". Retrieved 2015-07-04, from http://w