UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays in empirical macroeconomics Wei, Mengying 2020

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2020_november_wei_mengying.pdf [ 3.7MB ]
Metadata
JSON: 24-1.0394831.json
JSON-LD: 24-1.0394831-ld.json
RDF/XML (Pretty): 24-1.0394831-rdf.xml
RDF/JSON: 24-1.0394831-rdf.json
Turtle: 24-1.0394831-turtle.txt
N-Triples: 24-1.0394831-rdf-ntriples.txt
Original Record: 24-1.0394831-source.json
Full Text
24-1.0394831-fulltext.txt
Citation
24-1.0394831.ris

Full Text

Essays in Empirical MacroeconomicsbyMengying WeiB.Sc. Economics, University of Amsterdam, 2013M.A. Economics, University of British Columbia, 2014a thesis submitted in partial fulfillmentof the requirements for the degree ofDoctor of Philosophyinthe faculty of graduate and postdoctoral studies(Economics)The University of British Columbia(Vancouver)October 2020c©Mengying Wei, 2020The following individuals certify that they have read, and recommend tothe Faculty of Graduate and Postdoctoral Studies for acceptance, the disser-tation entitled:Essays in Empirical Macroeconomicssubmitted by Mengying Wei in partial fulfillment of the requirements forthe degree of Doctor of Philosophy in Economics.Examining Committee:Henry Siu, EconomicsSupervisorMichael Devereux, EconomicsSupervisory Committee MemberSanghoon Lee, BusinessUniversity ExaminerJack Favilukis, BusinessUniversity ExaminerAdditional Supervisory Committee Members:Paul Beaudry, EconomicsSupervisorViktoria Hnatkovska, EconomicsSupervisory Committee MemberiiAbstractThis thesis presents three chapters in empirical macroeconomics. The firstchapter studies how the mortgage expansions in the early 2000s affect U.S.regional economies by estimating its impact on the local labor market from2003 to 2017. Using a plausible exogenous measure of the credit supplyshock, I find that counties with higher credit supply shocks have not seensignificant changes in local unemployment but have shown slower wagegrowth. While the high-credit counties did not experience significantly dif-ferent changes in local labor markets in the expansion period, they did ex-perience larger increases in unemployment in the recession, but also recov-ered faster after the recession, summing to a zero net effect in the long run.Meanwhile, these counties experienced a slowdown in wage growth sincethe recession, resulting in a depressed wage level until 2017. Additionally,the wage decline was accompanied by a decrease in the employment shareof young firms.In Chapter 2, I propose a mechanism to explain how mortgage marketfluctuations affected the labor market, slowed down wage growth, and ledto labor reallocation. I introduce two financial constraints, one on the house-hold side and the other on the production side, both tied to the collateralvalues of houses. I show that changes in household borrowing constraintsaffect housing prices and thereby affect firms’ financial condition. Whenworking capital constraint binds, mortgage market fluctuations affect firms’labor demand, which led to labor reallocation from financially constrainedto unconstrained firms and a decline in wage.In Chapter 3, we study how small and micro enterprises (SMPE) re-iiispond to the policy in reducing the corporate income tax rate in China. Us-ing gradual increases in the qualifying threshold for SMPEs during 2010-2016 as a natural experiment, we find that the rate cut led to significantincreases in sales growth, investment, and productivity of affected SMPEfirms. We further show that the rate cut induced micro-sized firms to enterthe market.ivLay SummaryMy dissertation consists of three chapters in empirical macroeconomics.First, I investigate how mortgage expansion during the 2000s affected theU.S. local labor markets. Using a mortgage supply shock, I find that high-shock counties have not experienced differences in unemployment, but haveslowed down wage growth and a decrease in the employment share of youngfirms. Second, to explain the empirical findings, I propose a mechanismwhere the mortgage market and labor market are connected through hous-ing prices. The contraction in the mortgage market leads to declines in hous-ing prices and deteriorates the firm’s financial condition, which affects itslabor demand and leads to labor reallocation. Third, we study how tax in-centives targeted at small firms affect their performance in China. Using thecorporate income tax cut as a natural experiment, we find that the rate cutencouraged small firm growth in terms of revenue, investment, and pro-ductivity.vPrefaceChapter 1 and 2 are original, unpublished, independent work by the au-thor, Mengying Wei. Chapter 3 is an unpublished working paper that I co-authored with Wei Cui, Weisi Xie, and Jing Xing. The authors contributedequally to the project overall.viTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviIntroduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Mortgage Expansion and Long-run Labor Market Outcome . . . 51.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . 111.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . 131.3.1 Research Design . . . . . . . . . . . . . . . . . . . . . 131.3.2 Identifying Credit Supply Shock . . . . . . . . . . . . 141.3.3 Lender Characteristics . . . . . . . . . . . . . . . . . . 181.3.4 Credit Supply Shock . . . . . . . . . . . . . . . . . . . 201.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22vii1.4.1 Relationship between Credit Shock and Housing Mar-ket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241.4.2 Relationship between Credit Shock and Labor Market 261.4.3 Relationship between Credit Shock and Firm Dynamism 321.4.4 Pre-trend . . . . . . . . . . . . . . . . . . . . . . . . . . 341.4.5 Placebo Test . . . . . . . . . . . . . . . . . . . . . . . . 341.4.6 Sensitivity Test . . . . . . . . . . . . . . . . . . . . . . 351.4.7 Two-stage Least Square Regression . . . . . . . . . . . 351.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361.6 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381.7 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402 Credit Shocks, House Prices, and Labor Market . . . . . . . . . . 602.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 622.2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 622.2.2 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 662.2.3 Transmission Mechanism: from housing to labor mar-ket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 672.2.4 Quantitative Analysis . . . . . . . . . . . . . . . . . . 682.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 722.4 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 742.5 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773 China’s Proactive Fiscal Policy: Assessing the Early Impact ofTax Cuts on Small Firms . . . . . . . . . . . . . . . . . . . . . . . . 783.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.2 Policy Background . . . . . . . . . . . . . . . . . . . . . . . . 823.3 Theoretical Framework . . . . . . . . . . . . . . . . . . . . . . 863.3.1 Production . . . . . . . . . . . . . . . . . . . . . . . . . 863.3.2 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . 903.3.3 Policy Shock . . . . . . . . . . . . . . . . . . . . . . . . 903.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . 91viii3.4.1 Baseline Estimations . . . . . . . . . . . . . . . . . . . 913.4.2 Matched Difference-in-differences Estimations . . . . 943.5 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963.5.1 Confidential Corporate Tax Returns . . . . . . . . . . 963.5.2 Measuring Total Factor Productivity and Firm-levelInvestment Rate . . . . . . . . . . . . . . . . . . . . . . 973.5.3 Property of Matching . . . . . . . . . . . . . . . . . . 973.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983.6.1 Effects on Output Growth, Investment and Productivity 983.6.2 Substitution between Capital and Labor . . . . . . . . 1013.6.3 Firm Entry . . . . . . . . . . . . . . . . . . . . . . . . . 1033.6.4 Stay Small or Grow? . . . . . . . . . . . . . . . . . . . 1063.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1073.8 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1083.9 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132A Supporting Materials . . . . . . . . . . . . . . . . . . . . . . . . . 140A.1 Appendix to Chapter 1 . . . . . . . . . . . . . . . . . . . . . . 140A.1.1 Selected Lenders . . . . . . . . . . . . . . . . . . . . . 140A.1.2 Relationship between Credit Supply Shock and Hous-ing Price Changes . . . . . . . . . . . . . . . . . . . . . 144A.1.3 Relationship between Credit Supply Shock and Em-ployment Changes . . . . . . . . . . . . . . . . . . . . 145A.1.4 Relationship between Credit Supply Shock and Em-ployment Changes . . . . . . . . . . . . . . . . . . . . 148A.1.5 Relationship between Credit Shock and Weekly Wage 149A.1.6 Relationship between Credit Shock and EstablishmentGrowth: 2003-2016 . . . . . . . . . . . . . . . . . . . . 150A.1.7 Alternative Credit Shock I . . . . . . . . . . . . . . . . 151A.1.8 Alternative Credit Supply Shock II . . . . . . . . . . . 152ixA.2 Appendix to Chapter 2 . . . . . . . . . . . . . . . . . . . . . . 157A.2.1 Steady State Conditions . . . . . . . . . . . . . . . . . 157A.2.2 Parameter Settings . . . . . . . . . . . . . . . . . . . . 157A.3 Appendix to Chapter 3 . . . . . . . . . . . . . . . . . . . . . . 158A.3.1 The Total Factor Productivity Estimation . . . . . . . 158A.3.2 Additional Figures . . . . . . . . . . . . . . . . . . . . 163xList of TablesTable 1.1 County summary statistics . . . . . . . . . . . . . . . . . . 40Table 1.2 Lender characteristics in 2000. . . . . . . . . . . . . . . . . 41Table 1.3 Lender characteristics in 2002–2006. . . . . . . . . . . . . . 41Table 1.4 Relationship between lender local market shares and locallabor market outcomes: 2003–2006 . . . . . . . . . . . . . 42Table 1.5 Summary statistics: county economic characteristics in 2000 43Table 1.6 Effect of credit supply shock on mortgage market . . . . . 44Table 1.7 Effect of credit supply shock on housing price . . . . . . . 45Table 1.8 Effect of credit supply shock on the unemployment rate . 46Table 1.9 Effect of credit supply shock on unemployment change:sub-period . . . . . . . . . . . . . . . . . . . . . . . . . . . 47Table 1.10 Effect of credit supply shock on labor participation andemployment . . . . . . . . . . . . . . . . . . . . . . . . . . 48Table 1.11 Effect of credit supply shock on private employment: 2003-2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49Table 1.12 Effect of credit supply shock on weekly wage: 2003–2017 . 50Table 1.13 Effect of credit supply shock on weekly wage: subperiod . 51Table 1.14 Effect of credit supply shock on industry-specific wage:2003-2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52Table 1.15 Effect of credit supply shock on quarterly earning by edu-cation groups: 2003–2016 . . . . . . . . . . . . . . . . . . . 53Table 1.16 Effect of credit shock on employment composition: youngvs. old firm . . . . . . . . . . . . . . . . . . . . . . . . . . . 54xiTable 1.17 Effect of credit shock on firm size: 2006–2016 . . . . . . . 55Table 1.18 Relationship between credit shock and local labor market:1990-1999 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56Table 1.19 Relationship between credit shock and local labor market:2001–2003 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57Table 1.20 Effect of credit shock on the labor market: 2003–2017 . . . 58Table 1.21 OLS and 2nd Stage Regressions . . . . . . . . . . . . . . . 59Table 2.1 Parameter settings . . . . . . . . . . . . . . . . . . . . . . . 77Table 2.2 Quantitative results . . . . . . . . . . . . . . . . . . . . . . 77Table 3.1 Numbers of the treated and control firms. . . . . . . . . . 117Table 3.2 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . 118Table 3.3 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . 119Table 3.4 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . 120Table 3.5 Matching properties: means of matching covariates beforeand after matching . . . . . . . . . . . . . . . . . . . . . . 121Table 3.6 The impact of tax cuts on sales growth . . . . . . . . . . . 122Table 3.7 The impact of tax cuts on investment rate . . . . . . . . . 123Table 3.8 The impact of tax cuts on productivity . . . . . . . . . . . 124Table 3.9 Substitution between capital and labor . . . . . . . . . . . 125Table 3.10 The levels of employment and registered capital upon firmestablishment . . . . . . . . . . . . . . . . . . . . . . . . . . 126Table 3.11 The impact of corporate income tax cuts on firm entry . . 127Table 3.12 Stay small or grow? . . . . . . . . . . . . . . . . . . . . . . 128Table A.1 Top 10% largest selected lenders. . . . . . . . . . . . . . . 141Table A.2 Top 10% fastest growth selected lenders. . . . . . . . . . . 141Table A.3 Effect of local market shares on income growth: 2003–2006 142Table A.4 Effect of local market shares on local private employmentgrowth: 2003–2006 . . . . . . . . . . . . . . . . . . . . . . . 143Table A.5 Effect of credit supply shock on housing price: 2003–2016. 144Table A.6 Effect of credit supply shock on local employment rate andlabor market participation rate changes: 2003–2017. . . . . 145xiiTable A.7 Effect of credit supply shock on local private employmentchange: 2010-2017. . . . . . . . . . . . . . . . . . . . . . . . 146Table A.8 Effect of credit supply shock on weekly wage with com-muting zone fixed effects. . . . . . . . . . . . . . . . . . . . 148Table A.9 Effect of credit supply shock on employment share changeby young and old firms: 2003–2006. . . . . . . . . . . . . . 149Table A.10 Effect of credit supply shock on the number of establish-ment growth: 2006–2016. . . . . . . . . . . . . . . . . . . . 150Table A.11 Effect of credit supply shock on the local labor market:2003–2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . 151Table A.12 Effect of credit supply shock on the local labor market:2003–2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . 152Table A.13 Effect of credit supply shock on the number of mortgageexpansion: 2003–2010. . . . . . . . . . . . . . . . . . . . . 153Table A.14 Effect of credit supply shock on the local labor market:2003–2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . 154Table A.15 Effect of mortgage expansion on local unemployment ratechange: 2003–2017 . . . . . . . . . . . . . . . . . . . . . . 155Table A.16 Effect of mortgage expansion on weekly wage change: 2003–2017156Table A.17 Parameter settings for comparative statics . . . . . . . . . 157xiiiList of FiguresFigure 1.1 National trend in mortgage origination . . . . . . . . . . 38Figure 1.2 Credit supply shock. . . . . . . . . . . . . . . . . . . . . . 39Figure 2.1 Steady state and parameter settings . . . . . . . . . . . . 74Figure 2.2 Transition path with positive shock on φb . . . . . . . . . 75Figure 2.3 Transition path with negative shock on φb . . . . . . . . . 76Figure 3.1 The taxable income ranges of affected firms. . . . . . . . 108Figure 3.2 Salience of the tax rate cuts for SMPEs . . . . . . . . . . . 109Figure 3.3 The distributions of taxable income around the SMPE qual-ifying thresholds . . . . . . . . . . . . . . . . . . . . . . . 110Figure 3.4 Average log TFP of the matched and unmatched samples. 111Figure 3.5 Dynamic effects of the corporate income tax rate cut onsales growth . . . . . . . . . . . . . . . . . . . . . . . . . . 112Figure 3.6 Dynamic effects of the corporate income tax rate cut onfirm-level investment rate . . . . . . . . . . . . . . . . . . 113Figure 3.7 The dynamic effects of the corporate income tax cut onfirm-level TFP . . . . . . . . . . . . . . . . . . . . . . . . 114Figure 3.8 Time series pattern of firm entries . . . . . . . . . . . . . 115Figure 3.9 Time series pattern of firm entries . . . . . . . . . . . . . 116Figure A.1 Effect of credit supply shock on private employment change:2006-2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . 147Figure A.2 The distributions of total assets . . . . . . . . . . . . . . . 163xivFigure A.3 The distributions of firm size across financial constrainttertiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164xvAcknowledgmentsI would like to express my deepest gratitude to Henry Siu, Paul Beaudry,and Michael Devereux for their invaluable support, excellent guidance, andinfinite patience throughout my years at UBC. They were extremely gener-ous with their time and encouragement, and taught me how to be a bet-ter economist and researcher. I am also grateful for the support and ad-vice of Viktoria Hnatkovska and Michal Szkup, who devoted a lot of timein helping me. I was also incredibly lucky to have many talented and en-couraging peers at UBC: Aruni Mitra, Adlai Newson, Matthew Courchene,Michael Wiebe, Sev Chenyu Hou, Hao Li, Xiaodi Ni, Tsenguun Ekhbataar,and Natasha Kang. All aforementioned helped me more than they canimagine.Finally, I would like to thank my parents for their unconditional love andsupport throughout my life.xviIntroductionThis dissertation consists of three chapters in empirical macroeconomics.These chapters touch upon areas in household debt, employment, firm in-vestment, and productivity. In these chapters, I primarily use the cross-sectional method to identify the effects of economic shocks to regional econo-mies and enterprises. Using the regional data, the first two chapters investi-gate how mortgage fluctuations influenced the U.S. regional labor markets.In the third chapter, I study how preferential tax treatments stimulate thesmall firm performance.The first chapter asks the question that whether credit-induced boomand bust has any long-term impact on the real economy. To answer thisquestion, I study how mortgage expansion in the early 2000s affected theU.S. local labor market from 2003 to 2017. Great Recession, originated fromthe collapse of the housing market, destroyed over six million jobs in theU.S. during the 2007-2010 period. Many works have suggested the sever-ity of the economic downturn has a strong relationship with the amount ofhousehold debts buildup before the crisis (Jorda` et al. (2015); Mian and Sufi(2018)). Instead of focusing on the recession, this chapter tries to examinehow local labor markets have been affected by mortgage market fluctuationsthroughout the entire episodes of mortgage expansion, contraction, and re-covery.Focusing on U.S. counties, I construct a plausibly exogenous measure ofthe local credit supply shock to capture the supply-driven mortgage expan-sion during the 2002-2006 and use it to explain changes in local labor mar-ket outcomes from 2003 to 2017. Using the Bartik technique (Bartik, 1991),1I measure the county-specific credit shock by interacting the heterogeneouslending strategies of multi-market lenders with a county’s initial exposureto these lenders. This strategy relies on the fact that lenders operate acrossmultiple U.S. counties. Identification requires the initial market shares ofmulti-market lenders to be uncorrelated with the omitted variables, i.e., lo-cal labor market trends. Thus, I require that multi-market lenders did notlocate in regions that were experiencing positive local labor market trends,after controlling for observed county characteristics. I provide supportiveevidence to show that the credit shock is not correlated with the pre-existingtrends in regional labor markets.Result from a regression analysis shows that counties received largercredit supply shocks have not seen significantly different changes in lo-cal unemployment rates or labor participation rate but have shown slowerwage growth. To understand how credit shock affected local labor marketin different stages of credit fluctuations, I analyze three distinct sub-periods:credit expansion (2003-2006), contraction (2006-2010), and recovery (2010-2017). I find that high-credit counties did not experience significantly dif-ferent changes during the expansion period, but did experience larger in-creases in unemployment and larger declines in wage growth during therecession. After the recession, the they experienced a faster recovery in un-employment rates but not wages.To investigate the persistent effect of credit shock on wage growth, Istudy how credit shock affected local wage growth by industries and ed-ucation groups. As some might wonder, the effect of credit shock on wagegrowth might be driven by a specific industry, e.g., real estate, due to thenature of the recession. I show that the credit shock has similar effects onlocal wage growth across construction, financing, real estate, and all otherindustries. Another concern is that the effect of credit shock may be specificto the low-skill workers in the construction sector, which spilled over to allother sectors. I show that coefficient estimates that are very close across dif-ferent education groups. These results suggest that the persistent negativeeffects of a credit shock on wage growth cannot be explained by industry orskill-specific mechanisms, indicating a more general mechanism of credit2shock. Focusing on the labor demand side, I find that high-credit countieshave seen a larger employment shift from young to old firms, together witha larger decline in the total number of establishments. This evidence sug-gests that credit supply shock has negatively affected the local labor demandsince the recession, which could explain the decline in wage growth.In the second chapter, I propose a mechanism to explain the empiricalfindings in chapter one. The framework has three types of agents: workers,patient and impatient entrepreneurs. Workers and impatient entrepreneursare the borrowers and the patient entrepreneurs are the savers in the econ-omy. All agents face borrowing constraints tied to housing value. In addi-tion, entrepreneurs face working capital constraint that they need to financewage bills with intra-period borrowings, tied to housing value as well. Pro-duction technology requires labor input and features decreasing return toscale.By relaxing and tightening borrowing constraints, this framework canmatch the empirical findings: positive credit supply shock does not affectlabor markets, and the negative credit supply shock leads to wage declinesand labor reallocation. The nature of credit effect asymmetry is capturedby occasionally binding working capital constraints. A relaxation in bor-rowing constraint (positive credit supply shock) leads to housing price ap-preciation as the housing demand increases for impatient agents. But it maynot increase the labor demand if the working capital constraint is alreadyslack for all entrepreneurs. Housing value appreciation will not further im-prove the labor market condition. When borrowing constraint is tightenedup (negative credit supply shock), borrowers are forced to deleverage, driv-ing down the housing price. A significant drop in housing price restrictsthe working capital constraint for impatient entrepreneurs and leads to la-bor reallocation from constraint (impatient) to unconstrained (patient) en-trepreneurs. With diminishing marginal labor productivity, labor realloca-tion leads to wage decline.The third chapter, joint work with Wei Cui, Weisi Xie, and Jing Xing,focuses on the China’s recent ”proactive fiscal policy” emphasizes cuttingtaxes for small businesses. We study a key component of this policy, which3sharply reduced the corporate income tax rate for small- and micro-profitenterprises (SMPE). Using gradual increases in the qualifying threshold forSMPEs during 2010-2016 as a natural experiment, we analyze how the ratecut affects newly qualified SMPE firms’ performance based on confidentialcorporate tax returns. We find that the rate cut led to significant increases insales growth, investment and productivity of affected SMPE firms. There isevidence for substitution between capital and labor following the policy re-forms. We further show that the rate cut induced micro-sized firms to enterthe market. These findings are consistent with predictions from our theo-retical model. Our study offers an assessment of an important but under-studied aspect of China’s recent tax policy, and contributes to the discussionon the effectiveness of tax incentives in stimulating small business growth.4Chapter 1Mortgage Expansion andLong-run Labor MarketOutcome1.1 IntroductionA long literature in macroeconomics has focused on studying the role thefinancial sector plays in the real economy.1 Since the recent financial crisis,this topic has received increasing attention in economic research. The crisis,originating from the collapse of the U.S. housing market, has triggered thedeepest recession since the Great Depression, with a 6% decline in the em-ployment rate and the loss of millions of jobs. The slow recovery in the U.S.economy that followed the recession, especially for employment, has raisedconcerns of economic stagnation (Ball, 2009; Fernald et al., 2017; Summers,2014).2 Ultimately, this debate has led researchers to revisit an old question:does a credit boom and bust have a long-term impact on the real economy?This chapter answers this question by examining the long-term impact1Fisher (1933) proposes the debt-deflation mechanism to explain the Great Depression.2More work can be found in Haltmaier (2013), Cerra and Saxena (2008), and Ball (2014),which all have documented cross-country analysis on the long-term losses in output andproductivity after the financial crises.5of mortgage expansion on U.S. local labor markets. Focusing on U.S. coun-ties, I construct a plausibly exogenous measure of the local credit supplyshock during the mortgage boom, and use it to explain changes in locallabor market outcomes from 2003 to 2017.3 Results from a regression anal-ysis suggest that counties with larger credit supply shocks have not seensignificant changes in local unemployment rates or labor participation ratebut have shown slower wage growth. To understand the relationship be-tween credit shocks and local labor market responses in different phases ofcredit fluctuations, I analyze three distinct sub-periods: credit expansion(2003-2006), contraction (2006-2010), and recovery (2010-2017). I find thathigh-credit counties did not experience significantly different changes dur-ing the expansion period, but did experience larger increases in unemploy-ment and larger declines in wage growth during the recession. This wasfollowed by a faster recovery in unemployment rates, but not wages, afterthe recession. In addition, I find that high-credit counties also experienced agreater change in labor reallocation from the beginning of the recession on-ward: larger declines in the young firm employment share (firms less thanfour years old), together with a larger increase in the old firm employmentshare (firms more than five years old).To identify the supply-driven credit growth, I exploit the fact that largelending institutions operated in multiple counties before the housing boom.Using a Bartik variable, originally proposed in (Bartik, 1991), I measurethe county-specific by interacting the heterogeneous lending strategies ofmulti-market lenders with a county’s initial exposure to these lenders. Asdiscussed by Goldsmith-Pinkham et al. (2018), identification relies on theassumption that initial market shares are uncorrelated with any omittedvariables, (e.g. trends in local labor market conditions). 4 Therefore, Irequire that multi-market lenders did not locate in regions that were ex-periencing positive local labor market trends (conditional on the observedcounty characteristics, which I can control for).3Credit and mortgage are used interchangeably in this chapter.4See Adao et al. (2019) and Borusyak et al. (2018) for more discussion about shift-share de-signs.6One obvious concern is that large lenders have private information aboutlocal economic trends and, as a result, make strategic decisions with respectto their branch locations on the basis of this information. If this was thecase, we would expect large lenders to locate in regions with better eco-nomic performance. I argue that this is not supported by the empirical ev-idence, as I do not find any positive relationship between my credit shockmeasure and local labor market outcomes preceding the recession. The re-quired identifying assumption would also be violated if these large lenderssystematically chose to operate in subprime regions, which have worse eco-nomic performance in the long run. I also show that this is not supportedby county characteristics, as credit shocks are not systematically correlatedwith worse economic outcomes nor higher subprime populations in 2000,the year chosen to measure lenders’ market share. As supportive evidence,I compute each county’s initial exposures to the most expanding (top 20%of selected lenders) and least expanding (bottom 20% of selected lenders)lenders and test whether these market shares can explain local labor markettrends during the expansion period. I do not find any evidence that theselenders chose systemically different locations based on local labor markettrends.This chapter contributes to the literature by directly linking the expan-sionary credit shock with local labor market changes, as opposed to linkingcredit contraction and labor market changes, which most of the Great Reces-sion literature has focused on. Mian and Sufi (2014) examine the impact ofcredit contraction on the local demand and employment in the non-tradablesectors. Another strand of literature investigate how credit contraction re-stricts firm’s credit access and its employment demand (Chodorow-Reich,2014; Chodorow-Reich and Falato, 2017; Greenstone et al., 2020). However,credit contraction is not an exogenous event to U.S. economies. Arguably,the relaxation in banking regulation, the rise in private labeled securitiza-tion, and the growth in subprime mortgages all have contributed to the ac-celeration of mortgage expansion and sown the seeds for the subsequentcollapses in mortgage and housing markets.5 Empirically, several papers5Favara and Imbs (2015) have examined how relaxing regulation in inter-state bank branch-7have documented a strong correlation between household debt accumula-tion and the severity of the subsequent economic downturn (Jorda` et al.(2015); Mian and Sufi (2018)). However, very few studies have directlymade the causal link between mortgage expansion and economic output.6Instead of focusing on the recession, this chapter tries to provide a compre-hensive evaluation on how the supply-driven credit expansion affected locallabor markets throughout a boom-bust episode including the expansionary,contraction, and recovery periods.This chapter is closely related to Yagan (2019), who employs the lon-gitudinal administrative data to study the long-term impact of the GreatRecession on the local labor markets. He finds that individuals who stayedin areas with larger unemployment were more likely to be unemployed in2015. The difference in our findings is mainly caused by the different timeperiods we study. Yagan (2019) focuses on the employment change from2007 to 2015, whereas my work extends the time period to 2017. As I willshow in the empirical analysis, the high-credit counties accelerated the em-ployment recovery after 2015.To understand the nature of credit effect asymmetry, one needs to un-derstand how credit supply transmits to local labor markets. This chapterdoes not specify a particular channel through which a credit shock affectslocal labor markets. In the recession, the negative impact of a credit sup-ply shock on local labor markets could work through many different chan-nels: local mortgage markets, housing markets, or balance sheets’ effecton financial institutions. These mechanisms have been empirically exam-ing affects mortgage origination and housing price growth. Di Maggio and Kermani (2017)documents the impact of preemption of anti-predatory lending on the lending strategies ofnational banks. Demyanyk and Van Hemert (2011), Dell’Ariccia et al. (2012), and Mian andSufi (2009) discuss the contribution of subprime lending to the mortgage expansion preced-ing the financial crisis. Nadauld and Sherlund (2013), and Keys et al. (2010) document theimpact of securitization on subprime lending.6Di Maggio and Kermani (2017) exploit the impacts of the federal preemption of nationalbanks from anti-predatory lending laws on local mortgage markets and labor marketsthroughout the mortgage boom and bust periods. Gilchrist et al. (2018) identify the non-local mortgage supply shock and use it to evaluate the impact expansionary and contractionshocks on regional economic outcomes during the expansion and contraction period, sepa-rately.8ined in the Great Recession literature: credit contraction leads to declinesin housing prices and household net worth, affecting household demandand employment in the non-tradable sector (Di Maggio and Kermani, 2017;Mian et al., 2013); contraction in business lending due to bank’s liquidity cri-sis leads to job losses (Chodorow-Reich, 2014; Chodorow-Reich and Falato,2017; Greenstone et al., 2020); and contraction in household credit supplyleads to household deleverage and affects local demand and employment(Garcia, 2020; Gilchrist et al., 2018; Mondragon, 2018). In theory, the creditsupply shock could affect local labor markets through all the channels de-scribed above.To understand the persistent effects of a credit shock on wage growth,I investigate the industry-specific local wage growth. I find that the creditshock has similar effects on local wage growth across construction, financ-ing, real estate, and all other industries. Similarly, I report coefficient esti-mates that are very close across different education groups. This suggeststhat the persistent negative effects of a credit shock on wage growth cannotbe explained by industry or skill-specific mechanisms. Focusing on the la-bor demand side, I find that high-credit counties have seen a larger employ-ment shift from young to old firms, together with a larger decline in the totalnumber of establishments. This evidence suggests that credit supply shockhas negatively affected the local labor demand since the recession, whichcould explain the decline in wage growth. This finding is related to theaccelerated decline in firm entry after the Great Recession documented bySiemer (2012) and Davis and Haltiwanger (2019). Davis and Haltiwanger(2019) find that the decline in the young firms’ employment share is ex-plained by depressed housing prices using land supply elasticity as an in-strumental variable.All of these empirical findings are robust to different measures of creditsupply shock when I restrict the analysis to include only the larger lenders.The credit supply shock has also shown that pre-existing local labor markettrends do not exist (measured between 1990–1999) in counties with largercredit supply shocks. Another concern is that high credit shock regionshave systematically different responses to aggregate economic fluctuations9due to some unobserved characteristics. I conduct a placebo test on the2001 recession episode and find credit shock does not explain the local labormarket changes in the 2001 recession.The structure of the chapter is organized as follows. Section two de-scribes the data used in the analysis. Section three discusses the empiricalstrategies. Section four presents the empirical results. The conclusion isdrawn in the last section.1.2 DataI construct a wide range of economic variables at the county level. Themain data source is the home mortgage dataset, taken from Home Mort-gage Disclosure Act (HMDA), an application-level dataset published byFederal Financial Institutions Examination Council (FFIEC). The FFIEC re-quires mortgage lenders that have offices in the metropolitan areas and totalassets above a certain threshold to disclosure detailed mortgage informa-tion every year.7 According to Dell’Ariccia et al. (2012) and Avery et al.(2007), HMDA covers about 80%-90% of all mortgages written during the2000s. For this chapter, I will measure county mortgage growth with mort-gages for home purchase purpose only, as opposed to mortgages for homeimprovements or refinancing.8The Housing Price Index (HPI) is constructed by the Federal HousingFinance Agency (FHFA). It computes the housing price index using data onsingle-house conforming loans taken from the Federal Home Loan Mort-gage Corporation (Freddie Mac) and the Federal National Mortgage As-sociation (Fannie Mae). The price index is constructed with a weighted-repeated sales methodology. An alternative measure is the Zillow homeprice index, which covers fewer counties than HPI. I will use the Zillow in-dex as a robustness check in the empirical section.7Mortgage lenders include both depository and non-depository institutions. The non-depository institutions need to report all loans in MSAs where they have more than fiveapplications.8With HMDA, it is hard to measure credit growth with refinancing mortgages, as its repay-ment schedule is unobserved.10To measure unemployment changes in county-level labor markets, I col-lect the average annual unemployment and labor force data from Local AreaUnemployment Statistics (LAUS). The private employment and weekly wageare collected from the Quarterly Census of Employment and Wage (QCEW),which provides employment and wage information by 4-digit NAICS in-dustry code. Note that both employment and weekly wage are computedas annual average. The private employment measures the annual averageof monthly employment levels for a given year. The average weekly wage iscomputed as dividing annual payroll by annual employment and the totalnumber of weeks in the year. As a robustness check, I also include earningsby education attainment data, taken from the Quarterly Workforce Indica-tor (QWI), which provides local labor market characteristics by industry,worker demographics, employer size, and age. The source data of QWIis Longitudinal Employer-Household Dynamics (LEHD) linked employer-employee micro-data. The last part of the paper focuses on the employer-side characteristics. I compute the employment by firm age, taken fromQWI as well. The establishment by employment size data is taken fromCounty Business Pattern (CBP), which shares the same data source as QCEW.The remaining county economic and demographic characteristics are asfollows: poverty rate and household median income are from the 2000 Cen-sus; population and demographic data are from the U.S. Census; incomedata are from Statistics of Income published by the U.S. Internal RevenueService; and the subprime population is from Equifax New York FederalReserves.1.2.1 Summary StatisticsTable 1.1 presents the summary statistics for local labor markets from 2003to 2017, excluding Hawaii and Alaska. I divide the period into three sub-periods: mortgage expansion (2003–2006), contraction (2006–2010), and re-covery (2010–2017). All variables are measured by the change between thestart and end of the period, except for the mortgage. Mortgage change forthe expansion period is measured as the difference between average yearly11mortgage origination (in dollar value) in 2003–2006 and mortgage origina-tion in 2000; for the contraction period as the difference between averageyearly mortgage origination in 2007–2010 and its 2006 level; and for the re-covery period as the difference between the average mortgage in 2011–2015and its 2010 level. Note that in the contraction period, housing price, wage,and establishment changes are taken over the period 2007–2010, as 2007 isthe start of the turning point for these variables.Mortgage origination almost doubled during 2003–2006. Figure 1.1 plotsthe time series of total mortgage origination and home purchase mortgageonly. As we can see, the fastest growth happens during 2000–2002. In thisperiod, the mortgage growth is commonly viewed as driven by interest ratedecline, as the federal funds rate declined from 6.5% to 1.7%. After 2002,mortgage origination remained at a high level till 2006. Note that mort-gage origination is the flow variable. The high growth in mortgage during2003–2006 has been arguably viewed as supply-driven mortgage expansion.Following the literature, this chapter will take 2003–2006 as the mortgageexpansion period. During the contraction period, mortgage originationsdeclined by 24% annually and kept falling by 11% per year in the recoveryperiod.9As for labor markets, unemployment rates declined by 1% on averageduring the expansion period, increased by 5% in the recession, and declinedby 5% again in the recovery period. Local private employment on averageincreased 5% during the expansion period, followed by a 6% decline in therecession and 8% growth afterward.10 Dynamics of local wage growth areless obvious than employment changes across different periods. The annualwage growth is 4% during the expansion period, which slowed down to1.75% in the recession and recovered to 2.4% after the recession. Similarly,establishments experienced an annual growth of 1.3%, followed by a 0.3%decline in the recession and 0.4% growth after the recession.9Note these are county level mortgage statistics which may not fully represent the nationaltrend.10The employment change does not take into account population growth.121.3 Empirical Strategy1.3.1 Research DesignTo understand the local labor market changes in the last credit boom, bustand recovery episodes, let me decompose it into four components: one reg-ulated by the common national trend, one affected by credit fluctuations,one driven by its underlying trend and the rest driven by other idiosyncraticshocks during this period. The equation is summarized as follows:∆yi,2017−2003 = β∆xi+∆ypi,2017−2003+ c+ εiwhere ∆yi,2017−2003 represents the actual change from 2003 to 2017 for lo-cal economic variable and notation i denotes the county observation. ∆xistands for the variation in credit fluctuation during this period, and β is theparameter of interest, capturing the long-term impact of credit fluctuationson the local labor market from 2003 to 2017. ∆ypi,2017−2003 represents the po-tential local labor market trend, in the absence of the credit fluctuations. Forinstance, job losses in manufacturing towns are driven by industry-specifictrends.11 The main challenge for identifying β is to isolate the correlationbetween credit fluctuations and regional economic trends. As ∆ypi,2017−2003is unobserved, we can only run the regression if it is uncorrelated with thecredit fluctuations.Before introducing the research design, let me discuss the potential cor-relation between ∆ypi,2017−2003 and ∆xi. Focusing on the credit expansion pe-riod in 2003–2006, the predominant narrative is that relaxing the mortgagelending standard, caused by financial innovation, an increase in private se-curitization, and a relaxation of banking regulation, led to the mortgageexpansion and the subsequent crisis.12 The influential work by Mian and11See Charles et al. (2016) for a discussion of the relationship between declining manufactur-ing employment and local housing market boom12Favara and Imbs (2015) use relaxation in geographical restrictions on banks to show mort-gage expansion due to an increase in mortgage supply. Di Maggio and Kermani (2017)estimate the effect of mortgage growth due to the preemption of national banks from anti-predatory lending. Keys et al. (2010) document the effect of securitization in subprime mort-13Sufi (2009) shows that mortgage growth has negatively correlated with theincome growth across zip-code levels within the same metropolitan areas.If regions with poor economic conditions have suffered more in the pastcredit boom-busts, β suffers from negative bias. On the other hand, there isa growing literature supporting the alternative view of credit demand story.Adelino et al. (2015) show the mortgage origination increased for borroweracross all income levels and credit scores.13 They have also suggested theoverall positive relationship between mortgage and income growth at indi-vidual levels. Again, if the mortgage credit expansion was mostly driven byincome growth and improvement of economic condition, the estimates willbe upward biased.To solve this problem, I need to find an exogenous force that affects creditexpansion but is not related to the long-run regional trend. The solution isto measure the supply shift in mortgage expansion.1.3.2 Identifying Credit Supply ShockIdentifying the supply effect from an equilibrium variable is a commonchallenge that many economic researchers face. To construct an exogenouscredit supply shock, I will rely on different lending strategies taken by largelending institutions that operate in multiple areas.14 The intuition behindthis identification strategy is that some lenders are more lenient in the lend-ing criteria, and areas that have access to such lenders are considered toreceive a positive credit supply shock and are more likely to observe fastermortgage growth.So, the local exogenous credit supply shock is constructed as nationalgrowth of bank credit interacted with its initial share in the local mortgagegage on mortgage default.13Albanesi et al. (2017) document the role of prime borrowers in the 2001–2006 mortgageexpansion and argue that mortgage growth in the subprime group is confounded with thelife cycle demand for borrowers who were young at the start of the boom14Lending institutions include depository and non-depository institutions. In this chapter, Iwill use the word “lender” to refer to both.14markets. The specification is as follows:∆zi =∑jsi j∆Z j (1.1), where ∆Z j is mortgage growth by bank j and si j denotes the initial marketshare of bank j in county i. The strategy described here follows the Bartik(1991), which uses the aggregate industry shocks and the share of theseindustries in the local market to predict the local demand shock. As a pop-ular method, this method has been employed to produce many influentialworks (Autor et al., 2013; Blanchard and Katz, 1992; Card, 2001). In theGreat Recession literature, this method is also often used to identify thecredit supply shock (Chodorow-Reich, 2014; Gilchrist et al., 2018; Glancy,2017; Greenstone et al., 2020).Most works state the identifying assumptions for using the Bartik tech-nique as the exogeneity of national shocks. However, as pointed out bythe recent work of Goldsmith-Pinkham, Sorkin, and Swift (2018, GPSS),using the Bartik technique to construct exogenous credit shock is equiva-lent to using the local shares as the exogenous variation. As ∆Z j, the aggre-gate growth of bank j, is common to all county observations, the underlyingvariation in this constructed credit shock relies on the variation in the localshares, i.e., areas that received the shock should not be systematically differ-ent from areas that did not receive the shock. Numerically, using the Bartiktechnique is equivalent to applying generalized moment of method (GMM)with vectors {si j} j∈J being instruments and {∆Z j} j∈J being a weighting ma-trix. Therefore, the sufficient condition for Bartik relies on the assumptionthat local shares of every lender {si j}i need to be uncorrelated with localeconomic trends ∆ypi . In other words, the identification assumption requireslenders not to locate their branches based on correctly predicting the futurelocal labor market trends, and this assumption needs to be applied to eachlender j.E(si j∆ypi ) = 0, ∀ j ∈ J.15As an illustrative example, the Lehman Brothers bank is employed inthis exercise to construct credit supply shock. It grew by 398% annuallyin mortgage origination during 2002–2006, suffered severe losses in the re-cession and declared bankruptcy in 2008.15 As its credit expansion is largelybank-specific and not driven by specific local demands, areas that have higherexposures to Lehman Brothers would experience higher credit supply shock.The identification requires these areas not to have systematically differenteconomic structures than areas with low exposure to the Lehman Brothersbank in 2000.The identifying assumption proposed by GPSS is sufficient but not nec-essary. An alternative identification strategy exists, proposed by Borusyak,Hull, and Jarave (2018, BHJ), as well as by Adao, Kolesar, and Morelas(2019, AKM). If the initial shares are related to local economic growth, theconstructed credit shock remains exogenous as long as the national growth∆Z j is exogenous to the average local economic trends that it is exposed to.However, this requires a different convergence condition. Instead of con-verging with an increasing number of observation i, this requires a law oflarge number on shock ∆Z j must hold.16 As the convergence relies on thegrowing number of national shocks ∆Z j, one needs to adjust the inferenceaccordingly.In this chapter, I will focus on the approach proposed by GPSS and arguethe identifying assumption on local market shares si j. To ensure lenders’initial local market shares are uncorrelated with local labor market trends,lenders should at least operate in multiple counties. A lender that only op-erates in one county would be perfectly correlated with a county-specificlabor market trend if it exists. However, due to historical regulations, thebanking markets have been highly geographically segmented in the U.S.1715The mortgage expansion of Lehman Brothers is calculated as annual mortgage originationduring 2002–2006 relative to its 2000 level.16The Herfindahl index of the average expected shares converges to zero, i.e., ∑i(∑ j=1 s2i j)→ 0.17The McFadden Act of 1927 permits states to restrict branching for national banks. The BankHolding Company Act in 1956 restricts entry by out-of-state banks and bank holding compa-nies. Starting from 1994, the passage of the Interstate Banking and Branching Efficiency Actstarted the deregulation waves for bank branching. See Rice and Strahan (2010) for morediscussions.16In 2000, the median county coverage per lender is 7 (calculated by the au-thor using the HMDA dataset) To avoid this issue, I construct the selectioncriteria to choose large lending institutions that operate in multiple regions.The selection criteria are as follows:• It needs to operate in 100 or more counties in 2000.• It needs to operate in more than one census region.• It kept operating during 2000–2006.The first two requirements are used to select lenders with broad geo-graphic coverage. Note that large lenders are chosen based on their mort-gage market coverage in 2000. That is because the initial bank share is de-fined in 2000, which is considered as the beginning of the housing marketboom.18 The last criterion requires the selected lenders to remain operatingduring the credit boom period. This requirement is needed as the aggre-gate bank credit growth will be computed as the average mortgage increasefrom 2002–2006.Note that some lenders are subsidiaries of commercial banks or bankholding companies. These criteria are applied at the reporting lender level,instead of the parent company level. This is because, in the late 1990s andearly 2000s, the U.S. banking industry has experienced a historically highnumber of mergers and acquisitions activities.19 In the literature, the stan-dard approach (Bernanke et al., 1991; Greenstone et al., 2020) treats theacquiring and acquired lenders as the same entity throughout the sameperiod. However, one main reason for mergers and acquisitions to hap-pen is to enter new geographical markets for profit-seeking purposes. Thischallenges the identifying assumption that lenders locating in counties withgrowing economic trends or growing mortgage demands are more likely tobe merged. If this is true, a positive relationship between bank shares andlocal economic trends is generated. So, instead of focusing on the parent18As shown in Figure 1, the first wave of housing boom started in 2001, triggered by interestrate decline.19According to Pilloff (2010), there are 3517 completed bank mergers from 1994 to 2003.17company, this chapter uses lender-level aggregation. Note that if one hold-ing company consists of two selected lenders, the identifying assumptionthat holds at the lender level will also hold at the holding company level.One might question if the selection criteria are set arbitrarily. Indeed,there is no theoretical reasoning to justify these selection criteria. The goalis to select large lenders that do not concentrate their mortgage supply in afew areas, in which case bank credit growth is more likely to be driven bylocal demands. Note that it is not an issue if some lenders satisfy the iden-tifying assumption but are excluded from the selected group. As pointedout by GPSS, the identifying assumption needs to be applied to every se-lected lender. In the empirical section, I will show that the main results arerobust to credit shock constructed with stricter selection criteria. Selectingfewer lenders if they satisfy the identifying assumption is also not an issue.With too few lenders, the explanatory power of credit shock on actual localmortgage growth will be affected, and thereby the relevancy condition.1.3.3 Lender CharacteristicsIn this part, I will compare lender characteristics in the mortgage marketbetween the selected lenders and those not selected. Based on the selectioncriteria listed above, I select 215 out of 7585 lending institutions in 2000,which covers 36% of total mortgage origination in 2000. In the Appendix, Ilist the top 10% largest and fastest-growing lenders in the selected lenders.Among them, about 30-40% ceased mortgage business during the financialcrisis. Table 1.2 describes the average size of mortgage markets and geo-graphical coverage for each group of lenders. The median county coverageis seven for the non-selected lenders and 333 for the selected ones. In termsof state coverage, the selected lenders operate in at least three states, withmedian coverage of 32 states. The non-selected lenders show a median cov-erage of only one state. I can conclude the selected lenders are sufficientlylarger and cover more regions than the non-selected ones. This differenceis also reflected in the mortgage application and origination volumes: theselected lenders, on average, receive more mortgage applications and orig-18inate more mortgages. Note that there are a few non-selected lenders thatsatisfy the selection criteria for being large lenders (in the max column ofnon-selected lenders). These lenders were not selected because they didnot operate consecutively in the housing boom period 2002–2006.20 As Iwill discuss later, the aggregate bank mortgage growth employed an av-erage mortgage growth rate in 2002–2006, so it is important to make surelenders operated consecutively during this time period.Table 1.3 shows the mortgage market characteristics of two groups oflenders during 2002–2006. Overall, most lenders experienced a boom in themortgage market, with increasing geographical coverage and mortgage ap-plications. The selected lenders had even faster mortgage growth, with mar-ket share increasing from 36% in 2000 to 51% during 2002–2006. Althoughthe selected lenders are, on average, growing faster than non-selected ones,not all selected lenders were successful in the mortgage market. Some hadsignificantly decreased their mortgage market coverage from more than 100counties to 5.69 counties, as shown in the min column of selected lenders.These lenders are going to contribute negative mortgage supply shock whenpredicting the credit shock. This suggests that among the initial large lenders,there are still significant variations in the aggregate lending growth.Although the selected lenders are on average large and expanding in-stitutes, a concern remains that the selected lenders might have better pri-vate information about the future local economic trends and choose theirlocations accordingly. To resolve this concern, I select the top 20% (ap-proximately 40 lenders) and the bottom 20% selected lenders based on theirmortgage growth rates (computed as average annual mortgage originationduring 2002–2006 relative to 2000 level) and compute the total local marketshare each group covered in 2000. Using the market share of each group,I test whether regions with different exposures to the fastest (slowest) ex-panding lenders have systematically different labor market trends duringthe expansion period. Table 1.4 shows the test results for changes in theaverage weekly wage and unemployment rate for each group of lenders.20During 1994–2003, the U.S. banking industry experienced a big wave of merger and acqui-sition activities, due to Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994.19The left two columns present the relationship between local market sharesof bottom 20% lenders and local labor market outcome and the right twocolumns show the relationship for top 20% lenders. This regression in-cludes the same control variables as in the actual regression models. Thesetwo tables show that both the unemployment rate and weekly wage changesare not explained by the mortgage market shares of the fastest and slowestgrowing lenders. This statement also holds for changes in private employ-ment and income per capita, which can be found in the Appendix.1.3.4 Credit Supply ShockTo construct mortgage supply shock, I use the average annual mortgageorigination of selected lenders during 2002–2006 relative to their base yearlevel of 2000 as the measure bank.21 This number measures the yearly growthrate of mortgage origination during the expansion period.∆Z j = log(Z¯ j,2002−2006)− log(Z j,2000)With national mortgage growth, the credit supply shock is measured as thesum of interaction between a bank’s local mortgage market share in 2000and its aggregate growth in mortgage supply during 2002–2006.22 In thischapter, I will not specify the source of difference in the mortgage suppliesacross lenders because, as discussed before, the exogeneity condition is im-posed on the bank’s initial market share instead of the bank’s credit supplyshock. The national mortgage growth is employed to construct the creditshock to ensure the credit shock can predict local mortgage growth.2321I choose the time period of 2002–2006 to measure mortgage expansion. This is because mort-gage growth accelerated in different years for different types of lenders. Some mortgagecompanies and national banks have faster mortgage origination starting from 2002, whereassome thrift banks started to grow in 2004.22Note that I follow the literature to leave out the actual mortgage origination zi j whenconstructing the credit supply shock for county, i.e., ∆Z j,−i = log(Z¯ j,−i,2002−2006) −log(Z j,−i,2000).This eliminates the finite sample bias.23The national mortgage growth includes mortgages for all purposes: home purchase, homeimprovement, and mortgage refinance. This is to be consistent with the measure of lender’sinitial shares in the local mortgage markets. The local mortgage market shares measurethe exposure of each county to the lender’s supply shock. It is more accurate to measure20As supportive evidence for the identification, Table 1.5 summarizes thecounty characteristics in 2000, the initial year of bank market share, to showthat the credit shocks have not systematically favored good or bad economicregions. Counties are split into two groups based on the credit shock theyreceived: one with above-median credit shock and one with below-medianshock. As shown in the table, all the listed economic characteristics arenot statistically different across the two groups. In particular, the subprimerates are almost the same across the two groups. As argued by Mian andSufi (2009), the mortgage expansion disproportionately favored the low-income areas with high subprime credit populations. If one believes theseregions had slower potential growth than high income and prime regions,failing to disentangle the credit growth driven by subprime regions willoverestimate the long-term impact of credit fluctuation on local labor mar-kets. If the mortgage expansion had mostly been driven by the local demandof prime group with high income, as Adelino, Schoar, and Severino (2017)suggested, it is again important to ensure the credit supply shock does notfavor regions with good economic conditions. As Table 1.5 shows, the twogroups are very similar in terms of establishment per capita, employmentrate, and unemployment rate.The above-median credit supply group has, on average, higher weeklywage, income per capita, and median household income, but the differencesare not statistically significant. This table also shows the construction andmanufacturing employment shares. As regions have been exposed to differ-ent industry compositions, industrial trends could drive regional economicgrowth. If credit shock occurred to regions with high employment sharesof a particular industry, the estimation would be contaminated by indus-trial trends. This concern is particularly valid for manufacturing regions, asthe manufacturing industry suffered significant job losses during the sametime period (Charles et al., 2016). The last credit boom and bust had par-the lender’s local market share with its total mortgage origination, instead of just for homepurchase. In the appendix, I measure the national mortgage growth and its local marketshares with only home purchase mortgages and show that all the results are robust to thismeasure.21ticularly hit the construction sector and affected regions with a high con-struction share. The impact of the credit boom and bust should not onlybe attributed to the construction fluctuations. This table suggests construc-tion share is close to the same for the two groups while the manufacturingindustry is, on average, higher in below-median credit regions. This mightraise some concern about underestimating the impact of the credit boomand bust.Given the constructed credit shock, Figure 1.2 plots the map of creditsupply shock after controlling for the set of county characteristics shownin Table 1.5. The shock is normalized by its standard deviation. This mapshows little regional concentration but sufficient variation across countieswithin states. In the next section, the regression model will include statefixed effects to control for state-related economic trends or economic fluc-tuations due to state policy changes.1.4 ResultsIn this section, I discuss the results based on the identification strategy de-scribed before. The regression specifications are as follows:∆yi,2017−2003 =βi∆zi+β2wi+δs+ εiwhere i denotes county observation and s denotes for state. The dependentvariable defines the percentage change in yi over 2003–2017. wi includes thecounty characteristics listed in Table 1.5 as control variables and δs repre-sents the state fixed effect. ∆zi is the constructed mortgage supply shock.This regression uses the reduced form IV specification, which does notspecify the endogenous independent variable and directly uses credit sup-ply shock as the explanatory variable. The reasons are twofold. First, byspecifying the endogenous variable, the estimate will be specific to a par-ticular causal statement between an explanatory variable and the variableof interest. As the credit supply shock is constructed using the Bartik tech-22nique, the natural choice would be actual mortgage growth in 2002–2006.By doing so, the estimate is specific to the impact of mortgage expansion onlocal labor markets. However, this is not the only possible choice. Anotheroption is to use housing price rise preceding the recession as the explana-tory variable, which explores how real estate appreciation affects the localeconomy. These two mechanisms could happen simultaneously and are notmutually exclusive. However, it would be ideal for estimating the overallimpact of credit expansion and the housing market fluctuations on the realeconomy.Second, the relationship between the credit shock and the outcome vari-ables are of research interest. By using the credit shock as the explanatoryvariable, this regression empirically answers the question that how localeconomies respond to credit supply shocks. This is closely related to the the-oretical literature in studying how changes in borrowing constraints prop-agate in the real economy and shape the labor market.24. This question hasbeen studied intensively in the theoretical framework, but this chapter triesto answer it empirically.Note that the identification assumption remains the same with two-stageleast square (2SLS). Using the reduced-form IV makes it difficult to inter-pret the coefficient estimates. Throughout this chapter, I will only answerthis question qualitatively. As readers might still want to see the 2SLS esti-mate, the second stage results are included at the end of the empirical sec-tion using county mortgage growth as the independent variable.The regressions are operated on U.S. counties for 48 states, excludingHawaii and Alaska. The regressions are weighted by the total populationin the 2000 census. Extreme observations within the 1% tail of distributionare excluded from the regression.25 The standard errors are clustered at thestate level to allow for within-state correlation due to state-related institu-tional differences or spatial correlations.24Theoretical literature can be found in Justiniano et al. (2019), Midrigan and Philippon(2011), Korinek and Simsek (2016), Guerrieri and Iacoviello (2017), and Eggertsson andKrugman (2012)25Counties excluded from the regression are in the 1% tail of the dependent variable, whichmeans in each regression, there are different counties being excluded.231.4.1 Relationship between Credit Shock and Housing MarketBefore exploring the relationship between credit shock and the labor marketoutcome, I first show that the predicted credit supply shocks have strong ef-fects on local housing markets by showing their impact on mortgage growthand housing price in both the housing boom and bust periods. This sectionpresents evidence on how credit shock affects local housing markets in dif-ferent periods.Effect on Mortgage GrowthFirst, I show that the predicted credit supply shock has significant impactson the local mortgage markets. Table 1.6 presents the estimates of creditsupply shock on the local mortgage changes in the housing boom and bustperiods. In columns one and two, the dependent variable is defined as themortgage origination from 2003–2006 relative to its 2000 level. In columnsthree and four, the dependent variable is the mortgage origination from2007–2010 relative to its 2006 level.Note that this regression model requires a slightly stricter identifying as-sumption than the one discussed above, i.e., the initial bank location shouldnot be correlated with the regional trend in each sub-period (2003–2006and 2007–2010), as opposed to the long-run regional trend. Although theassumption is slightly more restrictive, it is highly unlikely that the creditshock does not satisfy the assumption with each sub-period but remainswith the longer time horizon. One could view two assumptions to be equiv-alent.The estimates suggest that a one percent increase in credit supply shockpredicts a 0.4–0.74% increase in the county mortgage origination during theboom period. In column three and four, the negative estimate suggests a onepercent increase in bank credit supply lead to a 0.45–0.72% decline in themortgage growth during the bust period. It is not surprising that the creditsupply shock had a positive impact on local mortgage markets in the boomperiod. The subsequent negative effect is not straightforward and needsmore detailed examinations. As it suggests, areas that had larger credit24supply shock before 2007 experienced larger mortgage market contractions.One explanation is the credit supply contraction from the lender side in thebust period, as the lenient lenders were more financially stressed during thefinancial crisis and had to reduce mortgage supplies. This mechanism hasbeen explored and supported by a few works, ranging from the impact ofthe bankruptcy event of Lehman Brothers (Chodorow-Reich, 2014) and Wa-chovia (Mondragon, 2014) to mortgage supply contraction by multi-marketlenders (Garcia, 2019; Gilchrist et al., 2018). Another explanation could bethat the areas with higher credit supply shock accumulated more householddebt and were more responsive to interest rate fluctuations or adverse eco-nomic shocks. In the recession, deterioration in local economic conditionsled to declines in mortgage demand.Effect on Housing PriceAfter exploring the effects of credit shock on mortgage markets, I proceedwith the housing price changes. Table 1.7 shows the impact of credit shockon housing prices. The housing price index is taken from the FHFA. Thealternative housing price measure uses the Zillow housing price index, andthe results are consistent with Table 1.7, shown in the Appendix. The firstcolumn shows the overall effect of credit shock on housing price growthfrom 2003 to 2017. The negative estimates suggest the housing market hasnot yet recovered from the last recession episode. Repeating the exercisewith each sub-period of housing boom 2003–2006, bust 2006–2010, and re-covery 2010–2017, I find that the credit supply shock has predicted the hous-ing price fluctuations in each sub-period. Specifically, a 1% rise in bankcredit supply shock leads to around a 0.11–0.32% rise in housing price dur-ing the boom and a subsequent 0.2–0.41% decline in housing price duringthe recession, followed with little change after the recession. The effect ofcredit shock appears to be slightly larger in the recession than the expansionperiod, causing the overall effect to be negative. Like mortgage growth,I find the credit supply shock has led to boom and bust dynamics on thehousing price. Together with the mortgage growth estimates, these results25suggest that regions that had larger housing and mortgage markets boomalso experienced the larger market disruption in the recession. 261.4.2 Relationship between Credit Shock and Labor MarketGiven the results shown in the previous section, I have established the rela-tionship between credit supply shock and the housing market dynamism incredit expansion and contraction periods. I will continue to examine the im-pact of credit shock on local labor markets. In this section, I will show howunemployment rates and wages respond to the credit shock in the long-termand as well as in each sub-period.Relation between Credit shock and UnemploymentTable 1.8 presents the regression estimates of credit shock on the long-termunemployment rate change, from 2003 to 2017. The effect of credit supplyshock on the unemployment rate change is negligible in the long run. A 10percent increase in credit shock during 2002–2006 leads to a 0.002–0.009 per-centage point decrease in the unemployment rate from 2003 to 2017. Notethat the estimates have not changed significantly as county characteristics,industry controls, and state fixed effects are sequentially added in the re-gression. State fixed effects help to control for state-level differences in un-employment policies. Adding state fixed effect reduces the coefficient byfour times, suggesting the local unemployment rates vary systemically bystate, consistent with the findings by Blanchard and Katz (1992). Althoughthe coefficients are not precisely estimated, with small standard errors andtight confidence intervals, the results suggest the effect of credit shock onunemployment changes is negligible. As readers might find it hard to inter-pret this coefficient, I will show the 2SLS results at the end of the empiricalsection, using county-level mortgage growth as the explanatory variable.Having shown the negligible impact of credit shock on unemploymentchanges from 2003 to 2017, one might ask whether the ”zero” effect is a26Note that this chapter cannot answer the question that how the housing market started tocollapse in the first place. The evidence presented here is cross-section and it suggests thathigh-credit counties experienced larger disruption in the local housing market.26result of net ”zero”, i.e., credit shock had different effects during the boomand the bust, and summed to zero; or the credit shock has never affected un-employment throughout this time period. To answer this question, the un-employment series is broken into three sub-periods, and the effect of creditshock on unemployment fluctuations is estimated in each sub-period.In Table 1.9, the first two columns show credit shock cannot explainthe unemployment changes during the housing boom, suggesting no posi-tive spillovers from local mortgage markets to labor market unemployment.This result also rejects the hypothesis that credit supply shock happened inareas with better or worse local labor market trends. In the recession pe-riod, the positive estimates suggest that counties with positive credit supplyshock experienced higher increases in unemployment rates. A 10 percentincrease in credit shock increased unemployment rates by 0.16 to 0.5 per-centage points. After the recession, the effect of credit shock on unemploy-ment rate change is reversed. The coefficient estimates are almost the sameas the one estimated in recession, with the opposite sign. This suggests thatareas had higher credit supply shock have experienced faster recovery inunemployment after the recession. If I sum the coefficients in the recessionand recovery period, the overall effect of credit supply shock on unemploy-ment fluctuations is close to zero. Note this does not mean that the labormarket has fully recovered from the Great Recession. However, long-runchanges in local labor markets cannot be explained by credit supply shockin the early 2000s.The sub-period analysis shows the effect of credit expansion on local un-employment rates is a business cycle phenomenon, i.e., its effect is negativein the recession and positive in the recovery period. The alternative phe-nomenon is labor market hysteresis, i.e., the negative effect of labor marketshock on unemployment persists after the recession, which is commonlyfound in European economies (Blanchard and Summers, 1987). The busi-ness cycle effect found in Table 1.9 is consistent with the state-level unem-ployment patterns documented by Blanchard and Katz (1992). In addition,they find that unemployment and employment respond differently to re-gional shocks: the effects of labor market shocks on employment tend to be27permanent. In the next part, I will show such patterns are not found in thelocal labor markets.As unemployment rates do not measure extensive margin adjustmenton the labor market, I employ the dependent variable as changes in em-ployment rate and labor participation rate to show that results are consistentwith the unemployment dynamics. Table 1.10 shows the overall changes inthe labor force participation rate and employment rate from 2003 to 2017,explained by credit supply shocks. All estimates are positive insignificant,suggesting that there is no clear evidence of extensive labor market adjust-ment.27 Table 1.11 shows the sub-period analysis for average monthly pri-vate employment changes. The estimates are consistent with the resultsshown in Table 1.9, that areas with higher credit shock suffered more joblosses in the recession and recovered faster afterward. This is different fromwhat Yagan (2019) found. Using longitudinal administrative data, he findsthat individuals who stayed in areas with larger unemployment rise in therecession are more likely to be unemployed in 2015. The main difference be-tween my results and his is my analysis focuses on longer periods. Privateemployment accelerated the recovery after 2015. I present yearly changesin local private employment in response to credit shocks from 2006 to 2017,attached in the Appendix. Although the regression results do not suggestthe labor market adjustment through extensive margin, one may suspect thelabor market adjusts through migration channel, as documented by Blan-chard and Katz (1992). Unfortunately, this work cannot exclude geographicmobility mechanism due to data availability. 28Tables 1.9 - 1.11 show that exogenous credit supply shock has caused un-employment rate fluctuations in recession and recovery periods. These re-27Both measures are divided by total population. Results on the labor force and employmentdivided by the working-age population between 16-65 can be found in the Appendix.28American Community Survey (ACS) provides the migration information at county level.However, for the full sample of counties, ACS only reports the five-year average estimateson migration, which is imprecise for analyzing the yearly changes in local labor force. Alter-native source on migration data comes from Income Revenue Survey (IRS), which is basedon the address changes reported on individual income tax return filed with the IRS. Unfor-tunately, this data is not representative and shows opposite pattern as the migration data inACS.28sults suggest that the negative effect of credit shock on local unemploymentis not permanent but a business cycle phenomenon, followed by positiveeffects after the recession.Relationship between Credit Shock and WageThe previous section has shown that credit shock does not affect the localunemployment changes from 2003 to 2017. In this section, I will show howcredit supply shock affects wage growth from 2003 to 2017. The primarydependent variable is the average weekly wage obtained from QCEW. Ta-ble 1.12 shows the long-run impact credit supply shock has on local wagegrowth, in that regions with larger credit shock have slower wage growthfrom 2003 to 2017. This result is robust to different regression specifica-tions, and the estimate becomes more significant after controlling for statefixed effects. Overall, a 10 percent increase in credit shock contributes to a0.93–2.6 percent decline in weekly wage growth. Although the magnitudeof the credit effect on wage growth is not very large, keep in mind that thiscredit shock is measured as yearly growth in the mortgage origination. Thetotal effect of credit expansion on wage growth will be presented with 2SLSregression, using mortgage growth as the explanatory variable.The sub-period analysis is reported in Table 1.13. As shown in the firsttwo columns, credit shock has little effect on wage growth before the re-cession. Consistent with the coefficient estimate of credit supply shock onunemployment change, these results suggest credit supply shock cannot ex-plain local labor market trends before the recession. Its negative impact onwages started since the recession, with a 10 percent increase in credit shock,leading to a 0.5 percent decline in weekly wage growth. After the reces-sion, the estimates are no longer precise. However, this result shows thatcounties with higher credit shock did not experience faster recovery afterthe recession, unlike the unemployment rate, which recovered faster afterthe recession.The negative relationship between credit expansion and wage growthhas not been discussed, except for Gilchrist et al. (2018) and Beraja et al.29(2019). Beraja, Hurst, and Ospina (2019) use ACS micro-level data to con-struct hourly wages of prime-age male workers and establish the strong re-lationship between wage growth (both nominal and real) and employmentgrowth at the state level in the Great Recession. Gilchrist et al. (2018) es-timate the wage to employment changes in response to household creditcontraction at both county and commuting zone levels. They find a morethan a 1 percent decline in wage for every 1 percent decline in employmentdue to credit contraction shock. Note that Mian and Sufi (2013) also esti-mated wage growth in response to a decline in housing net worth. Theyfind a small but statistically significant decline in payroll growth control-ling for industry composition. One reason I find a stronger and prolongednegative effect on wage growth is by using county-level variations and in-clude all county observations. The land elasticity index constructed by Saiz(2010) only varies at the MSA level, but the wage growth appears signifi-cant variation across counties within the same MSA. Using the instrumentalvariable only at the MSA level will miss significant wage variations at thecounty level.As wage growth is in nominal terms, it is concerned that the negativecoefficient captures slower price growth instead of wage growth. Unfortu-nately, inflation measures at the county level do not exist. If inflation variesat the state level, the inflation effect on nominal wage growth will be elim-inated by the state fixed effect. I have also controlled for commuting zonefixed effects and find the coefficient estimate of credit shock on wage growthremains at the same scale. The results can be found in the Appendix.Relationship between Credit Shock and Wage by IndustrySome might wonder if the effect of credit shock on wage growth is driven byan industry-specific trend, e.g., disruption in the construction sector. Highcredit shock regions have larger exposure to the construction industry. Al-though I have controlled for industry compositions in 2000, it is more con-vincing to show that the effect of credit on wage growth is prevalent acrossdifferent industries. I split 23 two-digit industries into three groups: con-30struction, finance, and real estate, and the rest of industries excluding man-ufacturing, and compute the industry-specific weekly wage. 29As Table 1.14 shows, the negative coefficient is most significant for thefinance and real estate industries, which is not surprising, given the natureof credit shock. Although the standard error becomes larger than before,the coefficient estimate on the rest of the industries remains the same asthe aggregate effect shown in Table 1.12. The coefficient estimates for allthree industry-specific wage growth are very close, too. Suppose the effectof credit shock on wage growth is construction or finance industry-specificand spilled over to other industries through labor supply channel, i.e., la-bor supply moves credit-shock affected the industry to other unaffected in-dustries. We should expect smaller effects of credit shock on unaffectedindustries if the labor supply is not perfectly mobile across industries. Thishypothesis is not supported by the coefficient estimates in Table 1.14.Relationship between Credit Shock and Wage by EducationAnother concern with the wage variable is that the average weekly wage isnot a good measure for the actual wage rate, as it does not consider workercharacteristics like gender, age, or education. If high credit shock regionshave different labor demographics than the low shock regions, the resultwill be driven by the aggregate trend in different working classes, insteadof a credit shock. Table 1.15 shows the average quarterly payroll changesby three education groups: workers with education below 12 years (lessthan high school), between 13–15 years (high school and some college),and above 15 years (bachelor and higher).30 Table 1.15 suggests the nega-tive wage effects are persistent across different education groups, althoughthe magnitudes of coefficients are different. The effect of credit shock onwage growth does not change monotonically with worker’s education at-tainment level. Both the lower educated workers (less than high school)and the higher educated workers (college degree and higher) are more af-29The declining trend in the manufacturing industry has deteriorated since the recession. SeeCharles et al. (2016).30Samples are restricted to workers more than 25 years old.31fected by the credit shock. Note that the end of the data period is 2016 dueto data availability.To summarize, I have shown the predicted credit supply shock does notaffect the long-term unemployment rate changes, but negatively affected thelong-term wage growth. The sub-period analysis suggests the credit expan-sion did not have positive effects during the housing market boom but neg-atively affected both unemployment and wage in the recession. The effect ofcredit shock diverged after the recession: it accelerated recovery speed onthe unemployment rate but persistently slowed down wage growth. Thenegative effect of credit shock on local wage growth is not specific to a par-ticular sector or education group. With all the evidence shown until now,I can conclude that credit shock has negatively affected local labor marketson the intensive margin in the long run. In the next section, I will showmore evidence and propose one plausible adjustment mechanism.1.4.3 Relationship between Credit Shock and Firm DynamismTo explain the long-term effect of credit shock on the local labor market,I investigate the local labor demand side by exploring the change in em-ployment composition by firm ages. As the decline in business dynamismhas been happening for over 30 years, the net startup rate first decreasesto negative since the Great Recession (Siemer, 2012). A number of workshave documented the strong positive relationship between housing priceand firm startup rate at MSA and state level (Davis and Haltiwanger, 2019;Gourio et al., 2016; Siemer, 2012). Motivated by their work, I will examine ifthis result holds with credit supply shock at the county level. This will helpto understand the labor demand-side changes and the adjustment mecha-nism in the local labor market. To examine the employment compositionshift, I use the dependent variable as the change in the employment shareof young firms below four years old relative to total county employment, aswell as the change in the employment share of old firms above five yearsold. Data is taken from Quarterly Workforce Indicator (QWI), which ag-gregates based on the source data Longitudinal Employer-Household Dy-32namics (LEHD) and covers 95% of private-sector jobs.Table 1.16 presents the employment share changes since 2006. Note thatthe time period used here is 2006–2016, which is the latest data is available.It is not surprising that estimates are close to symmetric with opposite signs.The noticeable thing is that regions that had larger credit supply shock haveseen a decline in young firm employment. This could happen for two rea-sons: if the employment share of old firms is constant, this is driven by thedeclining number of new firms or shrinking employment of new firms; orif the employment share of young firms is constant, the old firms have ex-panded after the recession. If the second case were right, we should havefound that regions had larger credit shock experienced faster employmentgrowth in the long run, which is not the case in the employment section.Given the negligible effect of credit shock on long-run employment changes,this points to shrinking employment in young firms. Unfortunately, with-out more detailed data, I could not distinguish whether the decline comesfrom the intensive margin (employment of young firms) or extensive mar-gin (number of young firms). However, there is some supportive evidencethat regions with larger credit shock have slowed the growth rate in localestablishments, shown in the Appendix.31 As a robustness validation, Ialso check the relationship between credit shock and the employment sharecomposition before the recession, shown in the Appendix. The estimates arenoisy, but the effect of credit shock on employment shares for young and oldfirms have the opposite signs before the recession. This suggests, at least,there is no evidence of pre-trend started before the recession.As supportive evidence, I also estimate the effect of credit shock on changesin the number of establishments by firm size. From columns 3 and 4 in Ta-ble 1.17, we can see firms with 10–100 workers have experienced the largestdecline in the number of establishments, with similar but insignificant ef-fects on large firms with more than 100 workers. The number of smallfirms with 1–10 workers had not changed significantly in the high creditshock counties, once controlling for state fixed effects. Unfortunately, with-31Data is taken from QCEW, which reports the total number of private establishments in thecounty.33out an establishment-level dataset, it is not possible to tell whether the de-cline in small to medium-size firms is driven by shrinking employment ormarket exit. Note that firm size is an endogenous state, which is subjectto a firm’s operational choice, whereas firm age is an exogenous variable.As pointed out byDavis and Haltiwanger (2019), contraction in small busi-ness loans has heterogeneous employment effects on small firms vs. youngfirms. Small firms are mature, stable, and have less external financing de-mand.To summarize, firm-side evidence suggests counties with larger creditsupply shock experienced larger declines in the employment share of youngfirms and slower growth in the number of establishments since the reces-sion. Given this evidence, I will propose a simple mechanism to explainhow credit shock fluctuations can lead to a decline in wage growth, youngfirm shares, with an unchanged employment rate.1.4.4 Pre-trendThe identifying assumption for credit supply shock is that multi-marketlenders cannot predict local labor market trends. One way to test this as-sumption is to check whether credit shock is correlated with pre-existingtrends in local labor markets. To do so, I estimate the correlation betweencredit shock and the unemployment rate, weekly wage, and employmentrate change from 1990 to 1999, as well as housing price during 1996-1999.32As Table 1.18 shows, credit supply shock cannot predict the local labor mar-ket trend during the 1990s. This serves as supportive evidence to show thatconditional on the observable county characteristics multi-market lendersdid not choose to locate in the growing local economies.1.4.5 Placebo TestOne concern regarding the credit supply shock is that large lenders tend tolocate in metropolitan areas, which are more responsive to the aggregateeconomy. If this were true, the effect of credit shock on labor market out-32The housing price change is restricted to between 1996 and 1999 due to data availability.34comes is merely a reflection of aggregate economic fluctuations. To test thishypothesis, I estimate the relationship between credit shock and local labormarket outcomes in the 2001 recession. Table 1.19 shows that credit shockcannot explain the local labor market outcomes in the 2001 recession.1.4.6 Sensitivity TestThis section shows the main results are robust to the predicted credit shockwith a shorter list of multi-market lenders, subject to stricter selection cri-teria. In the benchmark regression, I restrict lenders to those operating inmore than 100 counties and one census region. One might be concernedwith the geographical selection bias that lenders operating at the borderof census regions are more likely to be selected. To resolve this concern, Imodify the selection criteria to be lenders operating in more than 100 coun-ties and ten states in 2000 and keeping operation during 2000–2006. Thisrestricts to 193 selected lenders. I construct the credit supply shock withthese lenders following the same method to construct credit supply shock.Table 1.20 shows the regression results for changes in the unemploy-ment rate and weekly wage from 2003 to 2017 and changes in the old firmemployment share from 2006 to 2016, explained by the credit supply shock.The coefficients are very close to the benchmark estimates in Table 1.8, 1.12and 1.15. The sub-period analysis can be found in the Appendix and areconsistent with the benchmark results as well. In the Appendix, I show thatthe main results are also robust to credit shock constructed with an evenshorter list of lenders, which operated in more than 200 counties in 2000.1.4.7 Two-stage Least Square RegressionTo interpret the coefficient estimates of credit shock on unemployment andwage changes, I perform two-stage least square (2SLS) regressions usingthe explanatory variable as average yearly growth in new home mortgageorigination. To perform this test, the identifying assumption requires thatcredit supply shock cannot affect local labor markets other than throughhousehold mortgage channel. As shown in Table 1.6, the F statistics for first-35stage regression is 8.16 after controlling for state fixed effects, raising theconcern of weak instrument. To improve the first stage prediction, I modifythe credit shock by replacing the lender’s total mortgage growth with homepurchase mortgage growth.33 Table 1.21 shows the ordinary least square(OLS) regression estimates in the odd columns and 2nd stage regressionestimates in the even columns. The first stage is presented in the Appendix.As a robustness check, I show that the reduced-form estimates with modi-fied credit shock are consistent with the benchmark results shown so far.As we can see, the OLS regression suggests that mortgage expansionduring 2003–2006 is positively correlated with changes in the local unem-ployment rate from 2003 to 2017. The correlation is statistically significant,but not economically significant, with one standard deviation increase inmortgage expansion predicting a 0.13 percentage point increase in the un-employment rate from 2003 to 2017. Meanwhile, it appears to be uncorre-lated with wage growth from 2003 to 2017. On the contrary, the 2SLS resultssuggest the supply-driven mortgage expansion does not explain local un-employment rate changes but explains wage growth slowdown, with onestandard deviation increase in mortgage expansion predicting a 6.4 percentdecline in wage growth. The comparison between OLS and 2SLS regres-sion results suggest that mortgage expansion could be positively correlatedwith the local labor market trends. If local mortgage expansion were partlydriven by increasing housing demand due to a growing regional economy,the OLS estimates would under-estimate the impact of mortgage expansion.The sub-period regressions are available in the Appendix.1.5 ConclusionWhat is the long-run impact of credit boom-bust on the real economy? Thischapter tries to answer this question through the lens of the U.S. local la-bor markets. Focusing on the U.S. mortgage expansion in the early 2000s,this chapter constructs the exogenous mortgage supply shock by exploiting33The total mortgage includes mortgages for home purchase, home improvement, and refi-nancing.36different lending strategies taken by multi-market lenders and a county’sinitial exposure to these lenders. This chapter finds that counties receivedlarger credit shock do not have different outcomes in the local unemploy-ment but have slower wage growth from 2003 to 2017. Breaking the timeinto expansion (2003–2006), contraction (2006–2010) and recovery period(2010–2017), credit supply shock does not explain local unemployment andwage changes in the expansion period, but contributes to the increase inunemployment rate and decline in wage growth in the contraction period,followed by faster recovery in unemployment but not wage in the recoveryperiod.To explain the long-lasting negative effect of credit shock on local wagegrowth, I investigate the changes in industry-specific and education-specificwages in response to local credit shock. The estimates suggest that the nega-tive effect of credit shock on wage growth cannot be explained by industry-specific or skill-specific mechanisms. In addition, this chapter finds thatcounties with larger credit shocks had larger declines in the young firmemployment share and increases in the old firm employment share sincethe Recession. This evidence suggests credit shock might affect local labormarkets through labor demand sides.In the next chapter, I will propose a theoretical framework to rationalizethe empirical findings.371.6 FiguresFigure 1.1: National trend in mortgage originationNote: this figure plots the total mortgage and home purchase mortgage orig-ination from 2000 to 2015. The mortgage origination is defined defined as thelog form of dollar value. Total mortgage includes the mortgages for home pur-chase, home improvement, and refinancing purposes. Source: HMDA.38Figure 1.2: Credit supply shock.Note: this graph plots the residual of credit shock after controlling for countycharacteristics shown in Table 3, state fixed effects, and 23 two-digit industryemployment share in 2000. Control variables are used in the regression exerciseas well. The credit shock residual is normalized by its standard deviation. Themap sorts the counties into six percentiles based on credit supply shocks theyreceived. Darker colours indicate larger supply.391.7 TablesTable 1.1: County summary statisticsmean sd p10 p50 p90Dependent Variable, changes from 2003 to 2006∆ home mortgage 0.97 0.39 0.53 0.95 1.44∆ house price 0.20 0.13 0.07 0.16 0.40∆ unemployment rate -0.01 0.01 -0.02 -0.01 0.00∆ employment 0.05 0.10 -0.05 0.04 0.15∆ weekly wage 0.12 0.07 0.05 0.11 0.18∆ establishment 0.04 0.08 -0.04 0.03 0.13Dependent Variable, changes from 2006 to 2010∆ home mortgage -0.24 0.34 -0.62 -0.22 0.12∆ house pricea -0.08 0.13 -0.22 -0.06 0.03∆ unemployment rate 0.05 0.02 0.02 0.04 0.07∆ employment -0.06 0.11 -0.18 -0.06 0.05∆ weekly wageb 0.07 0.07 0.00 0.06 0.14∆ establishmentc -0.01 0.06 -0.08 -0.02 0.05Dependent Variable, changes from 2010 to 2017∆ home mortgaged -0.11 0.35 -0.45 -0.16 0.30∆ house price 0.11 0.13 -0.03 0.09 0.29∆ unemployment rate -0.05 0.02 -0.08 -0.05 -0.02∆ employment 0.08 0.15 -0.08 0.07 0.23∆ weekly wage 0.17 0.09 0.09 0.17 0.26∆ establishment 0.03 0.12 -0.09 0.02 0.16N 3108Note: this table provides summary statistics for three periods of time:2003–2006, 2006–2010, and 2010–2017. All the variables are taken the first dif-ference between the start and the end of the period, except for home mortgage.In each section, home mortgage changes are computed as average yearly changeduring 2003–2006 relative to 2000, 2007–2010 relative to 2006, and 2011–2017 rel-ative to 2010, respectively. a Housing price change is taken between 2007 and2010. b Weekly wage change is taken between 2007 and 2010. c Establishmentchange is taken between 2010 and 2010. d Mortgage change is taken between2010 and 2015, due to data availability.40Table 1.2: Lender characteristics in 2000.non-selected lenders selected lendersmin p50 max min p50 maxno. of counties 1.00 7.00 3097.00 103.00 333.00 2840.00no. of states 1.00 1.00 51.00 3.00 32.00 51.00no. of regions 1.00 1.00 4.00 2.00 4.00 4.00application per bank(log $) 1.61 9.15 17.75 9.36 13.13 17.78application per bank (log no.) 0.00 4.83 13.25 4.91 8.62 13.01origination per bank(log $) 1.61 8.99 17.53 8.34 12.70 17.47origination per bank(log no.) 0.00 4.67 12.53 3.78 8.11 12.61sharea 0.64 0.36N 7370 215Note: this table presents the lender characteristics in 2000 for selected and non-selected groups. The selected lenders satisfy the restriction for lending in morethan 100 counties and 1 census region in 2000 and remained operating during2000–2006. aIt is defined as the total dollar value of mortgage originated byselected (other) lenders relative to the total mortgage origination in 2000.Table 1.3: Lender characteristics in 2002–2006.non-selected lenders selected lendersmin p50 max min p50 maxno. of counties 1.00 10.70 3074.62 5.69 512.92 3045.50no. of states 1.00 2.00 51.00 1.77 37.00 51.00no. of regions 1.00 1.00 4.00 1.00 4.00 4.00application per bank (log $) 2.30 9.83 19.57 8.89 14.26 19.15application per bank (log no.) 0.00 5.06 14.35 3.26 9.24 13.54origination per bank (log $) 0.92 9.68 19.37 8.85 13.84 18.80origination per bank (log no.) 0.00 4.90 14.16 3.10 8.80 13.16sharea 0.49 0.51N 10975 216Note: this table presents the lender characteristics during 2002–2006 for se-lected and non-selected groups. The selected lenders satisfy the restriction forlending in more than 100 counties and 1 census region in 2000 and remainedoperating during 2000–2006. aShare is defined as the average total dollar valueof mortgage originated by selected(other) lenders relative to the total mortgageorigination during 2002–2006.41Table 1.4: Relationship between lender local market shares and locallabor market outcomes: 2003–2006Unemployment rate: 2003-2006∆y2003−2006 ∆y2003−2006 ∆y2003−2006 ∆y2003−2006market l 0.0207 0.000716(0.0160) (0.00526)market h -0.0140 -0.00286(0.0225) (0.00736)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes YesN 2538 2538 2642 2642adj. R2 0.176 0.544 0.176 0.535Weekly wage: 2003-2006∆y2003−2006 ∆y2003−2006 ∆y2003−2006 ∆y2003−2006market l -0.00161 0.0310(0.0368) (0.0307)market h 0.0199 -0.0169(0.0574) (0.0532)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes YesN 2556 2556 2644 2644adj. R2 0.147 0.221 0.143 0.220Note: this table reports regression estimates of lender market share on the lo-cal unemployment rate and weekly wage changes during 2003–2006. Variable“market l” represents the local mortgage market shares for slowest expandinglenders and “market h” represents the same for the fastest expanding lenders.Regressions are weighted by the 2000 population. Standard errors are clusteredat the state level. County characteristics include the fraction of subprime pop-ulation, establishment, employment rate, unemployment rate, weekly wage,average annual income, median household income and poverty rate in 2000.Industry composition includes 2000 employment share in a county for 23 two-digit industries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels,respectively.42Table 1.5: Summary statistics: county economic characteristics in 2000below median above medianmean sd mean sdsubprime ratea 33.52 8.29 32.73 6.37establishment per capita 0.02 0.01 0.03 0.01employment rate 0.46 0.06 0.49 0.05unemployment rate 0.05 0.02 0.04 0.01weekly wage (log) 6.15 0.18 6.45 0.26annual income per capita (log) 9.72 0.21 10.07 0.30household median income (log) 10.40 0.19 10.69 0.23poverty rate 0.15 0.06 0.12 0.05construction employment share 0.06 0.05 0.07 0.03manufacturing employment share 0.27 0.18 0.16 0.11N 1550 1587Note: this table presents summary statistics of county characteristics in 2000 forhigh-credit and low-credit counties. All the county observations are dividedinto two groups based on the credit shocks they receive. Summary statisticsare weighted by the 2000 population.aThe subprime rate is defined as the fraction of the population with a creditscore below 660.43Table 1.6: Effect of credit supply shock on mortgage marketMortgage expansion: 2003-2006 Mortgage contraction: 2007-2010∆y2003,2006−2000 ∆y2003,2006−2000 ∆y2007,2010−2006 ∆y2007,2010−2006credit 0.739∗∗∗ 0.402∗∗∗ -0.719∗∗∗ -0.455∗∗∗(0.155) (0.141) (0.140) (0.105)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes YesN 3038 3038 3036 3036adj. R2 0.249 0.419 0.216 0.437Note: this table reports regression estimates of credit supply shocks on themortgage growth during 2003–2006 and 2007–2010. Regressions are weightedby the 2000 population. Standard errors are clustered at the state level. Countycharacteristics include the fraction of subprime population, establishment, em-ployment rate, unemployment rate, weekly wage, annual income per capita,median household income, and the poverty rate in 2000. Industry composi-tion includes 2000 employment share in a county for 23 two-digit industries:Agriculture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade,Retail Trade, Transportation, Information, Finance, Real Estate, ProfessionalServices, Management, Administrative Services, Education, Health Care, En-tertainment, Accommodation and Food Services, Other Services. *, **, and ***indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.44Table 1.7: Effect of credit supply shock on housing priceHousing Price: 2003-2017∆y2003−2017 ∆y2003−2006 ∆y2006−2010 ∆y2010−2017credit shock -0.0941 0.322∗∗∗ 0.113∗∗ -0.408∗∗∗ -0.200∗∗∗ 0.0561 0.0199(0.0593) (0.0669) (0.0417) (0.0900) (0.0499) (0.0913) (0.0401)county controls Yes Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yes Yesstate FE Yes Yes Yes YesN 2645 2617 2617 2629 2629 2608 2608adj. R2 0.602 0.319 0.702 0.337 0.688 0.175 0.520Note: this table reports regression estimates of credit supply shock on housingprice changes from 2003 to 2017. Regressions are weighted by the 2000 popu-lation. Standard errors are clustered at the state level. County characteristicsinclude the fraction of subprime population, establishment, employment rate,unemployment rate, weekly wage, annual income per capita, median house-hold income, and the poverty rate in 2000. Industry composition includes 2000employment share in a county for 23 two-digit industries: Agriculture, Mining,Utilities, Construction, Manufacturing, Wholesale Trade, Retail Trade, Trans-portation, Information, Finance, Real Estate, Professional Services, Manage-ment, Administrative Services, Education, Health Care, Entertainment, Accom-modation and Food Services, Other Services. *, **, and *** indicate significanceat the 0.1, 0.05, and 0.01 levels, respectively.45Table 1.8: Effect of credit supply shock on the unemployment rateUnemployment rate: 2003-2017(1) (2) (3)credit shock -0.00791 -0.00937 -0.00213(0.00537) (0.00515) (0.00403)county controls No Yes Yesindustry controls No Yes Yesstate FE YesN 3055 3050 3050adj. R2 0.006 0.179 0.471Note: this table reports regression estimates of credit supply shock on the un-employment rate changes from 2003 to 2017. Regressions are weighted by the2000 population. Standard errors are clustered at the state level. County char-acteristics include the fraction of subprime population, establishment, employ-ment rate, unemployment rate, weekly wage, annual income per capita, me-dian household income, and the poverty rate in 2000. Industry composition in-cludes 2000 employment share in a county for 23 two-digit industries: Agricul-ture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade, RetailTrade, Transportation, Information, Finance, Real Estate, Professional Services,Management, Administrative Services, Education, Health Care, Entertainment,Accommodation and Food Services, Other Services. *, **, and *** indicate sig-nificance at the 0.1, 0.05, and 0.01 levels, respectively.46Table 1.9: Effect of credit supply shock on unemployment change: sub-periodUnemployment rate: 2003-2017∆y2003−2006 ∆y2006−2010 ∆y2010−2017credit shock -0.00472 0.000335 0.0495∗∗∗ 0.0157∗∗∗ -0.0528∗∗∗ -0.0173∗∗∗(0.00374) (0.00265) (0.00937) (0.00510) (0.0106) (0.00612)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3033 3033 3041 3041 3045 3045adj. R2 0.186 0.524 0.367 0.653 0.401 0.723Note: this table reports regression estimates of a credit supply shock onthe unemployment change before the recession (2003–2006), in the recession(2006–2010), and after the recession (2010–2017). Regressions are weighted bythe 2000 population. Standard errors are clustered at the state level. Countycharacteristics include the fraction of subprime population, establishment, em-ployment rate, unemployment rate, weekly wage, annual income per capita,median household income, and the poverty rate in 2000. Industry composi-tion includes 2000 employment share in a county for 23 two-digit industries:Agriculture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade,Retail Trade, Transportation, Information, Finance, Real Estate, ProfessionalServices, Management, Administrative Services, Education, Health Care, En-tertainment, Accommodation and Food Services, Other Services. *, **, and ***indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.47Table 1.10: Effect of credit supply shock on labor participation and em-ploymentEmployment rate Labor participation rate∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017credit shock 0.00958 0.00135 0.00545 0.00158(0.0150) (0.0141) (0.0160) (0.0141)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes YesN 3045 3045 3045 3045adj. R2 0.085 0.151 0.081 0.157Note: this table reports regression estimates of credit supply shock on the la-bor participation and employment rate change. Regressions are weighted bythe 2000 population. Standard errors are clustered at the state level. Countycharacteristics include the fraction of subprime population, establishment, em-ployment rate, unemployment rate, weekly wage, annual income per capita,median household income, and the poverty rate in 2000. Industry composi-tion includes 2000 employment share in a county for 23 two-digit industries:Agriculture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade,Retail Trade, Transportation, Information, Finance, Real Estate, ProfessionalServices, Management, Administrative Services, Education, Health Care, En-tertainment, Accommodation and Food Services, Other Services. *, **, and ***indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.48Table 1.11: Effect of credit supply shock on private employment: 2003-2017Employment change: 2003-2017∆y2003−2006 ∆y2006−2010 ∆y2010−2017credit shock 0.0239 0.0144 -0.147∗∗∗ -0.0602∗∗ 0.104∗∗∗ 0.0852∗∗∗(0.0263) (0.0227) (0.0238) (0.0278) (0.0292) (0.0311)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3027 3027 3024 3024 3023 3023adj. R2 0.125 0.206 0.140 0.235 0.204 0.273Note: this table reports regression estimates of bank credit supply shock on theaverage monthly private employment change before the recession (2003–2006),in recession (2006–2010), and after the recession (2010–2017). Regressions areweighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include the fraction of subprime population, estab-lishment, employment rate, unemployment rate, weekly wage, annual incomeper capita, median household income, and the poverty rate in 2000. Industrycomposition includes 2000 employment share in a county for 23 two-digit in-dustries: Agriculture, Mining, Utilities, Construction, Manufacturing, Whole-sale Trade, Retail Trade, Transportation, Information, Finance, Real Estate, Pro-fessional Services, Management, Administrative Services, Education, HealthCare, Entertainment, Accommodation and Food Services, Other Services. *,**, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.49Table 1.12: Effect of credit supply shock on weekly wage: 2003–2017Weekly wage rate: 2003-2017(1) (2) (3)credit shock -0.260∗∗∗ -0.112∗∗ -0.0933∗∗∗(0.0705) (0.0547) (0.0265)county controls Yes Yesindustry controls Yes Yesstate FE YesN 3031 3027 3027adj. R2 0.088 0.268 0.364Note: this table reports regression estimates of bank credit supply shock onthe weekly wage change from 2003 to 2017. Regressions are weighted by the2000 population. Standard errors are clustered at the state level. County char-acteristics include the fraction of subprime population, establishment, employ-ment rate, unemployment rate, weekly wage, annual income per capita, me-dian household income, and the poverty rate in 2000. Industry composition in-cludes 2000 employment share in a county for 23 two-digit industries: Agricul-ture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade, RetailTrade, Transportation, Information, Finance, Real Estate, Professional Services,Management, Administrative Services, Education, Health Care, Entertainment,Accommodation and Food Services, Other Services. *, **, and *** indicate sig-nificance at the 0.1, 0.05, and 0.01 levels, respectively.50Table 1.13: Effect of credit supply shock on weekly wage: subperiodWeekly wage rate: 2003-2017∆y2003−2006 ∆y2006−2010 ∆y2010−2017credit shock -0.0120 -0.00838 -0.0609∗∗∗ -0.0521∗∗∗ -0.0300 -0.0252(0.0157) (0.00961) (0.0164) (0.0181) (0.0305) (0.0176)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3024 3024 3025 3025 3029 3029adj. R2 0.128 0.202 0.198 0.243 0.120 0.198Note: this table reports regression estimates of credit supply shock on aver-age weekly wage changes before the recession (2003–2006), in the recession(2006–2010), and after the recession (2010–2017). Regressions are weighted bythe 2000 population. Standard errors are clustered at the state level. Countycharacteristics include the fraction of subprime population, establishment, em-ployment rate, unemployment rate, weekly wage, annual income per capita,median household income, and the poverty rate in 2000. Industry composi-tion includes 2000 employment share in a county for 23 two-digit industries:Agriculture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade,Retail Trade, Transportation, Information, Finance, Real Estate, ProfessionalServices, Management, Administrative Services, Education, Health Care, En-tertainment, Accommodation and Food Services, Other Services. *, **, and ***indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.51Table 1.14: Effect of credit supply shock on industry-specific wage:2003-2017Weekly wage rate by industry: 2003-2017Construction Finance and Real Estate Otherscredit -0.121∗ -0.0859∗∗ -0.142∗∗∗ -0.0924∗∗ -0.127∗ -0.105∗∗(0.0695) (0.0371) (0.0512) (0.0422) (0.0629) (0.0466)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yesstate FE Yes Yes YesN 2597 2597 2475 2475 3027 3027adj. R2 0.061 0.131 0.032 0.080 0.128 0.186Note: this table reports regression estimates of credit supply shock on industry-specific average weekly wage changes from 2003 to 2017. Regressions areweighted by the 2000 population. Standard errors are clustered at the statelevel. Industry composition includes 2000 employment shares in a county for 23two-digit industries, excluding construction, manufacturing, finance and realestate. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respec-tively.52Table 1.15: Effect of credit supply shock on quarterly earning by edu-cation groups: 2003–2016Annual earning by education attachment: 2003-2016education below 12 years education 12-15 years education above 15 yearscredit -0.114∗∗∗ -0.0565∗∗ -0.0770∗∗ -0.0371∗∗ -0.113∗∗∗ -0.0646∗∗∗(0.0426) (0.0227) (0.0386) (0.0173) (0.0408) (0.0232)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 2937 2937 2943 2943 2867 2867adj. R2 0.246 0.397 0.254 0.418 0.178 0.331Note: this table reports regression estimates of credit supply shock on the quar-terly earning change by education groups from 2003 to 2016. Regressions areweighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include the fraction of subprime population, estab-lishment, employment rate, unemployment rate, weekly wage, annual incomeper capita, median household income, and the poverty rate in 2000. Industrycomposition includes 2000 employment share in a county for 23 two-digit in-dustries: Agriculture, Mining, Utilities, Construction, Manufacturing, Whole-sale Trade, Retail Trade, Transportation, Information, Finance, Real Estate, Pro-fessional Services, Management, Administrative Services, Education, HealthCare, Entertainment, Accommodation and Food Services, Other Services. *,**, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.53Table 1.16: Effect of credit shock on employment composition: youngvs. old firmYoung firm employment share Old firm employment share∆y2006−2016 ∆y2006−2016credit -0.0472∗∗∗ -0.0222∗∗ 0.0545∗∗∗ 0.0348∗∗∗(0.0160) (0.0101) (0.0103) (0.00918)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 2604 2604 2744 2744adj. R2 0.111 0.154 0.087 0.108Note: this table reports regression estimates of credit supply shock on theemployment share changes by young (age<4) and old firms (age>5) during2006–2016. Regressions are weighted by the 2000 population. Standard er-rors are clustered at the state level. County characteristics include the fractionof subprime population, establishment, employment rate, unemployment rate,weekly wage, annual income per capita, median household income, and thepoverty rate in 2000. Industry composition includes 2000 employment share ina county for 23 two-digit industries. *, **, and *** indicate significance at the 0.1,0.05, and 0.01 levels, respectively.54Table 1.17: Effect of credit shock on firm size: 2006–2016Small firm Small-medium firm Large firm1-10 workers 10-100 workers more than 100 workerscredit -0.0495 0.00585 -0.178∗∗∗ -0.114∗∗∗ -0.166∗∗ -0.107(0.0427) (0.0399) (0.0599) (0.0327) (0.0810) (0.0894)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE No Yes No Yes No YesN 3069 3069 3053 3053 2701 2701adj. R2 0.154 0.292 0.097 0.165 0.045 0.064Note: this table reports regression estimates of credit supply shock on thechange in the number of establishments by firm size during 2006–2016. Re-gressions are weighted by the 2000 population. Standard errors are clusteredat the state level. County characteristics include the fraction of subprime popu-lation, establishment, employment rate, unemployment rate, weekly wage, an-nual income per capita, median household income, and the poverty rate in 2000.Industry composition includes 2000 employment share in a county for 23 two-digit industries: Agriculture, Mining, Utilities, Construction, Manufacturing,Wholesale Trade, Retail Trade, Transportation, Information, Finance, Real Es-tate, Professional Services, Management, Administrative Services, Education,Health Care, Entertainment, Accommodation and Food Services, Other Ser-vices. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respec-tively.55Table 1.18: Relationship between credit shock and local labor market:1990-1999Unemployment rate Weekly wage Employment rate Housing price∆y1990−1999 ∆y1990−1999 ∆y1990−1999 ∆y1996−1999credit shock -0.00724 -0.0190 -0.0123 0.0329(0.00466) (0.0174) (0.0133) (0.0191)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes Yes Yes YesN 3042 3047 3039 2037adj. R2 0.321 0.147 0.198 0.354Note: this table presents regression estimates of credit supply shock on thechanges in the unemployment rate, weekly wage, employment rate from 1990to 1999, and changes in housing prices from 1996 to 1999. Regressions areweighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include the fraction of subprime population, estab-lishment, employment rate, unemployment rate, weekly wage, annual incomeper capita, median household income, and the poverty rate in 2000. Indus-try composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels,respectively.56Table 1.19: Relationship between credit shock and local labor market:2001–2003Unemployment rate Weekly wage∆y2001−2003 ∆y2001−2003 ∆y2001−2003 ∆y2001−2003credit 0.00251 0.00409 0.00726 0.00750(0.00410) (0.00268) (0.0128) (0.0121)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 3042 3042 3027 3027adj. R2 0.096 0.348 0.056 0.106Note: this table presents regression estimates of credit supply shock on theunemployment rate and weekly wage changes from 2001 to 2003. Regressionsare weighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include the fraction of subprime population, estab-lishment, employment rate, unemployment rate, weekly wage, annual incomeper capita, median household income, and the poverty rate in 2000. Indus-try composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels,respectively.57Table 1.20: Effect of credit shock on the labor market: 2003–2017Unemployment rate Weekly wage Old firm employment share∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2006−2016 ∆y2006−2016credit -0.00980 -0.00131 -0.0952 -0.0829∗∗ 0.0556∗∗∗ 0.0383∗∗∗(0.00579) (0.00353) (0.0482) (0.0276) (0.0101) (0.00995)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE No Yes No Yes No YesN 3050 3050 3027 3027 2744 2744adj. R2 0.180 0.470 0.264 0.362 0.088 0.109Note: this table reports regression estimates of credit supply shock on thechange in the unemployment rate and weekly wage from 2003 to 2017 andchange in the employment share for old firms (age > 5). Regressions areweighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include the fraction of subprime population, estab-lishment, employment rate, unemployment rate, weekly wage, annual incomeper capita, median household income, and the poverty rate in 2000. Indus-try composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels,respectively.58Table 1.21: OLS and 2nd Stage RegressionsUnemployment rate Weekly wageOLS 2SLS OLS 2SLS∆mortgage2003,2006−2000 0.00227∗∗ -0.00320 -0.00522 -0.161∗∗(0.000833) (0.00675) (0.00567) (0.0607)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE Yes Yes Yes YesKleibergen-Paap F statistic 23.17 21.75N 3049 3049 3026 3026adj. R2 0.473 0.455 0.361 0.177Note: this table reports the OLS and 2nd stage estimates of mortgage expansionon the unemployment rate and weekly wage rate change from 2003 to 2017. Re-gressions are weighted by the 2000 population. Standard errors are clusteredat the state level. County characteristics include the fraction of subprime popu-lation, establishment, employment rate, unemployment rate, weekly wage, an-nual income per capita, median household income, and the poverty rate in 2000.Industry composition includes 2000 employment share in a county for 23 two-digit industries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels,respectively.59Chapter 2Credit Shocks, House Prices,and Labor Market2.1 IntroductionFrom Chapter 1, I find that credit supply shock does not explain unemploy-ment rate changes in U.S. local labor markets but explains the decline inlocal wage growth from 2003 to 2017. The negative effect of credit supplyshock on wage growth is robust across different industries and different ed-ucation groups, which makes it hard to explain this phenomenon throughindustry- or skill-specific mechanisms. Suppose the decline in local wagegrowth is driven by adverse shocks in the local construction sector due tothe housing market collapse. One should find the negative effects of creditshock on local wage growth to be stronger in the construction sector thanin the other sectors. Alternatively, if the credit shock has stronger effects onlow-educated laborers working in the construction sector, the coefficient es-timates should be largest in the low-educated wage regressions. These haveall been rejected by the empirical results.The empirical results of credit shock on wage growth point to a moregeneral explanation. If the price of labor has become cheaper in the high-credit regions, what stops labor demand from responding? I find some evi-dence from firm sides: counties with a larger credit supply shock witnessed60a larger increase in the employment share of old firms and a larger decline inthe employment share of young firms since the recession, accompanied bya decrease in the total number of firms. This evidence suggests a plausiblemechanism working from credit shock to the local labor demand side.In this chapter, I propose a mechanism to show how credit shocks affectthe labor market through its effect on the housing market. The frameworkintroduces two types of entrepreneurs with different discount factors, andtwo financial constraints: borrowing and working capital constraints. Bothborrowing and working capital constraints are tied to housing values. Pro-duction technology requires labor input and features decreasing return toscale. By relaxing and tightening borrowing constraints, this simple frame-work can match the empirical findings: positive credit supply shock doesnot affect labor markets, and the negative credit supply shock leads to wagedeclines and labor reallocation. Changes in borrowing constraint affect thelabor market through its effects on housing price. The nature of this crediteffect asymmetry is captured by the occasionally binding working capitalconstraint: a relaxation in borrowing constraint does not affect the labormarket if the working capital constraints are already slack; a tightened bor-rowing constraint restricts working capital constraint and leads to labor re-allocation from constrained to unconstrained entrepreneurs, resulting in la-bor productivity loss.This mechanism provides a plausible explanation to rationalize the em-pirical findings documented in chapter one. It explains the negative effectsof the credit shock on local wage growth with employment reallocation.The reallocation process causes labor productivity loss as the productionfunction features decreasing return to scales, which suggests the optimallabor allocation to be uniformly distributed across all firms. As financiallyunconstrained firms absorb the excess labor supply after a credit tighten-ing, it reduces labor productivity and wage in the equilibrium. Note thatthis may not be the only mechanism to explain how credit shocks affect thelabor market and lead to wage decline. Alternative explanations include thedecline in worker’s wage bargaining power due to the decreasing number offirms; the labor productivity loss due to long-term unemployment; the de-61cline in firm productivity due to reducing R&D expenditures. All the abovemechanisms are not mutually exclusive and could have all contributed tothe decline in local wage growth in the high-credit shock regions. It is be-yond the scope of this chapter to distinguish the contribution of differentmechanisms.This model is built on a large body of literature that studies the macroe-conomic implications of credit constraints. Among them, a large numberof works focus on firm-side frictions that to affect investment and aggre-gate economies (Bernanke et al., 1999, 1991; Jermann and Quadrini, 2012;Kiyotaki and Moore, 1997). In my model, this is achieved through work-ing capital constraints and its effect on labor demand. The working cap-ital constraint is often used to study the effects of financial constraints inthe open economies (Arslan et al., 2020; Neumeyer and Perri, 2005; Uribeand Yue, 2006).1 Among the recent studies, there is a growing amount ofresearch linking financial frictions with productivity losses (Buera et al.,2015; Gilchrist et al., 2014; Khan and Thomas, 2013). In my model, laborproductivity loss happens with labor reallocation due to firm side financialconstraints.2.2 Model2.2.1 SetupConsider a discrete-time, infinite horizon economy, populated by workersand entrepreneurs. There are two types of entrepreneurs: patient and im-patient, defined by their discount factor. The impatient entrepreneur (witha low discount factor) shares the same discount factor with workers. The1Arslan et al. (2020) develop a general equilibrium model with housing tenure decisions andlong-term mortgages, working capital constraints faced by firms, and the leverage constraintimposed on financial intermediaries, to show how the change in credit-supply generates aboom-bust cycle similar to the one observed in the U.S. around 2008. Although the mech-anism proposed in this chapter shares many similarities with their work, this chapter fo-cuses on explaining how credit supply shocks shape the labor market after the recession,whereas they study the general equilibrium effects of credit supply shock on the aggregateeconomies.62economy features two types of goods: housing and non-durable goods.Housing h is supplied with a fixed stock and is non-depreciable. Non-durable good c is produced and consumed every period. Production takeslabor as input. Workers provide labor and earn wages. Entrepreneurs onlyprovide production technology and claim production residuals.To allow for financial friction to play a role in this framework, I introducetwo constraints: one is the borrowing constraint faced by all agents, theother one is the working capital constraint, only faced by entrepreneurs.I will show how changes in borrowing constraints can affect the workingcapital constraint, and thereby production.WorkersThe worker’s problem is standard. They maximize lifetime utility subject tobudget constraint and borrowing constraint:∞∑t=0β tl u(ct ,ht)s.t. ct + ptht+1+bt+1 = bt(1+ rt)+wt + ptht−bt+1 ≤ φbptht+1, where βl denotes for low discount factor, ct is the consumption at time t,ht is the housing stock, bt is the bond holding, pt is the housing price, rt isthe interest rate applied at time t. The inequality equation represents theborrowing constraint faced by workers. It regulates the maximum amountof debt agents can borrow. φb determines the fraction of debt with respect tocurrent housing stock, known as loan to value ratio. This parameter deter-mines the tightness of borrowing constraints. Note that, at time t, workersnot only choose between consumption and saving but also need to makeportfolio choice on (bt+1,ht+1). The current utility form does not allow forendogenous labor supply just for simplicity. But it is possible to endogenouslabor supply, and the main results hold.Define λw,t as the Lagrange multiplier associated with borrowing con-63straint. The first-order conditions are given as follows:uc,t = βl(1+ rt+1)uc,t+1+λw,t (2.1)ptuc,t = βluh,t+1+βluc,t+1pt+1+ ptφbλw,t (2.2)The first equation is the consumption Euler equation, and the second oneis the trade-off between consumption today and saving in housing. In bothequations, the right-hand side has the term λw,t , which represents the shadowvalue of the borrowing constraint. If the worker is borrowing constrained(λw,t > 0), there is an extra benefit ptφbλw,t for holding one unit of housing,as workers can increase their borrowing by φbpt amount. In the steady state,workers are always borrowing constrained because the steady state interestrate is determined by the high discount factor. So, in the steady state, impa-tient agents always prefer to borrow to consume today, which drives themto the borrowing constraint.EntrepreneursEntrepreneurs hire workers to produce consumption goods. Productionscale is subject to working capital constraint, i.e., the wage bill needs to bepaid before production happens. The wage bill is financed by within-periodborrowing without any interest cost. As in the standard financial friction lit-erature 2, assume there is limited enforcement on the debt payment. So, toensure the debt is repaid, entrepreneurs need to use their house as collat-eral. They can borrow up to φh fraction of house value, where φh captures theamount of debt that can be recovered with collateral if a default happens.2See Kiyotaki and Moore (1997) and Iacoviello (2005).64Then, the entrepreneur’s problem is specified as follows:∞∑t=0β ti u(ct ,ht), i ∈ {l,h}s.t. ct + ptht+1+bt+1 = bt(1+ rt)+pit + ptht−bt+1 ≤ φbptht+1pit = lγt −wt ltwt lt ≤ φhpthtThe working capital constraint restricts the production scale for each en-trepreneur and gives them an additional incentive to hold housing stocks.One thing to point out is that the borrowing and working capital constraintsare associated with different timings of housing values. At the beginning ofperiod t, the entrepreneur borrows against the current housing stock to hireworkers and conduct production. At the end of the period t, productionis distributed, and entrepreneur makes a saving decision bt+1 for the nextperiod t+1, subject to the current value of next period’s housing stock ht+1.The production function is in the Cobb-Douglas form, with the decreas-ing return to scale. The decreasing return to scale assumption is essentialin this framework, as it requires the optimal production (labor) allocationto be equally distributed among the entrepreneurs. If the production is aconstant return to scale, the working capital constraint will not affect laborproductivity and wage, as the marginal product of labor is constant. En-trepreneurs optimal choices are governed by following equations:uc,t = βi(1+ rt+1)uc,t+1+λi,t (2.3)ptuc,t = βiuh,t+1+βipt+1uc,t+1+ ptφbλi,t +βiφhpt+1µi,t+1 (2.4)uc,t(γlγ−1t −wt) = wtµi,t (2.5)λi,t denotes the Lagrange multiplier of borrowing constraint for type i en-trepreneur, and µi,t denotes the Lagrange multiplier of working capital con-straint. The first equation is the consumption Euler equation, the same asworkers. The second one governs the optimal housing stock. The right-65hand side of the equation represents the marginal benefit of holding an ex-tra unit of the house. As compared to the optimal condition for worker,entrepreneurs have an extra incentive to hold housing, as one unit of thehouse can expand the production by φhpt+1, which increases marginal util-ity by µi,t+1 for time t+1. The last equation represents the optimal choice oflabor demand, given the working capital constraint.Note that both constraints are not necessarily binding at the same time.In the steady state, the impatient entrepreneur will always face the bind-ing borrowing constraint, with the same argument as workers. The patiententrepreneur will be the saver in the economy, with slack borrowing con-straints. Depending on the parameter settings, both of them may or maynot face working capital constraint binding.Market clearingThe model is a closed economy. Bonds, durable and non-durable goodsmarket clear simultaneously.bw,t L¯+bl,t +bh,t = 0 (2.6)hw,t L¯+hl,t +hh,t = H¯ (2.7)cw,t L¯+ cl,t + ch,t = f (ll,t)+ f (ll,t) (2.8)Note that H¯ is the total stock of housing and L¯ is total number of workers.2.2.2 EquilibriumThis economy is deterministic and does not have any stochastic shocks. Inthe steady state, the patient entrepreneur provides saving and impatient en-trepreneurs and workers will borrow against their housing stock up to thelimit. There exists equilibrium allocation that {hl,hh,hw,cl,ch,cw,bh,bl,bw}together with equilibrium prices {w,r, p} satisfying the equation (2)-(9), re-spectively (the subscript l and h denotes the choices for patient and impa-tient entrepreneurs, respectively).66In the steady state, the interest rate is determined by patient agent:r =1βh(2.9)All the impatient agents will borrow up to the limit. The patient entrepreneurprovides funds for all the loans. The steady state can be found by solvingthe optimal housing distribution. The steady-state conditions can be foundin the Appendix.2.2.3 Transmission Mechanism: from housing to labor marketThe essence of this model is two financial frictions: borrowing constraintand working capital constraint. The goal of the model is to show how fluc-tuations in the borrowing constraint can affect the labor market through itseffect on the housing market. Housing value regulates both maximum debtand production levels for entrepreneurs. Because borrowing constraints arealways binding for impatient agents in the steady state, a change in borrow-ing constraint affects their incentive to hold the house and leads to housingprice fluctuations. Housing price changes affect the working capital con-straint and transmit the credit shock to the labor market. This statement isnot always true. When a working capital constraint is sufficiently slack, achange in housing prices will not affect the labor market. This is shown inthe Figure 2.1.Given different values of φb and φh, the top panel plots the steady-statewages, and the bottom panel plots the steady-state outputs. Figure 2.1 showshow borrowing constraints and working capital constraints affect the labormarket allocations in the steady state. When the working capital constraintis tight, relaxing borrowing constraints will always increase wage and out-put because relaxing borrowing constraint leads to housing value apprecia-tion, which in turn relaxes the working capital constraint. Because produc-tion is decreasing return to scale, labor reallocation from unconstrained toconstrained firms improves production efficiency and increases total out-put. Note that the parameters are set such that working capital constraint isalways slack for the patient entrepreneur. If the working capital constraint67is binding for both entrepreneurs, relaxation in working capital constraintwill still increase wages, but labor allocation is unclear. This positive trans-mission from housing to the labor market disappears when φh is sufficientlyhigh. For example, when φh = 0.6, changes in borrowing constraint barelyaffect production because the working capital constraint is slack for bothentrepreneurs, and optimal production is reached.This figure shows the central mechanism of the model: credit shockleads to housing market fluctuations and transmits to labor markets whenthe working capital constraint is binding. This can explain the asymmet-ric effects of credit shock: positive credit shock does not affect local labormarkets, but negative credit shock disrupts labor markets.2.2.4 Quantitative AnalysisIn this exercise, I will use changes in borrowing constraints to match mort-gage market changes in the empirical part and show how much it explainslabor market fluctuations. The utility form follows standard constant rela-tive risk aversion (CRRA) setting: u(ct ,ht)= log(ct)+αhlog(ht). The time pe-riod is set as annual frequency, consistent with the empirical section. Param-eter values are presented in Table 2.1. The patient agent’s discount factor isset to match the equilibrium interest rate as 2%. Following the literature, theimpatient agent’s discount factor is set as 0.95.3 The elasticity for housinggood is commonly set to match the share of housing expenditure. However,housing is non-depreciable in this framework. So, I will follow the conven-tional setting as Iacoviello (2005) and Justiniano et al. (2019) to set αh to be0.1. The parameter φb governs the maximum of loan to value ratio, whichis 0.8 for conventional loans4. Parameter φh regulates the working capitalconstraint. As shown in the last section, the value of φh determines thesteady-state wage and labor productivity. As data on working capital loansare not available for both in aggregate and regional economies, I choose φhto make the impatient entrepreneur at the margin of slack working capi-3See Iacoviello (2005) for empirical evidence.4Conforming loans that meet Fannie Mae and Freddie Mac underwriting guidelines are lim-ited to an LTV ratio that is less than or equal to 80%68tal constraint, such that any negative credit shock will push the impatiententrepreneur to a binding working capital constraint. In the quantitativeexercise, this answers at most how much of the percent of wage decline canbe explained by credit shock. The last part of the parameter is labor share,housing stock, and labor supply. Labor input parameter γ is set to match70% labor income in the U.S. economy. Empirically, on average, there areten employees per firm at the county level. So, the total labor supply is setto be 20. The total housing unit is set so that the mortgage debt payment toincome ratio is close to 5.56%, matched with aggregate moment in 2000.To match the model with empirical findings, I introduce two unexpectedshocks to the system: one positive credit shock and one negative creditshock. First, I calibrate an increase in φb by 0.15 to match a 40% mortgagegrowth during 2003–2006, as shown in Table 1.6. After that, I decrease theφb by 0.3 to match a 45% decline from 2006–2010. Note that both shocks areconsidered as exogenous and permanent to all agents. The second shockoccurs when the new steady state is reached after the first shock. 5Table 2.2 shows that model moments are qualitatively matched withdata moments. During the expansion period, housing price increases by17%, while the empirical result in Table 1.7 suggests an 11.3% increase. Themodel over-predicts the housing price change, which might due to the fixedhousing supply. In the real world, housing supply responded to increasinghousing demand. Wage and employment share are held constant, whilethe data moments show slight but statistically insignificant changes. Notethat the employment share of the impatient entrepreneur is used to com-pare with young firm employment shares, as this model does not includefirm entry and exit channels. Debt payment to income ratio is computedas the interest payment to wage ratio. Its data moment is taken from theaggregate data due to data availability. While data suggests 1.5 percentage5The positive credit shock is supported by much empirical evidence in documenting the relaxin lending standard in the early 2000s. However, the credit crunch process is complex andhard to explain with any simple mechanism. One might want to endogenize the processof credit contraction. It is impossible to endogenize it in this simple framework withoutfinancial intermediaries. To keep this model simple, I will assume the exogenous process oftightening in borrowing constraint.69points increase in debt service, this model predicts a 2.2 percentage pointsincrease. However, it is unclear how interest payment to wage is changedat the county level.In the recession period, changes in debt and housing prices in the modelare matched well with data moments. This model only predicts wage de-cline by 0.88%, while the regression estimate is 5.21%. Note that all thedata moments are taken from empirical estimates in the recession period2006–2010 (decline in the employment share of young firms also happensin the recession). It also under-predicts the changes in employment share,with a 1.3% decline in the model and 2.22% in the data. Even if the em-ployment share is perfectly matched with data moment, this model onlypredicts a 1.3% decrease in wage. Labor reallocation effect can explain atmost 20% of the wage decline in the data, suggesting there are other chan-nels to depress wage growth.Relaxing borrowing constraintIn this section, I present the transition path of the economy in response toa 0.15 unexpected permanent increase in φb. Suppose the economy startsfrom a steady state. Figure 2.2 plots present how the economy transits tothe new steady state. 6A permanent relaxation in borrowing constraints increases housing de-mand and drives up housing prices. Because now agents can borrow moreagainst their houses, it increases the marginal utility of housing for bor-rowing constrained agents. In response, impatient agents increase housingstocks. Along the transition, housing price increases monotonically startingfrom period 3. In period 2, housing prices experience a small drop becausethe unexpected housing appreciation increases the agent’s net wealth. Tosmooth consumption, impatient agents start to increase consumption andhousing from period 2, financed by debt, which drives up the interest rate.The patient agent decreases his housing stock to provide more loans. Thisportfolio adjustment drives down the housing price in period 2. As the de-6The transition period is set to be 30. It does not affect the new steady state or the transitionpath if the total time period is extended.70mand for loans slows down, the patient agent gradually increases his hous-ing stock and house price increases.Focusing on the workers’ consumption path, it seems counter-intuitivethat the new steady-state consumption is lower than the previous one. Thatis because workers are impatient and always prefer consumption to sav-ing. Although the interest rate has increased, βl(1+ rt+1) < 1 remains truealong the transition. Therefore, all impatient agents follow the decreasingconsumption path. In the new steady state, due to higher debt and con-stant wage, consumption is permanently lower. This is true for the impa-tient entrepreneur, as well. The opposite statement holds for the patiententrepreneur.The key point in this exercise is that fluctuations in the housing and bondmarkets do not affect the production sector. Wage and employment alloca-tion are constant along the transition because the working capital constraintis slack in the initial steady state. Efficient production has already beenreached. A relaxation in borrowing constraint further relaxes the work-ing capital constraint but cannot improve production efficiency any more.This is consistent with the empirical finding in the pre-recession period thatpositive credit shock does not predict county employment or wage growth.When working capital does not restrict production frontier, changes in thecredit market cannot affect production.Tightening borrowing constraintThe starting point is the steady state reached in the previous exercise. Af-ter an unexpected decline in φb by 0.3, the transition path in Figure 2.3 lookslike the opposite of Figure 2.2. Tightening in the borrowing constraint forcesall borrowers to deleverage by selling houses and cutting consumption. Inresponse, the interest rate and housing price decline. Note that the inter-est rate is assumed to be non-negative. This framework does not introducemoney and inflation. So, the real interest rate is equal to the nominal in-terest rate. A zero lower bound is imposed on the real interest rate. Thisassumption is not crucial; it only affects the transition path, not the steady71state.Now, tightening in borrowing constraint affects the production side. Acredit crunch leads to a decline in wage and employment for the impatiententrepreneur because the impatient entrepreneur is forced to sell housesto repay his debt. Together with the declining house price, these lead totightening in the working capital constraint, which restricts the productionpossibility for the impatient entrepreneur and forces him to reduce em-ployment. The unconstrained patient entrepreneur absorbs the excess la-bor supply. Due to the diminishing marginal product of labor, this reducesthe equilibrium wage. This effect is strongest in period 2, with a 10% de-cline in wage and a 30% decline in the impatient firm employment. Labormarket conditions gradually recover from the credit crunch as the impa-tient entrepreneur starts to accumulate housing stock again. In the newsteady state, the impatient entrepreneur hires fewer workers, and wage ispermanently lower than before. This suggests the tightening in borrowingconstraint has permanently restricted production scale for the impatient en-trepreneur and causes production loss. Note that decreasing return to scaleassumption is important in this framework. It generates production effi-ciency losses due to labor reallocation.2.3 DiscussionIn this chapter, I propose a simple model to explain the negative effect ofcredit shock on wage growth and young firm employment share. Introduc-ing two financial frictions, one on the household side and the other on theproduction side, shows how changes in household borrowing constraint af-fect the collateral value of housing and thereby affect firms’ financial condi-tion. This effect transmits to the labor market when a firm’s working capitalconstraint is binding. For simplicity, this model excludes firm-entry andexit channel. The financially constrained firm (impatient entrepreneur) ismapped to young firms in the empirical section. The model also shuts downthe endogenous labor supply channel. The primary mechanism remainswith endogenous labor supply.72Consistent with empirical findings, the model also explains how creditmarket fluctuations can have stronger effects on the production and labormarket in bad times than good times. The mechanism shown here is in thesame spirit as Guerrieri and Iacoviello (2017), which shows the asymmetriceffect of the house price change on consumption growth. The main differ-ence here is that I introduce the working capital constraint, which allows thecredit market to affect the real economy through its effect on labor demand.In their work, the asymmetric mechanism works through the consumptionchannel.The financial frictions imposed in this model also share similarities withMendoza (2010), which uses a collateral constraint to explain Sudden Stopsof emerging economies after the financial crisis. Instead of introducing twofrictions, he assumes the sum of working capital and intertemporal debts islimited by the collateral value of capital. With binding collateral constraints,a negative exogenous shock can force the economy to deleverage and re-duces access to working capital. My model shares a similar effect of creditshock on labor demand. The benefit of introducing two separate frictions isthat it allows the credit shock to affect the housing and bonds market sep-arately without affecting the labor market and generate asymmetric effectson the real economy.This chapter provides a plausible mechanism to explain how credit fluc-tuations affect the labor markets and lead to wage decline. However, thismay not be the only mechanism. There are many other possible explana-tions. For instance, the decreasing number of firms has given the existingfirms higher bargaining power over workers or productivity loss in workersdue to unemployment. These mechanisms are not exclusive and can all helpto explain the slowdown in wage growth. The model proposed here servesas one plausible mechanism. Future work needs to be done to quantifyingdifferent working channels.732.4 FiguresFigure 2.1: Steady state and parameter settingsThe top and bottom panels plot the steady-state wage and output with differentvalues of φb and φh, respectively. The utility form is u(ct ,ht)= log(ct)+αhlog(ht).The rest of the parameters can be found in the Appendix.74Figure 2.2: Transition path with positive shock on φbNote: transition path after increasing φb by 0.15. The first row presents changes in wage,housing price and interest rate along the transition. The second row plots changes inworker’s consumption, saving and housing stock. The third row plots changes in employ-ment, bonds, and housing stock for the impatient entrepreneur. The last row plots a transi-tion for patient entrepreneur’s employment, bonds, and housing stock.75Figure 2.3: Transition path with negative shock on φbNote: transition path after decreasing φb by 0.3. The first row presents changes in wage,housing price and interest rate along the transition. The second row plots change in worker’sconsumption, saving and housing stock. The third row plots changes in employment, bonds,and housing stock for the impatient entrepreneur. The last row plots transition for patiententrepreneur’s employment, bonds, and housing stock.762.5 TablesTable 2.1: Parameter settingsDescription Parameter valuehigh discount factor βh 0.98low discount factor βl 0.95weight on housing service αh 0.1borrowing constraint φb 0.8working capital constraint φh 0.65labor share γ 0.7total housing stock H¯ 10total labor supply L¯ 20Table 2.2: Quantitative resultsMoment Model DataExpansion 2003-2006∆bw 38.5% 40.2%∆w 0 -0.8%a∆p 17% 11.3%∆ll/L¯ 0 0.7%b∆rbw/w c 2.2% 1.5%Contraction 2006-2010∆bw -47.6% -45.5%∆w -0.88% -5.2%∆p -20.7% -19.7%∆ll/L¯ -1.3% -2.2%∆rbw/w -3.8% -1%Note: this table shows the model and data moments with respect to creditshocks. The top section shows the model and data moments in response toa positive credit shock. The bottom section shows model and data momentsafter a negative credit shock. a The data moment is statistically insignificant.b The data moment is statistically insignificant. c This data moment for debtservice to income ratio uses aggregate data moment, obtained from the FederalReserve Board. It is unavailable at the county level.77Chapter 3China’s Proactive Fiscal Policy:Assessing the Early Impact ofTax Cuts on Small Firms3.1 IntroductionSmall businesses are often thought to be essential to job creation, innovationand investment (Decker et al., 2014). Consequently, in addition to loweringbarriers to entry to the formal sector, many governments around the worldhave implemented various incentive schemes for small firms, including thegrant of preferential tax treatments. Developmental economics has paid ex-tensive attention to policies that induce firms to move into the formal sector(La Porta and Shleifer, 2014; Ulyssea, 2019), but little is yet known abouthow tax incentives targeted at small firms affect their performance. Whilerecent studies suggest that small firms respond more strongly to general in-vestment incentives in ways policymakers intend (Agrawal et al., 2020; Liuand Mao, 2019; Zwick and Mahon, 2017), strong concern remains that taxpreferences specifically granted to small firms may be distortionary, includ-ing by encouraging them to stay small (Hsieh and Olken, 2014).In this study, we analyze the effect of an important tax preference granted78to small firms in China in recent years. Ever since 2008, China has pursued aseries of expansionary, or “proactive”, fiscal policies: first in reaction to theGlobal Finance Crisis, then to address declining economic and investmentgrowth (Bai et al., 2016), and most recently, to support the economy dur-ing the U.S.-China trade war and recovery from the COVID-19 pandemic.Yet since 2014, cutting the tax burdens of small firms has become increas-ingly central to this proactive fiscal agenda. A prominent component of theagenda is a series of temporary corporate rate cuts for “small and micro-profit enterprises” or SMPEs, which has reduced the statutory corporateincome tax rate for SMPEs by as much as three fourths (from an originalstatutory rate of 20%). Strikingly, by 2019, the government estimated thatclose to 95% of all firms in China enjoy one way or the other these SMPE taxbenefits.Unlike many other tax policy changes adopted during recent years inChina, such as the reform of the VAT (Chen et al., 2019), the income taxrate cuts for SMPEs took place within a relatively stable corporate incometax regime put in place in 2008. And unlike turnover and payroll taxes, de-sign of the income rate cuts was not intended to address compliance issuesor focus on particular behavioral margins (such as fixed asset investment),but works more like a pure transfer from the government to businesses.Using confidential corporate tax returns from firms in a large and relativelyprosperous Chinese province, we conduct an analysis on how the earlierinstallment in the SMPE rate cuts affected the performance of small busi-nesses. Specifically, we examine how the rate cuts affect productivity andinvestment, as well as the impact of the tax cuts on firms’ entry decisions.We first develop a theoretical model where individuals choose betweenbeing an entrepreneur and being a wage earner. This general equilibriummodel illustrates channels through which lowering the corporate incometax rate affects the behavior of small firms. We then examine, based on con-fidential corporate tax returns for the period 2010-2016, the consequencesof lifting the qualifying threshold for a lower tax rate using a difference-in-differences approach. In our benchmark analysis, we use newly qualifiedSMPE firms as the treatment group, and compare them with always-large79firms. To increase the comparability between the two groups, we performpropensity score matching based on their pre-treatment characteristics.We find that newly qualified SMPE firms enjoyed a substantial increasein their sales growth, investment rate and total factor productivity. The ef-fect on investment is especially large, yielding much larger elasticity of in-vestment with respect to the user cost of capital than previous studies re-port. Consistent with the conjecture that that the tax cuts reduce the cost ofcapital and increase the cost of labor, we find evidence for substitution be-tween capital and labor, especially for the 2014 and 2015 treatment cohortswhich had less time to adjust factor inputs. Further, based on firms’ registra-tion data, we show that the tax rate cuts led to more entries of micro-sizedfirms. We also find around 25% of the 2012 treatment cohort grew abovethe SMPE threshold by the end of our sample period, alleviating the con-cern that the tax incentives would keep firms small.To our knowledge, ours is among the first that analyze China’s recent taxpolicy targeted specifically at small firms, which represents a crucial eco-nomic policy tool for the Chinese government both to navigate adverse do-mestic and international economic environments and to enhance regime le-gitimacy.1 In other recent studies of Chinese tax policy reforms, even whereresearchers had access to some data from small firms, such firms were ei-ther used as a control group (Liu and Mao, 2019), or dropped from analyses(Chen et al., 2019). More generally, difficulties in accessing firm-level datahave hindered studies of the impact of tax policy in small firms, even thoughthere are good reasons to believe that small firms may respond to tax incen-tives differently from large ones (Zwick and Mahon, 2017). Our analysisleverages unique tax return data that covers the lower portions of the firm-size distribution, data that is well-tailored to studying the important policyquestion of how small firms respond to corporate tax incentives.Our study first contributes to the limited literature that explicitly ex-1Policies targeted at small firms have become especially important in post-COVID19 eco-nomic recovery. In an interview with the press on May 28, 2020, Chinese Premier Li Ke-qiang promoted China’s proactive tax policies targeted at small firms since 2015 as well-positioning China to deliver targeted stimulus in 2020.80amines tax policies that aim to stimulate small firms. Maffini et al. (2019)examines how first-year capital allowances affect investment of qualifiedsmall firms in the UK. A small number of studies analyze the effective-ness of R&D tax credit on small firms’ R&D activities (Agrawal et al., 2020;Dechezlepreˆtre et al., 2016; Koga, 2003; Lokshin and Mohnen, 2012). Com-pared with accelerated depreciation or R&D tax credit, a cut in the statutorycorporate income tax rate is more salient and potentially less distortionary.Harju et al. (2020) analyze the impact of a 2014 corporate rate reduction(from 24.5% to 20%) in Finland on small firms, and found substantial pos-itive effect on firm sales and variable costs, though no significant effect oninvestment. By contrast, we find positive policy impact on all these marginson Chinese small firmsRelatedly, our study also contributes to the growing literature on theeffectiveness of tax incentives in developing countries (Chen et al., 2019,2018). Simpler tax incentives may be more suited to a developing countrycontext. For instance, Cui et al. (2020) find that the majority of eligible firmsdid not take up a recent Chinese corporate tax incentive in the form of accel-erated depreciation and it failed to stimulate firms’ investment. Moreover,smaller firms were less likely to claim the accelerated depreciation benefits.This raises the question whether simpler tax incentives may be more effec-tive in developing countries. Studies of the impact of a direct tax cut onsmall firms in developing countries are few. Pham (2020) analyzes a tem-porary corporate income tax cut for Vietnamese firms that are below certainsize thresholds, and finds that the tax cut increased investment but only inthe reform year. The tax cuts for small firms we study, although originallyintroduced as temporary, has at least for now become a relatively perma-nent element of Chinese tax policy.We also contribute to the literature that examines the impact of tax poli-cies on firms’ productivity and firm entry. Empirical evidence on these mat-ters is scarce. Arnold et al. (2011) find that corporate income tax rate isnegatively associated with industry-level total factor productivity, but findno impact of the corporate tax rate on small and young firms’ productiv-ity. However, their identification relies on the variation across industries81in terms of their tax base. In our study, we identify the impact of tax in-centives on productivity using the difference-in-differences estimator. Ournatural-experiment approach allows us to pin down a significant effect onfirm productivity. Chen et al. (2018) and Liu and Mao (2019) also use thenatural experiment approach to study the impact of tax policy reforms onfirms’ productivity. Chen et al. (2018) analyze the impact of a lower statu-tory corporate tax rate on Chinese high-technology firms’ R&D activities.They find that the lower tax rate leads to substantial increase in R&D in-vestment and productivity for qualified firms. Liu and Mao (2019) showthat the reform of China’ value added tax that allowed input VAT credit tobe claimed for fixed asset purchases significantly increased firms’ produc-tivity. As noted above, none of these studies focuses on small firms, whichare increasingly emphasized by governments in China and elsewhere.Finally, our study, by examining a tax policy reserved for low profitfirms, contributes to the important debate about the proper treatment ofsmall firms in developing countries. Recent scholarship has emphasized thesubstantially lower productivity of small firms compared to large firms, andquestioned the wisdom of targeting preferential policies (including low taxrates) at small firms (Hsieh and Olken, 2014; La Porta and Shleifer, 2014).While such skepticism is often well justified, we provide evidence that taxpolicy tools may nonetheless stimulate small firm productivityThe remainder of the study is as follows. Section 2 discusses the policybackground. Section 3 provides our theoretical model. Section 4 illustratesour empirical strategies. Section 5 describes the data and main results areprovided in Section 6. Section 7 concludes.3.2 Policy BackgroundWe study a Chinese government policy that extended a corporate incometax cut to an ever-larger population of firms. Under the 2008 Enterprise In-come Tax Law, “small and micro-profit enterprises” (SMPEs) are entitledto a 20% tax rate, as compared to the regular 25% rate on taxable income.SMPEs are defined in State Council regulations as firms with (i) annual tax-82able income not exceeding 300,000 yuan, (ii) not more than 80 employees(or 100 employees for industrial firms), and (iii) total asset of not more than10 million yuan (or 30 million yuan for industrial firms). In response to theGlobal Financial Crisis, China’s Ministry of Finance announced in Decem-ber 2009 that for any firm that qualified as an SMPE but also had taxableincome not in excess of 30,000 yuan, only half of its taxable income neededto be included in computing its income tax liability. The tax rate on suchfirms was thus effectively reduced to 10%.2This rate reduction initially was to apply only in 2010—the first year forour data—but was subsequently renewed for 2011. Moreover, beginning in2012, the taxable income threshold under which the half-income-inclusionrule applied was raised several times: the threshold was (i) 60,000 yuan for2012-3; (ii) 100,000 yuan for 2014; (iii) 200,000 yuan for the first three quar-ters of 2015; and (iv) 300,000 yuan for the 4th quarter of 2015 and 2016.3By the end of 2016—the last year of our data—the half-income inclusionregime had completely eclipsed the 20% regime for SMPEs. Figure 3.1 il-lustrates this gradual increase in the taxable income threshold for SMPEsduring our sample period. Meanwhile, the qualifying thresholds in termsof total asset and employees remained intact.While the half-income-inclusion rule for SMPE firms has always beenannounced as a “temporary measure”, it has become an important generaltax reduction measure receiving great political emphasis. A turning pointcame in 2015, when Premier Li Keqiang decided to make corporate tax re-duction a central component of the government’s policy to encourage en-trepreneurship. As a result, the threshold for the tax preference was raisedtwice that year. While we study the impact of the rate reduction for SMPEfirms during the 2010-2016 period, it is worth noting that in 2017, the gov-ernment raised the taxable income threshold for the half-income-inclusionrule to 500,000 yuan (for 2017-19), and in 2018, to 1 million yuan (for 2018-2If the SMPE firm already qualified for some other preferential statutory rate, such as the15% rate for high-and-new-technology enterprises of HNTEs, the half-income inclusion ap-proach could lead to an even lower corporate tax rate, i.e. 7.5%.3The four threshold increases after 2011 were announced on November 29, 2011, April 8,2014, March 13, 2015 and September 2, 2015, respectively.832020). In 2019 (for the years 2019-2021), the asset and employee thresholdswere also lifted to 300 employees and 50 million yuan in total assets, regard-less of sector. Moreover, firms that fell under such new employee and assetthresholds and that earned less than 1 million yuan could include only onequarter of their income—reducing their tax rate to 5%–while those with in-come between the 1 and 3 million range can claim half-income inclusion ora 10% rate.4 Chinese President Xi Jinping spoke of these policies as deliv-ering “inclusive” tax cuts. China’s political leadership appeared to begin toadopt the view that to maintain stability in economic growth, it is no longersufficient to channel resources to large and political connected firms (Baiet al., 2016). Expansionary or “proactive” fiscal policy must target smallfirms.Our study examines the earlier phases of this strand of China’s “proac-tive” tax policy by focusing on the rate reduction for SMPEs introduced in2012, 2014, and March 2015. This choice is due to our data and methodol-ogy—we apply a difference-in-differences approach to firm-level data from2010 to 2016. But one implication of the policy history above is that it isplausible that by 2014 or 2015, if not by 2012, some firms may have come toregard the rate reduction as semi-permanent. In terms of potentially con-founding policies enacted during the same period, three are notable. First,China gradually rolled out the integration of its VAT with the Business Tax,a turnover tax on services, between 2012 and 2016 (Cui, 2014). Because weinvestigate the impact of income tax reductions on firm productivity, in ourstudy we choose to focus on manufacturing firms, which were not directlyimpacted by the 2012-2016 VAT reform (and for which it is reasonable to as-sume that indirect impacts through affected suppliers and customers weresimilar across our treated and control groups). Second, under the corpo-rate income tax regime, the government enacted accelerated depreciationpolicies in 2014 and 2015 (Cui et al., 2020; Fan and Liu, 2020). These poli-4In the rest of the chapter we will refer to the policy as a cut of the corporate tax rate to 10%.This is largely accurate for the period we study, although the partial-income-inclusion rulecould also apply to taxpayers facing the 15% (before 2016) and 25% (after 2016) statutoryrates.84cies affected manufacturing firms large and small, and thus again can beassumed to affect our treated and control groups equally. Third, the lawon the personal income tax (PIT) was amended in 2011, which (i) loweredthe tax rate for lower levels of wage income, (ii) slightly raised tax at thehighest levels of wage income, and (iii) lowered tax on sole proprietor orpartnership income. However, the top corporate tax rate remained lowerthan the highest marginal rates on wage (45%), non-wage labor compensa-tion (32%), or self-employment income (35%).5 Corporations in China areallowed a wide range of deductions in computing income, while deductionsare limited for wage earners. Income from sole proprietorships or partner-ships is taxed currently but losses cannot flow through to reduce other (e.g.wage) income (Cui, 2007). Therefore, the PIT changes are unlikely to havechanged the relative benefits of earning income through the corporate formfor entrepreneurs with the potential of earning high income.6To examine whether the tax rate cut and the changes in qualifying thresh-olds for SMPEs are salient, Figure 3.2 provides the search intensity for thefollowing key words (in Chinese) from Baidu, the most popular search plat-form in China: “preferential corporate income tax rate for SMPEs”, and “thequalifying criteria for SPMEs”. To provide a benchmark, we also illustratethe search intensity for the key word “tax reporting”. Figure 3.2 shows thatthe search intensity for the first two key words move closely with that for“tax reporting”, which partly explains the volatility in these series. How-ever, there are periods of intensive searches for the first two key words, no-tably during much of calendar years 2012, 2014 and 2015. This suggeststhat much attention has been paid to both items. As a comparison, we plotthe search intensity for the key word “accelerated depreciation for fixed as-sets” (AD), a tax incentive effective from 2014 that allows firms to opt to afaster depreciation schedule for eligible fixed asset purchases. It is striking5The personal income tax (PIT) rate in China on dividend and capital gain from ownershipof non-listed companies is 20%. Depending on the corporate tax rate, the aggregate incometax rate on taxable income earned through a corporation can thus range from 24% to 40%.6The most prominent type of incorporation to minimize PIT in China during the period westudy involved incorporation in jurisdictions outside the province we study that offered gen-erous tax exemptions.85that the tax cuts for SMPEs are substantially more searched for than AD,indicating the salience of the former.3.3 Theoretical FrameworkTo illustrate how corporate tax cut affects firms’ behavior, we present a modelfeaturing a small open economy with monopolistic competition and en-dogenous firm entry. In this model, we show how a decline in the corporatetax rate encourages entrepreneurship, investment and technology upgrad-ing by increasing firm profit, reducing cost of capital, and expanding pro-duction scale.3.3.1 ProductionFinal GoodsThis economy features intermediate and final goods productions. The finalgoods is traded in the competitive market at price P. It is produced with amixed variety of intermediate inputs, using the production technology:Yt =( i=n∑i=1yρi,t) 1ρ, (3.1)where ρ is the elasticity of demand, and ρ ∈ (0,1). The firm chooses theintermediate inputs to maximize its profit:maxyi,tPtYt −i=n∑i=1pi,tyi,t ,subject to constraint (3.1), where pi,t is the relative price of intermediategoods i. The aggregate price level is defined asPt =( i=n∑i=1pρρ−1i,t) ρ−1ρ.86Solving the maximization problem, we obtain:yi,t =(pi,tPt) 1ρ−1Yt . (3.2)Intermediate GoodsThe intermediate goods market is monopolistic competitive. Each firm adoptsthe production function as below:yit = θηit kαkit lαlitwhere θ represents the firm’s productivity. αk and αl are the shares of cap-ital and labor in total output, respectively. θ ,αk and αl ∈ (0,1) and η+αk+αl < 1.7 The firm can upgrade their productivity at a fixed cost of cθ , wherec is a positive constant and needs to be paid every period.In the beginning of each period, the firm hires labor through a compet-itive labor market at wage wt . The firm pays a corporate tax τt on its profit.The intermediate goods firm maximize its current profits:pii,t =maxlt ,θt(1− τt)(pi,tPtyi,t −wt li,t −δki,t − cθ),subject to its demand function (3.2). Note that wage is paid in final goodsand normalized by its price level Pt . In the end of each period, the after-tax profit is fully distributed to its owner.8 Solving the profit maximizationproblem, it yields the optimal conditions for labor and technology:wt = ραlY1−ρtyρi,tli,t(3.3)c= ρηY 1−ρtyρi,tθi,t(3.4)7This assumption is to ensure the positive profitability for intermediate firms.8In this model, we assume personal interest income tax is zero, so that it is optimal to dis-tribute the retained earnings to entrepreneur.87Each firm is owned by an entrepreneur. Entrepreneurs are infinitelylived and risk-neutral. Each entrepreneur decides how to allocate his re-sources among purchasing assets and investing in bonds. Their objective isto maximize lifetime dividends specified as:∞∑t=0β tdi,t ,subject to the budget constraint:di,t +bi,t+1+ ki,t+1 = pii,t + ki,t +(1+ rt)bi,t +Ti,twhere β is the discount rate, bi,t is the risk-free bonds with interest rate rt .In this economy, we take interest rate rt to be exogenous and constant.9 Ti,tis the lump sum rebate from the government. At the end of each period, theentrepreneur can choose to invest either in the risk-free bond bi,t+1, changecapital to ki,t+1, or pay himself dividend of di,t . Solving the dividends max-imization problem, it yields the break-even condition between saving inbond and investing in capital:rt+1 = (1− τt+1)(ραkY1−ρt+1yρ−1i,t+1ki,t+1−δ)(3.5)It suggests that the entrepreneur will continue investing in production tillits marginal return equals to the opportunity cost of capital.To show how investment is affected by other variables, we use the sym-metry of intermediate goods to substituteYt+1 = n1ρ yi,t+1 into equation (3.5),and re-arrange it to obtain the following:ln(ki,t+1yi,t+1)= ln(1− τrt+1+δ (1− τ))+1−ρρln(nt+1)+ ln(ραk).This equation shows that capital intensity increases as tax rate declines andthe number of firm increases. A tax rate cut leads to higher investment as9To ensure the existence of steady state, we assume that (1+ rt)β=1.88it lowers the marginal cost of capital. How in equilibrium the number offirms affects capital intensity is, however, less straightforward. The numberof firms captures the demand of good i relative to aggregate output Yt . Asthe number of firms goes up, the aggregate output increases and drives upthe demand for intermediate goods i. Notice that the final goods productionfunction features constant elasticity of substitution between intermediateinputs. The larger ρ is, the stronger the substitution effect is, and the smallereffect of aggregate output on good i.Similarly, for technology, we obtain the following expression:ln(θi,tyi,t)=1−ρρln(nt)+ ln(ρη)− ln(c).It shows that increasing the number of firms, resulting in increase in ag-gregate output, will encourage firms to upgrade their productivity to meetincreasing demand.Firm EntryThis economy consists of identical households with a mass of L. Amongthem, there are n units of entrepreneurs and L− n units of wage workers.Each worker provides one unit of fixed labor to earn wage. In the end ofeach period, each agent can choose to become a worker or an entrepreneur.Once becoming entrepreneur, he needs to give up the wage earnings as en-trepreneurs no longer provide labor.The free-entry condition implies that agents will enter the intermediategoods market so long as the entrepreneurial earning is larger than the totalof wage and earnings from savings:pit = wt + ktrt (3.6)The left-hand side of equation (3.6) shows the earnings of the individual ifhe chooses to become an entrepreneur, and the right-hand side indicates theopportunity cost that entrepreneur gives up: wage and interest earnings insavings.893.3.2 EquilibriumTo solve for the steady state, we need to find the set of (n,ki, li,wi,θi) that sat-isfies the optimal conditions (3.3)-(3.6). Combining with the condition forlabor input (3.3) and the no-arbitrary condition (3.6), the free-entry condi-tion can be shown as follows:w= pi− rkραln1−ρρyili= (1− τ)(1−ρ(η+αl+αk))n1−ρρ yi, (3.7)where the right-hand side of the equation shows the profit of running thefirm and the left-hand side shows the opportunity cost of labor and interestearnings.Given the number of labor supply L - n, in the equilibrium, each firmemploys labor input l = L−nn . Plugging it into equation (3.7), we obtain:l =Ln−1= ραl(1− τ)(1−ρ(η+αk+αl)) (3.8)This equation pins down the optimal number of firms by equating the gainsfrom entrepreneurship to labor income. Its numerator captures the laborshare of the production, whereas its denominator represents the entrepreneurshare.r = (1− τ)(ραkn1−ρρ θηkαk−1lαl −δ)(3.9)c= ρηn1−ρρ θη−1kαk lαl (3.10)Together with the optimal conditions for capital and productivity, the abovethree equations shows the steady state level for capital k, productivity θ , andthe number of firms n.3.3.3 Policy ShockSuppose that an unexpected reform reduces the corporate income tax rateτ permanently. The direct effect of the tax cut is to increase the marginal90return to capital and entrepreneur earnings. As a result, the tax cut willencourage investment. The increase in entrepreneur earnings encouragesworkers to leave labor force and become entrepreneurs. As more individu-als become firm owners, it increases the relative demand for each interme-diate goods, which further encourages investment and technology upgrade.That is, a reduction in the corporate income tax rate τ implies:1) The equilibrium number of firms will increase.2) The capital to output ratio ky and productivity to output ratio θy will increase.As the number of firms increases, the amount of labor declines and thecost of labor increases. As labor becomes more expensive, it increases thecost of production and hence, has negative feedback effect on the the pro-duction of the intermediate goods. How capital and productivity change inthe new steady state depends on whether the feedback effect dominates theprimary effect of the tax cut. In equilibrium, we conclude that:3) Provided that (r+δ )ρ(1−αl−η)> δ (1−η), the optimal level of capitalincreases, that is, dk/dτ < 0.4) Provided that rαk > (r+ δ )αl , the optimal level of productivity increases,that is, dθ/dτ < 0.3.4 Empirical Strategy3.4.1 Baseline EstimationsWe adopt a standard differences-in-differences (DID) strategy to identifythe effect of the preferential corporate income tax scheme for SMPEs onfirms’ performance. Since our data covers 2010-2016, we focus on identi-fying the impact of the tax rate changes on qualified firms in 2012, 2014 and2015. We cannot identify the effects of the 2010 and 2016 rate cuts since welack data to conduct the DID estimations. More specifically, we estimate thefollowing equation for each policy reform cohort c ∈ {2012,2014,2015}:Yi,t = βDIDc Ti,c ∗Postc+Xi,tγ+ϕi+ϕs,t + εi,t , (3.11)91whereYi,t is the outcome variable we are interested in, including firms’ salesgrowth, investment rate and total factor productivity. Ti,c is the indicator ofbeing treated by the policy reform in year c, and Postc is an indicator forpost-treatment years:Postc ={1 if t ≥ c,0 otherwise.We define “treatment” in our study by first introducing a related term,“exposure to a rate cut”. A firm is exposed to a rate cut in year c if: 1) itstaxable income falls between the old and the new qualifying thresholds inthe policy change year c; and 2) it experienced a reduction in the income taxrate in year c relative to the prior year.10 According to this definition, a firmis not exposed to a rate-cut policy in year c if its tax rate does not change,even if its taxable income falls into the new eligibility range. For example,this excludes firms whose taxable income grew from below 30k to between30k and 60k in 2012. For these firms, even if they satisfy criterion 1), therewas no change in their tax rate and hence, no change in the tax componentof the user cost of capital.11By our definition, it is possible for a firm to be exposed to multiple rate-cut policies in 2012, 2014 and 2015. For example, a firm whose tax incomewas in the range 30k-60k during 2010-2012 was exposed to 2012 policy re-form. If its taxable income rose to 70k in 2013 (rendering it ineligible in2013) and then to 80k in 2014, this firm would be exposed to the 2014 ratereduction, too. To identify the effect of each policy change separately, wethus define a firm as being ”treated” by the policy change in year c if it wasexposed to the policy change only in year c.12 In this way, changes in the10By limiting the pool of treated firms to ”exposure” so defined, we capture firms that cometo enjoy a rate reduction as the result of a policy change, and not as the result of incomechange under the same set of policies.11Note that these firms may still experience some shocks if they expect to move to a higher taxbracket, make investment plans accordingly, and do not anticipate the tax rate cut.12Thus we at once (i) assign firms to cohorts according to when they experienced their firstexposure to a rate cut, and (ii) exclude the small number of firms that, by virtue of fluctua-tion in reported income, become exposed for a second time. Treated firms so defined may92post-treatment performance of these firms should be attributed to the effectof the policy change in year c, rather than to the average effect of multiplepolicy changes.We exclude firms with annual total assets above 30 million yuan from thetreatment groups as these firms should be categorized as large firms. In theend, we obtain 1,116 firms that were exposed to the 2012 policy change, 138of which experienced exposure in later policy reforms; 960 firms were ex-posed to the 2014 policy reform, 10 of which experienced exposure in laterpolicy reforms; and 1,751 firms were exposed to the 2015 policy change,none of which were exposed to later policy reforms. Table 3.1 providessummary statistics for each treatment cohort, using both the full panel anda smaller balanced panel.We use firms that always paid income tax at the rate of 25% (that is,the “large” firms) in our sample to form the control group. Note that thesefirms were not exposed to any policy reform throughout the sample periodas their taxable income was always above the affected ranges and they neverexperienced any reduction in income tax rate. In principle, any firm facing aconstant income tax rate throughout the sample years can serve as a poten-tial control. That gives us an alternative choice for the control group: firmsthat always paid income tax at the rate of 10%. However, these “always-micro-firms” were exposed to the 2010 policy change.13 Nor can we usethe always ”medium-sized” firms as the control group since their tax ratedropped from 20% to 10% in 2016.One challenge to our identification strategy is that treated firms may ma-nipulate taxable income to be just below the qualifying threshold for SMPEs,which would lead to endogenous treatment status. In the left panels of Fig-ure 3.3, we plot the distribution of the taxable income (in logs) across firmsfor each cohort, along with the lower and upper thresholds of each treat-ment. There, we observe a prominent concentration of firms around thequalifying thresholds for all three cohorts. These bunching patterns sug-still benefit from the new policies raising eligibility thresholds.13One might also be concerned that this always-micro group comprises disproportionately offirms characterized by low productivity.93gest that some firms indeed manipulated their taxable income to qualify.14The other two qualifying criteria, namely total assets and employees, ap-pear to be more difficult to manipulate than taxable income. For example,we plot the distributions of total assets (in logs) across firms for the threereform years in Figure A.2, all of which appear to be smooth without ob-vious bunching. To address this problem of the non-random assignmentof treatment, we include in estimations a set of observable firm-level char-acteristics Xi,t15, as well as both firm-fixed effects (ϕi) and year-sector-fixedeffects (ϕs,t), to control for selection based on time-varying observables andunobservables at the firm- and year-sector level. As a further check, in ourbenchmark analysis, we drop firms that are within the upper and lower1.5% range of the upper taxable income thresholds of each policy reform asshown in the right panels of Figure 3.1.Under the assumption that, conditioning on the covariates and the fixedeffects included in equation (3.11), the error term is uncorrelated with thetreatment dummy, the conditional independent assumption (CIA) holds.βDIDc then captures the effects of the income tax rate cut on firms’ perfor-mance.3.4.2 Matched Difference-in-differences EstimationsThe standard DID approach we adopt so far introduces covariates linearlyas in equation (3.11) and this is subject to potential mis-specification prob-lems (Abadie, 2005). Recently, a growing body of literature has combinedthe method of matching with traditional DID to identify the causal effectsof non-random policy shocks (Arnold and Javorcik, 2005; De Loecker, 2007;Go¨rg et al., 2008; Volpe Martincus and Carballo, 2008). The basic idea is topair each treated firm of each reform cohort with one similar non-treatedfirm based on observed pre-treatment characteristics. The treatment ef-fects are then identified by comparing performance measures between the14These patterns are documented in more details by Hicks et al. (2018).15These covariates include: sales revenue, real capital stock, employment at time of registra-tion, costs of production in logs, and profitability, effective income tax rate and a dummy forloss-making.94treated and the matched non-treated ones, before and after each policy re-form. The most important advantage of matching is that it relaxes the lin-earity assumption imposed by the standard DID when estimating the condi-tional expectations of the outcome variables. Therefore, we perform a seriesof analysis using matched samples. Specifically, we apply the propensity-score matching (PSM) method as suggested by Rosenbaum and Rubin (1983),and estimate the following probit model for each reform cohort c∈{2012,2014,2015}:PSi,c = Pr{Ti,c = 1}=Φ{h(Zi,2011, ...,Zi,c−1),pii,2010, ...,pii,c−1}, (3.12)whereΦ(.) is the normal cumulative distribution function, and h(.) is a func-tion of a set of covariates and the outcome variables in the years precedingeach policy change, represented by Zi,t . The purpose of matching is to en-sure the common trend assumption holds for the treated and control firmsof each reform cohort in the pre-treatment period. Therefore, we matchfirms by their log-changes in the covariates and the outcome variables dur-ing the pre-treatment period. Explicitly:Zi,t = (logSalesi,tSalesi,t−1, logWagesi,tWagesi,t−1, logFixed asset i,tFixed asset i,t−1, logTFPi,tTFPi,t−1). (3.13)In addition, we match firms by their annual profitability pii,t , defined as theratio of total profit to sales revenue, prior to each treatment, as it has beenacknowledged by the literature as an important determinant of investment.We also include a set of dummies indicating firms’ industries and owner-ship types while estimating the probit function. The predicted probabilities(PSi,t) are then assigned to firms as their propensity scores.Next, we match each treated firm with its “nearest-untreated neighbor”firm without replacement based on their propensity scores. These firmsthat are matched to the treated firm then serve as the counterfactual in thePSM-DID analysis. Formally, the PSM-DID estimator is given by:95βMDIDc =1Nc∑j∈C[∆Y tj −∆Y oj], (3.14)where C is the set of matched firm-pairs of the reform cohort c; Nc is thethe number of matched firm-pairs of cohort c; ∆Y tj (∆Y oj ) is the before-afterchange in the outcome variables of the treated (control) firm of pair j.3.5 Data3.5.1 Confidential Corporate Tax ReturnsWe use confidential administrative tax return data and the firm registrationdata from one large and relatively prosperous Chinese province to analyzethe early impact of the tax cuts on newly qualified SMPEs. The de-identifiedtax returns cover the period 2010-2016, and are matched with informationfrom firms’ tax registration records, income statements and balance sheets.Several features distinguish our data from those used by other researchersstudying Chinese firms, notably the Annual Survey of Industrial Firms col-lected by the National Bureau of Statistics (often labeled as the ASIF).16 Un-like the ASIF data that consists of above-scale firms, our data covers firmsof all sizes. This is an important advantage of the tax returns, which allowsus to examine the impact of tax rate changes o SMPE firms. Second, thefirm registration data covers information such as the date of establishment,which allows us to examine the effect of tax cut on firm entry. Third, the taxreturns cover more recent years than the ASIF, which allows us to analyzethe impact of the tax cut on small firms during its early implementations.For regression analyses, we require firms to report necessary financialinformation, such as taxable income, total assets, wages and fixed assets.We end up with a much smaller regression sample due to missing observa-tions. Our definitions for the treatment and control groups further cut thesample size. Tables 2-4 provides summary statistics in terms of key vari-ables for the three treatment and control groups, respectively. For each pol-16See Brandt (2014) for detailed discussions of ASIF.96icy reform, we provide the summary statistics for the treatment and controlgroups both before and after the policy reform. Unsurprisingly, on aver-age, the treated firms tend to be smaller than their counterparts in terms oftotal assets, sales and total wage bills. Treated firms also tend to be less pro-ductive before treatment. The 2012 and 2014 treated groups exhibit muchlower investment rate than their counterparts (before matching), althoughthe 2015 treated group had similar investment rate to its control group be-fore the policy reform.3.5.2 Measuring Total Factor Productivity and Firm-levelInvestment RateTo analyze how the tax reforms affected newly qualified SMPEs’ productiv-ity, we first estimate a production function for each 2-digit Chinese IndustryClassification (CIC) sector, and calculate the firm-level total factor produc-tivity utilizing the estimates for the production function parameters. Wedocument the construction of the main variables used in our productivityestimations and more details of the estimation procedures in Appendix A.Our proxy for the nominal investment rate is the annual change in firm-level fixed assets, evaluated at the original purchasing price. As we don’tobserve capital expenditures in the tax return, this measure should approx-imate the true investment rate well if asset disposal is infrequent and smallin magnitude. Alternatively, we observe fixed assets net of accounting de-preciation from firms’ balance sheets. Presumably, changes in the net-of-depreciation fixed assets would better account for the effect of asset dis-posal. However, it would suffer from possibly larger measurement errorsdue to the inclusion of accounting depreciation.3.5.3 Property of MatchingTable 3.1 summarises the number of firms using both the full sample anda balanced sample that consists of firms operating consecutively from 2010through 2016, for each of the treatment cohorts. These numbers suggestthat, more than 85% of the firms exposed to the 2012 policy reform also qual-97ified as being treated in 2012 (that is, they were not exposed to subsequentpolicy changes). 98% of firms exposed to the 2014 policy change were notexposed to later changes and hence, were regared as being treated in 2014.None of the firms exposed to the 2015 reform were exposed to the 2016 pol-icy change and thus, they were all treated in 2015. We use the same set ofcontrol firms in the baseline DID estimations for all three cohorts, which in-cludes all large firms in our sample. Three sub-sets of these firms are usedas control groups in the PSM-DID regressions, each including large firmsmatched to treated firms of each treatment cohort. As a result, in the PSM-DID analysis, the number of control firms equals to the number of treatedfirms.To assess the quality of matching, we compare the treated and controlgroups in terms of the growth rates of TFP, investment rate, sales, and wagesin the year before treatment in Table 3.5. It shows that before matching, thecontrol group experienced significantly higher growth rate in all four co-variates before each cohort of policy reform. This is not surprising giventhat the control group is consisted of large firms. After matching, the treatedand control groups become much more comparable. This provides a ratio-nale for matching the two groups, so that the parallel trend assumption ismore likely to hold.3.6 Results3.6.1 Effects on Output Growth, Investment and ProductivityPooled EstimationsWe report the estimates of βDIDc in Equation (3.11) in Tables 3.6-3.8, wherethe dependent variable is the firm-level output growth, investment rate, andtotal factor productivity (in logs), respectively. For each outcome variable,we first estimate using the unbalanced panel and results are reported incolumns 1 and 2 of the relevant table, and alternatively using a balancedpanel in columns 3 and 4. In each table, we run the standard DID regres-98sions in columns 1 and 3. In columns 2 and 4, we instead run the PSM-DIDestimations. In all columns, we control for firm-level and sector-year fixedeffects, and a set of covariates that may affect the outcome variable.Table 3.6 shows that the tax policy reforms significantly increased out-put growth of treated firms relative to the control group. This result is ratherrobust to different samples and estimation methods we use. Based on col-umn 4 of Table 3.6, firms in the 2012 treatment cohort enjoyed a relative out-put growth by around 16% during 2012-2016. The 2014 and 2015 treatmentcohorts also experienced around 10% and 6.5% relative increase in outputgrowth during the shorter post-reform period we observe.Note that the point estimate for the estimated treatment effect on outputsales is larger for the 2012 treatment cohort than the later two cohorts. Thismay arise for two reasons. First, the 2012 treated group consists of smallerfirms which tend to respond more to tax incentives, assuming they are morefinancially constrained. Second, we have longer post-treatment period forthe 2012 reform and hence, the treated firms would have more time to re-spond.Our model predicts that a lower corporate income tax rate should alsolead to higher investment, as the cost of capital is lowered. Since we donot have accurate measure for capital expenditures in the data, we use theinvestment rate, proxied by the change in the log level of fixed asset capi-tal stocks from year t − 1 to year t, as the dependant variable in Equation(3.11). We report the estimation results in Table 3.7. There, we find posi-tive and significant effects of the policy reforms on the firm-level investmentrate for all three cohorts. The estimated treatment effects are qualitativelysimilar regardless of the samples or estimation methods we use. Based oncolumn 4 where we conduct the PSM-DID estimation on a balanced sample,the 2012 treated firms increased their investment by around 14 percentagepoints, which is significant at the 5 percent level. For the latter two treat-ment cohorts, the estimated increase in investment rate relative to their con-trol groups is lower, around 9-11 percentage points. The mean investmentrate before treatment is 7.5%, 16.7% and 20.6% for the three treatment co-horts. This implies an increase in investment rate by 186%, 54% and 56%99percent, respectively.We can convert these estimated effects into the elasticity of the invest-ment rate with respect to changes in the tax component of the user cost ofcapital. Assume the risk-adjusted interest rate is 7% and the present valueof tax deduction for 1 dollar of newly acquired fixed assets is 0.75217, cut-ting the statutory corporate tax rate from 20% to 10% implies a drop in thetax component of the user cost by around 3.2%. Thus, the implied elasticityof the investment rate with respect to the tax component of the user cost isaround 58, 17, and 18 for the 2012, 2014 and 2015 treatment cohorts, respec-tively. These are rather substantial elasticities relative to the literature thatfocus on normal-sized firms in developed countries, for example, between8-9 in Maffini et al. (2019) and 6.5 in Orhn (2018). However, ours is com-parable to the Pham (2019) who studies a temporary corporate income taxrate cut for eligible small firms in Vietnam.18According to our model, a tax cut would lead to higher productivity asfirms would find it less costly to upgrade technology, if output and after-tax profit increase while the cost of upgrading is fixed. Consistent withthis conjecture, in Table 3.8 we obtain robust evidence that treated firmsof the three policy reform cohorts experienced significant increase in theirTFP, relative to the control group. The magnitude of such increase is alsolarge: around 4% for the 2012 treated cohort, and around 3.5% for the 2014and 2015 treated cohorts. The results are strikingly similar whether we usebalanced panel or not, and whether we use the standard DID or the PSM-DID estimation methods. These effects, however, appear to be smaller thanthose found by Mao and Liu (2019). In that paper, the authors find thatChina’s 2009 transition from a production-based VAT system to a consump-tion based one increased firms’ productivity by around 8.9 percent.17This is the present value of deduction for the 10-year asset class that consists of productionequipment under the regular, not accelerated, depreciation schedule. See Cui et al. (2020)18Similar to Pham (2019), the corporate tax cut targets small firms and the lifting of the qual-ifying threshold was not announced as permanent.100Event StudiesIn addition to examining the pooled treatment effects of the tax policy re-forms, we conduct a series of “event studies” to investigate the effects ofeach rate cut on firms’ performance year by year. Specifically, we estimatethe following specification for each cohort c using both standard DID andPSM-DID:yi,t = ∑t 6=c−1βEventc,t Ti,c ∗Yeart +ϕi+ϕs,t + εi,t . (3.15)The dependent variables are the sales growth rate, the investment rate andthe logs of firm-level productivity. In Equation 3.15, the parameters βEventc,tmeasures the differences in firm performance between the treated and con-trol group annually. We use the year preceding each rate cut as the baseyear when estimating Equation 3.15.We plot the estimated ˆβEventc,t along with their 95% confidence intervalsfor each performance measure and each cohort in Figures 3.5-3.7. There isno significant difference in terms of each of these performance measuresduring the pre-rate-cut years for all three treatment cohorts. This indicatesthat the common trend assumption is satisfied in the data after matching.More, these figures show that treated firms experienced significant increasein all three performance measures relative to the control firms since thetreatment year. Note that, there is also no evidence that the positive effectof the 2012 policy reform faded even by 2016, especially on output growthand TFP. This suggests that the effect of the tax reforms is not temporary.Figures 3.6 indicates, nevertheless, that the effect on investment rate wasless persistent over time.3.6.2 Substitution between Capital and LaborOne interesting issue is whether the corporate income tax rate cut gener-ates different demands for capital and labor. According to our model, anindividual is induced by the tax rate cut to provide capital by becoming anentrepreneur, which reduces the amount of labor supply in the economy.101This, in equilibrium, would drive up wage while reducing the cost of cap-ital. If this is the case, we may observe substitution between capital andlabor.To test this hypothesis, we use the ratio of fixed assets to sales, the ratioof wages to sales, and the ratio of fixed assets to total wages (all in logs) asthe dependent variable in the DID estimations in Table 3.6. In panel A, wefind a significant increase in the ratio of fixed assets to total sales, for all threereform cohorts. This suggests that treated firms increased capital intensityfollowing the tax reforms. The estimated magnitude for such increases isalso large, between 10-13% depending on the specification.In the first two columns of panel B, we also observe an increase in theratio of wages to total sales for the 2012 treatment cohort, but the magni-tude is smaller than that in panel A. Thus, the 2012 treated firms appear toincrease both capital intensity and labor intensity, as their output grew. Forthe 2014 and 2015 treatment cohorts, however, the tax reforms had eitherinsignificant or negative impact on firms’ labor intensity. In panel C, we ex-amine the ratio of fixed assets to wages. There, we observe an increase inthis capital-labor ratio for all three treatment cohorts, which is particularlystrong for the latter two cohorts.Taking together, these results indicate possible substitution between cap-ital and labor, especially for the 2014 and 2015 treatment cohorts. Why is the2012 treatment cohort different? One possibility is that the 2012 treatmentcohort experienced stronger output growth post reform, which may causefirms to increase both capital and labor input. For the latter two cohorts, incontrast, the relatively smaller increase in output may strengthen the substi-tution between capital and labor. Another possibility is that the substitutioneffect is likely to dominate in the short run, especially if the adjustment costis higher for labor. When firms have more time to grow, they may be moreable to adjust both capital and labor.1023.6.3 Firm EntryFinally, we investigate the effects of the tax cuts on firms’ market entry de-cisions. Our general equilibrium model predicts that following a corporateincome tax rate cut, the return to capital increases, inducing more individ-uals to become entrepreneurs. To test this hypothesis, we use the firm reg-istration data for the universe of firms in the province of our study, whichcontains comprehensive information on firms’ establishment dates, size andthe numbers of employees when they establish. The registration data we ob-tain is a snapshot of all firms in 2017 in the province we examine and hence,it covers all firms established by then. We use the registration data insteadof the tax return because the latter excludes firms that report directly to thefederal-level tax administrator. As a result, we observe the establishmentdates of all firms only in the registration data, which is essential for the en-try analyses.One caveat of using the registration data, however, is that we do notobserve firms’ taxable income upon registration unless they also appear inthe tax return. Consequently, we cannot identify whether a firm in the reg-istration data is an SMPE or not. However, firms in the registration dataare categorized as “micro”, “small”, “median”, “large” or “unidentified”based on their size status upon registration. While the definitions for thesesize status differ from that used for tax purposes, there should be consid-erable overlap between the two classifications. Indeed, as Table 3.10 shows,the average and median levels of employment and registered capital uponregistration for “micro” firms are both much lower than the SMPE thresh-olds. Therefore, we utilize this information to measure the annual entry of“micro” firms and “small” firms between 2010 and 2016.Figure 3.8 illustrates the number of newly established firms clusteredby month for the four different types of firms we observe in the registrationdata. Note that the series for “micro” firms are much more volatile thanthat for other types of firms and generally speaking, there are substantiallymore new “micro” firms entering the market than other types. We also ob-serve an increase in the number of newly established “micro firms” during1032010-2012. There appears to be a drop in the number of new “micro” firmsin since 2013, but its number jumped again in 2014. Note that we do not ob-serve “unidentified” firms before 2013 and these firms appear to have ratherlow levels of employees and registered capital when they establish. Thus, itis possible that many of these “unidentified” firms are “micro” firms but didnot reported their status in the registration data. For other types of firms,we do not observe similar jumps during 2010-2016.To achieve identification, note that the impact of tax cuts on firm entriesis likely to be more profound in industries that are more financially con-strained. Tax cut induces firm entry because it increases after-tax cash flowto the entrepreneur and hence, can shorten the time needed to achieve theoptimal scale of operation. This effect is likely to be stronger for industrieswhere firms have to rely on more costly external finance. We therefore iden-tify the effect of the tax rate cut on firms’ entry decisions using the followingspecification:MC Sharer,s,t = δ1Ext.Fin.s×Postc+δ2ShareEr,s,t +ϕr,s+ϕr,t + εr,s,t . (3.16)The dependent variable MC Sharer,s,t is the share of newly established “mi-cro” firms in all firms in region r, industry s and year t. Ext.Fin.s is the indi-cator for financial constraint for industry s. We calculate the industry-levelexternal finance dependence using the Annual Survey of Industrial Firms(ASIF) conducted by the National Bureau of Statistics (NBS) of China. TheASIF data covers comprehensive financial statement information for all above-scale firms in the manufacturing sector.19 External finance dependence isdefined as the fraction of capital expenditures not financed by operatingcash flows. The ASIF data does not report firm-level capital expenditures,and we calculate it as the sum of increase in firms’ long-term investment,fixed assets and intangible assets as well as the firm’s current year capitaldepreciation, as in Feng, Li and Swenson (2012). Following Rajan and Zin-gales (1998), operating cash flow is defined as the sum of cash flow, plusinventory reductions, reductions in receivables and increases in the firm’s19“Above-scale” firms refer to firms with annual business income greater than 5 million RMB.104payables. We use the median level of this ratio for firms in each 2-digit CIC(Chinese Industry Classification) sector in 2009 to indicate the sector’s ex-ternal financing dependence.20The coefficient associated with the interaction term, Ext.Fin.s × Postc,captures the effects of the tax rate cut on “micro” firms’ entry decisions. Wecontrol for the ratio of all types of new entrants to the total number of firms,ShareErs,t , to control for the underlying common trend driving firm entries.We include the region-industry-fixed effects ϕr,s and the region-year-fixedeffects ϕr,t , to control for any unobserved shocks that are common to all re-gions and sectors over time, across space and industries. There may also beconcerns about simultaneous unobserved policy or macro economic shocksduring our sample period that might also affect firm entries. To address thisissue, we interact the industry-level share of skilled labor in total labor forceand capital intensity with the post-reform dummies in all regressions if theunobserved shocks work through the traditional sources of comparative ad-vantages.21Results in the first three columns of Table 3.11 indicate that raising thequalifying thresholds for a lower tax rate led more “micro” firms to enterthe market. For example, the share of newly established “micro” firms ison average 1.31% higher in an industry whose external finance dependencemeasure is one standard deviation higher than the mean after the 2014 ratereduction. The effect is larger for the 2014 and 2015 policy reforms. This isnot surprising as these latter reforms increased the SMPE qualifying thresh-old further and consequently, should induce more firm entry as it possiblyextends the period when a firm can enjoy the lower tax rate as it grows.20This measure of industrial financial vulnerability exhibits a stable pattern over time. In ro-bustness checks, we average this index over the years 2005-2009, and use it as the measurefor external finance dependence instead. All results remain qualitatively.21The industry-level factor intensities are calculated using the ASIF data set. Industry-levelskilled labor intensity is defined as the median share of workers with a college degree orabove in total labor force employed across firms within a 2-digit CIC sector. We calculatethese variables using the 2004 data, as information on employment by education levels isonly available for that year. Industry-level capital intensity is defined as the median ratio ofreal capital stock to total labor input across firms within a 2-digit CIC industry. We calculatethis measure using the 2009 ASIF data.105As the definition of “micro” firms differs from that of SMPEs, we usethe share of newly established “micro” and “small” firms in each region-industry pair as the alternative dependant variable. In columns (4) to (6),we find qualitatively equivalent results to those in the previous columns.However, the point estimates in columns 4-6 are smaller. This indicates thatthe reforms mainly induced “micro” firms to enter, which is unsurprisingas newly qualified SMPEs most likely register as “micro” firms.3.6.4 Stay Small or Grow?An important question is whether tax incentives targeting small firms wouldkeep them being small, rather than enhancing growth. For example, firmsmay lack incentives to grow large if that means loss of tax benefits. Thisis especially the case as the lower tax rate for SMPEs creates notches in thecorporate income tax schedule. Or, firms may manipulate their size statusto “look small”, if the benefit of doing so outweigh the costs. We investigatethis issue in this section.Table 3.12 shows that nearly half of the 2012 treated firms in our regres-sion sample increased their taxable income above the 2012 qualifying tax-able income threshold of 60K RMB by 2016. One concern is that our re-gression sample consists of firms with necessary financial information andhence, there may be potential selection issue. If we use the larger sample offirms that report taxable income in the tax returns, still close to 40% of firmsthat SMPEs in 2012 grew above 60K RMB by the end of our sample period.For both samples, a bulk of firms grew above the 60K threshold in 2013, justone year after the policy reform. Alternatively, we use the yearly qualifyingthreshold to examine whether a treated firm grew above this threshold ornot. Around 25% of the 2012 treated cohort grew above the SMPE thresholdby 2016, no matter which sample we examine. This is consistent with thehigher sales growth, investment rate and TFP growth we find among the2012 treatment cohort.There is also large percentages of firms grew above the qualifying taxincome threshold by the end of our sample period for the 2014 and 2015106treatment cohorts. However, since we observe shorter post-reform periodfor these latter two cohorts, the percentage of growers is smaller than thatfor the 2012 cohort. Possibly for the same reason, for the 2014 and 2015 treat-ment cohorts, much fewer SMPEs grew to become non-SMPEs by 2016. Thediscrepancy in the results based on the regression sample and the largersample also increases, indicating that growing firms are more likely to re-port financial data and hence enter our regression sample.3.7 ConclusionWe examine in this study how a lower corporate income tax rate targetingsmall- and micro-profit firms in China affects firms’ performance, includ-ing productivity, investment, sales growth and market entry. We utilize thegradual lifting of the qualifying threshold for SMPE firms from 30,000 RMBto 300,000 RMB during 2010-2016 as a natural experiment. We use confiden-tial corporate tax returns from firms located in a large Chinese province toconduct difference-in-differences analyses. We find that lowering the cor-porate tax rate led to substantial increases in newly qualified SMPE firms’total factor productivity, investment and sales growth. There is also evi-dence that firms substitute capital for labor following the tax cut. We findthat lowering the corporate income tax rate also encouraged more entry ofmicro-sized firms in industries facing a high degree of financial constraint.Our study provides an assessment of China’s corporate income tax cutstargeting small firms, an important element of its recent pro-active fiscalpolicy, on affected firms’ performance. Importantly, we show that thesetax incentives are particularly effective in stimulating firm-level investment,and there appears to be substitution between capital and labor, at least inthe short run. More, the tax cuts successfully encouraged firm growth, over-coming the concern that such policies may keep firms small. Findings fromour study contribute to the more general policy discussion about using taxincentives to stimulate small businesses.1073.8 FiguresFigure 3.1: The taxable income ranges of affected firms.Notes: This figure illustrates the tax policy reforms about the qualifying thresh-old in terms of taxable income for small and micro-profit (SMPE) firms during2010-2016.108Figure 3.2: Salience of the tax rate cuts for SMPEsNotes: This figure plots the Baidu search intensity for the key words (all inChinese): qualifying thresholds for SMPEs (blue), preferential corporate in-come tax policies for SMPEs (green), tax filing (yellow), and accelerated de-preciation for fixed assets (red). The period covers from January 1st, 2011 toDecember 31, 2016.109Figure 3.3: The distributions of taxable income around the SMPE qual-ifying thresholdsA. The 2012 cohortlog(30K) log(60K)0.1.2.3Density0 5 10 15 20Log of taxable income - 2012(a) Distributionlog(60K*(1-1.5%)) log(60K*(1+1.5%))05101520Density10.8 10.9 11 11.1 11.2Log of taxable income (50K ~ 70K) - 2012(b) Range of exclusionB. The 2014 cohortlog(60K) log(100K)0.05.1.15.2.25Density0 5 10 15 20Log of taxable income - 2014(c) Distributionlog(110K*(1-1.5%)) log(110K*(1+1.5%))0510152025Density11.4 11.45 11.5 11.55 11.6Log of taxable income (90K ~ 110K) - 2014(d) Range of exclusionC. The 2015 cohortlog(100K) log(200K)0.1.2.3Density0 5 10 15 20Log of taxable income - 2015(e) Distributionlog(200K*(1-3%)) log(200K*(1+3%))01020304050Density12.16 12.18 12.2 12.22 12.24 12.26Log of taxable income (190K ~ 210K) - 2015(f) Range of exclusionNotes: This figure plots the distributions of taxable income around the 2012,2014 and 2015 SMPE qualifying thresholds based on confidential corporate taxreturns. 110Figure 3.4: Average log TFP of the matched and unmatched samples.A. The 2012 cohort(a) Matched sample (b) Unmatched sampleB. The 2014 cohort(c) Matched sample (d) Unmatched sampleC. The 2015 cohort(e) Matched sample (f) Unmatched sample111Figure 3.5: Dynamic effects of the corporate income tax rate cut onsales growthA. The 2012 cohort(a) Standard DID (b) PSM-DIDB. The 2014 cohort(c) Standard DID (d) PSM-DIDC. The 2015 cohort(e) Standard DID (f) PSM-DIDNotes: This figure plots the estimated dynamic effects of the corporate incometax cut on the growth rate of sales for the 2012, 2014 and 2015 cohorts, re-spectively. For each cohort, we plot the estimated coefficients from both thestandard difference-in-differences regressions (left panels) and the matcheddifference-in-differences regressions (right panels). The year preceding the re-form year is used as the base year.112Figure 3.6: Dynamic effects of the corporate income tax rate cut onfirm-level investment rateA. The 2012 cohort(a) Standard DID (b) PSM-DIDB. The 2014 cohort(c) Standard DID (d) PSM-DIDC. The 2015 cohort(e) Standard DID (f) PSM-DIDNotes: This figure plots the estimated dynamic effects of the corporate in-come tax cut on firms’ investment rate for the 2012, 2014 and 2015 cohorts,respectively. For each cohort, we plot the estimated coefficients from both thestandard difference-in-differences regressions (left panels) and the matcheddifference-in-differences regressions (right panels). Investment rate is proxiedby the annual growth rate of fixed assets. The year preceding the reform yearis used as the base year. 113Figure 3.7: The dynamic effects of the corporate income tax cut on firm-level TFPA. The 2012 cohort(a) Standard DID (b) PSM-DIDB. The 2014 cohort(c) Standard DID (d) PSM-DIDC. The 2015 cohort(e) Standard DID (f) PSM-DIDNotes: This figure plots the estimated dynamic effects of the corporate incometax cut on firms’ TFP for the 2012, 2014 and 2015 cohorts, respectively. For eachcohort, we plot the estimated coefficients from both the standard difference-in-differences regressions (left panels) and the matched difference-in-differencesregressions (right panels). The year preceding the reform year is used as thebase year.114Figure 3.8: Time series pattern of firm entriesNotes: This figure plots the number of firms by registration month and by firmtype during 2005 and 2017.115Figure 3.9: Time series pattern of firm entriesNotes: This figure plots the normalized number of firms by registration monthand by firm type during 2005 and 2017.1163.9 TablesTable 3.1: Numbers of the treated and control firms.# of firms exposedto the policy reform # of treated firms # of control firmsUnbalanced panel2012 1,116 978 2,8092014 960 950 2,8682015 1,751 1,751 2,872Balanced panel2012 818 696 2,5082014 684 676 2,5082015 1,304 1,304 2,508Notes: This table reports the number of treated and control firms of each tax policyreform cohort, based on the unmatched regression sample. Firms are “exposed” thepolicy reform in year c if: 1) its taxable income falls between the old and the newqualifying thresholds in the policy change year c; and 2) it experiences a reductionin the corporate income tax rate in year c. Firms are “treated” in year c if they areonly exposed to the policy change in year c.117Table 3.2: Summary statisticsPanel A: The 2012 cohortBefore AfterObs Mean SD. Median Obs. Mean SD. Median1. Treated firmslog(TFP) 1,760 7.37 0.60 7.29 4,524 7.41 0.65 7.36log(Fixed assets) 1,760 14.12 1.18 14.15 4,524 14.28 1.19 14.33log(Sales) 1,760 15.49 0.82 15.45 4,524 15.06 1.10 15.11log(Total assets) 1,760 15.28 0.87 15.28 4,524 15.31 0.93 15.31log(Wages) 1,760 13.04 1.00 13.05 4,524 13.02 1.13 13.05log(Taxable income) 1,760 11.28 0.75 11.14 4,524 8.42 4.34 10.52ETR 1,760 0.21 0.02 0.200 4,524 0.10 0.06 0.10Profitability 1,760 0.02 0.02 0.01 4,524 0.01 0.02 0.004Rinvestment 782 0.08 0.24 0.02 4,524 0.12 0.78 0.004Rsales 782 0.05 0.34 0.02 4,524 -0.06 0.42 -0.082. Control firms - unmatchedlog(TFP) 5,430 7.59 0.63 7.48 13,822 7.64 0.62 7.54log(Fixed assets) 5,430 15.97 1.25 15.93 13,822 16.33 1.18 16.30log(Sales) 5,430 17.48 1.02 17.35 13,822 17.57 1.05 17.44log(Total assets) 5,430 17.26 1.10 17.14 13,822 17.41 1.10 17.31log(Wages) 5,430 14.72 1.06 14.69 13,822 15.12 1.04 15.09log(Taxable income) 5,430 14.19 1.09 13.98 13,822 14.20 1.08 13.99ETR 5,430 0.25 0.01 0.25 13,822 0.25 0.01 0.25Profitability 5,430 0.05 0.05 0.053 13,822 0.04 0.05 0.03Rinvestment 2,621 0.28 1.35 0.09 13,822 0.17 1.68 0.05Rsales 2,621 0.18 0.31 0.13 13,822 0.02 0.27 0.0013. Control firms - matchedlog(TFP) 1,562 7.67 0.68 7.58 3,845 7.72 0.69 7.65log(Fixed assets) 1,562 16.10 1.26 16.06 3,845 16.46 1.18 16.44log(Sales) 1,562 17.51 0.99 17.39 3,845 17.58 1.03 17.47log(Total assets) 1,562 17.31 1.11 17.19 3,845 17.48 1.11 17.34log(Wages) 1,562 14.84 1.06 14.79 3,845 15.20 1.08 15.16log(Taxable income) 1,562 14.23 1.12 14.00 3,845 14.23 1.12 14.02ETR 1,562 0.248 0.02 0.25 3,845 0.25 0.01 0.25Profitability 1,562 0.05 0.05 0.03 3,845 0.05 0.05 0.03Rinvestment 781 0.18 0.84 0.06 3,845 0.10 0.23 0.04Rsales 781 0.07 0.26 0.05 3,845 0.03 0.28 0.01118Table 3.3: Summary statisticsPanel B: The 2014 cohortBefore AfterObs Mean SD. Median Obs. Mean SD. Median1. Treated firmslog(TFP) 3,423 7.39 0.58 7.32 2,824 7.44 0.63 7.38log(Fixed assets) 3,423 14.34 1.12 14.36 2,824 14.54 1.09 14.56log(Sales) 3,423 15.74 0.78 15.70 2,824 15.35 0.94 15.39log(Total assets) 3,423 15.52 0.77 15.54 2,824 15.54 0.77 15.54log(Wages) 3,423 13.34 0.91 13.33 2,824 13.31 1.00 13.33log(Taxable income) 3,423 11.48 1.73 11.59 2,824 9.97 3.52 11.22ETR 3,423 0.20 0.04 0.20 2,824 0.09 0.04 0.100Profitability 3,423 0.02 0.02 0.01 2,824 0.02 0.02 0.01Rinvestment 2,473 0.17 0.74 0.02 2,824 0.17 1.04 0.01Rsales 2,473 0.02 0.34 -0.01 2,824 -0.07 0.35 -0.082. Control firms - unmatchedlog(TFP) 11,005 7.61 0.62 7.51 8,545 7.65 0.63 7.56log(Fixed assets) 11,005 16.07 1.22 16.03 8,545 16.40 1.18 16.37log(Sales) 11,005 17.52 1.03 17.39 8,545 17.54 1.06 17.41log(Total assets) 11,005 17.28 1.11 17.16 8,545 17.44 1.10 17.34log(Wages) 11,005 14.86 1.04 14.83 8,545 15.18 1.05 15.15log(Taxable income) 11,005 14.18 1.07 13.97 8,545 14.19 1.10 13.98ETR 11,005 0.25 0.013 0.250 8,545 0.250 0.01 0.25Profitability 11,005 0.04 0.05 0.03 8,545 0.05 0.05 0.03Rinvestment 8,137 0.24 1.32 0.07 8,545 0.10 0.23 0.04Rsales 8,137 0.08 0.29 0.05 8,545 0.02 0.27 -0.013. Control firms - matchedlog(TFP) 2,748 7.65 0.65 7.56 2,04 7.68 0.68 7.59log(Fixed assets) 2,748 16.15 1.17 16.11 2,043 16.43 1.13 16.40log(Sales) 2,748 17.55 1.02 17.43 2,043 17.49 1.06 17.40log(Total assets) 2,748 17.34 1.11 17.22 2,043 17.48 1.11 17.36log(Wages) 2,748 14.94 1.02 14.93 2,043 15.17 1.09 15.20log(Taxable income) 2,748 14.26 1.17 13.98 2,043 14.16 1.13 13.93ETR 2,748 0.25 0.01 0.250 2,043 0.25 0.01 0.25Profitability 2,748 0.05 0.05 0.03 2,043 0.05 0.05 0.03Rinvestment 2,061 0.15 0.67 0.05 2,043 0.08 0.22 0.03Rsales 2,061 0.02 0.28 -0.002 2,043 0.008 0.29 -0.02119Table 3.4: Summary statisticsPanel C: The 2015 cohortBefore AfterObs Mean SD. Median Obs. Mean SD. Median1. Treated firmslog(TFP) 8,032 7.42 0.55 7.36 3,494 7.52 0.60 7.47log(Fixed assets) 8,032 14.58 1.04 14.64 3,494 14.82 1.00 14.88log(Sales) 8,032 15.99 0.70 15.96 3,494 15.79 0.74 15.76log(Total assets) 8,032 15.76 0.73 15.78 3,494 15.86 0.70 15.87log(Wages) 8,032 13.62 0.84 13.64 3,494 13.66 0.86 13.66log(Taxable income) 8,032 11.99 1.62 12.15 3,494 11.51 2.16 11.94ETR 8,032 0.21 0.04 0.200 3,494 0.11 0.04 0.100Profitability 8,032 0.02 0.02 0.02 3,494 0.02 0.02 0.02Rinvestment 6,281 0.21 1.18 0.04 3,494 0.14 0.77 0.02Rsales 6,281 0.05 0.35 0.01 3,494 -0.05 0.31 -0.072. Control firms - unmatchedlog(TFP) 13,785 7.62 0.62 7.52 5,732 7.66 0.63 7.57log(Fixed assets) 13,785 16.11 1.22 16.08 5,732 16.42 1.18 16.41log(Sales) 13,785 17.53 1.03 17.40 5,732 17.52 1.06 17.38log(Total assets) 13,785 17.30 1.11 17.19 5,732 17.46 1.09 17.35log(Wages) 13,785 14.92 1.04 14.89 5,732 15.19 1.06 15.16log(Taxable income) 13,785 14.19 1.07 13.99 5,732 14.17 1.12 13.95ETR 13,785 0.25 0.01 0.25 5,732 0.25 0.01 0.25Profitability 13,785 0.04 0.05 0.03 5,732 0.05 0.05 0.03Rinvestment 10,913 0.20 0.87 0.06 5,730 0.05 0.13 0.03Rsales 10,913 0.08 0.29 0.04 5,732 0.01 0.27 -0.023. Control firms - matchedlog(TFP) 6,545 7.64 0.61 7.56 2,615 7.66 0.63 7.59log(Fixed assets) 6,545 16.15 1.16 16.13 2,615 16.44 1.11 16.44log(Sales) 6,545 17.56 0.99 17.44 2,615 17.50 1.02 17.38log(Total assets) 6,545 17.33 1.07 17.24 2,615 17.47 1.06 17.37log(Wages) 6,545 14.96 1.00 14.94 2,615 15.16 1.04 15.15log(Taxable income) 6,545 14.21 1.06 14.02 2,615 14.11 1.07 13.90ETR 6,545 0.25 0.01 0.25 2,615 0.25 0.01 0.25Profitability 6,545 0.05 0.05 0.03 2,615 0.05 0.05 0.031Rinvestment 5,236 0.18 0.93 0.05 2,615 0.04 0.12 0.03Rsales 5,236 0.04 0.27 0.01 2,615 -0.01 0.28 -0.03120Table 3.5: Matching properties: means of matching covariates before and aftermatchingUnmatched sample Matched sampleTreated Control Diff. Treated Control Diff.Panel A: The 2012 cohort∆log(TFP) 0.012 0.025 -0.013*** 0.012 0.012 0.000(-0.004) (0.005)∆log(Fixed assets) 0.093 0.147 -0.054*** 0.093 0.118 -0.023*(-0.016) (-0.015)∆log(Sales) 0.013 0.131 -0.119*** 0.013 0.033 -0.020(-0.011) (-0.015)∆log(Wages) 0.081 0.210 -0.129*** 0.081 0.104 -0.024(-0.017) (-0.017)Panel B: The 2014 cohort∆log(TFP) -0.005 0.006 -0.012*** -0.005 -0.006 0.001*(-0.004) (0.004)∆log(Fixed assets) 0.070 0.109 -0.039*** 0.070 0.075 -0.006(-0.012) (-0.014)∆log(Sales) -0.054 0.026 -0.080*** -0.054 -0.039 -0.015(-0.012) (-0.016)∆log(Wages) 0.030 0.117 -0.087*** 0.030 0.061 -0.031*(-0.015) (-0.018)Panel C: The 2015 cohort∆log(TFP) 0.013 0.030 -0.017*** 0.013 0.023 -0.010*(-0.005) (-0.006)∆log(Fixed assets) 0.091 0.112 -0.021** 0.091 0.091 -0.006(-0.010) (-0.012)∆log(Sales) -0.037 0.012 -0.049*** -0.037 -0.030 -0.007(-0.009) (-0.011)∆log(Wages) 0.076 0.128 -0.051*** 0.076 0.095 -0.018(-0.011) (-0.013)Notes: This table conducts the balance tests of the covariates used for matching thetreated and control groups. All variables are measured in the preceding year of each pol-icy reform. T-statistics are reported in the parentheses. ***p< 0.01, **p< 0.05, *p< 0.1.121Table 3.6: The impact of tax cuts on sales growthDependent variable: firm-level growth rate in salesPanel A: The 2012 cohortUnbalanced sample Balanced sampleDID PSM-DID DID PSM-DID(1) (2) (3) (4)Ti,2012×Post2012 0.121*** 0.171*** 0.110*** 0.164***(0.033) (0.043) (0.035) (0.044)Observations 21,856 8,348 19,002 7,584R-squared 0.224 0.224 0.207 0.212Panel B: The 2014 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2014×Post2014 0.155*** 0.109** 0.163*** 0.109***(0.039) (0.043) (0.042) (0.042)Observations 21,791 7,656 18,852 7,500R-squared 0.238 0.218 0.227 0.217Panel C: The 2015 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2015×Post2015 0.085*** 0.067*** 0.093*** 0.065***(0.021) (0.021) (0.020) (0.021)Observations 26,087 14,888 22,470 14,808R-squared 0.230 0.208 0.218 0.207Controls Effective tax rate & Indicator for loss-making firms.Fixed-effects Firm FE, Sector-Year FENotes: This table investigates the impact of the corporate income tax ratereductions on the growth rate of sales. Results from standard DID regres-sions are reported in columns (1) and (3), and the results from PSM-DIDregressions are presented in columns (2) and (4). Robust standard errorsare clustered on the sector-year level. ***p< 0.01, **p< 0.05, *p< 0.1.122Table 3.7: The impact of tax cuts on investment rateDependent variable: firm-level nominal investment ratePanel A: The 2012 cohortUnbalanced sample Balanced sampleDID PSM-DID DID PSM-DID(1) (2) (3) (4)Ti,2012×Post2012 0.136*** 0.138** 0.145*** 0.143**(0.043) (0.056) (0.045) (0.058)Observations 21,856 8,348 19,002 7,584R-squared 0.176 0.184 0.170 0.184Panel B: The 2014 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2014×Post2014 0.128** 0.088** 0.159*** 0.093**(0.058) (0.044) (0.0505) (0.045)Observations 21,790 7,656 18,852 7,500R-squared 0.193 0.223 0.185 0.223Panel C: The 2015 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2015×Post2015 0.122** 0.115* 0.109* 0.116*(0.052) (0.062) (0.059) (0.062)Observations 26,081 14,888 22,470 14,808R-squared 0.185 0.176 0.177 0.176Controls Growth rates in sales and wage bills;Effective tax rate & Indicator for loss-making firms.Fixed-effects Firm FE, Sector-Year FENotes: This table investigates the impact of the corporate income tax ratereductions on firm-level nominal investment rate. Results from standardDID regressions are reported in columns (1) and (3), and the results fromPSM-DID regressions are presented in columns (2) and (4). Robust stan-dard errors are clustered on the sector-year level. ***p < 0.01, **p < 0.05,*p< 0.1. 123Table 3.8: The impact of tax cuts on productivityDependent variable: firm-level TFP in logsPanel A: The 2012 cohortUnbalanced sample Balanced sampleDID PSM-DID DID PSM-DID(1) (2) (3) (4)Ti,2012×Post2012 0.039*** 0.043*** 0.048*** 0.042***(0.011) (0.012) (0.012) (0.013)Observations 25,760 9,818 22,169 8,848R-squared 0.968 0.959 0.969 0.957Panel B: The 2014 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2014×Post2014 0.038*** 0.034*** 0.033*** 0.038***(0.010) (0.012) (0.011) (0.012)Observations 25,661 8,942 21,994 8,750R-squared 0.974 0.967 0.977 0.969Panel C: The 2015 cohortUnbalanced sample Balanced sample(1) (2) (3) (4)Ti,2015×Post2015 0.038*** 0.035*** 0.035*** 0.036***(0.011) (0.012) (0.012) (0.012)Observations 30,725 17,372 26,215 17,276R-squared 0.974 0.974 0.977 0.975Controls Sales, Fixed assets, and Wage bills in logs;Effective tax rate & Indicator for loss-making firms.Fixed-effects Firm FE, Sector-Year FENotes: This table investigates the impact of the corporate income tax ratereductions on firm-level TFP. Results from standard DID regressions arereported in columns (1) and (3), and the results from PSM-DID regressionsare presented in columns (2) and (4). Robust standard errors are clusteredon the sector-year level. ***p< 0.01, **p< 0.05, *p< 0.1.124Table 3.9: Substitution between capital and labor2012 cohort 2014 cohort 2015 cohortDID PSM-DID DID PSM-DID DID PSM-DID(1) (2) (3) (4) (5) (6)Panel A:log(Fixed assets / Sales)Tc×Postc 0.150*** 0.128*** 0.178*** 0.110*** 0.122*** 0.099***(0.034) (0.038) (0.031) (0.032) (0.013) (0.015)Observations 25,760 9,818 25,661 8,942 30,725 17,372R-squared 0.855 0.856 0.864 0.860 0.858 0.855Panel B: log(Wages / Sales)Tc×Postc 0.086*** 0.106*** 0.021 0.006 -0.038*** -0.006(0.029) (0.035) (0.022) (0.029) (0.011) (0.013)Observations 25,760 9,818 25,661 8,942 30,725 17,372R-squared 0.809 0.776 0.825 0.807 0.814 0.817Panel C: log(Fixed assets / Wages)Tc×Postc 0.064** 0.022 0.157*** 0.104*** 0.160*** 0.104***(0.030) (0.033) (0.031) (0.031) (0.015) (0.015)Observations 25,760 9,818 25,661 8,942 30,725 17,372R-squared 0.823 0.832 0.827 0.834 0.826 0.822Controls Effective tax rate & Indicator for loss-making firms.Fixed-effects Firm FE, Sector-Year FENotes: This table examines possible substition between capital and labor followingthe tax policy reforms. Regression results are obtained using the unbalanced sam-ples. The odd columns report the standard DID estimates, while the PSM-DID esti-mates are reported in the even columns. ***p< 0.01, **p< 0.05, *p< 0.1.125Table 3.10: The levels of employment and registered capital upon firm es-tablishmentObs. Mean SD. Medium Min. Max.A. Micro-firm entriesEmployees 932,741 14.26 23.87 8 1 170Capital (in 1K RMB) 932,741 4,187 10,992 600 30 80,000B. Small-firm entriesEmployees 135,727 41.70 54.13 21 2 305Capital (in 1K RMB) 135,727 9,703 21,125 2,100 100 145,758C. Medium/Large-firm entriesEmployees 9,192 413.9 593.6 240 2 3,559Capital (in 1K RMB) 9,192 100,951 250,420 23,800 100 2.000e+06D.Unidentified-firm entriesEmployees 264,280 8.95 12.22 6 1 92Capital (in 1K RMB) 264,280 4,892 11,373 1,000 30 81,800Notes: This table presents the summary statistics for the levels of employmentand registered capital when firms were established during 2005-2017. We usethe snapshot of the 2017 registration data for the universe of firms in the provincewe study.126Table 3.11: The impact of corporate income tax cuts on firm entryShareMC ShareSMC(1) (2) (3) (4) (5) (6)Ext.Fin.s×Post2012 0.610** 0.554**(0.247) (0.250)Ext.Fin.s×Post2014 0.588*** 0.570***(0.158) (0.160)Ext.Fin.s×Post2015 0.660*** 0.643***(0.143) (0.144)ShareErs,t 0.846*** 0.848*** 0.849*** 0.881*** 0.884*** 0.885***(0.015) (0.014) (0.014) (0.017) (0.015) (0.015)Observations 2,566 2,566 2,566 2,566 2,566 2,566R-squared 0.984 0.984 0.984 0.987 0.987 0.986Other controls Post-dummies interacted with sectoral skill and capital intensity.Fixed-effects Region-Industry, Region-YearNotes: This table examines the effect of tax policy reforms on firm entry. Thedependent variable is the share of newly-established “micro” firm in all firms inan region-industry pair for columns (1)-(3), and the share of newly established“small” and “micro” firm in all firms within an region-industry pair for columns(4)-(6). We exclude firms with registered capital above 30 million RMB andmore than 100 employees upon registration. Robust standard errors are clusteredon the region-industry level. ***p< 0.01, **p< 0.05, *p< 0.1.127Table 3.12: Stay small or grow?2013 2014 2015 2016 Total PercentagePanel A: The 2012 cohortThreshold A1: 60K RMBTreated firms: 696 141 85 62 50 338 48.6%All SMPEs: 7,077 1,164 654 442 409 2,669 37.7%Threshold A2: Defined by policyTreated firms: 696 141 18 7 6 172 24.7 %All SMPEs: 7,077 1,164 370 145 143 1,822 25.7%Panel B: The 2014 cohortThreshold B1: 100K RMBTreated firms: 676 161 97 258 38.2%All SMPEs: 9,143 909 789 1,698 18.6 %Threshold B2: Defined by policyTreated firms: 676 15 10 25 3.7%All SMPEs: 9,143 365 277 642 7.0%Panel C: The 2015 cohortThreshold C1: 200K RMBTreated firms: 1,304 368 368 28.2%All SMPEs: 11,837 1,164 1,164 9.8%Threshold C2: Defined by policyTreated firms: 1,304 84 84 6.4%All SMPEs: 11,837 524 524 4.4%Notes: This table reports the number of firms that grow above the taxable incomethresholds of SMPEs in each year since each income tax policy reform. The firsttaxable income threshold of each panel is the qualifying threshold of SMPEs inthe reform year of each cohort. The second taxable income threshold of eachpanel is the threshold of SPMEs in place in each calendar year and is thereforetime-varying. 128ConclusionThis thesis consists of three chapters in studying financial economics, labormarket, and firm growth. Using micro-level data and cross-sectional meth-ods, I examine the impacts of economic shocks to regional economies andfirm growth.In Chapter 1, I study the role of credit in generating the boom and bustof economic activities, by investigating the U.S. mortgage expansion in theearly 2000s and its impact on the local labor markets. I construct the mort-gage supply shock by exploiting two sources of heterogeneity: differencein the lending behaviors of multi-market lenders and the exposure of eachcounty to these multi-market lenders. I find that credit supply shock did notexplain the local unemployment rate or employment rate changes from 2003to 2017, but it negatively affected the local wage growth. Next, I analyze theimpact of credit shock in each phase of credit expansion, contraction, andrecovery. I find that mortgage supply shock does not explain the local un-employment or wages in the expansionary period, but negatively affectedboth in the recession. After the recession, the high-credit regions experi-enced a faster recovery in unemployment, but not in wage, resulting in thelong-term negative impact on wages.To study how credit affected the wage growth, I investigate the effect ofcredit shock on wage growth by industry and education groups. One plau-sible explanation is that the decline in wage growth is driven by the sector-specific effect of credit shock, e.g. construction or finance sector, given thenature of the financial crisis. I show that this hypothesis is not supported bysector-specific regressions that the estimated coefficients are similar across129different sectors. Another concern is that credit shock negatively affectedthe low-skill workers concentrated in the construction sector, which spilledover to all sectors. This hypothesis is rejected by the education-specificwage regressions that the negative effect of credit shock are similar for bothlower educated workers (less than high school) and higher educated work-ers (bachelor and above).As the negative wage growth cannot be explained by sector- or skilled-specific mechanisms, I propose that credit shock affected the demand sideof the labor market. I find that high-shock regions experienced a labor re-allocation: rise in the young firm employment share and decline for the oldones, together with a decline in the total number of firms. I argue that thecredit market fluctuations affect the labor demand side of the market, whichleads to labor reallocation and declines in labor productivity.In Chapter 2, to explain the empirical results documented in the firstchapter, I propose a simple mechanism with two types of financial con-straints: one on the household side and one on the firm side. Both con-straints are tied to the collateral value of housing. A credit shock, referred toas a change in the household borrowing constraint, affects housing value asit affects the incentive to hold houses. This leads to changes in the workingcapital constraint on the firm side and affects the firm’s labor demand. Thismodel provides an explanation of how credit shock affects labor marketsthrough its effect on the housing market and labor demand. It can jointlyexplain the negative effect of credit shock on wage growth and labor real-location, as well as the asymmetric effect of credit shock on labor markets(positive credit shock does not affect the local labor market, but negativecredit shock does). The quantitative exercise suggests the credit channelcan explain up to 20% decline in wage growth in the recession. There areother possible mechanisms to explain the negative effect of credit shock onwage growth, e.g., change in worker-firm bargaining power. Future worksneed to be done to quantify the contribution of different mechanisms ondeclining wage growth.In Chapter 3, we turn to study how a lower corporate income tax ratetargeting small- and micro-profit firms in China affects firms’ performance.130Utilizing a gradual lifting of the qualifying threshold from 2010 to 2016,we find that lowering the corporate tax lead to increase in the total factorproductivity, investment and sales growth of newly qualified firms. In ad-dition, we find a substitution from labor to capital and an increase in thefirm entry. Our study provides an assessment of China’s corporate incometax cuts targeting small firms, an important element of its recent pro-activefiscal policy, on affected firms’ performance.131BibliographyAbadie, A. (2005). Semiparametric Difference-in-Differences Estimators.The Review of Economic Studies, 72(1):1–19. → page 94Ackerberg, D. A., Caves, K., and Frazer, G. (2015). IdentificationProperties of Recent Production Function Estimators. Econometrica,83(6):2411–2451. → page 158Adao, R., Kolesar, M., and Eduardo, M. (2019). Shift-Share Designs:Theory and Inference. The Quarterly Journal of Economics,134(4):1949–2010. → page 6Adelino, M., Schoar, A., and Severino, F. (2015). House prices, collateral,and self-employment. Journal of Financial Economics, 117(2):288–306. →page 14Agrawal, A., Rosell, C., and Simcoe, T. (2020). Tax Credits and Small FirmR&D Spending. American Economic Journal: Economic Policy, 12(2):1–21.→ pages 78, 81Albanesi, S., De, G., and Nosal, G. J. (2017). Credit Growth and theFinancial Crisis: A New Narrative. Technical report, National Bureau ofEconomic Research. → page 14Arnold, J. and Javorcik, B. (2005). Gifted Kids or Pushy Parents? ForeignAcquisitions and Plant Performance in Indonesia. CEPR DiscussionPaper, 3193. → page 94Arslan, Y., Guler, B., and Kuruscu, B. (2020). Credit Supply DrivenBoom-Bust Cycles. Available at SSRN 3573994. → page 62Autor, D. H., Dorn, D., and Hanson, G. H. (2013). The China Syndrome:Local Labor Market Effects of Import Competition in the United States.American Economic Review, 103(6):2121–2168. → page 15132Avery, R., Brevoort, K., and Canner, G. (2007). Opportunities and Issues inUsing HMDA Data. Journal of Real Estate Research, 29(4):351–380. →page 10Bai, C.-E., Hsieh, C.-T., and Song, Z. (2016). The Long Shadow of China’sFiscal Expansion. Brookings Papers on Economic Activity, 47(2):129–181.→ pages 79, 84Ball, L. M. (2009). Hysteresis in unemployment: old and new evidence.Technical report, National Bureau of Economic Research. → page 5Ball, L. M. (2014). Long-term Damage from the Greatt Recession in OECDCountries. Technical report, National Bureau of Economic Research. →page 5Bartik, T. J. (1991). Who benefits from state and local economicdevelopment policies? → pages 1, 15Beraja, M., Hurst, E., and Ospina, J. (2019). The aggregate implications ofregional business cycles. Econometrica, 87(6):1789–1833. → page 29Bernanke, B., Gertler, M., and Gilchrist, S. (1999). The financial acceleratorin a quantitative business cycle framework. Handbook of macroeconomics,(1):1341–1393. → page 62Bernanke, B. S., Lown, C. S., and Friedman, B. M. (1991). The creditcrunch. Brookings papers on economic activity, 1991(2):205–247. → pages17, 62Blanchard, O. and Katz, L. F. (1992). Regional evolutions. Brookings Paperson Economic Activity, 1992(1):1–61. → pages 15, 26, 27, 28Blanchard, O. J. and Summers, L. H. (1987). Hysteresis in Unemployment.Technical report, National Bureau of Economic Research. → page 27Borusyak, K., Hull, P., and Jaravel, X. (2018). Quasi-experimentalshift-share research designs. Technical report, National Bureau ofEconomic Research. → page 6Buera, F. J., Fattal Jaef, R. N., and Shin, Y. (2015). Anatomy of a creditcrunch: From capital to labor markets. Review of Economic Dynamics,18(1):101–117. → page 62133Card, D. (2001). Immigrant inflows, native outflows, and the local labormarket impacts of higher immigration. Journal of Labor Economics,19(1):22–64. → page 15Cerra, V. and Saxena, S. C. (2008). Growth Dynamics: The Myth ofEconomic Recovery. American Economic Review, 98(1):439–457. → page 5Charles, K. K., Hurst, E., and Notowidigdo, M. J. (2016). The Masking ofthe Decline in Manufacturing Employment by the Housing Bubble.Journal of Economic Perspectives, 30(2):179–200. → pages 13, 21, 31Chen, Z., Jiang, X., Liu, Z., Sua´rez Serrato, J. C., and Xu, D. (2019). Taxpolicy and lumpy investment behavior: Evidence from china’s vatreform. Technical report, National Bureau of Economic Research. →pages 79, 80, 81Chen, Z., Liu, Z., Sua´rez Serrato, J. C., and Xu, D. Y. (2018). Notching r&dinvestment with corporate income tax cuts in china. Technical report,National Bureau of Economic Research. → pages 81, 82Chodorow-Reich, G. (2014). The Employment Effects of Credit MarketDisruptions: Firm-level Evidence from the 2008–9 Financial Crisis. TheQuarterly Journal of Economics, 129(1):1–59. → pages 7, 9, 15Chodorow-Reich, G. and Falato, A. (2017). The loan covenant channel:How bank health transmits to the real economy. Technical report,National Bureau of Economic Research. → pages 7, 9Cui, W. (2007). The Prospect of New Partnership Taxation in China. TaxNotes International, 46(6). → page 85Cui, W. (2014). China’s Business-Tax-to-VAT Reform: an InterimAssessment. British Tax Review, 2014(5):617–641. → page 84Cui, W., Hicks, J., and Xing, J. (2020). Cash on the Table? ImperfectTake-up of Tax Incentives and Firm Investment Behavior. Working Paper.→ pages 81, 84Davis, S. J. and Haltiwanger, J. (2019). Dynamism Diminished: The Roleof Housing Markets and Credit Conditions. Technical report, NationalBureau of Economic Research. → pages 9, 32, 34De Loecker, J. (2007). Do exports generate higher productivity? Evidencefrom Slovenia. Journal of International Economics, 73(1):69 – 98. → page 94134De Loecker, J. and Warzynski, F. (2012). Markups and Firm-Level ExportStatus. American Economic Review, 102(6):2437–71. → page 158Dechezlepreˆtre, A., Einio¨, E., Martin, R., Nguyen, K.-T., and Van Reenen, J.(2016). Do tax incentives for research increase firm innovation? An RDdesign for R&D. Technical report, National Bureau of EconomicResearch. → page 81Decker, R., Haltiwanger, J., Jarmin, R., and Miranda, J. (2014). The Role ofentrepreneurship in US job creation and economic dynamism. Journal ofEconomic Perspectives, 28(3):3–24. → page 78Dell’Ariccia, G., Igan, D., and Laeven, L. (2012). Credit Booms andLending Standards: Evidence from the Subprime Mortgage Market.Journal of Money, Credit and Banking, 44(3):367–384. → pages 8, 10Demyanyk, Y. and Van Hemert, O. (2011). Understanding the subprimemortgage crisis. The review of financial studies, 24(6):1848–1880. → page 8Di Maggio, M. and Kermani, A. (2017). Credit-Induced Boom and Bust.The Review of Financial Studies, 30(11):3711–3758. → pages 8, 9, 13Eggertsson, G. B. and Krugman, P. (2012). Debt, Deleveraging, and theLiquidity Trap: A Fisher-Minsky-Koo Approach. The Quarterly Journal ofEconomics, 127(3):1469–1513. → page 23Fan, Z. and Liu, Y. (2020). Tax Compliance and Investment Incentives:Firm Responses to Accelerated Depreciation in China. Journal ofEconomic Behavior & Organization, 176:1–17. → page 84Favara, G. and Imbs, J. (2015). Credit Supply and the Price of Housing.American Economic Review, 105(3):958–992. → pages 7, 13Fernald, J. G., Hall, R. E., Stock, J. H., and Watson, M. W. (2017). Thedisappointing recovery of output after 2009. Technical report, NationalBureau of Economic Research. → page 5Fisher, I. (1933). The Debt-Deflation Theory of Great Depressions.Econometrica, 1(4):337. → page 5Garcia, D. (2020). Employment in the Great Recession: how importantwere household credit supply shocks? Journal of Money, Credit andBanking, 52(1):165–203. → page 9135Gilchrist, S., Siemer, M., and Zakrajsek, E. (2018). The real effects of creditbooms and busts. Technical report, Working Paper. → pages8, 9, 15, 29, 30Gilchrist, S., Sim, J., and Zakrajsˇek, E. (2014). Uncertainty, FinancialFrictions, and Investment Dynamics. Technical report, National Bureauof Economic Research. → page 62Glancy, D. (2017). Housing Bust, Bank Lending & Employment: Evidencefrom Multimarket Banks. Finance and Economics Discussion Series,2017(118). → page 15Goldsmith-Pinkham, P., Sorkin, I., and Swift, H. (2018). BartikInstruments: What, When, Why, and How. Technical report, NationalBureau of Economic Research. → page 6Gourio, F., Messer, T., and Siemer, M. (2016). Firm entry andmacroeconomic dynamics: A state-level analysis. American EconomicReview, 106(5):214–18. → page 32Greenstone, M., Mas, A., and Nguyen, H.-L. (2020). Do Credit MarketShocks Affect the Real Economy? Quasi-experimental Evidence fromthe Great Recession and” Normal” Economic Times. American EconomicJournal: Economic Policy, 12(1):200–225. → pages 7, 9, 15, 17Guerrieri, L. and Iacoviello, M. (2017). Collateral constraints andmacroeconomic asymmetries. Journal of Monetary Economics, 90:28–49.→ pages 23, 73Go¨rg, H., Henry, M., and Strobl, E. (2008). Grant Support and ExportingActivity. The Review of Economics and Statistics, 90(1):168–174. → page 94Haltmaier, J. (2013). Do recessions affect potential output? FRBInternational Finance Discussion Paper, (1066). → page 5Harju, J., Koivisto, A., and Matikka, T. (2020). The Effects of CporporateTaxes on Small Firms. VATT Working Papers, 129. → page 81Hsieh, C.-T. and Olken, B. A. (2014). The Missing ”Missing Middle”.Journal of Economic Perspectives, 28(3):89–108. → pages 78, 82Iacoviello, M. (2005). House Prices, Borrowing Constraints, and MonetaryPolicy in the Business Cycle. The American Economic Review,95(3):739–764. → pages 64, 68136Jermann, U. and Quadrini, V. (2012). Macroeconomic Effects of FinancialShocks. American Economic Review, 102(1):238–271. → page 62Jorda`, , Schularick, M., and Taylor, A. M. (2015). Leveraged bubbles.Journal of Monetary Economics, 76:S1–S20. → pages 1, 8Justiniano, A., Primiceri, G. E., and Tambalotti, A. (2019). Credit supplyand the housing boom. Journal of Political Economy, 127(3):1317–1350. →pages 23, 68Keys, B. J., Mukherjee, T., Seru, A., and Vig, V. (2010). Did SecuritizationLead to Lax Screening? Evidence from Subprime Loans. QuarterlyJournal of Economics, 125(1):307–362. → pages 8, 13Khan, A. and Thomas, J. K. (2013). Credit shocks and aggregatefluctuations in an economy with production heterogeneity. Journal ofPolitical Economy, 121(6):1055–1107. → page 62Kiyotaki, N. and Moore, J. (1997). Credit Cycles. Journal of PoliticalEconomy, 105(2):211–248. → pages 62, 64Koga, T. (2003). Firm size and R&D tax incentives. Technovation,23(7):643–648. → page 81Korinek, A. and Simsek, A. (2016). Liquidity trap and excessive leverage.American Economic Review, 106(3):699–738. → page 23La Porta, R. and Shleifer, A. (2014). Informality and Development. Journalof Economic Perspectives, 28(3):109–26. → pages 78, 82Levinsohn, J. and Petrin, A. (2003). Estimating Production FunctionsUsing Inputs to Control for Unobservables. The Review of EconomicStudies, 70(2):317–341. → page 158Liu, Y. and Mao, J. (2019). How Do Tax Incentives Affect Investment andProductivity? Firm-Level Evidence from China. American EconomicJournal: Economic Policy, 11(3):261–91. → pages 78, 80, 82Lokshin, B. and Mohnen, P. (2012). How effective are level-based R&D taxcredits? evidence from the netherlands. Applied Economics,44(12):1527–1538. → page 81Maffini, G., Xing, J., and Devereux, M. P. (2019). The Impact of InvestmentIncentives: Evidence from UK Corporation Tax Returns. AmericanEconomic Journal: Economic Policy, 11(3):361–389. → page 81137Mendoza, E. G. (2010). Sudden stops, financial crises, and leverage.American Economic Review, 100(5):1941–66. → page 73Mian, A., Rao, K., and Sufi, A. (2013). Household Balance Sheets,Consumption, and the Ecoomic Slump. Quarterly Journal of Economics,128(4):1687–1726. → page 9Mian, A. and Sufi, A. (2009). The Consequences of Mortgage CreditExpansion: Evidence from the U.S. Mortgage Default Crisis. QuarterlyJournal of Economics, 124(4):1449–1496. → pages 8, 13Mian, A. and Sufi, A. (2014). What Explains the 2007–2009 Drop inEmployment? Econometrica, 82(6):2197–2223. → page 7Mian, A. and Sufi, A. (2018). Credit Supply and Housing Speculation.Technical report, National Bureau of Economic Research, Cambridge,MA. → pages 1, 8Midrigan, V. and Philippon, T. (2011). Household leverage and therecession. → page 23Mondragon, J. (2018). Household credit and employment in the GreatRecession. Kilts Center for Marketing at Chicago Booth–Nielsen DatasetPaper Series, pages 1–025. → page 9Nadauld, T. D. and Sherlund, S. M. (2013). The impact of securitization onthe expansion of subprime credit. Journal of Financial Economics,107(2):454–476. → page 8Neumeyer, P. A. and Perri, F. (2005). Business cycles in emergingeconomies: the role of interest rates. Journal of Monetary Economics,52(2):345–380. → page 62Pham, A. (2020). Effects of Temporary Corporate Income Tax Cuts:Evidence from Vietnam. Journal of Development Economics, 146. → page81Pilloff, S. J. (2010). Bank merger activity in the United States, 1994-2003.DIANE Publishing. → page 17Rice, T. and Strahan, P. E. (2010). Does Credit Competition AffectSmall-Firm Finance? Journal of Finance, 65(3):861–889. → page 16138Rosenbaum, P. R. and Rubin, D. B. (1983). The central role of thepropensity score in observational studies for causal effects. Biometrika,70(1):41–55. → page 95Saiz, A. (2010). The Geographic Determinants of Housing Supply.Quarterly Journal of Economics, 125(3):1253–1296. → page 30Siemer, M. (2012). Firm Entry and Employment Dynamics in the GreatRecession. SSRN Electronic Journal. → pages 9, 32Summers, L. H. (2014). U.S. Economic Prospects: Secular Stagnation,Hysteresis, and the Zero Lower Bound. Business Economics, 49(2):65–73.→ page 5Ulyssea, G. (2019). Firms, Informality, and Development: Theory andEvidence from Brazil. American Economic Review, 108(8):2015–2047. →page 78Uribe, M. and Yue, V. Z. (2006). Country spreads and emerging countries:Who drives whom? Journal of International Economics, 69(1):6–36. →page 62Volpe Martincus, C. and Carballo, J. (2008). Is export promotion effectivein developing countries? firm-level evidence on the intensive and theextensive margins of exports. Journal of International Economics, 76(1):89– 106. → page 94Yagan, D. (2019). Employment hysteresis from the great recession. Journalof Political Economy, 127(5):2505–2558. → pages 8, 28Zwick, E. and Mahon, J. (2017). Tax policy and heterogeneous investmentbehavior. American Economic Review, 107(1):271–248. → pages 78, 80139Appendix ASupporting MaterialsA.1 Appendix to Chapter 1A.1.1 Selected LendersIn this section, I list the top 10 percent largest and most expanding lendersin mortgage origination during 2002-2006.140Table A.1: Top 10% largest selected lenders.Lender Name Mortgage Origination Status(dollar value)Washington Mutual Bank 8.5 ×1011 bankruptcyaBank of America 5.9 ×1011Wells Fargo Bank 5.3×1011World Savings Bank 2.4×1011 acquiredbGMAC Mortgage LLC 2.1 ×1011 bankruptcyFlagstar Bank 2.0×1011First Horizon Home Loan 1.9×1011Greenpoint Mortgage Funding 1.9×1011SunTrust Mortgage 1.9×1011PHH Mortgage Corporation 1.6×1011Note: The selected lenders satisfy the restriction for lending in more than 100counties and 1 census region in 2000 and remained operating during 2000–2006.a The holding company of GMAC Mortgage filed chapter 11 bankruptcy in 2012.b The acquired status refers to the World Savings Bank being acquired by WellsFargo in 2008 as part of Wachovia Corporation.Table A.2: Top 10% fastest growth selected lenders.Lender Name Mortgage Growth StatusEverbank 446%Wells Fargo Bank 425%Lehman Brothers Bank 398% bankruptcyMortgageit 381% acquiredaTD Banknorth 367%WMC Mortgage Company 354% bankruptcyDollar Mortgage Corporation 323%USAA Federal Savings Bank 314%The Huntingon National Bank 309%American Home Mortgage Corporation 302% bankruptcyNovastar Mortgage Inc. 301% bankruptcyNote: The mortgage growth rate is computed as the average annual mortgageorigination during 2002–2006 relative to mortgage origination in 2000. The se-lected lenders satisfy the restriction for lending in more than 100 counties and 1census region in 2000 and remained operating during 2000–2006. a The acquiredstatus refers to Mortgageit has been acquitted by Deutsche Bank in 2007.141Table A.3: Effect of local market shares on income growth: 2003–2006Income growth: 2003-2006∆y2003−2006 ∆y2003−2006 ∆y2003−2006 ∆y2003−2006market l 0.0267 0.0836(0.0589) (0.0473)market h 0.102 0.00372(0.0695) (0.0476)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 2566 2566 2665 2665adj. R2 0.266 0.440 0.276 0.440Note: this table reports regression estimates of lender market share on the lo-cal income growth during 2003–2006. Variable “market l” represents the localmortgage market shares for top 20% slowest expanding lenders and “market h”represents for top 20% fastest expanding lenders. Regressions are weighted bythe 2000 population. Standard errors are clustered at the state level. County char-acteristics include fraction of subprime population, establishment, employmentrate, unemployment rate, weekly wage, average annual income, median house-hold income and the poverty rate in 2000. Industry composition includes 2000employment share in a county for 23 two-digit industries: Agriculture, Mining,Utilities, Construction, Manufacturing, Wholesale Trade, Retail Trade, Trans-portation, Information, Finance, Real Estate, Professional Services, Management,Administrative Services, Education, Health Care, Entertainment, Accommoda-tion and Food Services, Other Services. *, **, and *** indicate significance at the0.1, 0.05, and 0.01 levels, respectively.142Table A.4: Effect of local market shares on local private employmentgrowth: 2003–2006Employment changes: 2003-2006∆y2003−2006 ∆y2003−2006 ∆y2003−2006 ∆y2003−2006market l 0.0883 0.0551(0.0770) (0.0877)market1 h 0.0739 -0.0535(0.0966) (0.0947)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 2558 2558 2644 2644adj. R2 0.136 0.228 0.139 0.229Note: this table reports regression estimates of lender market share on the aver-age monthly private employment change during 2003–2006. Variable “market l”represents the local mortgage market shares for slowest expanding lenders and“market h” represents for fastest expanding lenders. Regressions are weightedby the 2000 population. Standard errors are clustered at the state level. Countycharacteristics include fraction of subprime population, establishment, employ-ment rate, unemployment rate, weekly wage, average annual income, medianhousehold income and the poverty rate in 2000. Industry composition includes2000 employment share in a county for 23 two-digit industries. *, **, and *** in-dicate significance at the 0.1, 0.05, and 0.01 levels, respectively.143A.1.2 Relationship between Credit Supply Shock and HousingPrice ChangesIn this following table, I use the Zillow housing price index and estimate theeffect of credit supply shock on housing price changes from 2003 to 2016.Table A.5: Effect of credit supply shock on housing price: 2003–2016.Housing Price: 2003-2016∆y2003−2006 ∆y2006−2010 ∆y2010−2016credit shock 0.441∗∗∗ 0.216∗∗∗ -0.524∗∗∗ -0.307∗∗∗ 0.181∗∗ 0.0203(0.0914) (0.0571) (0.131) (0.101) (0.0785) (0.0752)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 946 946 978 978 1145 1145adj. R2 0.337 0.738 0.274 0.666 0.153 0.431Note: housing data is taken from the Zillow housing price index. Regressionsare weighted by the 2000 population. Standard errors are clustered at the statelevel. County characteristics include fraction of subprime population, establish-ment, employment rate, unemployment rate, weekly wage, average annual in-come, median household income and the poverty rate in 2000. Industry compo-sition includes 2000 employment share in a county for 23 two-digit industries:Agriculture, Mining, Utilities, Construction, Manufacturing, Wholesale Trade,Retail Trade, Transportation, Information, Finance, Real Estate, Professional Ser-vices, Management, Administrative Services, Education, Health Care, Entertain-ment, Accommodation and Food Services, Other Services. *, **, and *** indicatesignificance at the 0.1, 0.05, and 0.01 levels, respectively.144A.1.3 Relationship between Credit Supply Shock andEmployment ChangesThis section presents the effect of credit supply shock on labor force partic-ipation rate and private employment rate changes from 2003 to 2017. Theworking-age population is defined by population between 16 and 65 yearsold. It also presents the relation between credit supply shock and privateemployment changes from 2007 to 2017.Table A.6: Effect of credit supply shock on local employment rate andlabor market participation rate changes: 2003–2017.Employment rate :2003-2017 Labor participation rate:2003-2017∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017credit 0.0286 0.0150 0.0162 0.0110(0.0252) (0.0220) (0.0231) (0.0197)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 3044 3044 2946 2946adj. R2 0.078 0.146 0.081 0.154Note: this table presents the effect of credit supply shock on the changes in pri-vate employment rate and labor participation rate from 2003 to 2017. Employ-ment rate and labor force participation rate are computed with working-age pop-ulation between 16 and 65 years old. Regressions are weighted by the 2000 pop-ulation. Standard errors are clustered at the state level. County characteristicsinclude fraction of subprime population, establishment, employment rate, un-employment rate, weekly wage, average annual income, median household in-come and the poverty rate in 2000. Industry composition includes 2000 employ-ment share in a county for 23 two-digit industries: Agriculture, Mining, Utilities,Construction, Manufacturing, Wholesale Trade, Retail Trade, Transportation, In-formation, Finance, Real Estate, Professional Services, Management, Administra-tive Services, Education, Health Care, Entertainment, Accommodation and FoodServices, Other Services. *, **, and *** indicate significance at the 0.1, 0.05, and0.01 levels, respectively.145Table A.7: Effect of credit supply shock on local private employmentchange: 2010-2017.Employment Change: 2010-2017∆y2010−2013 ∆y2010−2013 ∆y2013−2015 ∆y2013−2015 ∆y2015−2017 ∆y2015−2017credit shock -0.0141 0.00755 0.0364∗∗∗ 0.0296∗ 0.0768∗∗∗ 0.0446∗∗∗(0.0172) (0.0228) (0.0129) (0.0151) (0.0170) (0.0147)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3019 3019 3032 3032 3037 3037adj. R2 0.098 0.161 0.150 0.179 0.180 0.236Note: this table presents the effect of credit supply shock on the private employ-ment changes during 2010–2017. Standard errors are clustered at the commutingzone level. Regressions are weighted by the 2000 population. County character-istics include fraction of subprime population, establishment, employment rate,unemployment rate, weekly wage, average annual income, median household in-come and poverty rate in 2000. Industry composition includes 2000 employmentshare in a county for 23 two-digit industries. *, **, and *** indicate significance atthe 0.1, 0.05, and 0.01 levels, respectively.146Figure A.1: Effect of credit supply shock on private employmentchange: 2006-2017Note: this figure plots the effect of credit supply shock on the yearly changes inlocal employments from 2006 to 2017, after controlling for county characteris-tics, industry composition, and state fixed effects. Standard errors are clusteredat the state level. County characteristics include fraction of subprime population,establishment, employment rate, unemployment rate, weekly wage, average an-nual income, median household income and poverty rate in 2000. Industry com-position includes 2000 employment share in a county for 23 two-digit industries.147A.1.4 Relationship between Credit Supply Shock andEmployment ChangesTable A.8: Effect of credit supply shock on weekly wage with commut-ing zone fixed effects.Weekly wage: 2003-2017∆y2003−2017 ∆y2003−2006 ∆y2006−2010 ∆y2010−2017credit -0.104∗∗∗ -0.0204 -0.0439 -0.0228(0.0337) (0.0187) (0.0252) (0.0239)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes YesCZ FE Yes Yes Yes YesN 3027 3024 3025 3029adj. R2 0.442 0.275 0.264 0.258Note: the commuting zone effects are included in the regression. Standard er-rors are clustered at commuting zone level. Regressions are weighted by 2000population. County characteristics include : fraction of subprime population,establishment, employment rate,unemployment rate, weekly wage, average an-nual income, median household income and poverty rate in 2000. Industry com-position includes 2000 employment share in a county for 23 two-digit industries.*, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.148A.1.5 Relationship between Credit Shock and Weekly WageTable A.9: Effect of credit supply shock on employment share changeby young and old firms: 2003–2006.Young firm employment share Old firm employment share∆y2003−2006 ∆y2003−2006credit 0.0216 0.00740 -0.0429∗∗ -0.0135(0.0162) (0.0107) (0.0161) (0.0119)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 2637 2637 2815 2815adj. R2 0.009 0.049 0.019 0.055Note: this table reports regression estimates of credit supply shock on the changein employment share for young firms (age<4) and old firms (age>5) from 2003to 2006. Regressions are weighted by the 2000 population. Standard errors areclustered at the state level. County characteristics include fraction of subprimepopulation, establishment, employment rate, unemployment rate, weekly wage,average annual income, median household income and poverty rate in 2000. In-dustry composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, re-spectively.149A.1.6 Relationship between Credit Shock and EstablishmentGrowth: 2003-2016Table A.10: Effect of credit supply shock on the number of establish-ment growth: 2006–2016.(1) (2)y ycredit -0.156∗∗ -0.111∗∗∗(0.0727) (0.0395)county controls Yes Yesindustry controls Yes Yesstate FE YesN 3055 3055adj. R2 0.087 0.420Note: the establishment growth is defined as the average annual growth ratein the number of establishments during 2006–2016 relative to average annualgrowth during 2003–2006. Regressions are weighted by the 2000 population.Standard errors are clustered at the state level. County characteristics includefraction of subprime population, establishment, employment rate, unemploy-ment rate, weekly wage, average annual income, median household income andpoverty rate in 2000. Industry composition includes 2000 employment share ina county for 23 two-digit industries. *, **, and *** indicate significance at the 0.1,0.05, and 0.01 levels, respectively.150A.1.7 Alternative Credit Shock IIn this section, I construct the credit supply shock with lenders that oper-ated in more than 200 counties and one census region in 2000 and remainoperation during 2000–2006. The national mortgage growth includes allmortgage origination: home purchase, home improvement and refinanc-ing. The regression estimates are reported in the following table.Table A.11: Effect of credit supply shock on the local labor market:2003–2017Unemployment rate Weekly wage Old firm employment share∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2006−2016 ∆y2006−2016credit -0.0112 -0.00191 -0.0818∗ -0.102∗∗∗ 0.0579∗∗∗ 0.0448∗∗∗(0.00569) (0.00354) (0.0491) (0.0294) (0.0121) (0.0101)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE No Yes No Yes No YesN 3050 3050 3027 3027 2744 2744adj. R2 0.179 0.471 0.263 0.364 0.087 0.109Note: this table reports regression estimates of credit supply shock on the changein the unemployment rate and weekly wage from 2003 to 2017 and change in theemployment share for old firms (age > 5). The credit supply shock is measuredwith lenders that operated in more than 200 counties and one census region in2000 and remain operation during 2000–2006. Regressions are weighted by the2000 population. County characteristics include fraction of subprime popula-tion, establishment, employment rate, unemployment rate, weekly wage, aver-age annual income, median household income and poverty rate in 2000. Indus-try composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, re-spectively.151A.1.8 Alternative Credit Supply Shock IIIn this section, I construct the credit supply shock with lenders that oper-ated in more than 100 counties and one census region in 2000 and remainoperation during 2000–2006. Both the national mortgage growth and localmortgage market shares are measured using only home purchase mortgagedata. The regression estimates are reported in the following table. In ad-dition to the reduced-form IV regressions, I present the estimate results forfirst and second stage regressions, using actual local mortgage growth dur-ing 2002-2006 as the explanatory variable.Table A.12: Effect of credit supply shock on the local labor market:2003–2017Unemployment Rate Weekly Wage Old Firm Employment Share∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017credit -0.00939 -0.00188 -0.0967∗ -0.0860∗∗∗ 0.0524∗∗∗ 0.0368∗∗∗(0.00536) (0.00346) (0.0488) (0.0279) (0.00996) (0.00946)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3050 3050 3027 3027 2744 2744adj. R2 0.179 0.471 0.266 0.364 0.088 0.109Note: effect of modified credit supply shock on unemployment rate, wage, andold firm employment share. /the credit supply shock is constructed by predict-ing the lenders’ aggregate growth in new home mortgage origination with theirinitial mortgage share in 2000. Lenders are selected to operate in more than 100counties and one census region in 2000. Regressions are weighted by the 2000population. Standard errors are clustered at the state level. County character-istics include fraction of subprime population, establishment, employment rate,unemployment rate, weekly wage, average annual income, median household in-come and poverty rate in 2000. Industry composition includes 2000 employmentshare in a county for 23 two-digit industries. *, **, and *** indicate significance atthe 0.1, 0.05, and 0.01 levels, respectively.152Table A.13: Effect of credit supply shock on the number of mortgageexpansion: 2003–2010.Mortgage expansion: 2003-2006 Mortgage contraction: 2006-2010∆y2003−2006 ∆y2003−2006 ∆y2006−2010 ∆y2006−2010credit shock 0.838∗∗∗ 0.560∗∗∗ -0.689∗∗∗ -0.431∗∗∗(0.156) (0.120) (0.137) (0.0983)county controls Yes Yes Yes Yesindustry controls Yes Yes Yes Yesstate FE No Yes No YesN 3038 3038 3036 3036adj. R2 0.263 0.426 0.218 0.437Note: credit supply shock is constructed by predicting the lenders’ aggregategrowth in new home mortgage origination with their initial mortgage share in2000. Regressions are weighted by the 2000 population. Standard errors areclustered at the state level. County characteristics include fraction of subprimepopulation, establishment, employment rate, unemployment rate, weekly wage,average annual income, median household income and poverty rate in 2000. In-dustry composition includes 2000 employment share in a county for 23 two-digitindustries. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, re-spectively.153Table A.14: Effect of credit supply shock on the local labor market:2003–2017Unemployment Rate Weekly Wage Old Firm Employment Share∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017 ∆y2003−2017credit -0.00939 -0.00188 -0.0967∗ -0.0860∗∗∗ 0.0524∗∗∗ 0.0368∗∗∗(0.00536) (0.00346) (0.0488) (0.0279) (0.00996) (0.00946)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes YesN 3050 3050 3027 3027 2744 2744adj. R2 0.179 0.471 0.266 0.364 0.088 0.109Note: effect of modified credit supply shock on unemployment rate, wage, andold firm employment share. A credit supply shock is constructed by predictingthe lenders’ aggregate growth in new home mortgage origination with their ini-tial mortgage share in 2000. Regressions are weighted by the 2000 population.Standard errors are clustered at the state level. County characteristics includefraction of subprime population, establishment, employment rate, unemploy-ment rate, weekly wage, average annual income, median household income andpoverty rate in 2000. Industry composition includes 2000 employment share ina county for 23 two-digit industries. *, **, and *** indicate significance at the 0.1,0.05, and 0.01 levels, respectively.154Table A.15: Effect of mortgage expansion on local unemployment ratechange: 2003–2017Expansion: 2003-2006 Contraction:2006-2010 Recovery: 2010-2017OLS IV OLS IV OLS IVx 0.00115∗∗ -0.00109 0.00213 0.0269∗∗∗ -0.000575 -0.0274∗∗(0.000546) (0.00403) (0.00132) (0.00884) (0.00159) (0.0113)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes Yes Yes Yes YesN 3031 3031 3031 3031 3038 3038adj. R2 0.509 0.506 0.640 0.540 0.701 0.597Note: this table presents the second-stage regression results of mortgage expan-sion on unemployment rate change from 2003 to 2017, using credit shock as aninstrumental variable. Regressions are weighted by the 2000 population. Stan-dard errors are clustered at the state level. County characteristics include fractionof subprime population, establishment, employment rate, unemployment rate,weekly wage, average annual income, median household income and povertyrate in 2000. Industry composition includes 2000 employment share in a countyfor 23 two-digit industries. *, **, and *** indicate significance at the 0.1, 0.05, and0.01 levels, respectively.155Table A.16: Effect of mortgage expansion on weekly wage change:2003–2017Expansion: 2003-2006 Contraction:2006-2010 Recovery: 2010-2017OLS IV OLS IV OLS IVx 0.000203 -0.00720 -0.00697 -0.0977∗∗ 0.00111 -0.0631(0.00445) (0.0316) (0.00550) (0.0401) (0.00562) (0.0405)county controls Yes Yes Yes Yes Yes Yesindustry controls Yes Yes Yes Yes Yes Yesstate FE Yes Yes Yes Yes Yes YesN 3019 3019 3017 3017 3019 3019adj. R2 0.194 0.193 0.214 0.113 0.164 0.122Note: this table presents the second-stage regression results of mortgage expan-sion on weekly wage growth from 2003 to 2017, using credit shock as an instru-mental variable. Regressions are weighted by the 2000 population. Standarderrors are clustered at the state level. County characteristics include fractionof subprime population, establishment, employment rate, unemployment rate,weekly wage, average annual income, median household income and povertyrate in 2000. Industry composition includes 2000 employment share in a countyfor 23 two-digit industries. *, **, and *** indicate significance at the 0.1, 0.05, and0.01 levels, respectively.156A.2 Appendix to Chapter 2A.2.1 Steady State ConditionsIn steady state, the model is described by the follow equations:r =1βh−1 (A.1)bl =−φbphl (A.2)bw =−φbphw (A.3)bh =−(bl+bwL¯) (A.4)w= min(φhphlll,γlγ−1l ) = min(φhphhlh, lγ−1h ) (A.5)ll = L¯− lh (A.6)βluwh +φbp(1−βl(1+ r))uwc = p(1−βl)uwc (A.7)βlulh+βlφh( fl/w−1)pulc+φb(1−βl(1+ r))pulc = p(1−βl)ulc (A.8)βhuhh+βhφh( fl/w−1)puhc = p(1−βh)uhc (A.9)hl+hh+hwL¯= H¯ (A.10)cwL¯+ cl+ ch = lγl + lγh (A.11)A.2.2 Parameter SettingsTable A.17: Parameter settings for comparative staticsDescription Parameter valuehigh discount factor βh 0.98low discount factor βl 0.95weight on housing service αh 0.1labor share γ 0.7total housing stock H¯ 5total labor supply L¯ 10157A.3 Appendix to Chapter 3A.3.1 The Total Factor Productivity EstimationTo obtain firm-level total factor productivity measures, we first estimate aproduction function, following the approach in De Loecker and Warzyn-ski (2012) and Ackerberg et al. (2015). Specifically, we start with a Cobb-Douglas production function:qi,t = βlli,t +βmmi,t +βkki,t +ωi,t + εi,t , (A.12)where qi,t , li,t , mi,t and ki,t are the log transformations of firm-level output,labor, intermediate input and capital, respectively. ωi,t is firm-level totalfactor productivity and εi,t represents idiosyncratic shocks to the firm-leveloutput. We follow Levinsohn and Petrin (2003) and specify the demand forintermediate input as:mi,t = m(li,t ,ki,t ,microi,t ,smalli,t , lossi,t ,ωi,t), (A.13)where microi,t is an indicator for being a micro-profit firm (SMPE), smalli,tis an indicator for being a small firm, and lossi,t indicates that a firm is in theloss-making position. We explicitly control for firm size and profit-makingstatus to account for the effects of both supply- and demand-shocks on thefirm’s decisions of the optimal input usage.microi,t = 1 ifTaxable incomei,t < 30K & t = 2010,2011Taxable incomei,t < 60K & t = 2012,2013Taxable incomei,t < 100K & t = 2014Taxable incomei,t < 200K & t = 2015Taxable incomei,t < 300K & t = 2016Total assetsi,t < 30/,Million ∀ t, and= 0 otherwise,158smalli,t = 1 ifTaxable incomei,t ∈ [30K, 300K)& t = 2010,2011Taxable incomei,t ∈ [60K, 300K)& t = 2012,2013Taxable incomei,t ∈ [100K, 300K)& t = 2014Taxable incomei,t ∈ [200K, 300K)& t = 2015Taxable incomei,t < 300K & t = 2016Total assetsi,t < 30Million ∀ t, and= 0 otherwise,Assuming that there exists a monotonic relationship between mi,t andωi,t , productivity can then be proxied by the inversion of function (A.13):ωi,t = h(li,t ,ki,t ,microi,t ,smalli,t , lossi,t ,mi,t), (A.14)The estimation proceeds in two steps. In the first step, we estimate:qi,t = φ(li,t ,mi,t ,ki,t ,microi,t ,smalli,t , lossi,t)+ εi,t ,where:φ(.) = βlli,t +βmmi,t +βkki,t +ωi,t= βlli,t +βmmi,t +βkki,t +h(li,t ,ki,t ,microi,t ,smalli,t , lossi,t ,mi,t).(A.15)Then we construct the estimate for productivity as:ωˆit = φˆi,t − (βlli,t +βmmi,t +βkki,t) (A.16)In the second stage, we rely on the law of motion for productivity specifiedas equation (A.17) below:ωi,t = g(ωi,t−1,microi,t ,smalli,t , lossi,t)+ εi,t (A.17)to recover the innovation of productivity εi,t(β ), given β = (βl,βm,βk). Wethen use the following moment condition to estimate the production func-tion parameters using General Method of Moments (GMM):159E = εi,t(β ) li,t−1mi,t−1ki,t = 0 (A.18)Lastly, we calculate the estimates of the firm-level TFP as:ωˆi,t = qi,t − (βˆlli,t + βˆmmi,t + βˆkki,t). (A.19)A. Gross output – qˆi,tWe use observed nominal sales revenue deflated by an output index asthe proxy for physical output. We obtain the nationwide sectoral producer-price index (PPI) at the 2-digit CIC level for 2010-2016 from the ChineseStatistics Yearbook and multiply it by the aggregate manufacturing PPI ofthe province to which our data belong relative to the country to constructthe province-sector specific output deflator:Pos,t = PPIs,t ×PPIProvince,tPPICN,t. (A.20)And then we calculate:qˆi,t =ri,tPoi,t. (A.21)B. Intermediate input – mi,tFirms in our sample do not report expenditure on material. However,it can be calculated as business costs net of labor costs and current depre-ciation according to common accounting rules. Next, we utilize the 2012input-output table of the province of our data to calculate the input defla-tor, which is the weighted average of the sectoral output deflators, using asweights the coefficients in the IO table. And then the nominal intermediate160expenditure is divided by the input deflator to obtain the firm-level usageof intermediate input.C. Employment – li,tOur data does not include information of total annual employment butonly employment at registration and total wage bill. We first obtain thedeflator of labor costs from the Provincial Statistics Yearbook, and use ittogether with the wage bill to calculate firm-level real wage growth rate rl,t ,which we will use as a proxy for the employment growth rate of firms. LetRLi,reg = Li,2010/Li,reg denote the ratio of a firm’s employment in 2010, thefirst year of our sample, and that at registration, the annual employment ofa firm can be inferred as:Li,t = Li,2010×t∏2011(1+ rl,t) = RLi,regLi,reg×t∏2011(1+ rl,t) for t ∈ [2011,2016].(A.22)The log-transformation then is:li,t = log(RLi,reg)+ li,reg+ log[t∏2011(1+ rl,t)]. (A.23)Although log(RLi,reg) is not observable, note that it is time-invariant andfirm-specific, and therefore can be controlled for using firm-fixed effectswhile estimating the production function.D. Real capital stock – ki,tSame as many other firm-level survey data, our sample does not in-clude information of fixed investment, but only the fixed assets at originalpurchase prices and current depreciation. We extend the methodology ofBrandt, Van Biesebroek and Zhang (2012) to infer the firm-level real capitalstock.161We begin by estimating the nominal and real capital stock of each yearin the year in which it was first established. We first use the data of manu-facturing firms of the province of our main dataset from the Annual Surveyof Industrial Firms (ASIF) database conducted by the National Bureau ofStatistics of China to estimate the average growth rate of the nominal capi-tal stock between 2005 and 2010 (or the first year a firm first appears in oursample) at the 2-digit CIC industry level. Combined with information ofthe firms’ fixed assets at original price and their birth years, we can calculatethe firm’s nominal capital stock in the year in which it was first established.The initial real capital stock is obtain by deflating the nominal capital stockwith the Brandt-Rawski deflator.1 Then then real capital stock for 2010, thefirst year of our sample, is calculated from the firm’s initial real capital stockusing the perpetual inventory method, with a depreciation rate of 9% andthe Brandt-Rawski deflator to deflate annual investment. And the same de-preciation rate and investment deflator is then combined with the annualchange in the firm’s fixed assets at original prices to infer the firm’s realcapital stock from 2011 to 2016.1The Brandt-Rawski deflator is only available up to 2007. We use the investment deflator forthe province of our main dataset from the Chinese Statistics Yearbook after 2007.162A.3.2 Additional FiguresFigure A.2: The distributions of total assets0.1.2.3.4Density12 14 16 18 20 22Log of total assets - 2012(a) 20120.1.2.3.4Density12 14 16 18 20 22Log of total assets - 2014(b) 20140.1.2.3.4Density12 14 16 18 20 22Log of total assets - 2015(c) 2015163Figure A.3: The distributions of firm size across financial constraint ter-tilesA. The 2012 cohort(a) 2012 (b) 2016B. The 2014 cohort(c) 2014 (d) 2016C. The 2015 cohort(e) 2015 (f) 2016164

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0394831/manifest

Comment

Related Items