UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays in corporate finance, labour economics, and political economy Xu, Sheng-Jun 2017

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

Item Metadata

Download

Media
24-ubc_2017_september_xu_sheng-jun.pdf [ 4.43MB ]
Metadata
JSON: 24-1.0349034.json
JSON-LD: 24-1.0349034-ld.json
RDF/XML (Pretty): 24-1.0349034-rdf.xml
RDF/JSON: 24-1.0349034-rdf.json
Turtle: 24-1.0349034-turtle.txt
N-Triples: 24-1.0349034-rdf-ntriples.txt
Original Record: 24-1.0349034-source.json
Full Text
24-1.0349034-fulltext.txt
Citation
24-1.0349034.ris

Full Text

Essays in Corporate Finance, LabourEconomics, and Political EconomybySheng-Jun XuB.Math, University of Waterloo, 2008B.B.A., Wilfrid Laurier University, 2008A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Business Administration)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)July 2017c© Sheng-Jun Xu 2017AbstractThis thesis presents a collection of essays on the intersection of finance, labour, and politicaleconomy. In Chapter 2, I exploit the 2003 reduction in the legislative cap for the H-1B visaprogram to show that a firm’s ability to hire skilled workers affects corporate investment. U.S.firms use the H-1B program to recruit foreign skilled (college-educated) workers, and I find thatthe reduction in the cap caused a significant decrease in investment for firms that were more relianton H-1B workers as a source of skilled labour. The effect persists for several years, and is morepronounced for firms hiring workers in “industrial” occupations compared with firms hiring workersin “knowledge” occupations.The remaining essays examine how political incentives affect the policies of U.S. public-sectordefined benefit pension plans. In Chapter 3, I present novel empirical evidence that “pensiondeficits”—the difference between liability accrual rates and asset accumulation rates—are systemat-ically higher in gubernatorial election years. This electoral cycle pattern is explained by systematicdips in governmental contributions, and plans that exhibit larger electoral cycles tend to experiencedeteriorating funding levels and lower economic growth. Falsification tests, including analysis ofprivate-sector DB pension plans and unexpected Governor transitions, indicate that non-politicalfactors are unlikely to explain the documented electoral cycles.In Chapter 4, I present a theoretical model detailing how electoral incentives induce incumbentpoliticians to borrow from public pension plans in a short-sighted manner at the expense of tax-payers. Using a career concerns model framework, I show this conflict is rooted in (1) moral hazardstemming from protections that insulate employees from the costs of unfunded pension liabilities,and (2) information asymmetry stemming from the opacity of public pension plans. The modelgenerates predictions consistent with empirical findings from Chapter 3. Specifically, electoral cy-cles in pension deficits are more pronounced for states that place the burden of funding unfundedpension liabilities on taxpayers, and for states with less transparent public pension systems. Fur-thermore, pension deficits are larger during elections that are more closely contested and duringgubernatorial terms in which the incumbent remains eligible to run for reelection.iiLay SummaryThis thesis constitutes a collection of essays on finance, labour, and political economy. In Chapter 2,I study a U.S. policy shift in 2003 that limited the ability for firms to hire foreign skilled (college-educated) workers through the H-1B visa program, and show that corporate investment is negativelyaffected by regulations that restrict firms’ ability to hire skilled workers.The remaining essays examine how electoral politics affect how governments fund public sectordefined benefit (DB) pension plans. In Chapter 3, I show that U.S. states tend to reduce publicpension contributions immediately prior to elections for state Governors. In Chapter 4, I present atheoretical model that explains the incentives that lead politicians to reduce public pension fundingin an election year. The model formalizes the idea that incumbent politicians conduct “hidden”borrowing through public pension plans in order to temporarily inflate their performance.iiiPrefaceThis dissertation, including the formulation of research questions, construction of theoretical mod-els, and execution of empirical investigations, is an original intellectual product of the author,Sheng-Jun Xu. The author received valuable advice and feedback from his dissertation committeemembers on both the theoretical modelling and empirical execution components of his research.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Skilled Labour and Corporate Investment. . . . . . . . . . . . . . . . . . . . . . . . 32.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2.1 Overview of the H-1B Program . . . . . . . . . . . . . . . . . . . . . . . . . 82.2.2 Identification based on the 2003 H-1B Cap Drop . . . . . . . . . . . . . . . . 92.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3.1 Sample Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3.2 Key Variables and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . 122.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.4.1 Effect of the 2003 H-1B Cap Drop . . . . . . . . . . . . . . . . . . . . . . . . 142.4.2 Investment Dynamics and Pre-Event Trends . . . . . . . . . . . . . . . . . . 152.4.3 Long Term Impact of the H-1B Cap Drop . . . . . . . . . . . . . . . . . . . 162.4.4 Comparisons Between Occupations and Industries . . . . . . . . . . . . . . . 162.4.5 Political Lobbying and Endogeneity of Investment . . . . . . . . . . . . . . . 202.4.6 Including Non-H1B Firms in the Sample . . . . . . . . . . . . . . . . . . . . 212.4.7 Alternative Definitions of H1B use . . . . . . . . . . . . . . . . . . . . . . . 222.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23v3 Politics and Hidden Borrowing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.2 State Defined Benefit Pension Plans . . . . . . . . . . . . . . . . . . . . . . . . . . . 443.2.1 State Pension Assets and Liabilities . . . . . . . . . . . . . . . . . . . . . . . 443.2.2 Governor Discretion over State Pension Policy . . . . . . . . . . . . . . . . . 453.2.3 Who Bears the Costs of Underfunded Public Pension Plans? . . . . . . . . . 473.2.4 State Pension Policy Opacity . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.3 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.4.1 State Pension Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.4.2 State Politics Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.4.3 Other Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5.1 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5.2 Electoral Cycles in State Pension Contributions . . . . . . . . . . . . . . . . 603.5.3 Electoral Cycles in State Pension Benefit Accruals . . . . . . . . . . . . . . . 623.5.4 Electoral Cycles and Employee Benefit Protection . . . . . . . . . . . . . . . 633.5.5 Electoral Cycles and Pension Plan Opacity . . . . . . . . . . . . . . . . . . . 643.5.6 Electoral Cycles and Political Factors . . . . . . . . . . . . . . . . . . . . . . 653.5.7 Consequences of Electoral Cycles in Pension Deficits . . . . . . . . . . . . . . 663.5.8 Falsification Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.5.9 Other Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704 Distortionary Reelection Incentives and Public Defined Benefit Pension Plans 924.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 924.2 Reelection Incentives and Pension Benefits . . . . . . . . . . . . . . . . . . . . . . . 944.2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.2.2 Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 964.2.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 964.3 Reelection Incentives and Pension Contributions . . . . . . . . . . . . . . . . . . . . 994.3.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 994.3.2 Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1004.3.3 Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014.4 Empirical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1024.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107viA Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115A.1 Proof of Lemma 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115A.2 Proof of Proposition 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116A.3 Proof of Lemma 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117A.4 Proof of Proposition 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118B Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120B.1 Variable Definitions for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120B.2 Variable Definitions for Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122C Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125C.1 Occupation Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125C.2 Actuarial Valuations Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127viiList of Tables2.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.2 Top H-1B Employers in 2001 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.3 H-1B Worker Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.4 Number of Applications by Occupation and Industry Group . . . . . . . . . . . . . . 302.5 Effect of the H-1B cap Drop on Investment . . . . . . . . . . . . . . . . . . . . . . . 312.6 Quarterly Investment Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.7 Long Run Effect of H-1B Cap Drop on Investment . . . . . . . . . . . . . . . . . . . 332.8 Effect of the H-1B Cap Drop on Investment by Occupational Category . . . . . . . . 342.9 Occupational and Industry Characteristics and the Effect of the H-1B Cap Drop onInvestment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.10 Characteristics Associated with Political Lobbying and the Effect of the H-1B CapDrop on Investment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.11 Effect of the H-1B Cap Drop on Investment Based on Unrestricted Sample of H-1Band Non-H-1B Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372.12 Effect of the H-1B Cap Drop on Investment Based on Alternative Definitions of H-1BExposure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.2 Average Payroll and Pension Policies by State . . . . . . . . . . . . . . . . . . . . . . 793.3 Electoral Cycles in Pension Deficits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803.4 Dynamics of Electoral Cycles in Pension Deficits . . . . . . . . . . . . . . . . . . . . 813.5 Electoral Cycles in Pension Contribution Rates . . . . . . . . . . . . . . . . . . . . . 823.6 Electoral Cycles in State Fiscal Outcomes . . . . . . . . . . . . . . . . . . . . . . . . 833.7 Electoral Cycles in Pension Benefit Accrual Rates . . . . . . . . . . . . . . . . . . . . 843.8 Benefit Protection Strength and Electoral Cycles in Pension Deficits . . . . . . . . . 853.9 Pension Plan Opacity and Electoral Cycles in Pension Deficits . . . . . . . . . . . . 863.10 Political Factors and and Electoral Cycles in Pension Deficits . . . . . . . . . . . . . 873.11 Consequences of Electoral Cycles in Pension Deficits . . . . . . . . . . . . . . . . . . 883.12 Electoral Cycles in Private-Sector DB Pension Policies . . . . . . . . . . . . . . . . . 893.13 Unexpected Governor Changes and Pension Deficits . . . . . . . . . . . . . . . . . . 903.14 Accounting for Geographic Clustering of State Electoral Cycles . . . . . . . . . . . . 91viiiList of Figures2.1 Trends in H-1B Petitions vs. Regulatory Cap . . . . . . . . . . . . . . . . . . . . . . 242.2 Timeline of H-1B Cap Drop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3 Aggregate U.S. Fixed Non-Residential Private Investment . . . . . . . . . . . . . . . 263.1 Illustrative Example of Institutional Timeline . . . . . . . . . . . . . . . . . . . . . . 723.2 Frequency of Gubernatorial Elections (2001 to 2015) . . . . . . . . . . . . . . . . . . 733.3 Geographic Variation in Political Institutions . . . . . . . . . . . . . . . . . . . . . . 743.4 Geographic Variation in Budgetary Institutions . . . . . . . . . . . . . . . . . . . . . 753.5 Geographic Variation in Public Pension Benefit Protection Legal Regimes . . . . . . 763.6 Geographic Variation in Transparency Indicators . . . . . . . . . . . . . . . . . . . . 774.1 Model Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104ixAcknowledgementsI am extremely grateful to my advisor Hernan Ortiz-Molina, whose guidance I relied on heavilythroughout my studies and thesis work. Hernan’s generosity with his time over countless meet-ings helped me push through many obstacles and difficult moments. I am also very thankful tomy committee member Elena Simintzi, whose practical advice I would have done better to followmore closely. I also thank Ralph Winter and Robert Heinkel for agreeing to join my disserta-tion committee and providing invaluable feedback with respect to the theoretical aspects of mydissertation.I am grateful to the rest of the UBC Finance division for their constructive feedback duringworkshops and seminars. I would like to especially thank Ron Giammarino, Will Gornall, JoyBegley, Markus Baldauf, Kai Li, Murray Carlson, Lorenzo Garlappi, and Carolin Pflueger forproviding in-depth feedback during one-on-one discussions. I also thank Joy Begley, Ira Yeung andRussell Lundholm from the Accounting division and faculty members from the Vancouver Schoolof Economics, including Joshua Gottlieb, Marit Rehavi, and Nicole Fortin, for their feedback andadvice during departmental workshops.I thank my wife Sophie for her endless patience, understanding, and support over the finalyears of my Ph.D. studies. I thank my Ph.D. classmates, who provided a sounding board forgauging early stage research ideas and a mutual support network during stressful times. I thank myfriends and fellow residents of St. John’s College in providing a social support network throughoutmy studies. I thank Carl Klarner for providing an update to his political datasets available onwww.klarnerpolitics.com. Lastly, I gratefully acknowledge financial support from the Social Sciencesand Humanities Research Council of Canada (SSHRC) CGS Doctoral Fellowship and the KillamDoctoral Scholarship.xDedicationI dedicate this work to my parents, Hongqi and Chen-Wei.xiChapter 1IntroductionFinancial markets, labour markets, and the political system are inextricably linked in the moderneconomy. Issues that that affect one area likely have implications for the others. In recent years,U.S. corporations have issued frequent complaints about a shortage of skilled workers, pointing toregulatory restrictions on the free movement of workers across borders as a factor in hamperingeconomic growth. At the same time, public pension plans, which provide promises of future retire-ment benefits to public sector workers, have grown increasingly underfunded. This has eroded thefinancial health of state and local governments, and prompted media commentators to speculatethat the insolvency of public pension plans may lead to the next financial crisis.In the collection of essays presented in this thesis, I describe how regulatory constraints in theU.S. labour market affect corporate decisions, and how institutional frictions in the U.S. politicalsystem affect public pension funding decisions. In Chapter 2, the first essay, I ask whether restric-tions on the ability for firms to hire skilled workers negatively impact corporate investment. Tothis end, I exploit a 2003 regulatory change that dramatically lowered the number of foreign skilledworkers that domestic firms in the U.S. are allowed to hire through the H-1B visa program. I com-pare the investment policies of firms that were differently affected by the quota reduction, and findthat firms that were more reliant on H-1B workers experienced a relative decline in their capitalexpenditure rates. This decline persisted for several years past the regulatory change date, and wasespecially pronounced for firms hiring workers in “industrial occupations” related to science andengineering fields.In the remaining essays, I study the political economy of public sector defined benefit (DB)pension plan policies. In Chapter 3, the second essay, I ask whether political incentives influencehow governments fund public DB pension plans. I empirically document an electoral cycle inwhich state pension deficits—the difference between the rate at which pension liabilities accrue andpension assets accumulate—are systematically higher in the year preceding a gubernatorial election.Results from follow-up empirical tests support the notion that incumbent Governors attempt tobolster their reelection chances through a form of “hidden borrowing” conducted through the statepension system.In Chapter 4, the final essay, I present a theoretical model that clarifies how electoral poli-tics distort government decisions over public pension policies. The stylized model illustrates thatreelection incentives can induce incumbent politicians to behave in a short-sighted manner whenrational voters are responsible for financing unfunded pension plans but cannot perfectly monitorthe government’s policies. By constructing a model of politically-motivated pension borrowing, I1am able to provide a theoretical basis for the empirical results presented in Chapter 3.The three essays in this thesis explore common themes relating to how financial and corporatedecisions are affected by labour market frictions and the democratic political process. Chapter 2 isself-contained, while Chapters 3 and 4 are closely interrelated and reference one another. More com-prehensive discussions regarding research questions, motivations, methodologies, and contributionsto the literature are left to the introduction sections within each individual chapter.2Chapter 2Skilled Labour and CorporateInvestment: Evidence from the H-1BVisa Program“In other businesses the capacity constraint is buildings, plant or equipment. In ourbusiness. . . it’s people.”—Jeff Owens, CEO of Advanced Technology Services, Inc.12.1. IntroductionComplaints about skill shortages have become a common mantra among business leaders. A globalsurvey conducted in 2012 found that one in four CEOs claimed they were unable to pursue a marketopportunity or had to cancel or delay a strategic initiative due to “talent constraints”.2 Lamentsabout the difficulties in hiring workers are especially ubiquitous concerning skilled occupations inspecialized fields. For example, a 2013 industry report estimates that a shortage of skilled workersin the oil and gas sector put $100-billion worth of industrial investment projects at risk,3 whilea separate industry survey found that 47% of Fortune 1000 firms reported business growth beingimpeded by unfilled jobs in technical occupations.4 These examples suggest that difficulties inhiring skilled workers can meaningfully inhibit demand for capital. Considering the fundamentalroles of capital and labour in economic production, it is surprising how little is known about howconstraints on firms’ access to skilled labour affect corporate investment.Corporate finance is largely concerned with how imperfect financial markets affect firms’ abilityto pursue attractive business opportunities. However without also considering the imperfections oflabour markets, one cannot gain a complete picture of how firms make decisions. Zingales (2000)encouraged finance researchers to investigate the increasingly important interplay between humanand physical capital to better understand the decisions of modern human-capital-intensive firms. Inthis essay, I study how labour market frictions affect demand for physical capital, with the broader1 Weitzman, Hal “Skills gap hobbles US employers” Financial Times 13, Dec. 2011.2 “15th Annual Global CEO Survey 2012” PWC Report. Web. www.pwc.com/ceo3 Hirtenstein, Anna and Shankleman, Jess “Skilled worker shortage threatens US$100-billion in U.S. energy projects”Bloomberg News 7 Mar. 2013.4 Bayer Corporation. “The Bayer Facts of Science Education XVI: US STEM Workforce Shortage—Myth or Reality?Fortune 1000 Talent Recruiters on the Debate.” Journal of Science Education and Technology 23 (2014): 617-623.3goal of advancing our understanding of how capital and labour markets interact.I study how restrictions on firms’ ability to hire skilled labour (defined as college-educated work-ers) impacts corporate investment decisions. It is not immediately obvious whether firms shouldincrease or decrease investment in response to an adverse skilled labour supply shock. In line withthe anecdotal accounts of hiring difficulties creating a drag on investment, there is a long litera-ture, starting with Griliches (1969), that explores the idea that skilled labour, relative to unskilledlabour, is more complementary to capital. In contrast, Autor (2003) argues that the boundarybetween labour and capital generally moves in the direction of capital taking over tasks formerlyperformed by labour, even in traditionally skilled occupations. For instance, recent advances inautomation and artificial intelligence may displace skilled workers in the modern economy, just asfactories and assembly lines once displaced skilled artisan workers in the distant past.My empirical tests are designed to shed light on whether an adverse shock to the supply ofskilled labour induces firms to cut investments (i.e. the complementarity hypothesis), as widelyclaimed by business leaders, or whether firms opt instead to increase capital investment in order tomitigate the shock through substituting one factor of production for another (i.e. the substitutionhypothesis). It is also possible that such shocks have no detectable effects on firms’ investmentdecision, especially if the shocks are small and thresholds for capital investments are high. Underthe null hypothesis, capital investment should remain unchanged after a shock to the supply ofskilled workers.While my research aims to improve our understanding of the fundamental relationship betweenlabour and capital, it also has practical implications from a policy perspective. The economic impactof employment-based immigration has long been a politically contentious subject, with protectionistadvocates arguing that the inflow of foreign workers negatively impacts the job opportunities andwages for domestic workers, while immigration reform advocates argue that restrictions on foreignworkers hurt the competitiveness of domestic businesses and limit their growth potential. However,the effect of immigration policy on capital investment is often overlooked by policymakers andlabour economists who are primarily concerned with employment and wage outcomes. My empiricalwork uncovers evidence that restrictions on skilled immigration do in fact impact capital growth atthe firm level, with accompanying estimates that quantify the economic significance of this impact.To establish the causal impact of changes to skilled labour supply on firm investment, onefaces the classic identification challenge of disentangling supply shocks from shifts in demand. Toovercome this challenge, I exploit an arguably exogenous change in skilled labour supply created bya change to the regulatory limit on the total number of foreign college-educated workers allowed tobe hired in the United States under the H-1B visa program. Specifically, the nation-wide numberof visas issued each year is limited by a regulatory cap, and a significant reduction in the cap tookeffect in 2003. The president of the American Immigration Lawyers Association predicted at thetime that “[the] immediate impact of not being able to obtain an H-1B approval. . . is that projectsare put on hold, capital expenditures are deferred and lives are thrown into chaos.”55 Gamboa, Suzanne “Limit reached for applications for skilled worker visas” Associated Press, 17 Feb. 2004.4The H-1B visa cap drop provides an ideal setting for my empirical investigation. As argued byBorjas (2001), immigrants constitute a relatively elastic supply of labour that serves to “grease thewheels” of the domestic labour market in large part due to their high intrinsic degree of mobility.This means that the relatively inelastic supply of domestic skilled workers often does not providea sufficient source of substitute labour, especially given the secular trend over the past severaldecades, as documented by Molloy et al. (2014), of declining mobility amongst domestic workerswith college educations. Therefore, an artificial restriction on the elastic supply of skilled immigrantworkers should be acutely felt by domestic firms facing an inelastic supply of domestic workers.My empirical approach is to compare changes in corporate investment between firms that weredifferentially affected by the 2003 cap drop. Firms which relied more heavily on H-1B workersex-ante as a source of skilled labour were more exposed to H-1B policy shocks and therefore shouldhave been more intensely affected by the cap drop. Accordingly, I set up a difference-in-differences(DD) estimation approach in which the 2003 H-1B cap drop forms the “treatment” event, andthe intensity of firm’s exposure to H-1B policy is measured based on firms’ ex-ante hiring rates ofH-1B workers during 2001. I compare quarterly investment rates for firms with differing levels ofH-1B exposure from one year prior to the cap drop (2002) and one year following the cap drop(2004). Under the complementarity hypothesis, firms more exposed to H-1B policy shocks shouldexperience declines in investment relative to firms less exposed to H-1B policy shocks, while thereverse should hold under the substitution hypothesis.In line with the complementarity hypothesis, the results show that the 2003 H-1B cap dropcaused more intense employers of H-1B workers to reduce capital expenditures relative to lessintense employers of H-1B workers. The results are both statistically and economically significant,implying that a firm in the 75th percentile of H-1B usage experienced a 10.1 percentage point dropin their quarterly investment rate relative to a firm in the 25th percentile of H-1B usage. Thiscorresponds to a 7 percent decline relative to the sample mean for the investment rate. I find thatthis result is robust to a variety of alternative definitions of firms’ H-1B usage rates.Crucially for identification, I find strong evidence in support of parallel trends, the key assump-tion behind the validity of my DD estimation approach, in that firms with differing exposures toH-1B policy did not experience diverging or converging trends in investment policy prior to the2003 event. This suggests that my results are not driven by pre-existing differences in investmentopportunity trends, and also that the event was not anticipated by firms prior to 2003. I also findthat the dynamics of quarterly investment from 2002 to 2004 generally conform to the timeline ofpolitical events surrounding the H-1B cap drop.Next, I examine whether the declines in investment for firms more reliant on H-1B workerspersists beyond the immediate one-year window used in my benchmark analysis. It is possiblethat complementarity holds in the short run, while over longer horizons, firms can substitute forthe restricted foreign H-1B workers using domestic workers—in which case the adverse effect oninvestment should attenuate over time—or to alter their production technology in substitutingcapital for labour—in which case the reverse substitution effect may occur. However, I find that5the impact on investment strongly persists for at least four years following the 2003 event, whichsuggests that it is difficult for firms to substitute for this elastic supply of skilled foreign labour.A potential concern regarding my empirical approach is that the 2003 H-1B cap drop came aboutdue to declining lobbying efforts from firms suffering declining investment opportunities. I mitigatethis concern by showing that the documented effect on investment was not confined to the high-tech sector, which was strongly associated with political lobbying on H-1B policy issues. I find nosignificant differences between industries inside and outside of the high-tech sectors. Furthermore,I find no significant differences between large firms, which tend to be more politically active inlobbying, and small firms, which tend to be less politically active.I also explore how the complementarity between skilled labour and capital investment dependson the specific occupational role of the labour and the specific industrial application of the capital. Ifind capital to be complementarity to workers in traditional “industrial” roles such as scientists andengineers, who are closely associated with working with physical capital, but not to workers in pure“knowledge” occupations such as computer programmers and accountants, who are more closelyassociated with working with ideas rather than physical machines.6 With respect to industries, Ifind limited evidence that the complementarity effect is stronger for manufacturing firms, in whichcapital expenditures tend to be directed towards more variable inputs such as equipment andmachinery, and weaker for service sector firms, in which capital expenditures tend to be directedtowards fixed overhead items such as buildings and offices.The literature on corporate investment is extensive. Research has shown investment to beimpacted by agency frictions (Panousi and Papanikolaou, 2012), information constraints (Foucaultand Fresard, 2014), behavioural biases (Malmendier and Tate, 2005), and real options (Carlsonet al. (2004), Bloom et al. (2006)). One of the most important areas of the investment literaturefocuses on the role of financial constraints. Starting with Fazzari et al. (1988) and revitalized byKaplan and Zingales (1997), researchers have long searched for evidence that financial constraintslimit the ability of firms to respond to investment opportunities. Recent works by Lemmon andRoberts (2010), Duchin et al. (2010), and Almeida et al. (2012) have empirically documented theadverse impact of financial sector shocks on corporate investment. My contribution to this literatureis to document how constraints on the supply of human capital, rather than financial capital, canaffect investment policy.My work also relates to the emerging body of finance research related to labour and employment.Previous works have explored both how labour market considerations affect financial and governanceoutcomes (Atanassov and Kim (2009), Chen et al. (2011), Simintzi et al. (2015), Matsa (2010),Agrawal and Matsa (2013)), as well as how financial factors affect employment outcomes (Benmelechet al. (2011), Mian and Sufi (2014)). While much of this literature looks at the impacts of unions,which is often associated with less-educated blue collar workers, I explicitly focus on highly-educatedworkers and their relationship to capital.6 For instance, software developers are tasked with the implementation of concepts and models, rather than theconstruction of physical machines.6My focus on skilled labour is motivated by the literature, starting with Griliches (1969), thatexplores the premise that skilled labour is relatively complementary to capital when compared tounskilled labour. Much of the evidence in this literature is descriptive and uses aggregate data, withno strong identification of causation at the micro level (DiNardo and Pischke, 1997). Lewis (2011)provides the most closely-related work to this essay, finding evidence that automation technologyis more complementary to medium-skilled workers relative to low-skilled workers. While his workexamines how low-skilled immigrants affect the technological mix used by manufacturing plants,I focus instead on high-skilled immigrants and their effect on total capital expenditures for firmsacross a broader set of industries. Furthermore, my dataset allows me to explore cross-sectionaldifferences with respect to occupations and industries, and my quasi-experimental setting allowsme to investigate the time horizon of the effect on investment.The H-1B visa program itself is increasingly attracting attention from academic researchers.Most research in this area focus on the effects of the program on the employment and wages ofdomestic workers (Lofstrom and Hayes (2011), Kerr et al. (2015b), Peri et al. (2014), Doran et al.(2014)), while others study the effect on patenting and innovation (Lewis et al. (2015), Kerr andLincoln (2010)). Rather than investigate the effect on domestic workers, I focus on the outcomeof capital investment, which is often neglected by labour and policy researchers working in thefield. In addition, I provide evidence on the horizon of the effect on investment, which may be animportant consideration for policymakers.Ghosh et al. (2014) and Ashraf and Ray (2016), who study the impact of skilled immigrant work-ers on innovation and productivity, also use the 2003 H-1B cap reduction as a quasi-experimentalsetting. However, they use data on labour condition applications (LCAs), which are noisy measuresof H-1B usage.7 In contrast, I use data on firms’ actual petitions to hire H-1B workers, which pro-vides additional detailed data on workers’ occupations and educational backgrounds. In addition,both Ghosh et al. (2014) and Ashraf and Ray (2016) use annual firm data, while I use quarterlydata in order to better isolate the timing of the policy shock. Nevertheless, my finding of a negativeeffect on firm investment is consistent with the negative effect on innovation and productivity thatthey document.The remainder of the essay is organized as follows. Section 2.2 provides a description of the2003 regulatory change in the H-1B visa program and how I exploit this quasi-experimental eventas part of my identification strategy. Section 2.3 describes the data and accompanying summarystatistics. Section 2.4 reports my main results as well as follow-up findings. Section 2.5 concludes.2.2. Empirical StrategyI base my identification strategy on the sharp 2003 drop in the legislative cap of the H-1B visaprogram. In this section, I explain the importance of the H-1B visa program as a source of skilledworkers for U.S. firms. I then briefly describe the history of the annual cap restricting the number7 Filing an LCA is a necessary step towards hiring an H-1B worker, but does not necessarily lead to a H-1B hire.7of foreign workers that can be hired through the program. Finally, I describe the sharp drop in theannual cap drop that occurred in 2003, and how I exploit this event as part of my identificationstrategy.2.2.1. Overview of the H-1B ProgramEstablished by Congress through the Immigration Act of 1990, the H-1B visa program allowsemployers to hire skilled foreign workers to legally work in the U.S. on a temporary basis. Visas areissued for a period of up to three years, with the possibility of a one-time extension for an additionalthree years. According to the U.S. government’s website, the program’s stated intent is to allowemployers to fill vacant positions for which “the nature of the specific duties is so specialized andcomplex that the knowledge required to perform the duties is usually associated with the attainmentof a bachelor’s or higher degree”.8 Since the late 1990s, H-1B workers have largely consisted ofworkers in technical occupations related to science and technology fields.9The H-1B visa program provides an economically significant source of skilled workers for U.S.firms. In 2003, U.S. employers hired 130,497 new H-1B workers, a substantial number when com-pared against the 442,755 new domestic bachelor’s degree holders in science and technology disci-plines.10 The H-1B program is also comparable in magnitude to that of employment-based legalpermanent residents (LPRs); LPRs are capped at 140,000 per year, and each foreign country isfurther limited to a maximum 7% of total worldwide admissions. Since the H-1B program placesno such per-country limit, it is often the only channel for firms to hire workers from countries withlarge emigrant populations such as India and China.11There is a legislative cap placed on the total number of H-1B visas issued per year, which appliesto new H-1B hires and not to extensions of existing visas. This cap has fluctuated throughout thehistory of the H-1B program, starting with 65,000 in 1992 and reaching as high as 195,000 during theearly 2000s. The cap was first raised to 115,000 through the American Competitive and WorkforceImprovement Act (ACWIA) in 1998, and then raised again to 195,000 in 2000 by the AmericanCompetitiveness in the Twenty-First Century Act (AC21). The sharpest change came in 2003,when the cap reverted from 195,000 to 65,000 upon the expiration of the AC21.12 This last eventforms the quasi-experimental event behind my identification strategy.8 More than 98% of all approved applicants in 2004 possessed at least a bachelor’s degree, with the remaining 2%coming from special exempt occupation such as fashion modelling.9 In 2003, the top five most common occupations for H-1B workers were found in systems analysis and programming(33.5%), college and university education (7.8%), accountants and auditors (4.8%), electrical/electrical engineering(3.9%), and computer-related occupations (3.1%), according to the USCIS Characteristics of Specialty OccupationWorkers (H-1B): Fiscal Year 2004.10 See Appendix Table 2-18 for “Science and Engineering Indicators 2012” by the National Science Foundation, availableonline at https://www.nsf.gov/statistics/seind12/appendix.htm.11 A 2015 National Foundation for American Policy policy brief reports that skilled workers from high-population coun-tries face expected wait times of 6-10 years in obtaining permanent residency. Unsurprisingly, India overwhelminglysupplied the largest share of new H-1B workers in 2003 at 46%, with China coming in at second at 8.7%.12 All changes to the H-1B cap up to this point focused exclusively on skilled immigration were not accompanied bypolicy changes relating to low-skilled immigration (Kerr et al., 2015a).82.2.2. Identification based on the 2003 H-1B Cap DropAs shown in Figure 2.1, the 2003 H-1B cap drop from 195,000 to 65,000 resulted in a bindingconstraint on H-1B hires. Before the drop, employers were effectively assured of securing visas,as the 105,185 initial petitions submitted in 2002 fell well below the higher cap. In 2004, thenewly-lowered cap of 65,000 was well above the 116,927 initial petitions submitted by employers.Consequently, employers found themselves rationed in their ability to hire H-1B workers due to thenewly binding cap.The main idea behind my empirical strategy is that the H-1B cap drop resulted in more severehiring constraints for firms that were ex-ante more reliant on H-1B workers as a source of skilledlabour. Accordingly, I employ a difference-in-differences (DD) estimation framework in which the2003 H-1B cap drop marks the onset of the treatment, and the intensity of treatment is determinedby the intensity at which firms employed H-1B workers in 2001. This follows the standard DDspecification in which treatment is continuous rather than binary as in Card (1992):CapExit = αi + λt + δ ·H1B usei · Postt +Xit−1β + it (2.2.1)where i indexes firms and t indexes time periods. CapEx represents capital expenditures, α rep-resents a firm-specific dummy, λ denotes a time-specific dummy, Post represents a dummy for the“post-treatment” period, X represents a vector of lagged firm-level control variables,  representsthe error term, and H1B use represents the intensity at which firms hired H-1B workers in 2001.The coefficient of interest is δ, which should be zero under the null hypothesis, negative under thecomplementarity hypothesis, and positive under the substitution hypothesis. This standard DDspecification controls for any time-invariant firm-level factors affecting investment, as well as forany time-specific shocks common to all firms.I define the post-treatment period to consist of the four quarters in 2004, as illustrated inFigure 2.2. Although the lower cap officially took effect in October 2003, firms were able tocontinue filing petitions for hiring new H-1B employees until several months later. It was not untilFebruary of 2004 that the United States Citizenship and Immigration Services (USCIS) announcedthat it would no longer accept new H-1B petitions for the coming fiscal year, marking the point atwhich firms were first subject to hiring restrictions due to saturation of the cap.13 Therefore, theimpact of the cap drop should have been fully felt by the beginning of 2004.Next, I define the pre-treatment period to consist of the four calendar quarters in 2002, alsoshown in Figure 2.2. Although the AC21 initially set the cap at 195,000 for a temporary period ofthree years, the previous trend of a rising cap created a reasonable expectation of permanence (Katoand Sparber, 2013).14 Media reports suggest that the business community had expectations of acontinued higher cap in early 2003, as the trade publication CIO Magazine reported in January that13 The government’s fiscal year starts in October and ends in September. The petition “window” never closed for twoyears preceding the 2003 cap drop.14 Consider that the previous cap increase from the ACWIA was also originally set to expire after three years beforebeing extended and raised by the AC21 in 2001.9“most expect the introduction of a bill that will either keep the cap high or eliminate it altogether”.15It was not until February 2003 that the congressional chairman of the House Judiciary Committeeindicated that the cap would revert back to 65,000 the following year.16 Therefore, the sharp H-1Bcap drop remained largely unanticipated by firms during the 2002 pre-treatment interval.In my main regression analysis, I restrict the sample to the pre-treatment and post-treatmentperiods, while dropping the 2003 calendar year—i.e. the “legislative shift” period. This is so thatδ captures the full extent of the effect of the H-1B cap drop on investment, and does not reflectany partial effects as expectations about the pending H-1B cap drop gradually built up throughout2003. In later tests, I include the full time series, including the legislative shift period, in order toexamine the quarter-by-quarter dynamics of investment around the event to verify that they alignwith the political timeline.As is the case with all natural experiments involving legislative action, there is the concern thatthe 2003 H-1B cap drop arose endogenously due to shifts in forward-looking economic demand. Inparticular, one must be wary of the possibility that declining expectations about future investmentopportunities resulted in reduced lobbying by firms looking to maintain a higher cap. In particular,the 2003 H-1B cap drop has been partly attributed to reduced lobbying efforts by informationtechnology firms following the dot-com crash that occurred in 2000-2001.Figure 2.3(a) reveals that the 2003 H-1B cap drop came at a time of growing rather than de-clining aggregate investment.17 Figure 2.3(b), which displays the same data series at quarterlyintervals, shows the same upward trajectory and also uncovers a sharp dip in aggregate investmentduring the latter parts of 2003 and early parts of 2004, precisely when the H-1B cap drop beganto take effect. To mitigate concerns that H-1B dependent firms faced declining investment oppor-tunities following the dot-com crash, I check to make sure investment rates between high H1B useand low H1B use firms did not diverge in the 2002 pre-treatment period.I further address concerns about political endogeneity by investigating whether the effects oninvestment are confined to sectors most closely associated with political lobbying on H-1B policy.As noted by the press around the time of the cap drop, firms in the high-tech sector were the mostsignificant political lobbyists on the issue of H-1B policy.18 If my results are driven by the declininginvestment opportunities of politically-active firms, for instance, then one would expect the relativeinvestment declines to be stronger in firms in the high-tech sector. I show this not to be the case,using various definitions of high-tech industries.Finally, plausibly exogenous political factors played a large role in influencing the regulatoryshift towards to a lower cap. In particular, the September 11, 2001 terrorist attacks created a fearof foreigners being admitted into the country. This created a political climate in which proponentsof more open immigration policies found it more difficult to influence politicians.19 In fact, the15 Overby, Stephanie “Cap on H-1B Visas Brought to Congress” CIO Maganize 1 Jan. 2003.16 “U.S. to tighten H1B visa norms” The Hindu Business Line 21 Feb. 2003.17 Note this series is taken from Bureau of Economic Analysis data, and represents aggregate investment activity acrossall sectors, not just Compustat firms in my sample.18 Sickinger, Ted “Congress Lowers Visa Cap for Foreign Tech Workers” The Oregonian, 29 Sep. 2003.19 In its September 2003 10-K filing, telecommunications company Wireless Facilities Inc. noted that “[immigration]10Department of Homeland Security, specifically formed in response to the 9/11 attacks, assumeddirect oversight of approving H-1B petition in March of 2003, taking over the mantle from thenewly-defunct Immigration and Naturalization Services agency. Combined with the non-decliningtrend in investments, the circumstances during this period make a strong case for the exogeneityof the H-1B cap drop with respect to firms’ changing investment opportunities.2.3. Data2.3.1. Sample ConstructionMy sample consists of firm-quarter observations from industrial firms, excluding utility firms (SICcode between 4900 and 4900), financial firms (SIC code between 6000 and 6999), and public sectorfirms (SIC code over 9000). All accounting and financial data come from the merged CRSP-Compustat Fundamentals Quarterly file. For my main results, I limit my sample to four quartersin the pre-treatment period (2002Q1-2002Q4) and four quarters of data in the post-event period(2004Q1-2004Q4).I employ data selection criteria standard in the investment literature (Almeida et al. (2004),Duchin et al. (2010), Almeida et al. (2012)) by discarding firms for which the total market capital-ization is less than $50 million as of the last quarter in the pre-treatment period (2002Q4). Thisserves to exclude the smallest firms with volatile accounting data and skewed investment patterns,resulting in a sample of 23,644 firm-quarter observations corresponding to 3,600 distinct firms.I further restrict the sample to firms that have submitted at least one H-1B petition during the2001 calendar year (i.e. “H-1B firms”), which results in a sample of 9,921 firm-quarter observationscorresponding to 1,395 distinct firms that make up 36.24% of the total market capitalization for allpublicly-listed companies listed on the Compustat quarterly database as of 2002Q4. Having H-1Bpetition-level data for all firms in the sample allows me to conduct cross-sectional heterogeneity testsbased on the observed characteristics of H-1B workers across firms. Furthermore, this restrictionensures that all firms in the sample have domestic operations in the U.S. and are potentially affectedby H-1B policy.Data on H-1B usage comes via a Freedom of Information Act (FOIA) request filed with theUSCIS. The data contains information from petitions submitted by firms to the USCIS duringthe final step of the H-1B approval process, and includes details about the sponsoring employer,prospective H-1B employee, and job position, including employee age, education level, job wage,and occupational category. I match the USCIS data to Compustat firms via company names. Dueto spelling mistakes and alternate variations of firm names in the USCIS data, I employ a matchingprocedure that incorporates fuzzy string matching as well as manual inspection of matches.20policies are subject to rapid change, and these policies have become more stringent since the terrorist attacks onSeptember 11, 2001.”20 Specifically, I first standardize the firm names found in both the Compustat and FOIA files, and employ a fuzzystring matching algorithm to arrive at a list of potential matches. I then inspect the list of potential matches to filterout the false positives.11By restricting the sample to only H-1B firms, my benchmark results are potentially subjectto selection bias. In order to address this, I include tests based on the expanded sample of firmsincluding “non-H-1B firms”—i.e. firms that did not submit a petition to hire H-1B workers in 2001.In particular, I show that non-H-1B firms experienced similar investment patterns to those of H-1Bfirms with marginally low exposure to H-1B policy, which suggests that the investment opportunitytrends of H-1B firms are similar to the investment patterns of non-H1B firms absent the effect of theH-1B cap restriction. The expanded sample includes an additional 13,723 firm-quarter observationscorresponding to 2,205 non-H1B firms.2.3.2. Key Variables and Descriptive StatisticsMy primary measure of H-1B usage intensity, H1B use, is defined as the total number of initialH-1B petitions filed during the 2001 calendar year by a given firm, scaled by the average numberof employees employed by the firm during the same interval. I use applications filed in 2001 inorder to create an ex-ante measure with respect to my sample period, as shown in Figure 2.2.This mitigates the possibility of H1B use being correlated with changing investment opportunitiessurrounding 2003. The distribution for H1B use exhibits a sizable degree of positive skewness, andtherefore it is winsorized at the 2% level at the upper tail.21The dependent variable in my analysis is CapEx , which is defined as the ratio of quarterlycapital expenditures to lagged total assets following the conventions of Baker et al. (2003) andRauh (2006).22 I include lagged control variables commonly found in the investment literature.These include Tobin ′s Q , defined as the ratio between the market value and book value of assets,ln(Size), defined as the natural log of total assets, Cash Flow , defined as the ratio between quarterlynet income before depreciation and total assets, Cash Holdings, defined as the ratio between cashholdings and total assets, and Leverage, defined as the ratio between long-term debt and totalassets. A detailed definition of these variables can be found in Appendix B.1. To mitigate theeffects of outliers, I winsorize each variable listed above at the 1% level at both tails. I furtherbound Tobin ′s Q to be no larger than 10, following Baker et al. (2003).Table 2.1 presents the descriptive statistics for the variables defined above. Panel A displays thedescriptive statistics for H-1B firms, while Panel B displays the descriptive statistics for non-H-1Bfirms. Note that in Panel A, H1B use exhibits positive skewness even after winsorization. In laterrobustness checks, I employ alternative measures of firm-level H-1B usage, including non-parametricmeasures in order to address potential concerns regarding skewness.In comparing Panel A and Panel B, it is apparent that H-1B firms are on average larger thannon-H1B firms, with lower average investment rates and leverage ratios but higher average Tobin’sQ, profitability, and cash holdings.23 This is consistent with anecdotes of large firms in high-growth21 The main results remain qualitatively unchanged if I winsorize by 1% at both tails like the other variables.22 Because capital expenditure is reported on a year-to-date basis in quarterly financial statements, the previous quarter’scapital expenditure is subtracted from the current quarter’s capital expenditure (capxy) for fiscal quarters 2, 3, and4 to arrive at the quarterly figure.23 The differences in means are significant for all variables.12technology industries being major employers of H-1B workers.24 Panel A in Table 2.2 shows a listof the top 10 H-1B employers (in terms of total petitions submitted to the USCIS) found in mysample; the list includes large technology companies such as Microsoft, Oracle, and Intel. PanelB in Table 2.2 shows that the firms most dependent on H-1B workers are not quite as large, butare also found in high-growth sectors such as telecommunications and high-tech manufacturing. Inlater tests, I include both H-1B and non-H1B firms in order to address concerns about selectionbias stemming from focusing only on H-1B firms.Table 2.3 presents statistics about the characteristics of the H-1B workers hired by firms in thesample during 2001. These characteristics include Wage, the listed wage of the H-1B employee,Occ Wage, the national average wage corresponding to the H-1B employee’s occupations accordingto BLS data, Occ Wage Growth, the national average wage net growth rate corresponding tothe H-1B employee’s occupation, Age, the age of the H-1B employee, Grad , a dummy variable forwhether the H-1B employee possesses a graduate-level education, and HQ State, a dummy variablefor whether the H-1B employee works in the same geographic state as firm headquarters. Detaileddefinitions for all worker characteristic variables can be found in Appendix B.1. All variablesare constructed using USCIS petition-level data, with the exception for HQ State, which usesadditional data that is publicly-available on the Department of Labor website.25Table 2.3 presents descriptive statistics about the characteristics of the top occupational groupsfound across the H-1B petitions in the sample. The occupational categories are taken directlyfrom the USCIS Dictionary of Occupational Codes, and a listing of more detailed occupationalsubcategories can be found in Appendix C.1.26 Panel A displays the mean values for the variousworker characteristic variables across occupational categories, while Panel B presents correlationsbetween the same set of variables. The results reveal that Computer workers constitute the youngestgroup of workers and are amongst the highest paid in terms of actual wages as well as being in thehighest-paid and highest wage growth occupations. Meanwhile, Science workers are among the leastwell-paid workers with the lowest rate of occupational wage growth, but also constitute the mostwell-educated category of workers. Science workers are also the most likely to be working in closeproximity to firm headquarters compared to the other occupations. Panel B from Table 2.3 showhigh correlations between wage and age as well as between occupational wage and occupationalwage growth.Table 2.4 presents a breakdown of H-1B petitions by occupational and industry group.27 Thevast majority of H-1B employees are found in Computer, Engineer, Science, Admin, and Manage-24 This could be due to the fact that large firms are better able to afford overhead costs associated with immigrationlawyer fees and overcoming regulatory hurdles in the H-1B hiring process.25 Before submitting petitions to the USCIS, firms must first submit a Labor Condition Application (LCA) to theDepartment of Labor attesting that the positions for prospective H-1B workers meet certain regulatory requirements.LCA data is are available online, and contain information on the geographic location of the prospective H-1B workers’work locations. I match application-level LCA data to firms in a similar manner as the petition-level USCIS data.26 The occupational codes are available online at http://www.uscis.gov/files/form/m-746.pdf. The physical sciences(Mathematics And Physical Sciences) and life sciences (Life Sciences) subgroups have been combined under a single“Sciences” group.27 Industry is defined at the SIC Division level. See note on next page for more details on this classification.13ment occupations, with Computer occupations accounting for the largest group by a wide margin.In terms of a breakdown by industry, the vast majority of firms operate within the Service andManufacturing industries. Manufacturing firms tend to employ the largest share of H-1B work-ers across the different occupational categories, with the exception of Computer workers, who aremore than twice as likely to be found in the Service rather than Manufacturing sector. Overall,the top H-1B occupations are found across all industries, with the exception of Construction andAgriculture,Forestry , which contain relatively few firms.The descriptive statistics are generally consistent with anecdotes of the H-1B program beingan important source of young IT workers from India in the software and other IT-related serviceindustries. However, the summary tables also show that manufacturing firms hire a significantnumber of H-1B workers, particularly in more traditional technical fields related to science andengineering. I later investigate heterogeneity across worker-level occupational characteristics aswell as across industry classifications. This is done in order to investigate how complementarityor substitutability between skilled labour and capital depends on the nature of the productionfunction, as well as to address various concerns about omitted variables driving my results.2.4. Results2.4.1. Effect of the 2003 H-1B Cap DropI run OLS regressions according to the Eq. 2.2.1 to estimate the impact of the 2003 H-1B capdrop on investment. Table 2.5 presents the results under various sub-specifications on the sampledescribed in the previous section. The coefficient estimate for the variable of interest H1B use×Postis negative and statistically significant under all specifications, implying that the H-1B cap dropinduced a relative investment decline in firms that were ex-ante more dependent on the program,in support of the complementarity hypothesis.Column (1) starts off with the most basic specification. This specification does not include anycontrol variables, but does include firm fixed effects, which control for time-invariant factors, andyear-quarter fixed effects, which control for time-specific macro shocks. The inclusion of controlvariables Tobin ′s Q , Cash Flow , ln(Size), Leverage, and Cash Holdings through columns (3)-(6)leaves estimates with slightly larger magnitudes and stronger statistical significance. The specifi-cations represented in columns (2), (4), and (6) further include industry-year-quarter fixed effects,where industry is defined at the SIC Division level.28 This addresses potential concerns that themain results are driven by time-varying industry-level demand shocks.Column (6) presents the most robust specification, which includes the full set of control variablesand firm and quarter fixed effects. The economic magnitude here is significant: the coefficient28 SIC Divisions consist of broad categories of economic activity identified in the SIC manual, corresponding to ranges oftwo-digit SIC codes (Kahle and Walkling, 1996). I use a coarse industry classification since finer industry definitionsresult in many industries for which my sample only contains a single firm, in which case all variation is subsumedby the industry-time fixed effects. In unreported tests, I use 2-digit, 3-digit, and 4-digit SIC classifications and theresults remain virtually unchanged.14estimate of -0.101 implies a firm in the 75th percentile of H1B use suffers 10.1 percentage pointdrop in their quarterly investment rate relative to a firm in the 25th percentile of H-1B use. Fora firm with the sample mean value for CapEx , this corresponds to a 7 percent relative decline inproportional terms. For a firm with the sample mean value for total assets, this corresponds to a$606,679 relative drop in capital expenditures in dollar terms.292.4.2. Investment Dynamics and Pre-Event TrendsThe validity of my DD estimation approach hinges on the “parallel trends” assumption, in thatfirms’ ex-ante H-1B exposure are not correlated with trends in investment policy in the lead up tothe 2003 H-1B cap drop. I provide evidence in support of this by estimating the following OLSregression:CapExit = αi + λt +∑τ<2004Q1δτ ·H1B usei · τt + δ ·H1B usei · Postt +Xit−1β + it (2.4.1)where τt represents a dummy variable that takes on a value of one when τ is equal to t, and τ takeson values from 2002Q2 to 2003Q4 inclusive.30 The sample consists of all quarters between 2002Q1and 2004Q4; data from 2003 is added back to the sample in order to analyze the full dynamicsof firm investment, including the legislative shift period. For the parallel assumption to hold, thecoefficients δτ should not exhibit any significance prior to 2003.The results are presented in Table 2.6, and provide support for the parallel trends assumption.Both specifications contain the full set of control variables as found in column (6) of Table 2.5,while including different sets of fixed effects. Under both cases, δt is statistically insignificant forall τ < 2003Q3, which means that, relative to the baseline period of 2002Q1, firms with differinglevels of H1B use did not experience differing trends in investment during the pre-treatment period.This implies that my results are not driven by differing trends in investment opportunities alreadyin place during the pre-treatment period, and that the pending cap drop was not anticipated byfirms prior to 2003.The timing of the detected effect also conforms to the legislative timeline outlined in Section 2.2.The first statistically significant coefficient in column (2) comes at the third quarter of 2003, meaningthat it took two quarters following the February 2003 congressional announcement before firmsimplemented changes in investment policy. This is consistent with the political timeline of thelegislative shift period: firms still had the opportunity to lobby for an extension of the higher capfollowing the congressional announcement, and therefore may have decided to refrain from major29 In a set of unreported tests, I estimate the same regression after collapsing the data along the time-series into apre-treatment mean and a post-treatment mean for CapEx and all control variables. This is done, based on therecommendations from Bertrand et al. (2004), in order to overcome concerns of serially-correlated standard errorsresulting in excessive rejection of the null hypothesis. The results are qualitatively similar when the data is collapsed,and remain statistically significant at the 1% level.30 The dummy variable for 2002Q1 is omitted since it is subsumed by the H1B use level term and intercept. Note thatI do not extend my sample back beyond 2002, as the September 11, 2001 terrorist attacks also impacted immigrationpolicy and therefore may confound the results.15shifts in investment policy until the lower cap became more certain. It was not until late Septemberof 2003 that the lower cap was officially finalized following a final congressional hearing on thesubject, which coincided with the first significant negative coefficient corresponding to 2003Q3.2.4.3. Long Term Impact of the H-1B Cap DropWhile the 2003 H-1B cap drop was of a permanent nature, the baseline results presented in Table 2.5are based on post-treatment investment policies at only a one year horizon.31 However, it is possiblethat the effect on investment attenuated gradually over time, as the supply of potential domesticskilled worker substitutes becomes less inelastic in the long-run. The negative effect on investmentmay even eventually reverse if firms gradually adjust their production technology to replace labourwith capital over longer horizons, as described by Autor (2003). Therefore, I extend the horizonpast 2004 to investigate whether the effect of the H-1B cap drop persists beyond the initial year byexpanding the sample to include data up to 2007.32 I collapse the quarterly data along the timedimension by calendar year and run the following regression:CapExit = αi + λt +2007∑k=2004δk ·H1B usei ·Year k t +Xit−1β + it (2.4.2)where Year k t represents a dummy variable that takes on a value of one when k is equal to t. Here,k takes on values from 2002 to 2007 inclusive, which allows me to estimate the effect on investmentfor four years following the 2003 cap drop.The results are presented in in Table 2.7, and reveal that the effect on investment is indeedpersistent for all four years following the 2003 cap drop, as δk remains negative and statisticallysignificant for all k > 1. Furthermore, they do not shrink in magnitude over time—in fact theyseem to grow larger. This is consistent with the anecdotal evidence of the increasing difficulties thatfirms faced in securing H-1B visas for their workers in the years following the cap drop. Over all,the results suggests that persistent rationing of foreign skilled workers is not gradually mitigated bydomestic replacements, and that firms are not able to adjust their production technology to directlyreplace workers with capital at the time scale examined here. Nevertheless, it is still possible thatreversals can occur in the very long run.2.4.4. Comparisons Between Occupations and IndustriesComplementarity between capital and skilled labour is a general and abstract concept, and itsmanifestation depends greatly on the characteristics of the specific production technology that31 In 2005, there was a small increase in the allowance for an additional 20,000 workers under the cap, but this was arelatively insignificant change in the cap compared to the 2003 decrease, and only applied to applicants holding aMaster’s or PhD degree from an U.S. institution. Furthermore, prospective legislation on skilled immigration becameincreasingly bundled with that on unskilled immigration in the years following the cap drop (Kerr et al., 2015a),making it politically difficult to enact policy changes to re-expand the H-1B program.32 I stop at 2007 due to the onset of the credit crunch and financial crisis that significantly impacted firm investmentin 2008 and 2009.16combines capital and labour to produce goods and services. In this section, I investigate how themain effect documented in this essay differ across firms with differing sets of worker and industrycharacteristics, in order to answer questions about how the complementarity between skilled labourand capital relates to occupational and industry characteristics.To give some historical context, Goldin and Katz (1998) and others point out that the relativecomplementarity between capital and skilled labour is a fairly recent phenomenon. During theindustrial revolution, for instance, physical capital in the form of factory systems, machinery, andmechanized equipment came to directly displace skilled artisans in occupations such as glassblower,shoemaker, and blacksmith. It is natural to expect that, in the modern economy, the degree ofcomplementarity between labour and capital depends critically on the specific occupational role oflabour as well as the specific industrial application of capital, and not necessarily on the abstractattribute of skill itself.First, I investigate whether the benchmark effects on investment from Table 2.5 differ acrossoccupational groups of H-1B workers in my sample. As shown in Table 2.4, the major occupationalcategories of worker found in the sample come from the fields of Admin, Computer, Engineer, Man-agement, and Science. One can further classify these occupational groups into broader categoriesin terms of their fundamental roles in modern economic production. First, workers in Science andEngineer occupations can be placed in the category of traditional “industrial workers”, with rolesthat are typically tied to physical production processes. Looking through the occupational sub-categories in Appendix C.1, employees in fields such as civil engineering, mechanical engineering,physics, chemistry, and biology typically work in close physical proximity to machinery, laborato-ries, and other physical hardware to produce physical products. Therefore, it is natural to expectthat workers in such occupations would exhibit strong complementarities to physical capital.Workers in Computer and Admin occupations, on the other hand, can be more appropriatelyplaced in the category of modern “knowledge workers”. Taking examples of computer programmers,computer technical support specialists, accountants, and public relations occupational subcategoriesfrom Appendix C.1, it is apparent that such workers are more closely associated with workingwith ideas and abstract models rather than heavy machinery and equipment, software rather thanhardware, and digital rather than analogue technologies.33 Rather than generating physicallycapital-intensive projects, the value of knowledge workers lies in the development of organizationalcapital, defined by Eisfeldt and Papanikolaou (2013) as intangible capital embodied in the firm’sspecialized labour inputs and distinct from physical capital. Furthermore, these workers often donot need to be in close physical proximity to their work, especially when augmented by moderntelecommunication technology. Therefore, one should expect that workers in such occupationswould exhibit weaker complementarities to physical capital.Finally, workers in Management occupations can be placed in their own category, as managers,who have been the focus of much academic research (see Bloom and Van Reenen (2007) for instance),33 (Martin and Moldoveanu (2003) note that, late in the 20th century, physical and financial assets came to be supplantedin importance by knowledge assets, including the know-how and experience of knowledge workers.17provide oversight over all aspects of the firms’ operations and strategic direction. Managers typicallypossess power over decisions on capital budgeting, but their decisions also depend on input fromemployees involved in day-to-day operations (Harris and Raviv, 2005). Therefore, it is not clearwhere Management workers fall on the occupational spectrum relative to industrial and knowledgeworkers, with respect to complementarity to physical capital.To empirically determine whether the results on investment from Table 2.5 follow the occupa-tional patterns described above, I break down the variable H1B use by occupational category andrun the following regression:CapExit = αi + λt +∑jδj ·H1B useij · Postt +Xit−1β + it (2.4.3)where H1B useij is defined as the total number of initial petitions submitted by firm i for workers inoccupational category j scaled by the average number of employees during 2001, and j takes on thevalues of Computer, Engineer, Science, Admin, and Management, as described in Table 2.3, as wellas the broader categories of Industrial and Knowledge, as defined above. The coefficients δj revealwhether the H-1B cap drop resulted in investment declines for firms that relied more intenselyon H-1B workers in occupation j relative to firms that relied less intensely on H-1B workers inoccupation j.The results are presented in Table 2.8. While the coefficient estimates δj are negative acrossall specifications, they are not statistically significant for workers in Knowledge, Computer andAdmin occupations (columns (4), (5), and (6)), which is consistent with knowledge workers notexhibiting strong complementarities to capital expenditures. On the other hand, the estimates arestatistically significant with respect to Industrial, Engineer, and Science occupations (columns(1), (2), and (3)), which is consistent with industrial workers exhibiting strong complementaritieswith capital-intensive projects.34 Finally, the coefficient estimate on Management is large, as seenin column (4).35 However, the estimate is imprecisely measured with large standard errors, whichmay be attributed to the relatively few H-1B workers in management roles as seen in Table 2.4.I further investigate how the benchmark effects on investment relate to other occupationaland industry characteristics. First, I test whether capital is more complementary to higher-skilledworkers who hold post-graduate degrees. Next, I investigate whether the geographic proximity ofworkers to firm headquarters has any bearings on the documented effects on investment, basedon the idea that workers closer to headquarters are more complementarity to capital due to theirrelative importance and/or advantages in monitoring and information acquisition. Finally, I testwhether the documented complementarity effects are found to be stronger in manufacturing-sectorfirms or in service-sector firm.34 In terms of economic magnitude, a 1-standard deviation change in H1B useij corresponds to a 4.39% and 9.41%decline in investment relative to the sample for j = Engineer and j = Science, respectively.35 The results for which all occupational categories are omitted to conserve space. When all categories are included,the coefficient estimates on the industrial occupation interaction variables remain significant, the coefficient es-timates on the knowledge occupation interaction variables remain insignificant. and the coefficient estimate onH1B usei,Management is no longer significant18I run the following triple-differences OLS regression, in which the DD interaction term H1B usei·Postt from my baseline specification is interacted with various dummy variables related to workerand industry characteristics:CapExit = αi+λt+δ ·H1B usei ·Postt+θ ·Wi ·Postt+γ ·H1B usei ·Wi ·Postt+Xit−1β+it (2.4.4)where Wi takes on the form of various dummy variables related to H-1B worker and industrycharacteristics. This test is used to determine whether the documented complementarity effectsare stronger for more-educated vs. less-educated workers (High Gradi indicates firm i is abovethe sample median in terms of Grad), for older vs. younger workers (High Agei indicates whetherfirm i is above the sample median in terms of Age), for workers with higher vs. lower wages(High Wagei indicates whether firm i is above the sample median in terms of Occ Wage), forworkers in close proximity vs. distant proximity to firm headquarters (Near HQi indicates whetherfirm i is above the sample median in terms of HQ State), for manufacturing vs. non-manufacturingfirms (Manufacturing i indicates whether firm i is in the manufacturing sector), and for servicevs. non-service firms (Servicesi indicates whether firm i is in the services sector). The reportedcoefficient γ reveals whether the effect of the H-1B cap drop on investment is different across firmswith different sets of worker and industry characteristics.Results for the triple differences regressions are presented in Table 2.9. First, taking education,age, and occupational wage as proxies for worker skill and experience, one would expect γ tobe negative in columns (1) to (3) if relatively more skilled workers are more complementarity tocapital when compared to relatively less-skilled workers. However, the only significant result of thethree is for the triple interaction involving High Grad . Furthermore, these results may be drivenby differences in worker characteristics between knowledge and industrial workers—recall fromTable 2.3 that Science workers tended to be better-educated and less highly-paid, while Computerworkers tended to be younger and more highly-paid.36 Therefore, there is limited evidence tosuggest the abstract attribute of skill by itself has a strong bearing on the degree of complementaritybetween college-educated workers and capital.Next, one may expect that workers in closer proximity to company headquarters to have alarger effect on firm decisions, in which case γ should be negative in column (5). This may be dueto advantages in monitoring and information acquisition, as described by Giroud (2013), or dueto the possibility that the roles of workers hired in close proximity to headquarters are less easilyrelocated and outsourced. The estimate from column (5) provides some evidence in support ofeither explanation, as the coefficient corresponding to H1B usei ·NearHQi · Postt is negative andsignificant at the 5% level. Again, this result cannot easily be disentangled from the fact that Scienceand Engineer workers tend to work closer to headquarters, as shown in Table 2.3.37 Industrial36 When running the same triple-differences regression using H1B usei,Industrial instead of H1B usei, the triple inter-action term for High Grad remains significant at the 1% level, suggesting the results are not totally driven bycorrelations between occupational categories and education levels.37 When running the same triple-differences regression using H1B usei,Industrial instead of H1B usei, the triple inter-action term for Near HQ remains significant at the 10% level, suggesting the results are not totally driven by19workers need to be in close proximity to physical capital, while knowledge workers are more easilyrelocated as their productive activities can be more easily augmented by modern telecommunicationtechnologies.Finally, column (5) and (6) from Table 2.9 present evidence of a differential effect on serviceand manufacturing industries.38 The coefficients corresponding to H1B use ·Manufacturing ·Postand H1B use · Services · Post are shown to be weakly significant in the negative and positivedirections, respectively. Given data limitations that prevent direct tests on different types of capitalexpenditures, these tests provide an indirect way to compare industries with different expenditurepatterns.39 According to U.S. census data, manufacturing firms tend to spend a significantlygreater proportion of their capital expenditures on equipment (e.g. computers, industrial machines,and communications equipment) relative to structures (e.g. offices, commercial buildings, andtransportation facilities) when compared to firms in the service sector.40 Therefore, the results areconsistent with the notion that skilled workers do not necessarily spur investments in structuressince they do not take up a lot of physical space, but rather tend to spur investment in machinesand equipment, which is consistent with the findings from Lewis (2011). As seen in Table 2.4,workers in Science and Engineer occupational fields are largely concentrated in the manufacturingsector, while the services sector is dominated by workers in Computer occupations.412.4.5. Political Lobbying and Endogeneity of InvestmentFirms in the high-tech industries, and in particular large firms in the information technology (IT)sector, were the most prominent corporate lobbyists for extending the H-1B cap at a higher level.If my results are driven by correlations between investment trends and lobbying efforts, then effectsof the H-1B cap drop on investment should be more pronounced for firms in these politically-activeindustries. In order to address this concern, I demonstrate that the previously documented effectson investment are not concentrated in industries most involved in H-1B lobbying, which mitigatesconcerns regarding the endogeneity of the H-1B cap drop driving my results. To this end, I splitthe sample by industry characteristics and conduct a triple-differences regression similar to the onepresented in the preceding section:CapExit = αi+λt+δ ·H1B usei ·Postt+θ ·Ii ·Postt+κ ·H1B usei ·Ii ·Postt+Xit−1β+it (2.4.5)correlations between occupational categories and worker proximity.38 As seen in Table 2.4, the vast majority of firms in the sample are found within these two major industry groups.39 Compustat does not break down capital expenditures in further detail.40 According to the U.S. Census 1998 Annual Capital Expenditures Survey, manufacturing sector firms spent $4.21 onequipment for every dollar spent on structures, while service sector firms only spent $1.66 on equipment for everydollar spent on structures. Note that 1998 was the last year that the survey results were broken down by SICcategories.41 Indeed, when running the same triple-differences regression using H1B usei,Industrial instead of H1B usei, the tripleinteraction term for both Manufacturing and Services are no longer statistically significant.20where Ii takes on the form of the following dummy variables related industry characteristics: IT i,which indicates whether firm i is in the information technology sector,42 New Econi, which indi-cates whether firm i is in the “new economy” sector,43 High TQi, which indicates whether firm iis in an industry with above-median average Tobin ′s Q , High RDi, which indicates whether firmi is in an industry with above-median average R&D spending, and High Sizei, which indicateswhether firm i is in an industry with above-median average total assets. All industry dummies aredefined using data from the pre-treatment period, and more detailed definitions for these variablescan be found in Appendix B.1.Table 2.10 reveals the estimates for κ are not statistically significant across all specifications.Therefore, the documented complementarity effects on investment are consistently found acrossIT and non-IT sector firms, as well as across new economy and old economy firms. Splitting thesample by other industry-level characteristics associated with the high-tech sector—i.e. high-growthversus low-growth industries, high R&D versus low R&D industries—also reveals no significancedifferences. Finally, splitting the sample according to asset size, which is a strong predictor ofimmigration-related political lobbying according to Kerr et al. (2014), also reveals no significantdifferences. These results suggest that my main results are not driven by the correlation betweeninvestment demand and lobbying activity by the most politically-active firms. The findings alsosuggest that skilled workers play an important role in implementing a wide range of investmentprojects, and not only those based on R&D intensive technologies in high-tech sectors.2.4.6. Including Non-H1B Firms in the SampleAll analysis presented so far is based on the sample restricted to firms that have submitted at leastone petition during 2001 to hire H-1B workers. This gives rise to the potential for selection bias,due to the fact that selection into H-1B and non-H-1B firms is non-random—i.e. the it error termin Eq. 2.2.1 may be correlated with H1B usei · Postt, conditional on firm i being an H-1B firm.Therefore, I estimate Eq. 2.2.1 based on the expanded sample that includes both sets of H-1B andnon-H-1B firms, where H1B usei is set at zero for all non-H1B firms.The results are presented in column (1) from Table 2.11, and show the estimate for δ to bestatistically significant at the 1% level and similar in economic magnitude to the benchmark resultsfrom Table 2.2.1. This suggests that my earlier results are not driven by selection effects. I alsorun the following regression:CapExit = αi + λt + η ·H1B i · Postt + δ ·H1B usei · Postt +Xit−1β + it (2.4.6)where H1B i represents a dummy variable indicating whether firm i is an H-1B firm. The coefficient42 The IT sector is defined according to BEA classification, which is found online at http://www.bea.gov/industry/xls/GDPbyInd_VA_NAICS_1998-2011.xls.43 Defined to be any industry that “involves acquisition, processing and transformation, and distribution of information”as in Nordhaus (2002). This includes SIC code 35 (industrial machinery and equipment), 36 (electronic and otherelectric equipment), 48 (telephone and telegraph), and 873 (software).21η captures the differential effect of the 2003 H-1B cap drop on non-H-1B workers versus H-1B firmswith marginal exposure to H-1B policy (i.e. the extensive margin),44 while δ still captures thedifferences in investment rate changes between firms of varying ex-ante H-1B dependence (i.e. theintensive margin). If the status of being an H-1B firm has no bearing on investment policy changesaround the 2003 event, then there is no reason to expect a significant difference between non-H-1Bfirms and firms that were marginal employers H-1B workers, in which case η should be zero.The results are presented in column (3) of Table 2.11, and show the estimate for η to be botheconomically and statistically insignificant. Therefore, there is nothing about being in the categoryof H-1B firms that affects investment policy changes around the 2003 event, further mitigatingconcerns of selection bias. All results in Table 2.11 are based on the fully saturated specificationincluding all fixed effects and control variables.It is worth noting the question of whether to include non-H-1B firms in the sample depends onthe specific economic question being asked. Since the impact of H-1B policies falls on H-1B firmsrather than non-H-1B firms, it makes certain sense to focus on the former population. This framesthe question as whether restrictions on the H-1B visa cap affects the investment policies of existingemployers of H-1B workers—i.e. the firms most likely to be affected. The concern then becomesone of external validity, as we are limited in our ability to answer broader questions about howaccess to skilled labour affects investment more generally. The inclusion of non-H-1B firms in thesample does not completely address this issue, since the entire set of non-H-1B firms is unaffectedby the policy and thus reveals no information about how such firms may be affected by skilledlabour supply restrictions. Nevertheless, this is a common limitation facing any study that exploitsan event that affects a limited cross-section of the population.2.4.7. Alternative Definitions of H1B useI check the robustness of my main results by using alternative definitions of H1B use within theregression framework from Eq. 2.2.1. This serves to mitigate concerns that my earlier results aredriven by the skewed distribution of the H1B use variable. The results are presented Table 2.12,and show that the main results hold under a variety of alternative definitions. Column (1) presentthe baseline results using the original definition of H1B use, column (2) presents results usingln(H1B use), the natural log of H1B use, and column (3) presents results using High H1B use, adummy variable indicating whether H1B use is above the sample median. In all cases, the coefficientestimates remain statistically significant and similar in economic magnitude to the baseline results.I also employ definitions of H-1B policy exposure based on the wages of H-1B workers, asthese measures potentially capture additional information regarding firms’ exposure to H-1B policybeyond those found in H1B use. Column (4) presents results based on H1B wage, which is definedas the sum of the wages listed across USCIS petitions submitted by a given firm in 2001, scaled44 This is because H1B use is equivalent to H1B ·H1B use (since H1B use is set at zero for all non-H1B firms), so thatEq. 2.4.6 in effect forms a triple-differences specification, where the double interaction term H1B · Post capturesthe differences between non-H-1B and marginal H-1B firms, and the “triple interaction” term H1B · H1B use · Postcaptures any additional differences between high H-1B users and low H-1B users.22by the total imputed wage bill for that firm during the same year.45 Columns (5) and (6) presentresults based on ln(H1B wage), the natural log of H1B wage, and High H1B wage, a dummyvariable indicating whether H1B wage is above the sample median, respectively. In all cases, theestimates remain statistically and similar in economic magnitude to the baseline results.2.5. ConclusionIn this essay, I find evidence that firms’ access to skilled workers is an important determinant ofinvestment policy. The sharp 2003 drop in the regulatory cap for H-1B visas provided a quasi-experimental setting in which some firms were affected more than others due to differing rates ofex-ante reliance on the visa program. I find firms relying more heavily on H-1B workers experienceda sharper decline in their investment rates relative to firms relying less heavily on H-1B workers,which is consistent with the capital-skill complementarity hypothesis. The effect on investmentpersist for several years, and the evidence suggests that it is unlikely to be driven by pre-existingtrends in investment opportunities, endogenous public policy related to H-1B lobbying, or selectioneffects specific to H-1B-employing firms.I further find evidence suggesting the complementarity effect to be linked to the specific natureof the firm’s worker characteristics, with the effect more pronounced for workers in “industrial”occupations and less pronounced for workers in “knowledge” occupations. I also find the effects tobe more pronounced for firms hiring more highly-educated workers and workers in close proximityto company headquarters. Finally, there is limited evidence to suggest that the effects are morepronounced for manufacturing firms and less pronounced for service sector firms. These resultsimply that the complementarity between skilled labour and capital depends critically on the specificcharacteristics of the production technology combining capital with labour.In addition to helping advance our basic understanding of the relationship between labour andcapital, my research also provides policy implications relating to immigration policy. Specifically,my empirical findings suggest that, rather than focusing only on the immediate impacts on domesticemployment and wage growth when evaluating immigration policy, policymakers should also con-sider the long-term impact on capital investment and subsequent implications for overall economicgrowth. My results further suggest that policymakers should bear in mind the mix of occupationalroles likely to be affected by prospective legislation when evaluating policy, as restrictions on someoccupations may have a large impact on businesses while restrictions on other occupations maysimply result in the jobs relocating abroad.45 The imputed wage is calculated by multiplying the total number of firm-level employees by the national average wagefor the industry for the firm according to data from the Bureau of Labor Statistics.23Figure 2.1: Trends in H-1B Petitions vs. Regulatory Cap050,000100,000150,000200,000H−1B Visas2001 2002 2003 2004 2005YearH−1B Cap at end of year # of Initial H−1B PetitionsSource: United States Citizenship and Immigration Services24Figure 2.2: Timeline of H-1B Cap DropNote: Figure shows the timeline of the H-1B cap drop, which officially took effect on October 1, 2003. In February 2004, the United States Citizenshipand Immigration Services (USCIS) announced that it was no longer accepting petitions for the upcoming fiscal year.25Figure 2.3: Aggregate U.S. Fixed Non-Residential Private Investment−10−50510Growth Rate (% pts)2001 2002 2003 2004 2005YearInvestment (total) Investment (equip.)Investment (I.P.)Source: Bureau of Economic Analysis(a) Annual Aggregate Private Investment−20−1001020Growth Rate (% pts)2001Q12001Q22001Q32001Q42002Q12002Q22002Q32002Q42003Q12003Q22003Q32003Q42004Q12004Q22004Q32004Q42005Q12005Q22005Q32005Q4QuarterInvestment (total) Investment (equip.)Investment (I.P.)Source: Bureau of Economic Analysis(b) Quarterly Aggregate Private Investment26Table 2.1: Descriptive StatisticsThis table presents summary statistics for the main variables in my regression models. In Panel A, the sampleconsists of 1,395 industrial firms (excluding utilities, financials, and public-sector firms) over the 2002Q1-2002Q4(“pre-treatment”) and 2004Q1-2004Q4 (“post-treatment”) periods, for firms submitting at least one H-1B applicationduring 2001 (i.e. “H-1B Firms”). In Panel B, the sample consists of 2,205 industrial firms (excluding utilities,financials, and public-sector firms) over the same time interval as the sample from Panel A, for firms that didnot submit any H-1B applications during 2001 (i.e. “Non-H-1B Firms”). CapEx is quarterly capital expendituresscaled by lagged quarter-end total book assets (atq), Tobin ′s Q is the quarter-end market value of total assets(atq + prccq × cshoq − ceqq − txditcq) scaled by quarter-end book value of total assets (atq), ln(Size) is the naturallog of quarter-end total book assets (atq), Cash Flow is quarterly income before depreciation (ibq + dpq) scaledby lagged quarter-end total book assets (atq), Cash Holdings is quarter-end cash holdings (cheq) scaled by laggedquarter-end total assets (atq), and Leverage is quarter-end long-term debt (dltt) scaled by lagged quarter-end totalbook assets (atq). H1B use represents the total number of initial H-1B petitions filed during the 2001 calendar year,scaled by average number of employees (emp) during the same interval. Detailed definitions for all variables can alsobe found in Appendix B.1. All variables constructed using Compustat variables are winsorized at the 1% level atboth tails, Tobin ′s Q is bounded to be no larger than 10, and H1B use is winsorized at the 2% level at the uppertail.Panel A: H-1B firmsObservations Mean Std Dev P25 Median P75CapEx 9,921 0.011 0.013 0.004 0.007 0.014Tobin’s Q 9,921 2.174 1.487 1.237 1.712 2.577ln(Assets) 9,921 6.666 1.749 5.357 6.489 7.805Cash Flow 9,921 0.004 0.065 -0.002 0.017 0.032Cash Holdings 9,921 0.264 0.253 0.051 0.179 0.429Leverage 9,921 0.165 0.200 0.000 0.104 0.267H1B use 9,921 0.007 0.010 0.000 0.002 0.008Panel B: Non-H-1B firmsObservations Mean Std Dev P25 Median P75CapEx 13,723 0.015 0.023 0.004 0.008 0.018Tobin’s Q 13,723 2.117 1.762 1.133 1.510 2.316ln(Assets) 13,723 5.844 1.906 4.801 5.902 6.991Cash Flow 13,723 -0.001 0.109 0.005 0.020 0.034Cash Holdings 13,723 0.176 0.229 0.021 0.078 0.239Leverage 13,723 0.189 0.243 0.002 0.137 0.30027Table 2.2: Top H-1B Employers in 2001Panel A lists the top 10 firms in the sample in terms of total H-1B initial petitions submitted to the USCIS in 2001,according to data from the USCIS. Panel B lists the top 10 firms in the sample in terms of H1B use (defined inAppendix B.1), according to data from the USCIS and Compustat.Panel A: Top 10 H-1B employers in 2001 (ranked by total petitions submitted)Company Name SIC Division 2-digit SICInfosys Ltd Services Business ServicesMicrosoft Corp Services Business ServicesIntl Business Machines Corp Services Business ServicesCisco Systems Inc Manufacturing Industrial Machinery & EquipmentOracle Corp Services Business ServicesIntel Corp Manufacturing Electronic & Other Electric EquipmentMotorola Solutions Inc Manufacturing Electronic & Other Electric EquipmentLucent Technologies Inc Services Business ServicesWipro Ltd Services Business ServicesCompuware Corp Services Business ServicesPanel B: Top 10 most H-1B-dependent firms in 2001 (ranked by H1B use)Company Name SIC Division 2-digit SICTelecommunication Sys Inc Services Business ServicesBroadwing Corp Manufacturing Electronic & Other Electric EquipmentPharmacyclics Inc Manufacturing Chemical & Allied ProductsArray Biopharma Inc Services Health ServicesActuate Corp Services Business ServicesCatapult Communications Corp Manufacturing Instruments & Related ProductsAlliance Semiconductor Corp Manufacturing Electronic & Other Electric EquipmentEnzon Pharmaceuticals Inc Manufacturing Chemical & Allied ProductsMaxygen Inc Services Engineering & Management ServicesWink Communications Inc Services Business Services28Table 2.3: H-1B Worker CharacteristicsThis table presents a breakdown of worker characteristics across H-1B applications from firms in the sample. PanelA lists the major occupational groups for H-1B workers hired by firms in the sample during the 2001 calendar year,and for each occupational group, the corresponding Count (i.e. the total number of applications) and mean valuesfor the following set of worker-level characteristics: Wage (listed wage of H-1B employee), Occ Wage (the nationalaverage wage from BLS corresponding to H-1B employee’s occupation), Occ Wage Growth (the national averagewage growth rate from BLS corresponding to H-1B employee’s occupation), Age (the age of H-1B employee), Grad (adummy variable for whether H-1B employee possesses a graduate-level education), and HQ State (a dummy variablefor whether H-1B employee works in the same geographic state as firm headquarters). Panel B presents a correlationtable for the same set of variables for which occupational averages are reported in Panel A. Detailed definitionsfor all variables can be found in Appendix B.1 and detailed definitions for occupational categories can be found inAppendix C.1. In Panel B, *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively.Panel A: Mean Characteristics by OccupationWage Occ Wage Occ Wage Growth Age Grad HQ StateComputers 72,989 72,393 0.061 29.85 0.660 0.436Engineering 73,951 63,867 0.036 30.71 0.738 0.608Admin 68,036 45,934 0.042 30.99 0.680 0.516Management 92,639 63,776 0.027 34.30 0.704 0.500Scientist 67,250 52,867 0.015 33.31 0.760 0.824Other 67,234 61,461 0.029 31.60 0.661 0.405Total 73,257 68,337 0.052 30.46 0.683 0.496Panel B: Application Level Correlations of Worker CharacteristicsWage Occ Wage Occ Wage Growth Age Grad HQ StateWage 1.000Occ Wage 0.013* 1.000Occ Wage Growth 0.006 0.473*** 1.000Age 0.336*** -0.121*** -0.112*** 1.000Grad 0.070*** -0.107*** -0.140*** 0.068*** 1.000HQ State 0.098*** -0.038*** -0.071*** 0.026*** 0.084*** 1.00029Table 2.4: Number of Applications by Occupation and Industry GroupThis table presents the number of H-1B worker applications submitted during the 2001 calendar year by firms in the sample, broken down across major industryand occupational groups. Industries are defined at the SIC Division level while occupations are defined at the 3-digit Dictionary of Occupational Code level ,with “Occupations in Life Sciences” and “Occupations in Mathematics and Physical Sciences” combined under the “Science” category. Detailed definitions foroccupational categories can be found in the Appendix C.1.Computers Engineering Admin Management Scientist Other TotalAgriculture 3 0 1 1 0 0 5Construction 6 21 6 7 0 0 40Manufacturing 6,570 6,129 822 667 1,398 443 16,029Mining 15 24 20 7 20 2 88Retail 327 11 49 68 9 242 706Services 15,661 1,128 447 462 340 308 18,346Transportation 315 261 100 81 2 32 791Wholesale 97 15 24 14 9 14 173Total 22,994 7,589 1,469 1,307 1,778 1,041 36,17830Table 2.5: Effect of the H-1B cap Drop on InvestmentThe tables below report the estimation results from the OLS regression CapExit = αi + λt + δ · H1B usei · Postt +Xit−1β + it in which δ captures how the 2003 H-1B cap drop affects the investment policy of firms with differentlevels of H-1B usage intensity. The sample is limited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4)and four quarters following the H-1B cap drop (2004Q1-2004Q4), for firms submitting at least one H-1B applicationduring 2001. CapExit denotes firm i’s investment rate during quarter t, H1B usei denotes firm i’s H-1B usageintensity during 2001, and Postt represents a dummy variable that takes on a value of 1 if quarter t is in the post-treatment period 2004Q1-2004Q4. Xit denotes the set of quarterly firm-level control variables, which are all laggedby one quarter relative to the dependent variable CapExit. Detailed definitions for all variables can be found inTable 2.1 as well as Appendix B.1. F denotes firm fixed effects, T denotes year-quarter fixed effects, and I × Tdenotes industry-year-quarter fixed effects. Standard errors are corrected for heteroskedasticity and clustered at thefirm level. Standard errors are in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1%level, respectively.(1) (2) (3) (4) (5) (6)CapEx CapEx CapEx CapEx CapEx CapExH1B use × Post -0.059** -0.081*** -0.080*** -0.100*** -0.080*** -0.101***[0.029] [0.029] [0.029] [0.029] [0.029] [0.029]Tobin’s Q 0.002*** 0.002*** 0.002*** 0.002***[0.000] [0.000] [0.000] [0.000]Cash Flow 0.006* 0.006* 0.006* 0.006*[0.003] [0.003] [0.004] [0.004]ln(Assets) -0.000 -0.000 -0.000 -0.000[0.001] [0.001] [0.001] [0.001]Leverage -0.003 -0.003*[0.002] [0.002]Cash Holdings -0.002 -0.002[0.002] [0.002]Fixed Effects F, T F, I × T F, T F, I × T F, T F, I × TObservations 9,921 9,921 9,921 9,921 9,921 9,921Adjusted R-squared 0.007 0.012 0.024 0.029 0.025 0.03031Table 2.6: Quarterly Investment DynamicsThis table reports the estimation results from OLS regression CapExit = αi +γt +∑δt ·H1B usei · τt + δ ·H1B usei ·Postt +Xit−1β+ it. The sample is limited to the 2002Q1-2004Q4 interval, which includes the pre-treatment period(2002Q1-2002Q4), the legislative shift period (2003Q1-2003Q4), and the post-treatment period (2004Q1-2004Q4), forfirms submitting at least one H-1B application during 2001. CapExit denotes firm i’s investment rate during quartert, H1B usei denotes firm i’s H-1B usage intensity during 2001, and τt denotes a dummy variable for quarter τ , whereτ takes on values from 2002Q1 to 2004Q4 inclusive. Xit−1 denotes the set of quarterly firm-level control variables,which are lagged by one quarter relative to the dependent variable CapExit. Detailed definitions for all variables canbe found in Table 2.1 as well as Appendix B.1. Only δ and δt are reported to conserve space. F denotes firm fixedeffects, T denotes year-quarter fixed effects, and I × T denotes industry-year-quarter fixed effects. Standard errorsare corrected for heteroskedasticity and clustered at the firm level. Standard errors are in parentheses, with *, **,and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2)CapEx CapExH1B use × 2002Q2 (pre-treatment) -0.011 -0.012[0.028] [0.026]H1B use × 2002Q3 (pre-treatment) 0.027 0.016[0.049] [0.049]H1B use × 2002Q4 (pre-treatment) -0.032 -0.041[0.043] [0.043]H1B use × 2003Q1 (legislative shift) 0.003 -0.024[0.035] [0.037]H1B use × 2003Q2 (legislative shift) -0.017 -0.049[0.040] [0.041]H1B use × 2003Q3 (legislative shift) -0.056 -0.079**[0.040] [0.039]H1B use × 2003Q4 (legislative shift) -0.090** -0.107**[0.045] [0.046]H1B use × Post (post-treatment) -0.080** -0.103***[0.036] [0.036]Control Variables Yes YesFixed Effects F, T F, I × TObservations 14,811 14,811Adjusted R-squared 0.023 0.02532Table 2.7: Long Run Effect of H-1B Cap Drop on InvestmentThis table reports the estimation results from OLS regression CapExit = αi + γt +∑4k=−1 δk ·H1B usei · Y ear kt +Xitβ + it. The sample is limited to the 2001 to 2007 time interval, with k = 0 corresponding to the 2003 calendaryear, for firms submitting at least one H-1B application during 2001. CapExit denotes firm i’s quarterly investmentrate averaged over calendar year t, H1B usei denotes firm i’s H-1B usage intensity during 2001, and Year k t denotesa year dummy variable for k = 2003 + t. Xit denotes the set of firm-level control variables, which are lagged by onequarter relative to the dependent variable CapExit and then averaged over calendar year t. More detailed definitionsfor all variables can be found in Table 2.1 as well as Appendix B.1. Only coefficients δk are reported to conservespace. F denotes firm fixed effects, Y ear denotes year fixed effects, and I × Y ear denotes industry-year fixed effects.Standard errors are corrected for heteroskedasticity and clustered at the firm level. Standard errors are in parentheses,with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2)CapEx CapExH1B use × Year -1 0.013 0.001[0.029] [0.030]H1B use × Year 0 -0.021 -0.049[0.032] [0.034]H1B use × Year 1 -0.071** -0.102***[0.031] [0.032]H1B use × Year 2 -0.062** -0.088***[0.031] [0.032]H1B use × Year 3 -0.111*** -0.131***[0.035] [0.036]H1B use × Year 4 -0.107*** -0.120***[0.038] [0.039]Control Variables Yes YesFixed Effects F, T F, I × TObservations 8,613 8,591Adjusted R-squared 0.096 0.11033Table 2.8: Effect of the H-1B Cap Drop on Investment by Occupational CategoryThis table reports the estimation results from the OLS regression CapExit = αi + γt +∑δj · H1B useij · Postt +Xit−1β + it. The sample is limited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4) and four quartersfollowing the H-1B cap drop (2004Q1-2004Q4), for firms submitting at least one H-1B application during 2001. Onlycoefficients δj are reported to conserve space. CapExit denotes firm i’s investment rate during quarter t, H1B useijdenotes firm i’s H-1B usage intensity in hiring workers in occupation j (i.e. Industrial, Engineer, Science, Knowledge,Computer, Admin, or Management) during 2001, and Postt represents a dummy variable that takes on a value of1 if quarter t is in the post-treatment period 2004Q1-2004Q4. Xit−1 denotes the set of quarterly firm-level controlvariables, which are lagged by one quarter relative to the dependent variable CapExit: all specifications includeTobin ′s Q , Cash Flow , ln(Size), Cash Holdings, and Leverage as controls. Detailed definitions for all variables canbe found in Table 2.1 as well as Appendix B.1, and more detailed definitions for occupational categories can be foundin Appendix C.1. F denotes firm fixed effects and I × T denotes industry-year-quarter fixed effects. Standard errorsare corrected for heteroskedasticity and clustered at the firm level. Standard errors are in parentheses, with *, **,and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6) (7)CapEx CapEx CapEx CapEx CapEx CapEx CapExH1B use (Industrial) × Post -0.493***[0.120]H1B use (Engineer) × Post -0.264**[0.120]H1B use (Science) × Post -0.651***[0.235]H1B use (Knowledge) × Post -0.060[0.046]H1B use (Computer) × Post -0.055[0.048]H1B use (Admin) × Post -0.806[0.491]H1B use (Management) × Post -1.403*[0.724]Control Variables Yes Yes Yes Yes Yes Yes YesFixed Effects F, I × T F, I × T F, I × T F, I × T F, I × T F, I × T F, I × TObservations 9,921 9,921 9,921 9,921 9,921 9,921 9,921Adjusted R-squared 0.031 0.028 0.030 0.027 0.027 0.027 0.02834Table 2.9: Occupational and Industry Characteristics and the Effect of the H-1B CapDrop on InvestmentThis table reports the estimation results from the OLS regression CapExit = αi + γt + δ ·H1B usei · Postt + θ ·Wi ·Postt + λ · H1B usei ·Wi · Posti + Xitβ + it. Only coefficients δ and λ are reported in order to conserve space.The sample is limited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4) and four quarters following theH-1B cap drop (2004Q1-2004Q4), for firms submitting at least one H-1B application during 2001. CapExit denotesfirm i’s investment rate during quarter t, H1B usei denotes firm i’s H-1B usage intensity during 2001, and Posttrepresents a dummy variable that takes on a value of 1 if quarter t is in the post-treatment period 2004Q1-2004Q4.In columns (1)-(4), Wi represents a dummy variable that take on a value of 1 if firm i is above the sample median interms of the following variables listed and defined in Table 2.3: Grad (Wi = High Gradi), Age (Wi = High Agei),Occ Wage (Wi = High Wagei), and HQ State (Wi = Near HQi). In columns (5)-(6), Wi represents an indicatorvariable for whether firm i is in the manufacturing sector (Wi = Manufacturingi for SIC 2000-3999) or the servicessector (Wi = Servicesi for SIC 7000-8999). Xit−1 denotes the set of quarterly firm-level control variables which arelagged by one quarter relative to the dependent variable CapExit. All specifications include Tobin′s Q , Cash Flow ,ln(Size), Cash, and Leverage as controls. Detailed definitions for all variables can be found in Table 2.1 as well asAppendix B.1. F denotes firm fixed effects and I × T denotes industry-year-quarter fixed effects. Standard errorsare corrected for heteroskedasticity and clustered at the firm level. Standard errors are in parentheses, with *, **,and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)CapEx CapEx CapEx CapEx CapEx CapExH1B use × Post -0.006 -0.069** -0.126** -0.002 -0.040 -0.143***[0.049] [0.034] [0.062] [0.048] [0.045] [0.036]H1B use × High Grad × Post -0.123**[0.062]H1B use × High Age × Post -0.052[0.057]H1B use × High Wage × Post 0.034[0.069]H1B use × Near HQ × Post -0.146**[0.060]H1B use × Manufacturing × Post -0.105*[0.058]H1B use × Services × Post 0.109*[0.060]Control Variables Yes Yes Yes Yes Yes YesFixed Effects F, I × T F, I × T F, I × T F, I × T F, I × T F, I × TObservations 7,526 8,371 7,964 8,223 9,921 9,921Adjusted R-squared 0.035 0.032 0.031 0.032 0.030 0.03035Table 2.10: Characteristics Associated with Political Lobbying and the Effect of the H-1BCap Drop on InvestmentThis table reports the estimation results from the OLS regression CapExit = αi+γt+δ·H1B usei·Postt+pi·Ii·Postt =κ · H1B usei · Postt + Xitβ + it. Only coefficients δ and κ are reported in order to conserve space. The sample islimited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4) and four quarters following the H-1B cap drop(2004Q1-2004Q4), for firms submitting at least one H-1B application in 2001. CapExit denotes firm i’s investmentrate during quarter t, H1B usei denotes firm i’s H-1B usage intensity during 2001, and Postt represents a dummyvariable that takes on a value of 1 if quarter t is in the post-treatment period. In columns (1)-(2), Ii represents adummy variable that take on a value of 1 if firm i is in the information technology sector (Ii = ITi) and the “neweconomy” sector (Ii = New Econi). In columns (3)-(5), Ii represents a dummy variable that take on a value of 1 iffirm i is in an above-median industry (2-digit SIC) in terms of average values for the following variables: Tobin ′s Q(Ii = High TQi), R&D spending (Ii = High RDi), and total assets (Ii = High Sizei). Xit−1 denotes the set ofquarterly firm-level control variables, which are lagged by one quarter relative to CapExit: all specifications includeTobin ′s Q , Cash Flow , ln(Size), Cash Holdings, and Leverage as controls. Detailed definitions for all variables canbe found in Table 2.1 as well as Appendix B.1. F denotes firm fixed effects and I × T denotes industry-year-quarterfixed effects. Standard errors are corrected for heteroskedasticity and clustered at the firm level. Standard errors arein parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5)CapEx CapEx CapEx CapEx CapExH1B use × Post -0.125** -0.095** -0.124*** -0.096*** -0.099***[0.049] [0.038] [0.041] [0.037] [0.030]H1B use × IT × Post 0.031[0.061]H1B use × New Econ × Post -0.039[0.059]H1B use × High TQ × Post 0.045[0.054]H1B use × High RD × Post 0.008[0.055]H1B use × High Size × Post -0.017[0.159]Control Variables Yes Yes Yes Yes YesFixed Effects F, I × T F, I × T F, I × T F, I × T F, I × TObservations 9,921 9,921 9,921 9,921 9,921Adjusted R-squared 0.030 0.031 0.032 0.031 0.03036Table 2.11: Effect of the H-1B Cap Drop on Investment Based on Unrestricted Sample ofH-1B and Non-H-1B FirmsThe tables below report the estimation results from the OLS regression CapExit = αi + γt + δ · H1B usei · Postt +Xit−1β + it in column (1), and CapExit = α1 + γt + η ·H1B i ·Postt + δ ·H1B usei ·Postt +Xit−1β + it in column(3). The sample is limited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4) and four quarters followingthe H-1B cap drop (2004Q1-2004Q4) for all firms (including those that did not submit H-1B applications during2001). CapExit denotes firm i’s investment rate during quarter t, H1B i denotes that firm i submitted at least oneH-1B application during 2001, H1B usei denotes firm i’s H-1B usage intensity during 2001, and Postt represents adummy variable that takes on a value of 1 if quarter t is in the post-treatment period 2004Q1-2004Q4. Xit denotesthe set of quarterly firm-level control variables, which are all lagged by one quarter relative to the dependent variableCapExit. Detailed definitions for all variables can be found in Table 2.1 as well as Appendix B.1. F denotes firmfixed effects, T denotes year-quarter fixed effects, and I × T denotes industry-year-quarter fixed effects. Standarderrors are corrected for heteroskedasticity and clustered at the firm level. Standard errors are in parentheses, with *,**, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3)CapEx CapEx CapExH1B use × Post -0.079*** -0.070**[0.028] [0.030]H1B × Post -0.001* -0.000[0.000] [0.000]Control Variables Yes Yes YesFixed Effects F, I × T F, I × T F, I × TObservations 23,644 23,644 23,644Adjusted R-squared 0.028 0.028 0.02837Table 2.12: Effect of the H-1B Cap Drop on Investment Based on Alternative Definitionsof H-1B ExposureThis table reports the estimation results from the OLS regression CapExit = αi + γt + δ · H1B variablei · Postt +Xit−1β+it, where H1B variablei represents various alternative measures of firm i’s H-1B usage intensity during 2001.The sample is limited to four quarters prior to the H-1B cap drop (2002Q1-2002Q4) and four quarters following theH-1B cap drop (2004Q1-2004Q4), for firms submitting at least one H-1B application during 2001. CapExit denotesfirm i’s investment rate during quarter t and Postt represents a dummy variable that takes on a value of 1 if quartert is in the post-treatment period 2004Q1-2004Q4. Column (1) uses the same definition of H1B usei as found inTable 2.5, column (2) uses the natural log of H1B usei, and column (3) uses a non-parametric measure of whetherH1B usei is above or below its cross-sectional sample median. Columns (4)-(6) apply analogous measures of H-1Busage intensity calculated based on the cumulative wages of H-1B workers per firm rather than the number of hiresper firm. Xit−1 denotes the set of quarterly firm-level control variables, which are lagged by one quarter relativeto the dependent variable CapExit: all specifications include Tobin′s Q , Cash Flow , ln(Size), Cash Holdings, andLeverage as controls. Detailed definitions for all variables can be found in Table 2.1 as well as Appendix B.1. Fdenotes firm fixed effects and I × T denotes industry-year-quarter fixed effects. Standard errors are corrected forheteroskedasticity and clustered at the firm level. Standard errors are in parentheses, with *, **, and *** denotingsignificance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)CapEx CapEx CapEx CapEx CapEx CapExH1B use × Post -0.101***[0.029]ln(H1B usage) × Post -0.103***[0.029]High H1B usage × Post -0.002***[0.001]H1B wage × Post -0.072***[0.022]ln(H1B wage) × Post -0.074***[0.023]High H1B wage × Post -0.002***[0.001]Control Variables Yes Yes Yes Yes Yes YesFixed Effects F, I × T F, I × T F, I × T F, I × T F, I × T F, I × TObservations 9,921 9,921 9,921 9,892 9,892 9,892Adjusted R-squared 0.030 0.030 0.029 0.029 0.029 0.03138Chapter 3Politics and Hidden Borrowing:Electoral Cycles and State DefinedBenefit Pension Plans3.1. IntroductionPublic sector defined benefit (DB) pension plans allow governments to defer payment to theirworkers by offering guaranteed future retirement benefits. In the United States, the aggregateliability formed by state-level DB pension plan obligations is enormous, with Novy-Marx and Rauh(2011) estimating unfunded pension liabilities to be as high as $4.43 trillion as of 2009. Motivatedby the idea that politicians undertake opportunistic actions for politically motivated purposes,I investigate how electoral incentives can motivate incumbent state Governors to shape publicpension policies for their own benefit. Specifically, Governors may “borrow’ on behalf of taxpayersusing their discretionary power over how public pension assets accumulate and how public pensionliabilities accrue.On the asset side, Governors may divert governmental contributions intended to fund statepension plans towards more politically expedient uses, such as increasing public services, cuttingtaxes, or reducing the state budget deficit. Anecdotal accounts suggest this to be an attractiveoption. In the run-up to the 1990 gubernatorial election, New York Governor Mario Cuomo workedto lower contributions to state pension plans by $1.3 billion, using the funds to reduce the budgetdeficit instead. After Governor Cuomo secured his reelection bid, the New York State of Appealsruled in 1993 that the state had illegally borrowed from state pension funds, and ordered the stateto pay back the shortfall over the next 12 years.On the liability side, DB pension benefits promised to public sector employees represent debt-like obligations for the government. Incumbent Governors may be tempted to raise benefits inorder to gain political support from public sector labour unions. In the early 2000’s, CaliforniaGovernor Gray Davis pushed through numerous bills to increase state pension benefits, winningstrong support from public sector unions along the way at the expense of creating large unfundedpension liabilities for taxpayers. Pension benefits also provide employers with a potential bargainingchip that can be used to negotiate against wage increases. By raising promises of pension benefitsfor public sector employees, for instance, the incumbent administration can keep payroll growth incheck in the short run, freeing up funds for more immediate uses.39Spurred by the idea that political incentives are strongest immediately prior to an election, Iinvestigate whether the net amount of borrowing conducted through state pension plans is sys-tematically different in election years versus non-election years. To this end, I construct a novel“pension deficit” flow variable by taking the difference between the rate at which pension planliabilities accrue (benefit accruals) and the rate at which pension plan assets accumulate (contri-butions). My findings indicate that that state DB pension plan deficits (surpluses) are on average11% higher (lower) in gubernatorial election years compared to in non-election years. I include avariety of state-level and plan-level control variables, as well as plan fixed effects and year fixedeffects, to control for potential confounding factors.Separating the pension deficit measure into its two components (contributions and benefit ac-cruals), I find that the electoral cycles in pension deficits are largely explained by election yeardecreases in pension contribution rates, as the magnitude of election year “dips” in contributionsare almost identical to those of election year “spikes” in pension deficits. The significant electoralcycle in contribution rates are made possible by Governors’ significant powers over the state budgetprocess through which contributions are approved. Accordingly, I find contribution cutbacks to belarger for election years that coincide with the passage of a state budget relative to election yearsthat do not.In contrast to contributions, I find that benefit accruals do not exhibit a significant electoral cyclepattern. This may be attributed to the inflexibility of pension benefit policy, which is typically setthrough multi-year labour contracts and/or special statutory provisions, as well as to the fact thatbenefit accruals are imprecisely measured due to discretionary actuarial assumptions. Nevertheless,I find states with higher rates of public sector union membership experience significantly largerelection year increases in benefit accrual rates relative to states with lower union membershiprates, which suggests a motive to grant higher pension benefits in exchange for election year politicalsupport from powerful labour unions.To gain a deeper understanding of the Governor’s incentive to increase election year pensiondeficits, one must first understand how taxpayers and public pension employees are affected bypension funding policy, and in particular which stakeholder group bears the future burden of un-derfunded public pension plans. Higher pension deficits today necessarily imply future cuts togovernment spending, future increases in taxes, or future cuts to pension benefits. The first twooutcomes are at the expense of the wider taxpayer base, while the third outcome is at the expenseof public sector employees.If public sector employees bear the future burden of pension underfunding, then election yearspikes in pension deficits effectively constitute funds appropriated from public employees by incum-bent politicians to “buy” votes from the electorate. However, the empirical evidence contradictsthis interpretation, as I exploit cross-sectional variation in legal frameworks across states and findelectoral cycles in pension deficits to be concentrated in states in which public sector employeesenjoy stronger legal protection over their future DB retirement benefits. This suggests that publicpension plan participants with weak protection over their benefits have both the incentive and the40means to limit politically-motivated policies that devalue their future retirement benefits.46The implication is that strong benefit protection creates a moral hazard for employees to ignorethe consequences of pension borrowing, as taxpayers are left to bear the burden through higherfuture taxes or lower public services. This should not be an issue if rational and forward-lookingvoters can observe government pension policies and understand that higher pension deficits serveonly to “kick the can down the road”. Under such a scenario, election year spikes in pensiondeficits would be politically self-defeating for incumbent Governors if such policies were not in thebest interests of voting taxpayers who ultimately determine election outcomes.In reality, voters are unlikely to be able to perfectly monitor the government’s public pensionpolicies due to well established free rider problems inherent in political settings with diffuse voters,47as well as due to the inherent opacity of DB pensions plans that rely on complex actuarial methodsto evaluate and report on funding levels. Previous research has shown information asymmetry tobe an important factor in generating politically-motivated electoral cycles in fiscal deficits (Shi andSvensson (2006), Alt and Lassen (2006)), based on the idea that incumbent politicians attempt to“fool” voters with increased deficit spending only if voters cannot directly observe that the higherspending is financed through debt.Following this logic, state pension plans provide incumbent Governors with a particularly opaquechannel to finance politically-motivated expansionary activities in election years. Using an indexmeasure of state pension opacity based on journalist surveys, I find that electoral year spikes inpension deficits are significantly more pronounced in states with more opaque pension systemsrelative to states with more transparent pension systems, which supports the notion of state DBpension plans constituting a channel for “hidden” deficit financing prone to politically-motivatedmanipulations.I find additional evidence that election concerns drive pension borrowing decisions. First, theincentive to win additional votes should be stronger for closely contested elections that are near a“tipping point”, and I find election year pension deficit spikes are indeed larger for close electionsin comparison to lopsided ones. I also find that pension deficits are smaller when the incumbentGovernor is ineligible to run for reelection due to term limits. Lastly, I find no significant differencein electoral cycle patterns in pension deficits between Republican and Democrat Governors, whichsuggests that my results are not driven by the ideological preferences of one particular party’spartisan supporters.Next, I investigate whether systematic election year pension deficit spikes have real conse-quences. I find that state DB pension plans exhibiting larger electoral cycles in pension deficitstend to experience larger increases in unfunded liabilities over the 2001-2015 sample period. Onaverage, the electoral cycle in pension deficits can explain 6.65% of the average increase in pensionunderfunding, which suggests that election year pension deficit spikes are not totally offset by lower46 For example, employees can collectively exert political pressure through lobbying by their unions, exert economicpressure through collective bargaining, or directly influence pension policy through employee representation on statepension boards of trustees.47 See, for instance, Becker (1983).41rates of pension deficits during non-election years and play an economically significant role in ex-plaining the deteriorating funding status of state DB pension plans in recent years. Furthermore,I find suggestive evidence that states containing plans that exhibit larger electoral pension deficitcycles are associated with lower economic growth over the sample period.I run additional tests in order to rule out plausible alternative explanations for my findings.Most notably, I find no evidence of an electoral cycle pattern in pension deficits for private-sectorDB pension plans. Since private sector plans should be immune from political incentives relating togubernatorial elections, this finding supports the key assumption behind my main empirical test,in that pension policies unaffected by political incentives should exhibit no systematic election yeareffects. I also find no evidence of pension deficit increases during years in which states experienceunexpected governor turnovers. This mitigate concerns that my findings are driven by leadershiptransition effects unrelated to reelection incentives.At its root, my research is about how information frictions can lead to short-sighted decisionsin the context of a principal-agent relationship. This relates to the broad literature on managerialmyopia, including works by Stein (1988), Bebchuk and Stole (1993), and Nagarajan et al. (1995),who provide models of how hidden information problems can lead to myopic corporate decisions. Inparticular, Narayanan (1985) and Stein (1989) show myopic decisions can arise due to hidden actionproblems using reputation building models. However, the corporate finance literature has foundmixed success in finding empirical evidence in support of such theories. For example, Meulbroeket al. (1990) reject the prediction from Stein (1988) that takeover threats induce myopic corporatepolices.In this essay, I turn to the public sector to search for evidence of short-sighted decisions stem-ming from distortionary career concerns. As noted by Tirole (1994), career concern incentives asdescribed in Holmstro¨m (1999) should be especially strong in the public sector due to the lack ofhigh powered incentive contracts. Furthermore, while large block shareholders are able to concen-trate ownership and overcome agency-induced managerial myopia (Wahal and McConnell (2000),Edmans (2009)), diffuse taxpayers cannot accumulate votes in order to overcome the free rider prob-lem. Therefore, political elections and public sector defined pension policies provide a particularlyappropriate setting for an empirical investigation into distortionary incentives.The idea that political agency problems are most pronounced in election years comes fromthe political cycles literature, which examines politicians’ incentives to manipulate macroeconomicoutcomes for reelection purposes.48 My work delivers the insight that opaque public pension plansoffer governments a “hidden” way to finance expansionary election year policies,49 an interpretationthat potentially reconciles the finding of Poterba (1994) and Rose (2006), who empirically documentfiscal policies to be systematically more expansionary during gubernatorial election years, withthe findings of Peltzman (1992), who find that voters in gubernatorial elections tend to punish48 See Nordhaus (1975), Rogoff and Sibert (1988), Rogoff (1990), Alesina et al. (1997), and Persson and Tabellini (2002)for the major theories on what generates political cycles.49 This relates to work by Shi and Svensson (2006) and Alt and Lassen (2006), who find the budgetary transparencyhelps to mitigate electoral cycles in budget deficits in OECD countries.42budget deficits and reward fiscal conservatism. This also relates to the recent literature on financialinnovation regarding how the opacity of complex financial products can be exploited by politicians.For example, Pe´rignon and Valle´e (2017) find that local governments in France tend to increasetheir use of complex structured loans ahead of closely-contested elections as a way to temporarilyshroud deficits.My work also relates to the literature that explores how political elections affect financial mar-kets and firm behaviour. Prior research has identified political cycles in banking regulation (Brownand Dinc (2005), Cole (2009), Liu and Ngo (2014), Haselmann et al. (2015)), firm-level investment(Julio and Yook, 2012), discretionary accounting choices (Kido et al., 2012), and rates of job andplant creation (Bertrand et al., 2007). In the corporate governance literature on board elections,Fos et al. (2016) find temporal proximity to board election increases CEO turnover-to-performancesensitivity.Surprising, little work has been done to examine the impact of political incentives on publicpension funding decisions. The existing literature on this topic has identified various factors thataffect public pension funding levels. These factors include taxpayer mobility (Inman, 1982), union-ization rates (Mitchell and Smith, 1994), state demographics (Giertz and Papke (2007), Kelley(2014)), and state fiscal conditions (Chaney et al. (2003), Munnell et al. (2011b), Splinter (2015)).Elder et al. (2015) study how political polarization and electoral uncertainty can lead to greaterpension underfunding, but their results are based on noisy measures of pension funding and politi-cal conditions, and lack a clear empirical strategy to distinguish political causes from confoundingeconomic channels.50 By exploiting the exogenous scheduling of gubernatorial elections, I am ableto plausibly identify a strictly political motive behind how state governments fund their DB pen-sion plans. My work also contributes to the literature by (1) constructing a novel flow measure ofpension borrowing that accounts for the fact that pension deficits are jointly determined by con-tributions and benefits, (2) providing a testable conceptual framework relating to employee moralhazard and uninformed voters to explain the roots of political incentives regarding public pensionborrowing, and (3) using falsification tests that rule out alternative explanations for documentedelectoral cycle patterns.The remainder of this essay is organized as follows. Section 3.2 describes how the institutionalsetting of state DB pension plans gives rise to incentives for Governors to borrow through thepension system for politically-motivated purposes. Section 3.3 describes the empirical strategythat I employ to identify an electoral cycle pattern in pension deficits and the political incentivesbehind the pattern. Section 3.4 describes data used in the empirical analysis. Section 3.5 reportsand interprets the empirical results and supplementary findings. Section 3.6 concludes.50 For instance, Epple and Schipper (1981) make the point that governments may borrow through the public pensionsystem as a way to smooth taxes and public spending in response to economic shocks, to the benefit of taxpayers.433.2. State Defined Benefit Pension PlansIn this section, I outline the institutional setting surrounding state DB pension plans and detail theinstitutional roots behind Governors’ incentives to use public pensions for political purposes. First,I describe how the balance of of state pension assets and liabilities are determined by the flow ofcontribution and benefit policies over time. I then describe the Governor’s discretionary power overcontribution and benefit policies. Next, I explain how taxpayers and public employees are affectedby public pension underfunding and what that implies for the Governor’s political incentives. Last,I describe how the opacity of public pension plans can distort the Governor’s incentives.3.2.1. State Pension Assets and LiabilitiesI focus my analysis on defined benefits (DB) pension plans, which comprise the majority of all U.S.public-sector plans at the state level. According to the 2015 BLS Employee Benefits Survey, 84% ofall public-sector workers in state and local governments were eligible to participate in a DB pensionplan, and 89% of those eligible workers were active participants in those plans. At its core, a DBpension plans consists of a collection of liabilities, which represent promises of future benefits toemployees, and a collection of assets, which is accumulated in order to fund those promises beforethey become due.51In contrast to defined contribution (DC) plans, which provide benefits that fluctuate with themarket value of a plan’s assets, DB benefits are predefined in advance. Typically, a participatingemployee’s annual benefit is determined by the product of their average salary over the final 3-5years of employment, the number of years of employment, and a plan-specific accrual rate. Forexample, an employee with an average ending salary of $100,000 and possessing 20 years of servicewould receive a base annuity of $60,000 under a plan with an accrual rate of 3%.52As semi-fixed promises of future payment to employees, DB pension benefits constitute a debt-like liability for state retirement systems. Each year, state DB plans accrues new liabilities asactive employees gain an additional year of service, and a portion of existing pension liabilities isretired as benefits are distributed to retiring employees. Conceptually, the evolution of a DB plan’sliability from year t to year t+ 1 follows:Liabt+1 = Liabt(1 + rLiab) + Acct+1 − Benefitst+1, (3.2.1)where Liab denotes the stock of pension liabilities, rLiab denotes the discount rate used to calculatethe present value of future obligations, Acc denotes the present value of new benefits accrued, andBenefits denotes benefits paid.While Eq. 3.2.1 provides a conceptual representation of how DB pension liabilities change overtime, the practical process of accounting for DB pension liabilities is considerably more complicated.51 In this way public DB pension plans are pre-funded, which is in contrast to the pay-as-you-go funding scheme of U.S.Social Security, in which each generation takes on the full burden of paying for the previous generation’s benefits.52 Most plans apply a cost-of-living adjustment (COLA) add-on to adjust for inflation.44In order to estimate the expected present value of future benefits, a DB plan must make assumptionsabout future wage growth, mortality rates, inflation, discount rates, etc. In practice, state planshire specialized actuarial consultants to calculate DB pension liabilities via complicated actuarialmethods. These practical considerations relating to actuarial assumptions are accounted for in myempirical analysis, but are omitted here in order to highlight Acc as a conceptual flow measure ofpension liability accruals.On the asset side, contributions are set aside every year to match the steady accrual of bene-fits. The contribution funds are invested in marketable securities and held in trust until they aredistributed to plan beneficiaries. Conceptually, a DB plan’s assets evolves according toAssetst+1 = Assetst(1 + rAssets) + Contribt+1 − Benefitst+1, (3.2.2)where Assets denotes the stock of pension assets, rAssets reflects the rate of return on investment,Contrib denotes the flow of contributions into pension assets, and Benefits denotes benefits paidfrom pension assets.When a plan’s liabilities exceed its assets, the plan is considered to be underfunded, and theshortfall difference is termed the unfunded liability. Combining 3.2.1 and 3.2.2 allows us to expressthe evolution of a plan’s unfunded liability as follows:UnfLiabt+1 = UnfLiabt(1 + rLiab) + Acct+1 − Contribt+1 − (rLiab − rAssets)Assetst, (3.2.3)where UnfLiab denotes the stock of unfunded liabilities.Conceptually, UnfLiab represents the the “net” indebtedness of a pension plan, in the sense thatany accrued benefit obligations not covered by accumulated assets must be eventually be repaid.UnfLiab can be negative, in which case a plan’s assets are more than sufficient to cover its accruedliabilities and the plan is considered to be overfunded.The policy variable of interest is the difference between the accrued liability and the contributionamount—i.e. the “pension deficit” (“pension surplus”):PenDef t = Acct − Contribt. (3.2.4)At its core, PenDef represents the rate at which the government borrows through the state pensionsystem. Eq. 3.2.3 shows that, assuming rLiab = rAssets , a DB pension plan grows more underfunded(or less overfunded) at a rate that is increasing in PenDef . In this essay, I focus on how Governorscan manipulate PenDef through their discretion over contributions and benefit accrual policies.3.2.2. Governor Discretion over State Pension PolicyIn practice, both government employers and employees are responsible for funding state DB pen-sion plans. This means that Acc is split into two portions: the part for which government em-ployers are responsible (denote this AccGov) and the part for which employee members them-45selves are responsible (denote this AccMbrs). Similarly, Contrib consists of contributions fromthe government employers (denote this ContribGov) and contributions from employee members(denote this ContribMbrs). This means that the total pension deficit can be decomposed intoPenDef = PenDefGov + PenDefMbrs, wherePenDefGov t = AccGov t − ContribGov t, (3.2.5)represents the government’s share of the pension deficit, andPenDefMbrst = AccMbrst − ContribMbrst, (3.2.6)represents the employees’ share of the pension deficit.As chief executive of the state government, the Governor has powers to shape PenDefGov ona year-to-year basis. While other policymakers, such as state legislators, also play a role in theformulation of public pension policy, I focus on Governors due to their prominent roles in shapingthe state budget and their oversight over state administrative agencies. Furthermore, public officialswith political interests aligned with the Governor’s interests may also wield influence over pensionpolicy. For instance, members of the Governor’s cabinet, members of the Governor’s party in thestate legislature, and Governor-appointed members of the pension board all have incentives to keepthe incumbent Governor in office.53 In contrast to PenDefGov , PenDefMbrs tends to be relativelyinflexible, as employee contribution rates are typically set through collective bargaining agreementsand/or statutory provisions that require special legislative actions.Governors have significant discretion over ContribGov , which are typically approved as part ofthe budgeting and legislative appropriations process. According to Novy-Marx and Rauh (2014),pension contributions will eventually reach 14.1% of state and local budget revenues, absent signif-icant policy reforms. Historically, Governors have played a prominent role in the budget process,with the responsibility of submitting budget proposals and signing enacted budgets into law. Inmany states, Governors have the authority to veto line items and spend unanticipated funds withoutlegislative approval. In certain instances, such as in Illinois in 2006 and 2007, Governors have cutspecial deals with legislators to implement “pension holidays” that drastically reduced budgetarycontributions.There is a clear temptation for politicians to temporarily divert contributions away from statepension plans towards more immediately pressing needs. In recent years, for example, the Governorsof New Jersey54 and Connecticut55 both made cuts to state pension contributions, citing that thefunds were needed for the more urgent purpose of preventing immediate budget cuts. In certaincases, it may indeed be in the public’s best interest to use public pension plan funds as means to53 In 1993, New York State Comptroller H. Carl McCall was accused by his political opponents of giving an “election-yeargift” to his mentor Gov. Mario Cuomo by proposing a short-term reduction in state pension contributions.54 Zernike, Kate “Christie Vetoes 2015 Pension-Paying Budget” The New York Times 30 Jun. 2014.55 De Avila, Joseph, “Connecticut Governor, Unions Reach Deal to Restructure Pension Payments” The Wall StreetJournal 9 Dec. 2016.46prevent painful short-term budgetary cuts. The insight of this essay is that it is unclear why suchstopgap measures should be more prevalent during election years.Governors also play a role in determining pension benefits, albeit in a more limited capacity.Typically, pension benefits are set through long-term collective bargaining agreements or throughspecial legislative approval, which renders benefit policy less discretionary and flexible in comparisonto contribution policy. However, Governors can assert their influence over benefit policies throughtheir ability to set the legislative agenda and veto bills. In 2001, for instance, California GovernorGray Davis approved legislation that significantly increased the benefits for state employees, aftermaking public assurances that the increased benefits would put no additional pressures on thestate budget.56 By the time that it became clear that the higher pension obligations would imposesignificant fiscal burdens, Governor Davis had been re-elected to a second term in the 2002 election.Raising public pension benefits provides a channel for politicians to win the support frompolitically-powerful labour unions. For example, New York State Comptroller H. Carl McCallpursued such a strategy for a 2002 gubernatorial election bid, as media accounts at the time notedthat “Mr. McCall, who is planning a run for governor in 2002, has called for automatic pensionincreases, cementing his standing as a favorite of state workers and retirees.”57 Pension benefitincreases also serve as a potential bargaining chip that state governments can use to negotiateagainst wage concessions during labour negotiations with their employees. In fact, the relativegenerosity of public sector retirement benefits has been used to explain the earning differentialbetween public and private sector workers (Munnell et al., 2011a).3.2.3. Who Bears the Costs of Underfunded Public Pension Plans?As Eq 3.2.3 shows, unfunded liabilities are decreasing in Contrib, increasing in Acc, and decreasingin rAssets − rLiab . Therefore, a state DB plan looking to improve its funding situation must eitherraise contributions (which imposes a cost on taxpayers), lower benefits (which imposes a cost onemployees), or realize asset returns in excess of assumed discount rates. Thus, the political economyof the Governor’s decision regarding pension funding policies hinges crucially on how state pensiondebts are expected to be repaid.First, it is important to establish that reliance on excess realized returns to make up for unfundedliabilities tends to be a naive and unsustainable solution. The vast majority of state DB plansdiscount their liabilities at the expected rate of return on invested assets, usually in the 7-9% perannum range in accordance to equity-heavy portfolios. As Novy-Marx and Rauh (2011) and Brownand Wilcox (2009) point out, this severely undervalues pension liabilities, as DB liabilities shouldbe discounted at a lower rate that more appropriately reflects the underlying risk of quasi-fixedpension obligations.58 Even if one disregards the inappropriate discount rate, it is unrealistic for56 Crane, David “Dow 28,000,000: The Unbelievable Expectations of California’s Pension System” The Wall StreetJournal 19 May 2010.57 Perez-Pena “Legislators Back Pension Rises For Retired Public Employees” The New York Times 14 Jun. 2000.58 The mismatch of risk between a plan’s assets and its liabilities implies that taxpayers implicitly bear the cost of therisk premium (Bader and Gold, 2007).47state plans to expect to earn consistently above-market returns over the long run. Rauh et al.(2010) estimates that 20 states will run out of pension funds by 2025 given their current fundingpolicies, assuming average returns of 8%.Therefore, underfunded plans must eventually raise contributions or reduce benefits. Withrespect to benefits, it is generally difficult for state DB plans to cut state pension benefits thathave already accrued to employees (i.e. Liabt in Eq. 3.2.3), as accrued benefits represent debt-like obligations with strong legal protection in most states. With few exceptions, such as the2013 Detroit bankruptcy, public sector DB pension plans rarely “default” on their promises to paybenefits already accrued by employees.In certain states, government employers have more leeway to cut benefits that have yet toaccrue (i.e. reducing Acc going forward).59 At the extreme, some states operate under a “gratuity”principle, which allows employers to reduce public DB pension benefits at will. At the otherextreme, some states have constitutional provisions that prevent the state from reducing pensionbenefits that employees expect to earn over their employment tenures. According to Munnell andQuinby (2012), it is practically impossible to cut benefit accruals in such states without amendingthe state constitution.When benefit protection is weak, state employees have the incentive to monitor the governmentand prevent them from taking actions that would increase unfunded liabilities, as this would put theemployees’ retirement savings at risk. In comparison to diffuse voters who face the classic free riderproblem, employee members of state DB plans have more concentrated interests and are in a betterposition to take on a monitoring role through various institutional channels, including employeerepresentatives on pension boards of trustees and lobbying through public sector labour unions.When benefit protection is weak, however, employees are largely insulated from the consequences ofpension underfunding, and the burden falls upon taxpayers through future contribution increases.As noted in the previous section, state pension contributions come from both the government(ContribGov) and from employees (ContribMbrs). However, the employee’s share of contributionstends to be inflexible, as it is typically set through long-term labour contracts and/or requiresspecial legislative approval in a manner similar to benefit policies. Furthermore, just as employeescan pose legal challenges to attempts by state employers to cut pension benefits, they can also turnto the courts to prevent employers from raising employee contribution rates. For example, in 2012the Superior Court of Arizona ruled against S.B. 1614, a bill introduced in 2011 to reform the statepension system by increasing employee contributions, because it violated the pension protectionclause of the Arizona Constitution.In the end, increasing governmental contributions is the most plausible course of action forplans facing large unfunded liabilities. The unfunded liability for most state DB pension plans iseither implicitly or explicitly the obligation of the state government (Giertz and Papke, 2007), andsince governments are financed through tax revenue, taxpayers bear the ultimate burden of funding59 Since federal laws regulating pension benefits do not apply to state pension plans, individual states are responsiblefor the level of legal protection afforded to employees’ rights to state pension benefits.48these contributions. This sets up a potential agency conflict between incumbent politicians andtaxpayers, in that the government may borrow on behalf of taxpayers through the state pensionsystem in a manner that taxpayers would not choose for themselves. This conflict is discussed inmore detail in the following section.3.2.4. State Pension Policy OpacityWhen unfunded state pension liabilities represent a debt burden for taxpayers, the Governor’spension policy decisions should in theory be disciplined by forward-looking taxpaying voters whoanticipate that higher pension deficits incurred today necessarily imply future tax increases orspending cuts. Under the principle of Ricardian equivalence, voters with rational foresight willdiscount any current expansionary fiscal activity funded through pension deficits, and may evenpunish Governors for exhibiting fiscal imprudence (Brender and Drazen, 2008).However, state DB pension plans present a vulnerable target for political interference due totheir inherent opacity. This is because politicians have the incentive to manipulate voters’ per-ceptions of their governing abilities through “hidden” forms of borrowing that are not directlyobservable to the public. The existing literature has highlighted the importance of this informationasymmetry as the key friction in rationalizing the occurrence of political cycles in fiscal deficits(Alt and Lassen (2006), Shi and Svensson (2006)). Furthermore, Coate and Morris (1995) arguethat welfare transfers to political special interests tend to be funneled through non-transparentchannels.Voters pay limited attention to state pension finances due to free rider problems that arise whenthe future tax burden of current unfunded pension liabilities is dispersed across a large populationbase. For an individual voting taxpayer, it may not be worth the effort to delve into the details ofpublic pension plan reports and individual line items on the state budget in order to understandthe long run fiscal implications of pension contribution and benefit policies. For example, whenGovernor Cuomo raided New York state pensions in the early 1990’s, the New York Times reportedthat “there is no mystery in why politicians find the pension funds, which are worth more than $700billion nationally, such attractive targets. Reducing the amount a state gives to the funds is likelyto generate less protest from the voters than raising taxes.”60There are also various institutional reasons why public pension plans tend to be opaque tothe public. For instance, the complexity of actuarial methods used to report pension liabilitiesand determine contribution rates makes it difficult for the average voter to evaluate the long-termconsequences of pension policies. In order to estimate the expected present value of future benefits,a DB plan must make assumptions about future wage growth, mortality rates, inflation rates,discount rates, among a host of other economic and demographic factors. This makes it easy forgovernment employers to manipulate actuarial assumptions in order to “cover up” pension costs.6160 Verhovek, Sam Howe “The Region; States Are Finding Pension Funds Can Be a Bonanza Hard to Resist” The NewYork Times 22 Apr. 1990.61 Kido et al. (2012) find that state DB pension plans tend to underreport their unfunded liabilities in election yearsrelative to non-election years, and attribute their findings to politically motivated actuarial manipulations. Pension49Even if one takes the government’s financial reports at face value, institutional features of thestate budget process make it difficult for voters to observe the impact of pension policies in atimely manner. In particular, the protracted nature of the state budget and legislative appropri-ations process makes it difficult for the public to appreciate the long-term implications of statepension contributions in the short run. This implies that incumbent Governors have an especiallystrong incentive to borrow through the state pension system right before an election, with the un-derstanding that voters will likely not be able to fully appreciate the impact until after the electionis over. In the stylized model presented in Chapter 4, a temporary lag in voters’ ability to observethe impact of pension policies is sufficient to generate election year spikes in pension deficits.Figure 3.1 presents an example of a typical state budget cycle based on information provided bythe National Association of State Budget Officers (NASBO). Before the start of a given fiscal year,the Governor’s office adopts or amends a recommended contribution rate suggested by the pensionboard of trustees. After consulting with other governmental agencies, the Governor submits aproposed budget to the legislature, which is eventually finalized and signed into law just before thestart of the fiscal year. It is not until after the end of the fiscal year that a plan releases its auditedend-of-year financial statements.As Figure 3.1 shows, there is a one-year delay between when state governments set their pensioncontribution rates and when the impact on unfunded pension liabilities is reported. In addition,the impact on unfunded liabilities is generally not reported directly in the general fund budget—which covers the majority of state appropriation, expenditure and receipt transactions and is theprimary focus of public attention—but released separately via financial reports provided by thestate pension plans themselves. For instance, a 2010 New York Times report described how NewYork State officials regularly concealed costs by excluding expenses from the general fund, leadingthe State Comptroller Thomas DiNapoli to declare the state’s balance sheet to be unreliable.62In comparison to changes in pension contribution rates buried in the state budget, changesto state pension benefits are more likely to receive public scrutiny. However, the complexity ofthe actuarial valuation process may nevertheless serve to obfuscate the funding impact of benefitpolicy.63 For example, Senate Bill 400, the legislation that significantly increased benefits forCalifornia Public Employees’ Retirement System (CALPERS) participants in 2001, met with littleopposition in the state legislature after actuaries provided estimates that the investment earningson pension assets would be sufficient to cover the increased pension costs.To sum up, the information asymmetry problem stemming from the opaqueness of public pen-sion policy, combined with the moral hazard problem relating to employees being insulated from theconsequences of underfunding, create the incentives for Governors to use state pension borrowingreporting manipulations have also been documented in the private sector by Bergstresser et al. (2006) and Stefanescuet al. (2015).62 Confessore, Nicholas “Grab Bag of Gimmickry Hides State Deficit” The New York Times(City Room Blog) 6 Apr. 2010. Web. https://cityroom.blogs.nytimes.com/2010/04/06/albany-accounting-hides-deficit-size-comptroller-says/?src=mv.63 Glaeser and Ponzetto (2014) argue that “shrouded” public pension packages are better understood by public-sectorworkers than by than ordinary taxpayers.50for politically self-interested purposes at the expense of taxpayers. This intuition is formalized in astylized model presented in Chapter 4, which applies the career concerns framework of Holmstro¨m(1999) in a political setting. In the following section, I describe empirical tests used to determinewhether such distortionary reelection incentives play a significant role in driving state pensionpolicy.3.3. Empirical StrategyIn this section, I describe the empirical tests I use to evaluate how Governors’ reelection incentivesaffect governmental borrowing conducted through state DB pension plans. First, I look for anelectoral cycle in pension deficits to check whether governments increase their rates of borrowingthrough state pension plans in election years. To this end, I estimate the following OLS specification:PenDef it = α+ κi + λt + δ0 · Electionit +Xitβ + it (3.3.1)in which PenDef denotes the pension deficit, α denotes a constant intercept, κi denotes a plan-specific indicator, λt denotes a year-specific indicator variable, Xit denotes a column vector ofcontrol variables, Electionit denotes a dummy variable indicating whether an election occurs inperiod t in plan i’s state, and it denotes an unobservable mean-zero error term.We expect δ0 to be positive if pension deficits are higher in election years relative to non-electionyears. The null hypothesis is there should be no systematic electoral cycle patterns in pensiondeficits in the absence of political distortions. The credibility of this assumption is supportedby the fact that gubernatorial elections occur at pre-determined and fixed intervals and thereforeshould not be influenced by confounding factors. Furthermore,, the inclusion of year and plan fixedeffects implies that 3.3.1 essentially forms a repeated difference-in-difference estimation frameworkin which plans from states with offsetting electoral cycles serve as control groups for one another.In particular, plan-level fixed effects account for time-invariant differences between different plans,while the year fixed effects account for time-specific shocks that commonly affect all plans.I also estimate Eq. 3.3.1 using PenDefGov , the government’s share of the pension deficit,and PenDefMbrs, the employee share of the pension deficit, as the dependent variable. We ex-pect politically-motivated pension borrowing to be reflected through election year increases inPenDefGov , but Governors may also be tempted to increase PenDefMbrs as a form of election yearwealth transfer to public employees. For instance, a Governor may grant a special contributionholiday to employees in exchange for political support from unions or as a bargaining chip dur-ing election year wage negotiations. However,PenDefMbrs is relatively inflexible due to long-termlabour contracts and statutory contribution rates, as described in Section 3.2.While my baseline specification in Eq. 3.3.1 places the focus on the difference between electionyears and non-election years, it is not immediately obvious whether one should expect a sharpelection year spike in pension deficits or a more gradual increase in pension deficits throughout theelectoral cycle. The dynamics depend on whether the incumbent’s incentive to inflate performance51rises gradually as election year draws near, or whether the increased media scrutiny and voterattention in election years produce a sharp surge in the incumbent’s desire to inflate performancefor political purposes.64To investigate the full electoral cycle dynamics, I include dummy variables indicating one yearbefore the election (Electiont+1) and two years before the election (Electiont+2) in estimatingEq. 3.3.1. Positive coefficient estimates on these additional dummy variables would indicate thatincreases in pension deficits occur earlier in the electoral cycle. Note that the coefficient for thedummy variable for three years before the election (Electiont+3) is not included since it is absorbedby the intercept term, as each electoral cycle is at most four years long.Next, I separate PenDef into contributions (Contrib) and benefit accruals (Acc) and checkwhether the two components exhibit electoral cycle patterns by estimating Eq. 3.3.1 using Contriband Acc, respectively, as the dependent variable. On the contribution side, we expect the Gov-ernor’s budgetary discretion to drive election year reductions to contributions, particularly in thegovernment’s share (ContribGov). Thus, I further include an interaction term between Electionitand a dummy variable indicating the passage of a state budget (Budget Year it) to check whetherelection year reductions in ContribGov are more pronounced during budget years.On the benefits side, we expect Acc to be higher in election years relative to non-election years.However, the relative inflexibility of benefit policy makes it less likely for benefit accruals to exhibitsystematic electoral cycle patterns. Nevertheless, the incentive to grant higher benefits in exchangefor political support should be stronger in states with relatively powerful public sector labourunions. To test this empirically, I include an additional interaction term between Electionit andthe state-level public sector union membership rate (Pub Union Mbrshpit) in estimating Eq. 3.3.1with AccGov as the dependent variable.I conduct several follow-up tests to determine whether electoral cycles in pension deficits stemfrom a politically-motivated agency conflict between politicians and taxpayers. First, I exploit vari-ation in the strength of public pension benefit legal protection across states, and include interactionterms between Electionit and various measures of benefit protection strength in estimating Eq. 3.3.1.We should expect states that provide stronger benefit protection to exhibit more pronounced elec-toral cycles in pension deficits, as benefit protection insulates employees from the consequencesof underfunded pension plans and reduces their incentives to monitor the government’s pensionfunding policies.To highlight the importance of information asymmetry between Governors and taxpayers, Iinclude interaction terms between Electionit and measures of state pension transparency in esti-mating Eq. 3.3.1. We expect to find election year spikes in pension deficits to be larger for plansin states with more opaque pension systems, as the incentive to finance expansionary activitiesthrough pension borrowing depends on the incumbent’s ability to temporarily hide the pensionborrowing from taxpayers.64 In a theoretical context, a sharp election year spike may arise if the opacity of the incumbent’s actions with respect topublic pension policy is only temporary, or if the signal of the incumbent’s fiscal performance regarding his underlyingability is only informative for one period. This is discussed in greater detail in Chapter 4.52I investigate several political factors involving Governors’ reelection motives. First, I interactElectionit with the electoral margin of victory (VicMarginit) and include the term in estimatingEq. 3.3.1. Following the logic that electoral incentives are stronger for more competitive elections,we expect pension deficits to be higher for elections that are more closely contested. Next, Iexploit the existence of gubernatorial term limits by including Lame Duck it, a dummy variableindicating reelection ineligibility, in estimating Eq. 3.3.1. If reelection incentives drive pensionborrowing, then pension deficits should be higher during terms in which the incumbent Governoris reelection-eligible. Lastly, I include interaction terms between Electionit and a dummy variableindicating the incumbent Governor belongs to the Republican party (Republicanit), in order tocheck whether election year spikes in pension deficits can be explained by differences in partisanpreferences between Democrat and Republican voters.I perform several tests to evaluate the economic consequences of electoral cycles in pensionborrowing. First, I check whether election year spikes in pension deficits are associated withdeteriorating pension funding levels by estimating the following OLS specification:∆UnfundedLiabi = α+ δ · PenDefCyci + X¯β + i (3.3.2)where ∆UnfundedLiabi denotes the time series average for the annual change in the level of un-funded liabilities (scaled by payroll), PenDefCyci denotes the average time-series difference betweenelection year and non-election year pension deficits, X¯i denotes a set of control variables which havebeen averaged along the time series for plan i, and i denotes the residual error term.We expect the coefficient on PenDefCyci to be positive if larger electoral cycles in PenDef areassociated with larger increases in unfunded liabilities over the sample period. This would indicatethat state governments do not create sufficient buffers in non-election years to offset higher electionyear pension deficits, leading to steadily deteriorating funding levels over time.Note that estimating Eq. 3.3.1 with UnfundedLiabit as the dependent variable constitutes analternative way to test the impact of electoral cycles on the level of unfunded liabilities. However, asmentioned earlier, unfunded liabilities are self-reported and calculated using actuarial assumptionsand methodologies that can be manipulated, leading to under-reporting of unfunded liabilities inelection years (Kido et al., 2012).65 By taking the time-series average over the sample period inestimating Eq. 3.3.2, I circumvent this concern to a large extent, as it is much more difficult to hidefunding deterioration over a 15-year period.Ultimately, we are interested in whether electoral cycles in pension deficits lead to real economicconsequences. There is fierce debate in both policy and academic circles over how public debtimpacts economic growth. Our empirical setting allows me to ask a more specific question ofwhether “debts” incurred through the public pension system can have adverse effects on economic65 In unreported results, I find suggestive evidence that state plans overstate the value of plan assets in election years.Public pension plans use actuarial methods to smooth over fluctuations in asset values, and I find that the differencebetween the actuarial value and market value of plan assets is systematically larger in election years.53growth. To this end, I estimate the OLS following specification:ln(GDP Growth)j = α+ δ · PenDefCycj + j (3.3.3)where ln(GDP Growth)j denotes the average GDP log growth rate for state j over the sampleperiod, and PenDefCycj denotes the average PenDefCyci across sample plans in state j, weightedby plan liabilities. We expect a negative coefficient estimate on PenDefCycj if systematic electionyear spikes in state pension borrowing are associated with lower economic growth.I also estimate Eq 3.3.3 using ln(HPI Growth)j , the average log growth rate in house pricesfor state j over the sample period, as the dependent variable. This test is motivated by Epple andSchipper (1981), who show that public pension underfunding can be capitalized in house pricesthrough the market’s expectation of higher future property taxes. We should expect a negativecoefficient on PenDefCycj if systematic election year pension borrowing is capitalized throughfalling house prices.Lastly, I run several robustness tests to rule out alternative explanations for my main findings.Most importantly, I estimate 3.3.1 using a sample of private sector DB pension plans that shouldbe unaffected by Governors’ reelection incentives. This falsification test serves to address concernsthat, in the absence political incentives, electoral cycle patterns in DB pension plan policies maystill occur due to political cycles in economic conditions. For example, private firms may reduce DBpension contributions in election years due to systematic economic downturns that correlate withthe electoral cycle.66 I also address concerns that my results are driven by increased uncertaintysurrounding transition of political leadership67 by checking whether state pension deficits exhibitsystematic patterns following unexpected changes in the Governorship due to death, resignation,or impeachment.3.4. Data3.4.1. State Pension DataI investigate the annual pension deficit policies of state-administered defined benefits pension plansover the period 2001-2015. The primary source of public pension data comes from the Public PlansDatabase (PPD) maintained by the Center for Retirement Research. The PPD maintains datastarting in 2001 from 150 public pension plans, consisting of 115 plans administered at the statelevel and 35 administered at the local level, which covers 90% of public pension membership andassets in the United States.66 The existence of electoral cycles in aggregate output and employment at the national level is rejected by Alesina andRoubini (1992) who examine a sample of OECD countries. The authors find evidence of an electoral cycle patternin inflation, but their findings indicate that inflation tends to occur immediate after elections rather than beforeelections. I control for inflation assumptions in my empirical specifications.67 This is motivated by Julio and Yook (2012), who find that corporate investment tends to be lower during electionyears due to higher levels of political uncertainty.54The PPD data includes information on public pension contributions broken down by originatingsource. Using the PPD contribution measures, I construct ContribGov it, a measure of contributionsfrom the government, by aggregating regular contributions from employers (contrib ER regular)and contributions directly from the state (contrib ER state), and scaling by total covered payroll.68This represents the total discretionary governmental spending directed towards funding pensionplan i in year t, as a percentage of payroll. Scaling by payroll makes contribution rates comparablebetween plans of differing sizes, and follows public pension accounting conventions that expresspension costs as a fraction of payroll. I multiple these fractions by 100 in order to express them inpercentage terms for clearer exposition in tables.Next, I construct ContribMbrs it, a measure of contributions from participating employee mem-bers, by aggregating regular contributions from employees (contrib EE regular), contributions usedto purchase service credits (contrib EE PurchaseService),69 and other uncategorized contributionscoming from employees (contrib EE other), and scaling by total covered payroll. The aggregatecontribution rate Contribit is defined as the sum of ContribMbrs it and ContribGov it.I construct measures of benefit accruals based on normal cost rates, which are self-reportedfigures that represent the present value of benefits accrued by plan i in year t as a percentage ofpayroll. The normal cost rate is calculated by apportioning the total present value of an employee’sexpected benefits in retirement to each year of an employee’s work life, based on a specific actuarialcost method, and is reported in annual actuarial valuation reports. The PPD data provides boththe employer’s share of the normal cost rate (NormCostRate ERit), which I use as my measure ofthe government’s share of the normal cost rate, denoted AccGov it, as well as employees’ share ofthe normal cost rate (NormCostRate EE it), which I use as my measure of the employees’ share ofthe normal cost rate, denoted AccMbrs it. The total rate of benefit accruals, denoted Accit, is thesum of AccGov it and AccMbrs it.I define PenDef it, the pension deficit, as the difference between Accit and Contribit. Thismeasure represents the rate at which that the government effectively borrows from the state pensionplan, as described in Eq. 3.2.4 from Section 3.2. I further define PenDefGov it, the government shareof PenDef it, as the difference between AccGov it and ContribGov it, and define PenDefMbrs it, theemployee share of PenDef it, as the difference between AccMbrs it and ContribMbrs it.Since normal costs are actuarially-determined figures, I include observable actuarial assump-tions as control variables in order to account for changes in benefit accruals that come from ac-tuarial assumptions and not from changes in the underlying benefits. In particular, I control forcontemporaneous values of Discount Rate, the reported rate used to discount future benefit obli-gations, Inflation Rate, the assumed inflation rate used in the actuarial valuation of liabilities, andCostMthd EAN , a dummy variable that indicates whether the plan uses the Entry Age Normal(EAN) actuarial cost method in order to value its liabilities. The EAN method is the most common68 Covered payroll represents the total pensionable earnings among participants. Normalizing by payroll is standard inpublic pension accounting in order to make plans of different sizes comparable.69 Service credit contributions represents contributions made by employees to directly purchase accrued pension benefitsas a means to increase their accrued pension savings.55cost method, and also the most conservative one in terms of liability recognition. A more detailedexplanation of actuarial valuation methods can be found in Appendix C.2.In addition to contemporaneous actuarial control variables, I also include several plan-levelcontrol variables constructed from the PPD data. This includes lagged values of ln(Payroll),defined as the natural log of total payroll among plan participants, ln(Avg Salary), defined asthe natural log of average salary among plan participants, and Income, defined as the total non-contribution income (including investment income) scaled by payroll. In particular, ln(Payroll) andln(Avg Salary) control for variation in plan size and employee wage levels, while Income controlsfor changes to pension funding levels due to changes in investment returns.I keep observations which contain non-missing variables for my benchmark regression specifica-tions. This results in an unbalanced panel of 114 plans corresponding to 1,318 observations over15 years from all 50 states. I winsorize all continuous variables at the 1% level at both tails. Adetailed list of variable descriptions can be found in Appendix B.2.Table 3.1 presents the descriptive statistics for the variables used in my main regression specifi-cations. The table shows that on average, contribution rates are larger than benefit accrual rates,with the average Contrib at 17.979% of payroll and the average Acc at 12.5% of payroll. Thisresults in an average PenDef of -5.39% of payroll, indicating an average surplus. This surplus canbe attributed to the persistent underfunding of plans in my sample, which results in plans con-tributing more funds on average than accruing new liabilities in order to service the amortized costsof their unfunded liabilities. We see that the surplus is largely driven by the difference betweenContribGov and AccGov rather than the difference between ContribMbrs and AccMbrs. This isconsistent with the fact that employee shares of benefit accruals and contributions are usually set,either by contract or statute, to the same rate, while the burden of unfunded pension liabilitiesfalls upon the government.Table 3.2 presents a breakdown of pension plans by state. The number of plans in each stateranges from 1 to 5, with the average state containing 2.76 state-administered DB pension plans.Table 3.2 also includes summaries of the average size of pension plans in the sample in terms ofpayroll, as well as averages for Contrib, Acc, and PenDef . The table reveals there is substantialcross-state variation in terms of plan size as well as pension contribution and benefit policies.3.4.2. State Politics DataI obtain data on gubernatorial elections from Carl Klarner’s website (www.klarnerpolitics.com).70I supplement and verify Klarner’s Governors data set against information extracted from Bookof the States provided by the Council of State Government Knowledge Center. From these datasources, I also obtain data on gubernatorial election voting results, gubernatorial term limits, partyaffiliations of incumbent Governors, and Governors’ prior political experience. Data regardinginstitutional budgetary rules comes from the National Conference of State Legislatures website.70 I thank Carl Klarner for making early updates of his datasets available for use.56The schedule of U.S. gubernatorial elections is exogenous and set by law. Governors are electedto four-year terms in all states except for New Hampshire and Vermont, where each term is twoyears. Gubernatorial elections are held in early November in all states except for Louisiana, whichholds its elections in October. Figure 3.2 shows that gubernatorial elections are staggered over mysample period, with the majority of elections occurring two years offset from presidential elections.Figure 3.3(a) provides an illustrated map of how gubernatorial electoral cycles vary across states.I define Electionit as a dummy variable that indicates whether plan i is located in state that holdsan election in fiscal year t. Specifically, a plan-year observation is associated with Electionit = 1 ifand only if an election occurs between the start and end of fiscal year t. For example, a plan-yearobservation with fiscal year beginning in July 2006 is counted as an election year only if an electiontakes place in November 2006. This timing convention conforms to the timing of pension policychoices and election dates as illustrated in Figure 3.1, in the sense that the pension policy decisionoccurs prior to the election, and the impact on the pension plan’s funding status is revealed inaudited financial reports only after the election.I define VicMarginit as the margin of victory in percentage points between the winning guber-natorial candidate and the runner-up in year t for the state in which plan i is located. If no electiontakes place in year t, then VicMarginit is set to equal zero. I define Lame Duck it as a dummyvariable that indicates whether an incumbent Governor faces binding term limits in their currentterm.71 Figure 3.3(b) provides an illustrated map of states which impose gubernatorial term limits.I define Republicanit as a dummy variable that indicates whether the incumbent Governorbelongs to the Republic party,72 Budget Year it as a dummy variable that indicates whether a thestate passed a budget in year t, BalBudget i as a dummy variable that indicates the state is subjectto balanced budget restrictions, and LegisExpit as a dummy variable that indicates whether theincumbent Governor possesses prior experience as a member of the state legislature. Figure 3.4(a)provides a map illustrating the geographic distribution of states with biennial versus annual budgets,and Figure 3.4(b) provides a map of states with balanced budget restrictions.3.4.3. Other DataIn order to control for state-specific economic factors, I include lagged state-level control variablesin my empirical specifications. These include Deficit Shock it−1, which measures the unexpectedper capita deficit for a given state in year t − 1. This measure is constructed using data obtainedfrom NASBO’s Fiscal Survey of States following the methodology from Poterba (1994). In par-ticular, Splinter (2015) documents that states tend to reduce contributions towards public DBpension plans when they experience negative budgetary shocks. I also include State Unempit−1,the state unemployment rate taken from the Bureau of Labor Statistics Local Area UnemploymentStatistics, and Pub Union Mbrshpit−1, the state-level public sector unionization rate taken from71 The majority of states maintain term limits for their Governors, although the exact nature of the term limit candiffer from state to state.72 Republicans hold the Governor’s office in 52.35% of the plan-year observations in my sample.57Barry Hirsch and David Macpherson’s website www.unionstats.com, as additional lagged controlvariables. Descriptive statistics for these variables are included in Table 3.1, and a more detaileddescription of variable definitions is found in Appendix B.2.Data on legal protection for state employees’ pension benefits comes from Munnell and Quinby(2012). I define the Weak Protect i and Strong Protect i as dummy variables that indicate whetherplan i is located in a state that protects benefits under the gratuity principle and the constitu-tional protection principle, respectively. Some states offer benefit protection only to public sectoremployees that meet a certain threshold of employment tenure. For example, benefit protectionmay be offered only after a certain vesting period or after the employee is eligible for retirement.Accordingly, I define Unconditional Protect i as a dummy variable that indicates whether plan i islocated in a state that offers unconditional benefit legal protection.Figure 3.5(a) and Figure 3.5(b) provide illustrated geographic breakdowns of benefit protectionlegal regimes across states. Figure 3.5(a) shows several intermediate forms of benefit protectionregimes; some states protect benefits as explicit contractual arrangements (contract principle), somestates offer protection of benefits even where no contract has been explicitly stated (promissoryestoppel), and some states considers public pension benefits to be property that cannot be takenaway without due process (property principle). A comparison of Figure 3.5(a) and Figure 3.5(b)reveals the existence of states that provide unconditional but weak protection of state pensionbenefits (such as Texas), as well as states that provide strong protection of state pension benefitsthat are conditional on vesting or retirement eligibility (such as Michigan).I obtain data on institutional transparency from the State Integrity Investigation (SII), a jointdata project conducted by nonpartisan investigative news and open data organizations.73 The SIIprovides index measures based on surveys of experienced journalists that reflects the degree of stategovernment transparency and accountability across 13 different categories. I focus on the particularindices that fall under the categories of (1) state pension fund transparency and (2) state budgetprocess transparency.The SII pension transparency index is based on journalists’ survey responses to questions such aswhether “citizens can access information on state pension funds within a reasonable time period andat no cost,” and whether “state pension funds information is made available in open data format.”The score is on a scale from 0 to 100 and a higher score indicates a greater level of transparency instate pension fund management. The similarly-constructed budget transparency index is based onjournalists’ responses to questions such as whether “the state budgetary debate process is conductedin a transparent manner,” and whether “citizens can access itemized budget allocations within areasonable time period and at no cost.” Illustrated breakdowns of the geographic variation in statepension transparency and in budget process transparency scores are presented in Figure 3.6(a) andFigure 3.6(b), respectively.73 The State Integrity Investigation is a collaboration between the Center for Public Integrity, Global Integrity andPublic Radio International. The project was first carried out in 2011, and was updated in 2015 using more rigorousmethods that required reports to supply more specificity. I base my measure based on the 2015 scores. See https://www.publicintegrity.org/accountability/state-integrity-investigation/ for details.58I obtain data on state budgetary revenues and expenditures from the U.S. Census Bureau’sAnnual Survey of State Government Finances in order to check for electoral cycle patterns in severalvariables related to state fiscal policy. In particular, I construct per capita measures of tax revenues(Taxesit), general fund expenditures (Spendit), education expenditures (Edu Spendit), capitaloutlay expenditures (Cap Spendit), and police expenditures (Police Spendit). The final threeexpenditure variables listed represent items that are especially likely to be targeted for politically-motivated purposes.Lastly, I obtain data on state economic growth from the Bureau of Economic Analysis, anddata on state housing prices from the Federal Housing Finance Agency. Specifically, I constructln(GDP Growth)j as the time-series mean of the annual log growth rate of real GDP for state jover the sample period, and ln(HPI Growth)j as the time-series mean of the quarterly log growthrate of seasonally-adjusted house price index values (based on purchases only) for state j over thesample period. These variables allow me to check whether electoral cycles in pension deficits impactreal economic outcomes.3.5. ResultsIn this section, I present the results from estimating the empirical specifications outlined in Sec-tion 3.3 in order to show that political incentives distort how state governments borrow from stateDB pension plans. I also present supplementary tests and robustness checks to understand whetherthese findings are driven by contributions or benefit accruals, as well as to rule out alternative ex-planations for the documented electoral cycle patterns.3.5.1. Main ResultsTo estimate how pension deficits in election years differ from non-election years, I estimate 3.3.1using PenDef , PenDefMbrs, and PenDefGov , respectively, as the dependent variable, and presentthe results in Table 3.3. Columns (1), (3), and (5) do not include any control variables, whilecolumns (2), (4), and (6) include the full set of control variables described in the previous section.74The signs on the coefficients on the control variables lack statistical significance for the most partand are therefore difficult to interpret. All specifications presented contain year fixed effects andplan fixed effects. Standard errors are robust to heteroskedasticity and clustered at the state level.The estimates from columns (1) and (2) reveal a statistically-significant and positive relationshipbetween Electionit and PenDef it. The magnitude of the estimate is economically significant, asthe coefficient estimate in column (2) implies that governmental pension deficits as a percentageof payroll are on average 0.603 percentage points higher in election years relate to non-electionyears. Relative to the sample mean 5.392 percentage point surplus, this represents a 11.2% increase74 Note that the number of observations reported is less than than the full 1,316 sample size. This is due to the droppingof singleton groups (i.e. states with only one observation) during the estimation process. According to Correia (2015),maintaining singleton groups when fixed effects (in this case plan fixed effects) are nested within clusters (in this casestates) can overstate statistical significance and lead to incorrect inference.59(decrease) in pension deficits (surplus). With the sample average payroll at $4.67 billion per plan,this represents a difference of $28.15 million between election and non-election years in dollar terms.Columns (3) to (6) show that the electoral cycle pattern in PenDef is driven by the governmentshare of the pension deficit and not the employee share. The coefficient estimate on Electiontwhen PenDefMbrs is the dependent variable is small and statistically insignificant, while the sameestimate when PenDefGov is the dependent variable is significant and similar in magnitude to theestimates on PenDef in columns (1) and (2). This is consistent with expectations, as the Governorhas significantly greater discretion over the government’s share than over employees’ share of thepension deficit, as described in Section 3.2.Next, I estimate the same specifications as in Table 3.3 and include additional indicator variablesfor the other years in the electoral cycle. The results are presented in Table 3.4, which shows the fulldynamics of how PenDef , PenDefMbrs, and PenDefGov , respectively, vary over the electoral cycle.Column (1) shows that the pension deficit spike is confined to the final year of the electoral cycleas the coefficient on Electiont is significant while the coefficients on Electiont+1 and Electiont+2are not. Estimates from columns (2) and (3) reinforce the evidence provided by Table 3.3 inthat the election year effect is driven by discretionary governmental pension policies rather thanby inflexible employee contribution and benefit accrual rates. The magnitudes of the coefficientestimates on Electiont are similar to those found in Table 3.3, while the coefficients on Electiont+1and Electiont+2 are statistically insignificant and close to zero for all specifications.Given the increased voter engagement and media scrutiny of state politics in the lead-up to anelection, it is unsurprising that pension deficits experience a sharp increase in election years. Thesharp election year effect supports the temporary nature of the information asymmetry regardingpension policy, which renders policies undertaken in earlier years in the electoral cycle ineffectivein influencing voters’ perceptions by the time the election occurs. It is also consistent with theidea that the most recent fiscal performance is most predictive of an incumbent politician’s futureperformance, in which case voters rationally weigh the most recent fiscal year more heavily in eval-uating the incumbent candidate. Chapter 4 provides a more detailed discussion of the theoreticalbasis behind a sharp election year effect.3.5.2. Electoral Cycles in State Pension ContributionsSince pension deficits reflects the difference between benefit accruals and contributions, the docu-mented electoral cycles in PenDef can be explained by election year spikes in Acc, election yeardips in Contrib, or a combination of both. We begin by looking at contributions, as it constitutesthe more discretionary policy choice facing Governors. To this end, I estimate 3.3.1 using variouscontribution measures as the dependent variable and report the results in Table 3.5.Column (1) shows that Contrib experiences a statistically significant election year drop, which isabout equal in magnitude to the 0.603 percentage point increase in PenDef reported in Table 3.3.We see from columns (2) and (3) that the election year dips in Contrib are entirely explainedby election year dips in ContribGov . The evidence suggests that governments cut back on their60own share of pension contributions in election years, but do not provide election year contributionbreaks to employees. This is consistent with our earlier findings on pension deficits, and also in ourline with expectations relating to the Governor’s greater discretion over the government’s share ofpension contributions.I conduct additional tests to check whether larger election year contribution reductions areassociated with cases where the Governor possesses greater budgetary discretion. To this end, Iexploit the fact that 19 out of 50 U.S. states pass a state budget on a biennial rather than onannual basis. In general, annual budget cycles allow for more flexibility and responsiveness, whilebiennial budget cycles provide more opportunity for oversight.75 This means that Governors haveless discretion to influence election year pension contributions when the election coincides with anoff-budget year.I interact Electionit with Budget Year it, a dummy variable indicating a budget year, and includethe interaction term in Eq. 3.3.1. The estimation results are reported in column (4) of Table 3.5,which reveal a positive and significant coefficient estimate on Electionit × Budget Year it, and acoefficient estimate of zero on Electionit. This indicates that election year dips in governmentalcontributions are confined to budget years, thereby reinforcing the notion that budgetary discretionplays an important role in the Governor’s ability to borrow through state pension plans.I also exploit the fact that state budgets are passed via an appropriations process through thestate legislature. I interact Electionit with LegisExpit, a dummy variable that indicates whether theGovernor has prior experience as a member of the legislature, and include the interaction term inestimating Eq. (4.1). The results from column (5) of Table 3.5 reveal that the coefficient estimateon Electionit × LegisExpit is positive and statistically significant, which implies that Governorswho possess prior legislative experience leverage their experience to reduce contribution rates inelection years. Column (6) of Table 3.5 shows that the coefficients on Electionit × Budget Year itand Electionit × LegisExpit remain negative and statistically significant when both are included inthe empirical specification.If Governors cut back on state pension contributions in election years, what do they do withthe redirected funds? While we cannot directly track the redirected contribution funds dollar fordollar, we can look at overall electoral cycle patterns in state spending. The previous literature hasdocumented the occurrence of expansionary spending policies in election year, and I corroboratethose findings here by regressing various budgetary variables at the state level, including per capitaspending (Spend) and per capita tax revenue (Taxes), on the election year dummy variable and ahost of control variables.76The results are presented in Table 3.6, and while column (1) shows that an election year decreasein taxes is not statistically significant, column (2) shows that state spending tends to increase in75 See The Hon. Leon Panetta’s testimony before the House of Representatives Rules Committee (March 16, 2000), athttp://archives.democrats.rules.house.gov/archives/rules_hear09.htm.76 I also include the interaction term Electionit ×BalBudget i to compare states that allow budget deficits to be carriedover from year to year versus states that do not in order to account for the findings of Rose (2006), who show thatexpansionary spending in election years is attenuated by the presence of balanced budget requirements.61election year. These findings suggest that Governors look to expand budgetary expenditures duringelection years without raising taxes. I also examine budgetary expenditures on particularly visibleitems in columns (4)-(6) in Table 3.6. In particular, I find election year increases in particularlyvisible items, including per capita spending on education (Edu Spend), capital outlay projects(Cap Spend), and police (Police Spend).773.5.3. Electoral Cycles in State Pension Benefit AccrualsTurning to the liability side of the balance sheet, I estimate 3.3.1 using various measures of benefitaccruals as the dependent variable and present the results in Table 3.7. The positive coefficientestimate from column (1) shows that Acc tends to be higher in election years relative to non-electionyears, but the effect is not statistically significant. Results reported in column (2) and column (3)show similar findings if we use AccMbrs or AccGov as the dependent variable in the specification.The lack of significant election year effects in benefit accruals is consistent with the fact thatpension benefits are relatively inflexible as they are typically set according to multi-year labouragreements and/or require special legislative approval. Moreover, the normal cost is a noisy mea-sure of benefit accrual rates as it is determined via actuarial methods that incorporate manyassumptions about future economic and demographic conditions. Election-year increases in ben-efits may further be concealed by unobservable actuarial manipulations that understate electionyear election unfunded liabilities, as documented by Kido et al. (2012). Therefore, the coefficientsreported in Table 3.7 likely underestimate systematic election year increases in benefit accrual rates.Next, I examine instances in which we should expect to see larger and more significant elec-tion year increases in benefit accruals. In particular, we should expect larger election year benefitincreases in states with higher rates of public sector union membership if raising pension benefitsprovides a way for Governors to gain political support from labour unions in election years. Fur-thermore, we focus on AccGov rather than AccMbrs since it is self-defeating to make employeesthemselves responsible for paying for a benefit increase if the objective is to generate a welfaretransfer to workers.I interact the Electionit with Pub Union Mbrshpit and include the interaction term in estimating3.3.1 with AccGov as the dependent variable. The results are reported in column (4) of Table 3.7,and the positive and statistically significant coefficient estimate on Electionit×Pub Union Mbrshpitindicates that election year increases in state pension benefit accruals are indeed larger for plansin states with stronger public sector unions. In terms of economic magnitude, a plan in a state inthe 75th percentile of public sector union membership experiences a relative 0.33 percentage pointelection year in AccGov increase relative to a plan in a state in the 25th percentile of public sectorunion membership. Note that the negative coefficient on Electionit in column (4) suggests thatthe government may even lower its share of pension benefit accruals when public sector unions areespecially weak.77 Prior literature has found election year increases in police hiring (Levitt et al., 1997) and decreases in college tuitionrates (Reynolds, 2014).62These finding suggest an alternative interpretation to the results from Mitchell and Smith(1994), who find that higher state unionization rates are associated with lower levels of statepension funding. The authors speculate that this is due the government reducing contributions inresponse to upward pressures on salaries stemming from collective bargaining. Our results suggestthat the underfunding may also stem from public sector labour unions’ ability to increase benefitsfor their constituents by exploiting politicians’ reelection incentives, without bothering to considerhow those benefits will be funded.Since significant changes to state pension policies usually require legislative approval, I checkwhether Governors who possess legislative experience are more likely to increase benefit accrualrates in election years. To this end, I interact Electionit with LegisExpit and include it in estimating3.3.1 with AccGov as the dependent variable. The results in column (5) of Table 3.7 shows a positiveand significant coefficient on the interaction term. This suggests that legislative experience not onlyprovides Governors with more budgetary discretion over pension contributions, but also increasestheir ability to influence benefit policies. Column (6) of Table 3.7 shows that the coefficients onElectionit × Pub Union Mbrshpit and Electionit × LegisExpit remain positive and significant whenboth are included in the empirical specification.Overall, Tables 3.5 and 3.7 show that electoral cycles in pension deficits are primarily drivenby lower contributions in election years, but that in certain scenarios, the Governor may also faceelection year pressures to raise benefits. As expected, the pattern is found only in the governmentshare of contributions and benefit accruals, since these are the items over which the Governor hasdiscretion. Therefore, I focus on PenDefGov as the policy variable of interest in the followingsections.3.5.4. Electoral Cycles and Employee Benefit ProtectionIn order to understand the the political economy behind the electoral cycles documented thus far,we turn to an examination of the institutional factors that distort the incentives of incumbent Gov-ernors. First, we investigate the idea that opportunistic borrowing through state pension systemshinges on taxpayers rather than employees bearing the consequences of pension underfunding.Exploiting variation in state-level legal regimes, I interact Electionit with various indicatorsof benefit protection strength as described in Section 3.4 and include the interaction terms inEq. 3.3.1. The results are reported in Table 3.8, and show that election year spikes in pensiondeficits are significantly larger for states offering stronger legal protection as well as for statesoffering unconditional legal protection for state pension benefits. Note that the level effects for thelegal protection variables are not reported since they are time-invariant and thus absorbed by planfixed effects.The coefficient estimates on the interaction terms are economically significant. The coefficientestimate on Electionit× Strong Protect i in column (1) implies that state pension plans from statesthat provide constitutional protection of employee pension benefits experience a 1.817 percentagepoint (35.3% relative to the sample mean) election year increase in pension deficits relative to states63that do not. Similarly, states that operate under the gratuity principle experience a 1.679 percentagepoint (33.3% relative to the sample mean) election year decrease in pension deficits relative to statesthat provide stronger forms of protection. States that provide unconditional protection of statepension deficits experience a 1.009 percentage point (19.6% relative to the sample mean) electionyear increase in pension deficits relative to states that places tenure requirements on legal protectionof state pension benefits.These findings suggest that strong benefit protection which insulate employees from the futurecosts of underfunded pension plans creates a moral hazard them to ignore opportunistic electionyear pension borrowing. This creates the necessary conditions for an agency conflict betweenGovernors and taxpayers, in which the Governor borrows through the state pension system in amanner in which taxpayers may not choose for themselves.3.5.5. Electoral Cycles and Pension Plan OpacityIf taxpayers can perfectly observe governmental pension policies, then any pension policy decisionsnot in the best interests of taxpaying voters should be self-defeating from the incumbent Governor’sperspective. Thus, I investigate the idea that information asymmetry plays an important role ingenerating the distortionary reelection incentives that drive electoral cycles in pension deficits.I interact Electionit with measures of pension plan opacity and include the interaction termsin estimating Eq. 3.3.1. First, I interact Electionit with Opaque Pensions i, a dummy variableindicating if the SII state pension transparency index measure (as described in Section 3.4) is inthe bottom decile of the sample, and with Transparent Pensions i, a dummy variable indicating thesame index measure is in the top decile.Column (1) of Table 3.9 shows that the estimate on Electionit×Opaque Pensions i to be positiveand the estimate on Electionit × Transparent Pensions i to be negative. The point estimates arestatistically significant and indicate that pension plans in the bottom decile of pension transparencyexperience a 1.081 percentage point (21.5% relative to the sample mean) election year pensiondeficits increase relative to plans in the middle 80 percentile, while pension plans in the top decileof pension transparency experience a 1.228 percentage point (23.5% relative to the sample mean)election year pension deficit decrease. The economic magnitudes and significance of the estimatesdo not change much when both interaction terms are included together in one specification, asreported in column (4).Since, the state budget process ultimately determines pension contributions, I conduct a similartest using the SII indicator for the transparency of the state budget process. I interact Electionitwith Opaque Budget i, a dummy variable indicating whether the SII budget transparency indexmeasure is in the bottom decile of the sample, as well as Transparent Budget i, a dummy variableindicating whether the same index measure is in the top decile.Column (2) of Table 3.9 reveals that the estimate on Electionit×Opaque Budget i to be positivebut insignificant, while the estimate on Electionit×Transparent Budget i is negative and significant.The point estimate on the latter term indicates that states in the top decile of budget transparency64experience a 0.794 percentage point (15.6% relative to the sample mean) lower election year pensiondeficit spike relative to plans in the middle 80 percentile.Overall, the results reported in Table 3.9 support the idea that information asymmetry forms akey friction in generating the incentive distortions that drive election year spikes in pension deficits,and further suggest that pension transparency is more important than budgetary transparency.When all interaction terms are included in column (3), the coefficient estimates on the budgettransparency interaction terms are no longer significant while the estimates on the pension trans-parency interaction terms remain largely unchanged. A possible explanation is that nontransparentbudgetary process provide incumbent Governors with alternative channels to fund opportunisticelection year activities, such as delaying infrastructure investment.3.5.6. Electoral Cycles and Political FactorsI investigate various political factors to determine whether Governors’ reelection concerns drivetheir incentives to borrow opportunistically through state pension plans. First, I test whetherelectoral cycles in pension deficits are stronger for elections that are more closely contested. Tothis end, I include VicMarginit, an inverse measure of election closeness, in estimating Eq. 3.3.1.78.The results are presented in Table 3.10, and column (1) shows that the coefficient estimateon Electionit × VicMarginit is indeed negative and statistically significant. The point estimate of-2.232 implies that a close election in which the winning candidate barely edges out the runner-upcandidate is associated with an election year spike in pension deficits that is 0.446 percentage points(8.8% relative to the sample mean) higher than an election in which the winning candidate prevailsby a margin of 20 percentage points.Next, I include Lame Duck it, a dummy variable indicating whether binding term limits applyto the incumbent Governor, in estimating Eq. 3.3.1. The results are presented in column (3)of Table 3.10 and the negative estimate on Lame Duck it reveal that lame duck (i.e. reelection-ineligible) Governors incur lower pension deficits on average, which is consistent with the idea thatpoliticians who are unable to seek reelection have a weaker incentive to inflate their performancethrough concealed pension borrowing. Interesting, Besley and Case (1995) and Alt et al. (2011)find that taxes and spending are higher under lame duck Governors, which the authors attributeto reduced fiscal prudence stemming from a lack of electoral accountability. My findings suggestsa silver lining to the lower accountability associated with lame duck terms, as it may serve to limitdistortionary actions motivated by reelection ambitions.Surprisingly, the estimated coefficient on interaction term Electionit×Lame Duck it is positive,which implies that reelection-ineligible Governors incur higher pension deficits in election years.However, this result is potentially confounded by electoral competitiveness, as reelection-eligibleincumbent Governors tend to enjoy a significant electoral advantage (Ansolabehere and Snyder Jr,78 Since voting occurs only during election years, VicMarginit is set to zero for non-election years. This means that wedo not need to include the interaction term between VicMarginit and Electionit, since the coefficient on VicMarginitdirectly captures the marginal effect of election closeness conditional on the occurrence of an election year652002).79 Indeed, the statistical significance of the interaction term is statistically weak and disap-pears when the terms involving VicMarginit are included in the specification, as reported in column(5).Lastly, we check whether party affiliation have any effects on a Governor’s propensity to raisepension deficits during election years. U.S. politics is dominated by a two party system, and eachparty may wish to cater to its core constituency, with Democratic voters preferring higher spendingand Republican voters preferring lower taxes.80 Therefore, we must consider the possibility thatelectoral cycles in pension deficits, rather than being a sign of distorted political agency, simplyreflect the policy preferences of a partisan electorate.I interact Electionit with Republicanit, and include the interaction term in Eq. 3.3.1. Theestimation results are reported in column (3) of Table 3.10, and show that there is no statisticallysignificant effect of having an incumbent Republican Governor relative to having an incumbentDemocrat Governor. The estimate remains insignificant when interaction terms relating to otherpolitical variables are included in column (4). These results suggest that electoral cycles in pensiondeficits are not driven by policies catered to the political preferences of one particular party’spartisan base.3.5.7. Consequences of Electoral Cycles in Pension DeficitsThus far, we have shown that pension deficits tend to be higher in election years relative to in non-election years. The natural follow-up is to determine whether such electoral cycles lead to increasesin the level of unfunded liabilities over time. The more benign possibility is that governmentsaccumulate sufficient pension surpluses in non-election years to offset the increased election yearpension borrowing.81 The other possibility is that each successive incumbent chooses to “kick thecan down the road” by not accumulating sufficient buffers during non-election years.Following steps outlined in Section 3.3, I collapse my sample along the time series and estimateEq. 3.3.2 where the variable of interset is PenDefCyci, the average difference in election year pensiondeficits and non-election year pension deficits, and the dependent variable is ∆UnfundedLiabi, theaverage change in unfunded liabilities over the sample period. The estimation results are reportedin Panel A of Table 3.11, and show that the point estimate is positive across all specification andstatistically significant at the 1% level, even when state fixed effects are included in columns (2),(4), and (6). This indicates that a greater degree of electoral cyclicality in PenDef is associated79 Another possibility is that the incumbent’s party exerts greater influence towards the end of the incumbent’s lameduck term, and the party is strongly motivated to secure the election for the successor candidate, whose chances ofvictory are helped by burnishing in the incumbent party’s perceived performance.80 The previous literature has found mixed results in identifying partisan differences in opportunistic fiscal activitiesby the two major U.S. political parties. Poterba (1994) finds no difference in electoral cycles in fiscal policy at statelevel. Alesina et al. (1997) find Democrats tend to be associate with more expansionary monetary policy, but only infirst half of electoral cycle. Cunha et al. (2016) find that Democrats are more likely to exploit exogenous reductionsto credit constraints.81 Note that artificial cycles in pension borrowing may still be welfare-destroying in this scenario if taxpayers prefersmooth policy paths with respect to fiscal policy—in effect, politicians may be gambling with taxpayer dollars byputting the state balance sheet in a vulnerable state following every election.66with a larger increase in the level of unfunded pension liabilities over time, which implies thatstate governments do not “save up” in non-election years to sufficiently offset higher election yearpension deficits.Columns (1) and (2) report results using the baseline definition of PenDefCyci, which is thedifference, for each plan i, between the time series average of PenDef it conditional on t being anelection year and the time series average of PenDef it conditional on t being a non-election year.The point estimate of 1.306 in column (2) indicates that the average plan, which experiences a0.603 percentage point difference between election year and non-election year PenDef accordingto Table 3.3, experiences a 0.788 percentage point higher ∆UnfundedLiabi over the sample period.This accounts for 6.65% of the sample mean of ∆UnfundedLiabi (11.02 percentage points), whichimplies that the electoral cyclicality of pension deficits can explain an economically significantportion of the increasing level of unfunded pension liability over the sample period.Columns (3) and (4) report the same estimation results using a measure of pension deficitcyclicality that has been adjusted for aggregate time trends. Specifically, PenDefCycD i is definedin the same manner as PenDefCyci, but uses the estimated residual terms from the OLS regressionPenDef it = α + δ · t + it instead of the raw PenDef it when computing conditional time seriesaverages. The point of removing the linear time trend component is to ensure that the measure ofcyclicality is not influenced by some plans having their electoral cycles starting later in the sampleperiod relative to other plans. The coefficient estimates on PenDefCycD i are similar in magnitudeto those for PenDefCyci and remain statistically significant.Similarly, Columns (5) and (6) report the same estimation results using a measure of pensiondeficit cyclicality that has been adjusted for control variables, plan fixed effects, and time fixedeffects. Specifically, PenDefCycRi is defined in the same manner as PenDefCyci, but uses theestimated residuals from the OLS regression PenDef it = α+κi+λt+Xitβ+it instead of using theraw PenDef it when computing conditional time series averages. Again, the coefficient estimates onPenDefCycRi are similar in magnitude to those for PenDefCyci and remain statistically significant.There is heated debate about whether government debt affects economic growth.82 Thus, itis natural to ask whether large unfunded liabilities affects state economic growth. For example,expectations of future tax increases may imply lower expected firm profits and individual incomes inthe future, leading to lower investment and consumption. Large unfunded liabilities may also driveprofitable businesses and high-income individuals to relocate in order to escape local tax regimes.Lastly, states such as Illinois have struggled with indecision over what policies to use to addresslarge pension shortfalls, and policy uncertainty is also negatively associated with investment andgrowth (Gulen and Ion, 2015).I test whether electoral cycles in pension deficits are associated with changes in real economicoutcomes—in particular, growth rates in state GDP. I compute state-level measures of pensiondeficit cyclicality, following steps described in Section 3.3, and estimate Eq. 3.3.3. Panel B of82 For example, Rogoff and Reinhart (2010) find a negative relationship between national public debt and GDP growthat high levels of debt-to-GDP ratios, but their findings have been questions and debated over.67Table 3.11 reports the estimation results from using ln(GDP Growth), the state GDP log growthrate, as the dependent variable in columns (1)-(3). The negative coefficients in columns (1)-(3)suggest that larger electoral cycles in state pension deficits are associated with lower economicgrowth, although only two of the coefficient estimates are (weakly) significant and the sample sizeof 50 states is limited.Another real consequence of pension underfunding is the possibility of lower house prices. AsEpple and Schipper (1981) show, unfunded pension liabilities can be capitalized through houseprices if the housing market rationally impounds expectations of higher future taxes into currentprices. The negative coefficients in columns (4)-(6) of Panel B in Table 3.11, in which the houseprice index log growth rate ln(HPI Growth) is the dependent variable, are consistent with thisinterpretation, but the estimates are not statistically significant. One potential explanation for theweakness of this result (aside from the small sample size) is provided by Brinkman et al. (2016),who show that downpayment constraints in the housing market can dampen the capitalization ofunderfunded liabilities into house prices.We note that the evidence presented regarding real outcomes is only suggestive and does notnecessarily imply causal connections. In particular, reverse causality is a major concern whenexamining the relationship between unfunded pension liabilities and growth. However, it is lessclear why slower growth would lead to more pronounced electoral cycles in pension deficits. Inaddition, the described mechanisms behind how large public pension debts cause slower economicgrowth operate through the channel of rational taxpayer expectations, which may at first appear atodds with the underlying opaqueness of public pension systems expounded in this paper. However,it is reasonable for rational expectations to form over longer time horizons, and the 15 years that isincorporated into the cyclicality measure is much longer than the one year disclosure delay describedin Section 3.2.3.5.8. Falsification TestsMy benchmark empirical tests rely on the identifying assumption that, in the absence of politicaldistortions, pension policies should not exhibit any systematic electoral cycle patterns. A naturalway to test this assumption is to examine corporate DB pension plans in the private sector, whichshould be immune from political incentives relating to state gubernatorial elections. Therefore,running my benchmark tests on a sample of corporate DB plans provides a natural placebo test onmy main findings.I construct a sample of corporate DB pension plan policies using data from the CompustatPension Annual database (ACO PNFNDA). I construct the dependent variables and control vari-ables using the same method as in the public plan sample, with PenDefFirm, ContribF irm, andAccF irm as the dependent variables. Corporate plans face different reporting and regulatory stan-dards relative to public sector plans, so many variables may not be perfectly comparable betweenthe corporate sample and public plan sample.83 Compustat does not report the inflation assump-83 I scale the private pension deficit, contribution, and accrual variables by the payroll variable XLR in order to match68tions and the actuarial cost method made by corporate plans. However, it does include the wagegrowth assumption, which I include as an additional control variable.84 I assign each corporateplan to the state of its headquarters in order to match it to the gubernatorial election data.The results from estimating 3.3.1 on the sample of corporate DB plans are presented in Ta-ble 3.12. The results show no election year effect for any of the specifications, as all coefficientestimates for election year dummy variables are statistically insignificant. This result provides evi-dence in support for the assumption that pension policies unaffected by political incentives do notexhibit electoral cycle patterns, which implies that the electoral cycle patterns that I identify inpublic sector DB pension plans are driven by political incentives.I also exploit occurrences of sudden Governor changes due to death, resignation, or impeachmentin order to address the concerns that my results are driven by leadership transition effects unre-lated to reelection considerations. In particular, I address the concern that additional uncertaintyassociated with election years may affect public pension policies. For instance, the governmentmay choose to finance expansionary policies through pension borrowing in order to stimulate theeconomy in response to uncertainty-induced economic slowdowns.Following this logic, sudden and unexpected changes in Governors due to exogenous causesshould also be associated with periods of high political uncertainty. Therefore, I estimate thefollowing OLS specificationPenDef it = α+ κi + λt + ν0 ·Gov Changeit +Xitβ + it (3.5.1)in which Gov Changeit represents a dummy variable that indicates whether there was an unexpectedchange in the state Governorship in year t due to death, impeachment, or resignation.The estimation results are reported in Table 3.13, and show that sudden Governor changesdo not have detectable effects on PenDef , PenDefMbrs, or PenDefGov . The same is true if oneincludes a lagged value of Gov Change in the specification, as reported in columns (2), (4), and (6),in order to account for the possibility that political uncertainty over unexpected Governor changespersists for more than one year. These results suggest that it is anticipation of reelection prospects,rather than leadership transitions per se, that drives election year spikes in pension deficits. Notethat there are few occurrence of unexpected Governor changes in my sample (60 out of 1,318 plan-year observations in sample). This leads to large estimated errors that limit the statistical powerof the test.3.5.9. Other Robustness ChecksAs a final robustness check, I address concerns that my main results are driven by regional shocksthat affect a small number of states that share the same gubernatorial election schedules. As seenthe variable construction of their public plan counterparts. However, XLR is missing for the majority of firms andthereby significantly limits the sample size. If I scale by total employment (EMP), which has significantly fewermissing observations, I obtain qualitatively similar results.84 The PPD data also includes wage growth assumptions but it is missing for most of the sample.69in Figure 3.2, the majority of states hold their elections in years that are two years offset frompresidential elections (i.e. in 2002, 2006, 2010, etc.). The concern is that regional shocks thataffect the small number of states that are “off-cycle” from this dominant schedule drive my mainfindings. Due to the potential clustering of state election schedules, there is also the concern thatcorrelated pension policies across states could lead to correlated standard errors that understatestandard error estimates in my benchmark tests.To address these concern, I estimate my benchmark test following 3.3.1, but add region × yearfixed effects as well as cluster standard errors by year in addition to by state.85 The inclusionof region × year fixed effects controls for time-varying shocks at the census region level,86 whileclustering by standard errors by year accounts by correlation of standard errors across states withina given year. The results are reported in Table 3.14, and show that my main estimation resultsremain largely unchanged whether one includes region × year fixed effects, clusters by year and bystate, or does both. Figure 3.3(a) illustrates that on-cycle states and off-cycle states do not followobvious patterns of geographic clustering, which should further mitigate concerns that my resultsare driven by correlated pension policies across states that cluster together geographically.3.6. ConclusionIn this essay, I investigate an electoral cycle in the borrowing state governments conduct throughpublic DB pension plans. The premise is that state Governors, who possess discretion over publicpension policy, face incentives to increase “pension deficits” for politically motivated purposes. Theresult is a systematic pattern in which pension borrowing is higher during election years relativeto non-election years. I present empirical evidence that state DB pension plans increase their rateof borrowing during election years, and that this pattern is driven by election year reductions ingovernmental contributions. I run additional tests in order to rule out alternative explanations forthe documented electoral cycle patterns.I find strong empirical support that electoral cycles in pension deficits are rooted in an agencyconflict between politicians and taxpayers. In particular, election year spikes in pension deficits arelarger in states which place the burden of unfunded public pension liabilities on taxpayers ratherthan state employees, and which contain less transparent public pension system. I also find thatGovernors’ reelection incentives drive pension funding policy, as pension deficits are higher duringmore closely contested elections and during the terms of reelection-eligible incumbents.My work offers implications regarding potential policy remedies to address the distortionaryincentives underlying electoral cycles in pension deficits. One possibility is to place stricter re-strictions that limit governmental discretion over contributions. For example, Kentucky passedlegislation in 2013 that required state governments to follow up on their contribution promises.85 Clustering by year be problematic as the number of years in my sample is not large. This can lead to a downwardbias in the cluster-robust variance matrix estimate and consequently over-rejection of the null hypothesis. Therefore,I follow the suggestions of Cameron and Miller (2015) and use bootstrap clustering methods in order to estimatestandard errors when clustering by year.86 The U.S. consists of four census regions: Northeast, Midwest, South, and West.70Another potential solution is to address the underlying opacity of public pension plans. For exam-ple, the Governmental Accounting and Standards Board (GASB) recently passed new disclosurerules that placed stricter restrictions on the use of discount rates and actuarial smoothing method-ologies. Reforming pension systems by loosening protection over state pension benefits presentsanother option to mitigate the conflict between politicians and taxpayers. However, reducing ben-efit protection may have unintended effects on the labour supply decisions of for public sectoremployees, and therefore should be approached with great care.Lastly, my results suggest that electoral cycles in state pension borrowing have real conse-quences. In particular, I find that plans that exhibit larger election year spikes in pension deficitsalso experience larger increases in total unfunded liabilities over the sample period. This suggeststhat state governments do not accumulate sufficient buffers during non-election years to offsetthe higher election year pension borrowing. I also find suggestive evidence that states that con-tain plans that exhibit larger electoral cycles in pension borrowing also experience lower economicgrowth. However, much more work is needed to improve our understanding of how public pensionunderfunding affects the real economy.71Figure 3.1: Illustrative Example of Institutional Timeline72Figure 3.2: Frequency of Gubernatorial Elections (2001 to 2015)010203040Number of Gub. Elections200120022003200420052006200720082009201020112012201320142015Year73Figure 3.3: Geographic Variation in Political InstitutionsPres. Election +3Pres. Election +2Pres. Election +1Coincide w/ Pres. ElectionSource: Klarner Politics(a) Gubernatorial Electoral Cycles (as of 2015)Gub. Term LimitsNo Gub. Term LimitsSource: Klarner Politics(b) Gubernatorial Term Limits (as of 2015)74Figure 3.4: Geographic Variation in Budgetary InstitutionsBiennial BudgetAnnual BudgetSource: National Conference of State Legislatures(a) Annual vs. Biennial Budget Cycles (as of 2015)Can carry over deficitCannot carry over deficitSource: National Conference of State Legislatures(b) State Balanced Budget (No-Carry-Over Rule) Restrictions75Figure 3.5: Geographic Variation in Public Pension Benefit Protection Legal RegimesConstitutional ProtectionContract PrinciplePromestory EstoppelProperty PrincipleGratuity PrincipleSource: Munnell & Quinby (2012)(a) State Pension Benefit Legal Protection RegimesConditional ProtectionUnconditional ProtectionSource: Munnell & Quinby (2012)(b) State Pension Benefit Legal Protection Conditions76Figure 3.6: Geographic Variation in Transparency Indicators(78,96](66,78](62,66](50,62][30,50]Source: Center for Public Integrity State Integrity Investigation(a) State Integrity Investigation Transparency Score for State Pension Fund Management(88,98](78,88](74,78](66,74][52,66]Source: Center for Public Integrity State Integrity Investigation(b) State Integrity Investigation Transparency Score for State Budget Process77Table 3.1: Descriptive StatisticsThis table presents summary statistics for the variables in my benchmark regression specifications. The sampleconsists of 114 state-administered public pension plans (covering all 50 states) over the period 2001 to 2015. Contribdenotes the total pension contribution scaled by payroll, ContribMbrs denotes the employee pension contributionscaled by payroll, ContribGov denotes the governmental pension contribution scaled by payroll, Acc denotes thetotal benefit accrual scaled by payroll, AccMbrs denotes the employee benefit accrual scaled by payroll, AccGovdenotes the governmental benefit accrual scaled by payroll, PenDef denotes the pension deficit scaled by payroll,PenDefMbrs denotes the employee pension deficit scaled by payroll, PenDefGov denotes the governmental pensiondeficit scaled by payroll, Election denotes a dummy variable for a gubernatorial election year, ln(Payroll) denotes thenatural log of total payroll among plan participants, ln(Avg Salary) denotes the natural log of average salary amongplan participants, Income denotes non-contribution income scaled by payroll, Discount Rate denotes the assumeddiscount rate reported by the plan, Inflation Rate denotes the inflation rate assumed by the plan, CostMthd EANdenotes a dummy variable for Entry Age Normal being the actuarial cost method, Deficit denotes the per capitaunexpected state deficit, State Unemp denotes the state unemployment rate, and Pub Union Mbrshp denotes thestate unionization rate among public-sector workers. Detailed definitions for all variables can also be found inAppendix B.2. All variables except for Election are winsorized at the 1% level at both tails. Missing variablesaccount for differences in number of observations.Observations Mean Std Dev P25 Median P75Pension ContributionsContrib 1,316 17.979 9.469 11.866 16.667 22.428ContribMbrs 1,316 6.052 3.560 3.625 6.408 8.234ContribGov 1,318 11.911 8.604 6.493 10.200 14.290Pension AccrualsAcc 1,318 12.500 4.270 9.870 11.529 14.480AccMbrs 1,318 5.710 2.832 3.990 6.000 7.689AccGov 1,318 6.852 4.113 4.170 6.030 8.250Pension DeficitsPenDef 1,316 -5.392 8.258 -8.236 -3.938 -0.796PenDefMbrs 1,316 -0.319 1.848 -0.633 -0.139 0.126PenDefGov 1,318 -5.065 7.967 -7.966 -3.619 -0.429Electoral CycleElection 1,318 0.262 0.440 0.000 0.000 1.000Plan-Level Control Variablesln(Payroll) 1,318 7.851 1.107 7.190 7.902 8.564ln(Salary) 1,318 3.746 0.273 3.568 3.734 3.912Income 1,318 0.219 0.485 -0.048 0.296 0.525Discount Rate 1,318 0.079 0.004 0.075 0.080 0.080Inflation Rate 1,318 0.034 0.006 0.030 0.032 0.035CostMthd EAN 1,318 0.804 0.397 1.000 1.000 1.000State-Level Control VariablesDeficit Shock 1,318 -0.018 0.108 -0.069 -0.011 0.033State Unemp 1,318 0.063 0.020 0.048 0.059 0.075Pub Union Mbrshp 1,318 0.333 0.177 0.175 0.282 0.50978Table 3.2: Average Payroll and Pension Policies by StateThis table presents a state-by-state summary, including the number of plans for each state, as well as the averagePayroll, average Contrib, average Acc, and average PenDef for each state. Detailed definitions for all variables canbe found in Table 3.1 as well as Appendix B.2.Number of Plans Payroll Contrib Acc PenDefAK 2 1021.927 26.387 13.881 -12.656AL 2 4520.158 14.695 9.917 -4.778AR 2 1441.067 13.712 11.391 -2.329AZ 3 3458.666 19.558 15.492 -4.067CA 3 21108.221 16.731 18.415 0.378CO 3 2568.761 22.707 12.953 -9.754CT 3 3160.995 26.223 10.294 -15.191DE 1 1670.206 8.641 9.776 1.136FL 1 24803.133 10.635 10.084 -0.550GA 2 5944.057 14.492 10.742 -3.750HI 1 3504.771 15.822 10.391 -5.474IA 2 4544.873 17.596 14.508 -3.088ID 1 2469.950 17.345 14.204 -3.141IL 4 5418.373 26.020 16.094 -9.827IN 2 4302.572 14.900 8.882 -5.839KS 1 5815.819 12.321 8.345 -3.976KY 3 2474.669 18.614 11.087 -7.526LA 5 2153.382 30.756 15.132 -15.627MA 2 4868.025 23.433 11.528 -11.903MD 2 4772.687 14.643 10.924 -3.719ME 2 937.685 18.887 14.073 -4.814MI 3 4292.982 21.670 9.700 -11.499MN 4 3085.265 14.641 11.105 -3.381MO 5 1795.267 21.466 12.715 -8.747MS 1 5319.257 20.737 10.561 -10.176MT 2 851.190 16.865 10.967 -5.898NC 2 8071.481 11.771 12.003 0.232ND 2 569.737 13.559 9.584 -3.539NE 1 1436.878 17.937 11.438 -6.500NH 1 2431.064 17.213 10.520 -6.693NJ 3 9382.403 15.950 9.340 -6.612NM 2 2116.240 22.768 16.577 -6.191NV 2 2591.627 26.258 21.507 -4.749NY 3 12400.116 10.369 12.063 1.575OH 4 5428.947 22.380 15.106 -6.258OK 3 2042.291 22.326 14.089 -8.231OR 1 8281.775 8.409 8.120 -0.286PA 3 6382.234 13.381 14.296 0.914RI 2 931.877 20.295 11.878 -8.418SC 2 4047.379 18.680 11.277 -6.637SD 1 1346.206 13.401 11.771 -1.629TN 2 3902.333 12.996 9.611 -3.384TX 5 9267.403 12.331 10.346 -1.693UT 2 2139.902 20.159 16.843 -3.301VA 1 13771.987 8.664 9.330 0.666VT 2 441.174 12.471 8.978 -3.493WA 4 3392.746 7.137 10.897 3.715WI 1 11890.885 11.458 12.908 1.450WV 2 1203.019 29.414 9.673 -18.710WY 1 1473.984 11.922 11.161 -0.761Total 3 4667.729 17.979 12.500 -5.39279Table 3.3: Electoral Cycles in Pension DeficitsThis table reports the estimation results from the OLS regression PenDef it = α+κi +λt + δ0 ·Electionit +Xitβ+ itin columns (1) and (2). In column (3) and (4), PenDef is replaced by PenDefMbrs as the dependent variable, andin columns (5) and (6), PenDef is replaced by PenDefGov as the dependent variable. The variables of interest isElectionit and coefficient δ0 captures the relative difference in the outcome variable between election years and non-election years. Xit denotes the set of control variables, and is included in columns (2), (4), and (6). Control variablesincluded lagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub UnionMbrshp,as well as contemporaneous values of Discount Rate, Inflation Rate, and CostMthdEAN . Detailed definitions forall variables can be found in Table 3.1 as well as Appendix B.2. All specifications include both plan and year fixedeffects. Standard errors are corrected for heteroskedasticity and clustered at the state level. The sample consists of114 state-administered public pension plans for the period 2001 to 2015 described in Table 3.1. Standard errors arein parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)PenDef PenDef PenDefMbrs PenDefMbrs PenDefGov PenDefGovElection 0.581** 0.603*** -0.021 -0.012 0.605** 0.613***[0.218] [0.211] [0.091] [0.089] [0.231] [0.199]ln(Payroll) 12.704 -2.879** 15.560*[9.134] [1.353] [8.090]ln(Salary) -15.472* -1.986 -13.467[8.795] [1.679] [8.641]Income -0.843* 0.073 -0.904**[0.450] [0.122] [0.420]Deficit Shock 0.631 -0.208 0.827[2.863] [0.383] [2.761]State Unemp -30.748 -2.701 -28.202[29.112] [10.717] [24.828]Pub Union Mbrshp -6.051 -1.564 -4.493[8.992] [1.864] [8.012]Discount Rate 63.339 -0.198 63.692[101.704] [34.699] [111.216]Inflation Rate -53.375 11.455 -64.960[46.087] [21.108] [44.761]CostMthd EAN -2.899 -0.186 -2.713[2.873] [0.298] [2.603]Fixed Effects Plan, Year Plan, Year Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,312 1,312 1,312 1,312 1,314 1,314Adjusted R-squared 0.649 0.677 0.565 0.589 0.632 0.67280Table 3.4: Dynamics of Electoral Cycles in Pension DeficitsThis table reports the estimation results from the OLS regression Yit = α+κi+λt+∑2j=0 δj ·Electionit+j+Xitβ+it,where the outcome variable Yit is PenDef it in column (1), PenDefMbrsit in column (2), and PenDefGov it in column(3). The coefficients δj ’s captures how contribution rates are affected by proximity to gubernatorial elections ona year-to-year basis over the electoral cycle. All specification include the set of control variables Xit, includinglagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well ascontemporaneous values of Discount Rate, Inflation Rate, and CostMthd EAN . Detailed definitions for all variablescan also be found in Table 3.1 as well as Appendix B.2. All specifications include both plan and year fixed effects.Standard errors are corrected for heteroskedasticity and clustered at the state level. Standard errors are in parentheses,with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3)PenDef PenDefMbrs PenDefGovElection 0.759*** 0.074 0.686***[0.272] [0.139] [0.242]Election(t+1) -0.005 0.006 -0.008[0.398] [0.087] [0.373]Election(t+2) 0.368 0.198 0.176[0.323] [0.149] [0.274]ln(Payroll) 12.715 -2.873** 15.566*[9.131] [1.348] [8.092]ln(Salary) -15.559* -2.027 -13.513[8.712] [1.690] [8.558]Income -0.867* 0.062 -0.917**[0.455] [0.119] [0.430]Deficit Shock 0.690 -0.176 0.856[2.854] [0.390] [2.741]State Unemp -31.484 -3.088 -28.554[29.330] [10.866] [24.943]Pub Union Mbrshp -5.949 -1.514 -4.442[9.041] [1.871] [8.057]Discount Rate 61.416 -1.189 62.745[101.847] [35.326] [110.901]Inflation Rate -53.079 11.616 -64.809[46.011] [20.933] [44.827]CostMthd EAN -2.913 -0.194 -2.720[2.875] [0.300] [2.606]Fixed Effects Plan, Year Plan, Year Plan, YearObservations 1,312 1,312 1,314Adjusted R-squared 0.677 0.590 0.67281Table 3.5: Electoral Cycles in Pension Contribution RatesThis table reports the estimation results from the OLS regression Yit = α + κi + λt + δ0 · Electionit + Xitβ + it in columns (1) to (3), ContribGov it =α+ κi + λt + δ0 · Electionit + ρ · Budget Year it · Electionit + pi · Budget Year it +Xitβ + it in column (4), and ContribGov it = α+ κi + λt + δ0 · Electionit + ρ ·LegixExpit ·Electionit+pi ·LegixExpit+Xitβ+ it in column (5), where Yit represents various measures of pension contribution rates, BudgetY earit is a dummyvariable indicating whether there a state budget passed in year t, and LegisExpit is a dummy variable indicating whether the incumbent Governor has priorexperience in the state legislature. Column (6) reports the results from including all terms from columns (4) and (5). All specification include the set of controlvariables Xit, including lagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneousvalues of Discount Rate, Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as well as Appendix B.2. Allspecifications include both plan and year fixed effects. Standard errors are corrected for heteroskedasticity and clustered at the state level. Standard errors arein parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)Contrib ContribMbrs ContribGov ContribGov ContribGov ContribGovElection -0.618** 0.011 -0.628*** 0.200 -0.328 0.466[0.234] [0.086] [0.219] [0.218] [0.232] [0.280]Election × Budget Year -1.179*** -1.141***[0.366] [0.365]Budget Year 0.574*** 0.553***[0.176] [0.162]Election × LegisExp -1.830** -1.789**[0.893] [0.871]LegisExp 1.460* 1.454*[0.812] [0.811]Control Variables Yes Yes Yes Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,312 1,312 1,314 1,314 1,314 1,314Adjusted R-squared 0.726 0.871 0.692 0.692 0.694 0.69482Table 3.6: Electoral Cycles in State Fiscal OutcomesThis table reports the estimation results from the OLS regression Yit = α+ κi + λt + δ0 ·Electionit + ρ0 ·BalBudgeti ·Electionit +Xitβ + it, where BalBudgetitakes on a value of one if state i does not allow deficits to be carried over from one year to the next. In The outcome variable Yit is Taxesit (per capita tax revenue)in column (1), Spendit (per capita general fund expenditure) in column (2), Edu Spendit (per capita expenditure on education) in column (3), Cap Spendit(per capita expenditure on capital outlays) in column (4), and Police Spendit (per capita expenditures on police) in column (5). Xit denotes the set of controlvariables, which include lagged values of State Unemp, Pub Union Mbrshp, State GDP , Deficit Shock , and State Debt. All specifications include both state andyear fixed effects. Standard errors are corrected for heteroskedasticity and clustered at the state level. The sample consists of 50 states for the period 2001 to2015. Standard errors are in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5)Taxes Spend Edu Spend Cap Spend Police SpendElection -0.006 0.063*** 0.034** 0.014* 0.002**[0.024] [0.022] [0.013] [0.008] [0.001]Election × BalBudget 0.022 -0.026 -0.024* -0.005 -0.001[0.028] [0.025] [0.013] [0.008] [0.001]State Unemp 0.027 -0.051 -0.021* -0.007 -0.001[0.026] [0.034] [0.011] [0.008] [0.001]Pub Union Mbrshp 0.011 -0.000 0.002 -0.002 0.000[0.010] [0.008] [0.003] [0.001] [0.000]State GDP 0.111*** 0.074*** 0.020*** 0.017*** 0.000[0.025] [0.016] [0.005] [0.006] [0.000]Deficit Shock -0.367 0.272 0.047 0.024 -0.001[0.431] [0.212] [0.071] [0.028] [0.003]State Debt 0.016 0.113* 0.003 -0.003 0.003[0.052] [0.058] [0.024] [0.015] [0.002]Fixed Effects State, Year State, Year State, Year State, Year State, YearObservations 647 647 647 647 647Adjusted R-squared 0.888 0.968 0.946 0.890 0.91583Table 3.7: Electoral Cycles in Pension Benefit Accrual RatesThis table reports the estimation results from the OLS regression Yit = α + κi + λt + δ0 · Electionit + Xitβ + it in columns (1) to (3), AccGov it = α + κi +λt + δ0 · Electionit + ρ · Pub Union Mbrshpit · Electionit + pi · Pub Union Mbrshpit + Xitβ + it in column (4), and AccGov it = α + κi + λt + δ0 · Electionit +ρ · LegixExpit · Electionit + pi · LegixExpit + Xitβ + it in column (5), where Yit represents various measures of pension accrual rates, Pub Union Mbrshpit isthe state-level public sector unionization membership rate in year t, and LegisExpit is a dummy variable indicating whether the incumbent Governor has priorexperience in the state legislature. Column (6) reports the results from including all terms from columns (4) and (5). All specification include the set of controlvariables Xit, including lagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneousvalues of Discount Rate, Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as well as Appendix B.2. Allspecifications include both plan and year fixed effects. Standard errors are corrected for heteroskedasticity and clustered at the state level. Standard errors arein parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)Acc AccMbrs AccGov AccGov AccGov AccGovElection 0.063 0.001 0.062 -0.271* -0.011 -0.330**[0.080] [0.043] [0.067] [0.157] [0.078] [0.144]Election × Pub Union Mbrshp 1.058** 1.022**[0.489] [0.451]Pub Union Mbrshp 2.077 1.905[3.098] [2.969]Election × LegisExp 0.444** 0.435**[0.175] [0.182]LegisExp -0.421 -0.414[0.344] [0.338]Control Variables Yes Yes Yes Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,314 1,314 1,314 1,314 1,314 1,314Adjusted R-squared 0.860 0.917 0.863 0.864 0.864 0.86484Table 3.8: Benefit Protection Strength and Electoral Cycles in Pension DeficitsThis table reports the estimation results from the OLS regression PenDefGov it = α + κi + λt + δ0 · Electionit +ρ · Wi · Electionit + Xitβ + it in columns 1, 2, and 3, where Wi represents Strong Protect i (a dummy variableindicating whether a plan’s state provides constitutional protection of public pension plan members’ benefits) incolumn (1), Weak Protect i (a dummy variable indicating whether a plan’s state provides protection of public pensionplan members’ benefits under the gratuity principal) in column (2), and Unconditional Protect i (a dummy variableindicating whether a plan’s state provides unconditional protection of public pension plan members’ benefits) incolumn (3). Column (4) reports the estimation results from including all terms from columns (1), (2), and (3).All specification include the set of control variables Xit, including lagged values of ln(Payroll), ln(Avg Salary),Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate,Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as wellas Appendix B.2. All specifications include both plan and year fixed effects. Standard errors are corrected forheteroskedasticity and clustered at the state level. Standard errors are in parentheses, with *, **, and *** denotingsignificance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4)PenDefGov PenDefGov PenDefGov PenDefGovElection 0.320 0.731*** 0.039 -0.014[0.214] [0.200] [0.238] [0.166]Election × Strong Protect 1.817*** 1.525***[0.334] [0.336]Election × Weak Protect -1.679** -1.502***[0.687] [0.515]Election × Unconditional Protect 1.009*** 0.857***[0.327] [0.302]Control Variables Yes Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,314 1,314 1,314 1,314Adjusted R-squared 0.673 0.672 0.672 0.67385Table 3.9: Pension Plan Opacity and Electoral Cycles in Pension DeficitsThis table reports the estimation results from the OLS regression PenDefGov it = α+ κI + λt + δ0 · Electionit + pi ·Opaque Pensionsi ·Electionit + ρ ·Transparent Pensionsi ·Electionit +Xitβ+ it in column (1), and PenDefGov it =α+κI +λt+δ0 ·Electionit+pi ·Opaque Budget i ·Electionit+ρ ·Transparent Budget i ·Electionit+Xitβ+it in column(2), where Opaque Pensionsi is a dummy variable indicating whether a state is in the bottom decile in terms ofstate pension SII transparency score, Transparent Pensionsi is a dummy variable indicating whether a state is in thetop decile in terms of state pension SII transparency score, Opaque Budget i is a dummy variable indicating whethera state is in the bottom decile in terms of state budget SII transparency score, Tranpsarent Budgeti is a dummyvariable indicating whether a state is in the top decile in terms of state budget SII transparency score. Column (3)reports the estimation results from including all LHS terms from columns (1) and (2). All specification include the setof control variables Xit, including lagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp,and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate, Inflation Rate, and CostMthd EAN .Detailed definitions for all variables can also be found in Table 3.1 as well as Appendix B.2. All specifications includeboth plan and year fixed effects. Standard errors are corrected for heteroskedasticity and clustered at the statelevel. Standard errors are in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level,respectively.(1) (2) (3)PenDefGov PenDefGov PenDefGovElection 0.509** 0.681*** 0.518**[0.212] [0.236] [0.243]Election × Opaque Pensions 1.081** 1.123**[0.474] [0.466]Election × Transparent Pensions -1.228*** -0.782**[0.431] [0.327]Election × Opaque Budget 0.299 0.470[1.077] [1.096]Election × Transparent Budget -0.794** -0.556[0.392] [0.332]Control Variables Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, YearObservations 1,314 1,314 1,314Adjusted R-squared 0.672 0.672 0.67286Table 3.10: Political Factors and and Electoral Cycles in Pension DeficitsThis table reports the estimation results from the OLS regression PenDefGov it = α + κi + λt + δ0 · Electionit + ρ ·VicMarginit+Xitβ+it in column (1), PenDefGov it = α+κi+λt+δ0 ·Electionit+ρ ·VicMarginit+pi ·VicMarginit ·IncumbLosesit + Xitβ + it in column (2), PenDefGov it = α + κi + λt + δ0 · Electionit + ρ · Lame Duck it + pi ·Lame Duck it ·Electionit+Xitβ+it in column (3), and PenDefGov it = α+κi+λt+δ0 ·Electionit+ρ·Republicanit+pi ·Republicanit ·Electionit+Xitβ+it in column (4). VicMarginit is the margin of victory between the winning candidateand the runner-up in the gubernatorial election in year t if an election occurred and zero otherwise, IncumbLosesitis a dummy variable that indicates if the incumbent Governor loses reelection in year t, Lame Duck it indicateswhether the Governor faces binding term limits, and Republicanit indicates whether the Governor is a member ofthe Republican party. Column (5) reports the results from including all terms from columns (1), (2), (3), and(4). All specification include the set of control variables Xit, including lagged values of ln(Payroll), ln(Avg Salary),Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate,Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as wellas Appendix B.2. All specifications include both plan and year fixed effects. Standard errors are corrected forheteroskedasticity and clustered at the state level. Standard errors are in parentheses, with *, **, and *** denotingsignificance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5)PenDefGov PenDefGov PenDefGov PenDefGov PenDefGovElection 1.043*** 0.746** 0.373* 0.747*** 0.680*[0.303] [0.338] [0.208] [0.268] [0.381]VicMargin -2.232** -2.198** -2.384**[0.976] [1.018] [1.098]Lame Duck -0.913** -0.834**[0.391] [0.384]Election × Lame Duck 0.861* 0.764[0.464] [0.481]Election × Republican -0.218 -0.122[0.515] [0.500]Republican -0.556 -0.665[0.635] [0.631]gub election legCtrl 0.517 0.467[0.463] [0.459]Control Variables Yes Yes Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,280 1,280 1,318 1,318 1,280Adjusted R-squared 0.681 0.681 0.675 0.674 0.68387Table 3.11: Consequences of Electoral Cycles in Pension DeficitsPanel A reports plan-level cross-sectional regression estimation results from ∆UnfundedLiabi = α + δ · Zi + X¯iβ + i, where Zi represents PenDefCyci (theplan-level time series average of PenDef conditional on election year minus the plan-level time series average of PenDef conditional on non-election year) incolumns (1) and (2), PenDefCycDi (PenDefCyci adjusted for time trends) in columns (3) and (4), and Residual PenDefCycRi (PenDefCyci adjusted fortime-varying covariates, plan fixed effects, and time fixed effects) in columns (5) and (6). ∆UnfundedLiabi denotes the plan-level time series average for annualchanges in unfunded liabilities scaled by payroll. X¯i denotes the plan-level time-series averages for the set of control variables, which includes lagged values ofln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate, InflationRate,and CostMthd EAN . Panel B reports state-level cross-sectional regression estimation results from Yj = α+ δ · Zj + j , where j indexes states, Yj represents thetime-series average of log growth rates in state GDP in columns (1)-(3) and in house price index values in columns (4)-(6), and Zj represents the weighted averagesof PenDefCyci, PenDefCycDi, and PenDefCycRi, respectively (weighted by plan liabilities). Detailed definitions for all variables can be found in Table 3.1 aswell as Appendix B.2. State level fixed effects are included in columns (2), (4), and (6). Standard errors are corrected for heteroskedasticity and clustered at thestate level in Panel A. Standard errors are in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.Panel A: Changes in Unfunded Liabilities(1) (2) (3) (4) (5) (6)∆ UnfundedLiab ∆ UnfundedLiab ∆ UnfundedLiab ∆ UnfundedLiab ∆ UnfundedLiab ∆ UnfundedLiabPenDefCyc 1.338*** 1.306***[0.487] [0.396]PenDefCycD 1.361** 1.384***[0.523] [0.438]PenDefCycR 1.228*** 0.867***[0.425] [0.273]Control Variables Yes Yes Yes Yes Yes YesFixed Effects State State StateObservations 106 106 106 106 103 103Adjusted R-squared 0.330 0.614 0.324 0.620 0.348 0.591Panel B: State-Level Economic Outcomesln(GDP Growth) ln(GDP Growth) ln(GDP Growth) ln(HPI Growth) ln(HPI Growth) ln(HPI Growth)PenDefCyc -0.142* -0.031[0.076] [0.026]PenDefCycD -0.153* -0.031[0.077] [0.027]PenDefCycR -0.071 -0.016[0.065] [0.022]Observations 50 50 48 50 50 48Adjusted R-squared 0.048 0.056 0.004 0.010 0.007 -0.01088Table 3.12: Electoral Cycles in Private-Sector DB Pension PoliciesThis table reports the estimation results from the OLS regression Yit = α + κi + λt + δ0 · Electionit + Xitβ + it,where Yit is DefFirmit in column (1), ContribF irmit in column (2), and AccF irmit in column (3). All specificationinclude the set of control variables Xit, including lagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock ,State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate, and Wage Growth.Control variable coefficient estimates are not reported in order to conserve space. All specifications include both planand year fixed effects. Standard errors are corrected for heteroskedasticity and clustered at the state level. Standarderrors are in parentheses, with *, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3)DefFirm ContribFirm AccFirmElection 0.135 -0.352 -0.316[0.492] [0.299] [0.600]ln(Payroll) 18.500*** -1.946** 16.030**[6.467] [0.791] [6.439]ln(Avg Salary) -15.292** 2.453*** -12.179[7.167] [0.869] [7.651]Income -0.365*** -0.007 -0.377***[0.112] [0.031] [0.117]Deficit Shock 1.696 -1.863 0.631[6.569] [2.132] [5.955]State Unemp 0.719 -0.340 0.571[0.889] [0.257] [0.894]Pub Union Mbrshp -0.346 -0.014 -0.332[0.237] [0.059] [0.211]Discount Rate -1.479 0.134 -1.445[1.028] [0.181] [0.964]Wage Growth -0.429 0.129 -0.175[0.587] [0.178] [0.593]Fixed Effects Plan, Year Plan, Year Plan, YearObservations 2,430 2,431 2,439Adjusted R-squared 0.671 0.317 0.67089Table 3.13: Unexpected Governor Changes and Pension DeficitsThis table reports report the estimation results from the OLS regression Yit = α+κi+λt+ω0·Gov Changeit+Xitβ+itwhere Yit represents the outcome variable PenDef it in columns (1) and (2), PenDefMbrsit in columns (3) and (4), andPenDefGov it in columns (5) and (6). A lagged Gov Changeit−1 is added to the specification in columns (2), (4), and(6). All specification include the set of control variables Xit, including lagged values of ln(Payroll), ln(Avg Salary),Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate,Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as wellas Appendix B.2. All specifications include both plan and year fixed effects. Standard errors are corrected forheteroskedasticity and clustered at the state level. Standard errors are in parentheses, with *, **, and *** denotingsignificance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)PenDef PenDef PenDefMbrs PenDefMbrs PenDefGov PenDefGovGov Change 0.257 0.300 0.067 -0.005 0.188 0.303[0.888] [0.835] [0.132] [0.138] [0.887] [0.848]Gov Change(t-1) -0.133 0.222 -0.352[1.198] [0.179] [1.222]Control Variables Yes Yes Yes Yes Yes YesFixed Effects Plan, Year Plan, Year Plan, Year Plan, Year Plan, Year Plan, YearObservations 1,312 1,312 1,312 1,312 1,314 1,314Adjusted R-squared 0.676 0.676 0.589 0.589 0.671 0.67190Table 3.14: Accounting for Geographic Clustering of State Electoral CyclesThis table reports the estimation results from the OLS regression Yit = α+κi+λt+δ0 ·Electionit+Xitβ+it, where the outcome variable Yit is PenDef in columns(1) and (2), PenDefMbrs in columns (3) and (4), and PenDefGov in columns (5) and (6). All specification include the set of control variables Xit, includinglagged values of ln(Payroll), ln(Avg Salary), Income, Deficit Shock , State Unemp, and Pub Union Mbrshp, as well as contemporaneous values of Discount Rate,Inflation Rate, and CostMthd EAN . Detailed definitions for all variables can also be found in Table 3.1 as well as Appendix B.2. Standard errors are correctedfor heteroskedasticity and double clustered at the state and year level. Bootstrap clustering is applied due to the small number of years in the panel. Plan-leveland year-level fixed effects are included in columns (1), (3), and (5), while plan-level and region-year fixed effects are included in the remaining columns, whereregion represents the U.S. Census geographic grouping of U.S. states into Northeast, Midwest, South, and West regions. Standard errors are in parentheses, with*, **, and *** denoting significance at the 10%, 5%, and 1% level, respectively.(1) (2) (3) (4) (5) (6)PenDef PenDef PenDefMbrs PenDefMbrs PenDefGov PenDefGovElection 0.603** 0.524*** -0.012 0.026 0.613*** 0.499***[0.214] [0.107] [0.099] [0.089] [0.173] [0.162]Control Variables Yes Yes Yes Yes Yes YesFixed Effects Plan, Year Plan, Year × Region Plan, Year Plan, Year × Region Plan, Year Plan, Year × RegionCluster by State, Year State, Year State, Year State, Year State, Year State, YearBoostrap Cluster Yes Yes Yes Yes Yes YesObservations 1,312 1,312 1,312 1,312 1,314 1,314Adjusted R-squared 0.677 0.685 0.589 0.594 0.672 0.68391Chapter 4Distortionary Reelection Incentivesand Public Defined Benefit PensionPlans4.1. IntroductionPublic sector defined benefit (DB) pension plans require policymakers to make major financialdecisions over the use of public funds. These decisions include the granting of retirement benefitsto public sector employees, which represents a form of public borrowing, and the funneling ofcontributions into plan funds, which represents a form of public saving. Recent cases of severeunderfunding of U.S. state DB pension plans in Illinois, California, and Rhode Island show thatmismanagement of public DB pension plans can lead to the financial destabilization of state andlocal governments. At the same time, public pension policies are at the discretion of electedpoliticians, who face political incentives to win elections. In particular, politicians are often accusedof acting in a short-sighted manner by reducing funding for public pension plans and “kicking thecan down the road” by placing the burden on future administrations.In this essay, I develop a stylized model based on the framework of Holmstro¨m (1999) to showhow reputational concerns can distort public DB pension policy decisions in a political setting. Anincumbent politician makes public pension policy decisions on behalf of voting taxpayers, but ismotivated by reelection concerns in addition to caring about voters’ utility. Policy choices are notimmediately transparent to all voters, which results in the incumbent agent taking hidden actionsin an attempt to manipulate the election result.I first consider a scenario in which an incumbent politician grants promises of pension benefits topublic sector employees, but cannot prefund those promises with contribution savings. I show thatwhen the benefit policy is not fully transparent to voters, the incumbent has the incentive to raisethe benefit above the socially-optimal level in exchange for obtaining short-term wage concessionsfrom public sector employees. This in turn allows the incumbent to temporarily boost the outputof public goods in order to inflate his perceived economic performance in the eyes of uninformedvoters before the election.Next, I consider the scenario in which the incumbent agent chooses the amount of contributionsthat goes towards funding the pension benefit, which is exogenously set. Following the samereasoning as before, the incumbent has the incentive to reduce contributions prior to an election92in order to temporarily inflate his perceived performance, but only if pension policies are not fullytransparent to all voters. In addition, if the employee does not enjoy protection over pensionbenefits, then lowering contributions will be offset by higher wages that the worker demands inexchange for unfunded benefits, preventing the incumbent from inflating his performance. Thus,both the opacity of public pension policy as well strength of legal protection for employee benefitsare necessary conditions for reelection incentives to affect the incumbent’s decisions.In both scenarios, voters are rational and make the utility-maximizing choice between theincumbent and a challenger at election time. The incumbent agent, who cares about voter welfarebut also derives private benefits from holding political office, holds a temporary informationaladvantage over voters regarding public pension policy. The result is a “signal-jamming” equilibrium,in which the incumbent attempts to boost the signal of his governing ability during election yearby “borrowing” from the public pension plan to increase the provision of public goods, even thoughvoters are rational and anticipate the incumbent’s opportunistic behaviour in equilibrium.My model delivers several empirical prediction. Specifically, pension deficits—defined as thedifference between pension benefit accruals and contributions—should be higher in election years,and this pattern should be more pronounced in states with pension systems that are more opaque,for elections that are more closely contested, and for plans that provide stronger guarantees toemployees over their future benefit payments. I find evidence in support of these predictions for asample of U.S. state DB pension plans in a detailed empirical investigation presented in Chapter 3,.My work extends a long literature on the phenomenon of political cycles in economic policies,which examines the tendency for governments to enact expansionary policies immediately beforean election. Nordhaus (1975) first interpreted such cycles as the consequences of opportunisticpoliticians fooling irrational voters in order to win elections. Subsequent work by Rogoff and Sibert(1988) and Rogoff (1990) show that political cycles in fiscal policies may arise out of a signallingequilibrium in which incumbents signal their intrinsic competence to rational voters through incur-ring fiscal deficits. Alesina et al. (1997) later showed that political cycles in U.S. inflation rates canresult from expansionary monetary policies enacted during the terms of Democratic presidentialadministrations.Persson and Tabellini (2002), Alt and Lassen (2006), and Shi and Svensson (2006) provide theclosest work to my research, as they show that election year spikes in fiscal deficits can be rational-ized by models in which nontransparent fiscal policies allow politicians to undertake opportunistic“hidden borrowing” as a means to inflate their perceived performance. I apply the same basic in-tuition to the setting of public pension plan funds, and I further show that electoral cycle patternsin policy decisions can emerge when agents’ innate qualities remain constant over time, given thatthat the information asymmetry over policy is temporary. This assumption provides a differentmechanism for generating political cycles compared to the existing literature, and is motivated bythe institutional features of the public pension system as described in Section 3.2 from Chapter 3.My model also generates novel testable predictions relating to the closeness of elections and thestrength of legal protection over benefits.93The remainder of the essay is organized as follows. Section 4.2 describes a model of politically-motivated public pension benefit policies. Section 4.3 describes a model of politically-motivatedpublic pension contribution policies. Section 4.4 describes the empirical implications of the models.Section 4.5 concludes.4.2. Reelection Incentives and Pension BenefitsWe first consider the case in which an incumbent political agent make decisions over public sectoremployee wages and pension benefits, but do not allow them to make contributions to prefund thepension plan.4.2.1. SetupI adopt a two-period setting in which a political agent makes decisions that affect the welfare oftax-paying voters. In the first period (t = 1), the incumbent agent, denoted I, is assumed to be theleader with authority over policy decisions regarding granting defined benefits to a governmentalworker. An election occurs near the end of the period, in which voters decide whether to re-electagent I or a political challenger, denoted C, to become the leader in the second period (t = 2).87Voters and agents derive utility from consuming a public good in each period t. The publicgood, denoted gt, is net of taxes, which allows us to abstract from taxation policy. Voter utility,denoted Uv, is determined by the sum of the public goods produced during the two periods—i.e.Uv = g1 + g2.At t=1, the public good output is determined according tog1 ≡ ηI − w + 1 (4.2.1)where ηI denotes I’s fiscal competence, w denotes the employee wage bill, and 1 denotes a randomshock.Public sector employees are paid a wage w in wages period 1 and a pension benefit b in period 2.To abstract away from labour demand considerations, we assume public goods production requiresthe employment of a single worker. Furthermore, the incumbent is able to commit in period 1 topaying b in period 2. To employ the worker, the government must provide adequate compensationaccording to the worker’s participation constraint:u(w) + u(b) ≥ u¯ (4.2.2)where u¯ denotes the the worker’s reservation utility, and u(·) denotes a concave utility function suchthat u′(·) > 0 and u′′(·) ≤ 0. The concavity of u(·) implies that the employee prefers consumptionto be smoothed over the two periods.87 One can think of the agents as individuals or political parties in this setup.94The incumbent agent sets w and b at the beginning of period 1. An election takes place at theend of period 1, at which point voters decide whether to elect I or C as the leader. At t = 2, publicgood output is determined according tog2 ≡ θηI + (1− θ)ηC − b+ 2, (4.2.3)where θ ∈ {0, 1} takes on a value of 1 if the incumbent is re-elected and 0 otherwise, ηC denotes thechallenger’s fiscal competence, b denotes the promised pension benefit, and, 2 denotes a randomshock term that is independent from 1.We assume that political agents care about voters’ utility, but also derive positive benefits fromholding political office, such that the incumbent’s utility is defined as UI = Uv + θx, where x isassumed to be strictly positive and represents the “ego rents” of being in power, following Rogoff(1990).The fiscal competence (“ability”) parameter η captures the innate qualities of the politicalagent, such as how well he is able to eliminate wasteful spending or deal with unexpected fiscalshocks. As is standard in models of career concerns, ability is not directly observed, and voters andagents alike must make inferences about the incumbent’s ability through observing g1. We assumeηI and ηC to be invariant over time, with the following common prior distribution at the beginningof t = 1:ηi ∼ N(mi1, 1hi1), (4.2.4)for i ∈ {I, C}.The random output shocks t are also not directly observable, and are normally distributedaccording toi ∼ N(0, 1h1), (4.2.5)where ηI , ηC , 1 and 2, are independently distributed and unaffected by w and b.At the beginning of t = 1, the incumbent decides on public pension policies b and w. Next, g1 isrealized and observed by everyone, followed by an election in which voters decide whether to votefor I or C. Crucially, we assume that the representative voter, who casts the decisive vote in theelection, observes b and w before the election only with probability 1− ρ, while with probability ρshe does not observe b and w until after the election. The parameter ρ captures the degree of policyopacity. In the second period, the elected leader collects the ego rent x and repays the promisedbenefit b, but has no influence on public goods output g2 except through his ability.An illustrated timeline of the model is provided in Figure 4.1. The left column in the figureprovides a mapping between the model timeline and the institutional timeline described in the 3.1from the previous chapter. The decision point for b and w corresponds to the beginning of thestate budget process in a given fiscal year. The realization of g1 corresponds to the realization ofactual revenues and expenditures as the budget takes effect. Finally, the post-election revelationof pension policies b and w represents the end of the fiscal year, at which point the state releases95its independently audited financial reports.4.2.2. InferenceVoters form posteriors about the incumbent agent’s ability from observing output and pensionpolicies. Let mI2 and hI2 denote the mean and precision of the representative voter’s posteriorabout ηI , conditional on having observed g1 and w. If the representative voter observes w or bbefore the election, she will rationally form a posterior mean of mI2 at election time.88 Since priorsabout ability and output shocks are jointly independent and normally distributed, we can applyBayes’ law to express mI2 asmI2 = (1− µ)mI1 + µz, (4.2.6)where z ≡ g1 + w = ηI + 1 represent the period 1 signal of the I’s ability conditional on observingg1 and w, andµ ≡ hhI1 + h, (4.2.7)represents the relative weight of the signal.Let mˆI2 and hˆI2 denote the mean and precision of the representative voter’s posterior aboutI’s ability conditional on having observed g1 but not w. Thus, if the representative voter does notobserve w or b before the election, she will form a posterior mean of mˆI2 at election time. ApplyingBayes’ law, we express mˆI2 asmˆI2 = (1− µ)mI1 + µzˆ = mI2 + µ(w¯ − w), (4.2.8)where w¯ represent the representative voter’s conjecture about w, and zˆ ≡ g1 + w¯ = z − w + w¯denotes the period 1 signal of the incumbent’s ability conditional on observing g1 but not w.The precision of the representative voter’s posteriors about the incumbent’s ability evolvesdeterministically—i.e. hI2 = hˆI2 = hI1 +h—regardless of whether the she observes w or not. Sinceutility is linear in the incumbent agent’s ability, voters and agents only care about the posteriormean. From this point forward, reputation refers to the posterior mean of an agent’s ability, unlessstated otherwise. Since C cannot influence g1 in any way during the first two periods, there isno learning about the challenger’s ability—i.e. mC2 = mˆC2 = mC1 and hC2 = hˆC2 = hC1. It isonly through the incumbent’s power to enhance his reputation by manipulating w and b that thepossibility of a political agency conflict arises.4.2.3. EquilibriumWe solve the optimization problems facing voters and the incumbent agent, given each other’soptimal strategies. At election time, the representative voter understands that g1 is already set88 Note that if she observes b, she can “back out” b as we assume that she understands that the employee’s participationconstraint will be binding in equilibrium.96and therefore chooses θ to maximize expected period 2 utility g2:maxθEˆ1[ηC + θ(ηI − ηC)− b+ 2], (4.2.9)where Eˆ1[·] denotes the expectation function with respect to voters’ information set at electiontime.It follows that the representative voter’s optimal strategy followsθ =1 if mI2 −mC2 ≥ 00 if mI2 −mC2 < 0, (4.2.10)if she observes w or b before the election, andθ =1 if mˆI2 − mˆC2 ≥ 00 if mˆI2 − mˆC2 < 0. (4.2.11)if she does not.The intuition behind 4.2.10 and 4.2.11 is straightforward. The representative voter understandsthat b has already been set, and therefore bases her election decision entirely on comparing thereputations of I and C. The incumbent’s ability to influence this voting decision hinges on whetherthe representative voter is able to observe w before the election.Anticipating the voter’s decision process, the incumbent chooses w and b at the beginning ofperiod 1 according the following constrained optimization problemmaxb,wE1[ηI − w + 1 + ηC + θ(ηI − ηC + x)− b+ 2]subject to u(w) + u(b) ≥ u¯,(4.2.12)where E1[·] denotes the expectation function with respect to the incumbent’s information set atthe beginning of period 1.If the representative voter observes w or b before the election, we see from 4.2.7 that she can“back out” the true signal of the incumbent’s ability (z = ηI + 1), in which case the incumbent’schoices for w and b would have no effect on the election result. It follows from first order conditionsthat the incumbent’s optimal policy under full transparency (i.e ρ = 0) is characterized by w = b.It is immediately clear that w = b also characterizes the first-best policy from voters’ perspec-tive.89 Intuitively, the incumbent agent and voters face the same marginal benefits and marginalcosts to adjust w and b when election results are exogenous to w and b. In the absence of reelec-tion incentives, the incumbent minimizes spending on employee compensation on behalf of votingtaxpayers by offering wages and benefits that perfectly smooth the employee’s consumption overthe two periods.89 This is trivially obtained by solving for the w and b that maximizes Uv subject to 4.2.297If the representative voter does not observe w or b before the election, then we see from 4.2.8that the incumbent can use w to influence the signal of the incumbent’s ability (zˆ = z − w + w¯).In effect, the incumbent boosts his reputation by inflating output through paying a lower period 1wage. To see this, let us denote Ω = E1[w(ηI − ηC + x)], and express the partial derivative of Ωwith respect to w via the following lemma (see A.1 in Appendix A for proof).Lemma 1. Let Φ(v;µ, σ2) denote the probability density function for a normally distributed randomvariable V with mean µ and variance σ2. It follows that∂Ω∂w= −ρµφ(µ(w − w¯);m∆1 ,µhI1)(x+ µ(w − w¯)) (4.2.13)where m∆1 ≡ mI1 −mC1 denotes the difference between the common prior beliefs of I’s and C’sabilities, and µhI1 is the variance of mI2−mC2 given the incumbent’s information set at the beginningof period 1.Eq. 4.2.13 presents an intuitive representation of the incumbent’s reelection incentive. Thefirst term ρ captures the fact that w affects I’s election-time reputation only if the representativevoter does not observe w before the election, in which case the decrease in election probabilityis −µφ(µ(w − w¯);m∆1 , µhI1 ) and unambiguously negative. The x + µ(w − w¯) component can befurther decomposed into an ego rents term, x, which is unambiguously positive, and an “electiondistortion” component, µ(w − w¯), which is ambiguously signed. This distortion component maybe negative or positive, depending on the relative difference between w and w¯. For example, bylowering w when w < w¯, the incumbent creates additional states of the world in which he wins theelection even when he believes C to have a higher ability. Following the same logic, the incumbentcan eliminate such suboptimal states by lowering w when w > w¯.In equilibrium, voters conjecture correctly about w, which implies that w = w¯ andω∗ ≡ ∂Ω∂w∣∣∣∣w=w¯= −ρµφ(µ(0;m∆1 ,µhI1)x, (4.2.14)where ω∗ represents the equilibrium “election manipulation incentive” term.When voters form the correct conjecture about w, there is no election distortion and the onlymarginal effect on agent I’s utility is through the unambiguously positive expected ego rents chan-nel. If ρ is positive, then ω∗ < 0 and the incumbent agent faces an additional benefit from loweringw. In equilibrium, the incumbent does not gain any advantage, but still lowers w in order to“protect” his reputation.From Eq. 4.2.14, it is immediately obvious that ω∗ is decreasing in ρ, which captures the ideathat greater opacity leads to stronger election manipulation incentives. Moreover, ω∗ is increasingin m∆1 if m∆1 < 0 and decreasing in m∆1 if m∆1 > 0.90 This captures the idea that the election90 This stems from the characteristics of the normal probability density function. The same results should hold for similardistributions in which median is the same as the mode and the probability density function is strictly increasing tothe left of the median and strictly decreasing to the right of the median. I thank Masahiro Watanabe for pointingthis out.98manipulation incentive is greater when the election is “closer” in the sense that the differencebetween the prior reputations of the incumbent and the challenger is small.We obtain the following proposition (see A.2 in Appendix A for proof):Proposition 1. The equilibrium pension benefit, b∗, satisfies the following conditions:(a) Ceteris paribus, b∗ is increasing in ρ,(b) If ρ > 0, then ceteris paribus b∗ is decreasing in m∆1 for m∆1 > 0 and increasing in m∆1 form∆1 < 0, and(c) If ρ = 0, then ceteris paribus b∗ is unaffected by m∆1 .Part (a) of Proposition 1 formalizes the idea that a greater degree of opacity leads to a strongerincentive for the incumbent agent to increase pension benefits during election year. Part (b)formalizes the idea that the incentive to manipulate voters through election year pension borrowingis higher when the election is closer to a “tipping point” between the incumbent winning and thechallenger winning, while part (c) captures the idea that the manipulation incentive exists only ifthe pension system is not fully transparent.The general intuition behind Proposition 1 is that the incumbent wants to realize additionalshort term wage savings by providing higher pension benefits in order to inflate the signal ofhis period 1 performance. In the real world, short term wage savings constitute one of severalpotential channels motivating incumbent politicians to grant higher pension benefit. For example,the incumbent may wish to increase benefits to win direct political support from public sectorlabour unions. I focus on only the wage savings channel for the sake of model parsimony.4.3. Reelection Incentives and Pension ContributionsWe now consider the case in which unfunded benefits are not wholly guaranteed to employees, butthe incumbent agent can make contributions into the public pension fund in period 1. To shift theattention to contribution policy rather than benefits policy, we assume b has been set and cannotbe changed by the incumbent at the beginning of period 1. This assumption is justified by therelative inflexibility of pension benefit policy, which is explained in detail in Chapter 3.4.3.1. SetupThe basic framework of remains the same as in Section 4.2. Voter utility is Uv = g1 + g2 and theincumbent agent’s utility is UI = Uv + θx. There are two periods and an election occurs in period1. However, we modify the public goods output in the two periods to beg1 ≡ ηI − w − k + 1 (4.3.1)g2 ≡ θηI + (1− θ)ηC − pi(b− k) + 2, (4.3.2)99where k denotes the pension contribution in period 1, pi denotes the portion of the unfunded pensionliability (i.e. b− k) that is paid out of g2 to the employee in period 2, and the remaining variablesare defined as before.Due to the imperfect guarantee on the unfunded portion of the pension benefit, the worker’sparticipation constraint is nowu(w) + u(k + pi(b− k)) ≥ u¯ (4.3.3)where k+pi(b−k) reflects that fact that the employee is paid the entirety of the funded contributionk plus a portion pi of the unfunded benefit.91Note we allow for the possibility that b < k, in which case the fund is overfunded and theemployee receives a payment greater than b in the second period. This can be interpreted aspension beneficiaries “skimming” the surplus of overfunded public pension plan funds though benefitincreases. We also allow for the possibility for k < 0, which is difficult to interpret. We may insertan additional constraint that k ≥ 0, but the case of when this constraint binds is not economicallyinteresting, so for the sake of simplicity we assume the equilibrium is characterized by an interiorsolution at which k > 0.The timeline of the model again proceeds as illustrated in 4.1. The incumbent agent chooses wand k at the beginning of period 1. This is followed by the realization of g1 and then an electionbetween the incumbent and the challenger. The representative voter first observes w and k beforethe election with probability 1− ρ, and first observes w and k after the election with probability ρ.4.3.2. InferenceThe incumbent’s ability and the random shock terms follow the same distributions as in Section 4.2,which means that the representative voter’s inference of ηI is characterized by Eq. 4.2.6 if she firstobserves ρ prior to the election—i.e. mI2 = (1 − µ)mI1 + µz where z = ηI + 1. However, if shefirst observes ρ after the election, then her inference of ηI is characterized bymˆI2 = (1− µ)mI1 + µzˆ = mI2 + µ(w¯ − w + k¯ − k), (4.3.4)where w¯ represent the representative voter’s conjecture about w, k¯ represent the representativevoter’s conjecture about k, and zˆ ≡ g1 + w¯ = z − w + w¯ + k¯ − k denotes the period 1 signal of theincumbent’s ability conditional on observing g1 but not w or k.Eq. 4.3.4 indicates that the incumbent can manipulate w and k in order to inflate his rep-utation in the eyes of uninformed voters. As before, the precision of the representative voter’sposteriors about the incumbent’s ability evolves deterministically, and there is no learning aboutthe challenger’s ability.91 An alternative formulation is to make the benefit payment be b with probability pi and k with probability 1 − pi.However, this introduces addition complications relating to employee risk aversion, which we abstract away from bymaking the benefit payment deterministic.1004.3.3. EquilibriumAs in Section 4.2, the representative voter understands that she cannot affect g1, w, or k with herelection choice and therefore makes her decision based on 4.2.10 if she first observes w and k beforethe election, and based on 4.2.11 if she first observes w and k after the election.The incumbent agent anticipates the representative voter’s decision rule and optimizes accordingtomaxk,wE1[ηI − w − k + 1 + ηC + θ(ηI − ηC + x)− pi(b− k) + 2]subject to u(w) + u(k + pi(b− k)) ≥ u¯,(4.3.5)to determine his choices for w and k at the beginning of period 1.Again, let Ω = E1[w(ηI−ηC+x)] represent the incumbent’s marginal utility bonus from winningthe election. We can show that changing w and changing k have the same marginal effect on Ω, asstated in the following lemma (see A.3 in Appendix A for proof):Lemma 2. Let Φ(v;µ, σ2) denote the probability density function for a normally distributed randomvariable V with mean µ and variance σ2. It follows that∂Ω∂w=∂Ω∂k= −ρµφ(µ(w − w¯ + k − k¯);m∆1 ,µhI1)(x+ µ(w − w¯ + k − k¯)) (4.3.6)where m∆1 ≡ mI1 −mC1 denotes the difference between the prior beliefs of I’s and C’s abilities,and µhI1 is the variance of mI2 − mC2 given the incumbent’s information set at the beginning ofperiod 1.In equilibrium, voters correctly conjecture that w = w¯ and k = k¯, and so we can express theequilibrium election manipulation incentive ω∗ asω∗ ≡ ∂Ω∂w∣∣∣∣w=w¯,k=k¯=∂Ω∂k∣∣∣∣w=w¯,k=k¯= −ρµφ(µ(0;m∆1 ,µhI1)x. (4.3.7)As was the case in Section 4.2, the equilibrium election incentive ω∗ is nonpositive, and is strictlynegative if ρ > 0. This leads to the following proposition (see A.4 in Appendix A for proof).Proposition 2. The equilibrium pension contribution, k∗, satisfies the following conditions:(a) If pi > 0 then ceteris paribus k∗ is decreasing in ρ,(b) If pi > 0 and ρ > 0, then ceteris paribus k∗ is increasing in m∆1 for m∆1 > 0 and decreasingin m∆1 for m∆1 < 0, and(c) If pi = 0, then ceteris paribus k∗ is not affected by ρ nor m∆1 .Proposition 2 closely parallels Proposition 1 from the previous section. Specifically, election yearmanipulation incentives are increasing in the degree of opacity and in the closeness of the election.However, Proposition 2 also illustrates that the incentive to reduce k depends on a nonzero portion101of the pension benefit b being guaranteed. Intuitively, if pi = 0, then any reduction in k is perfectlyoffset by the worker demanding a higher w in period 1, which leaves the incumbent’s reputationunchanged in the eyes of the uninformed voter.The intuition underlying Proposition 2 mirrors the intuition underlying Proposition 1. Theincumbent prefers to redirect pension contributions into increasing pre-election public goods output,but lowering contributions is immediately offset by the employing making higher wage demands inresponse. The more insulated the employee is against losses from unfunded benefits, the less theoffsetting wage demands, and the greater the incentive to cut back on contributions.Just as in the previous section, we use employee wages as a parsimonious modelling mecha-nism, but alternative mechanisms are possible For example, rather than demanding higher wagesfrom underfunded pension plans, the incumbent may exert direct political pressure on the incum-bent. Regardless of the mechanism, higher benefit protection in essence create a moral hazard foremployees to abstain from disciplining the incumbent from cutting back on contributions.4.4. Empirical ImplicationsThe model delivers the insight that when pension plan policies are not fully transparent to voters,incumbent politicians have the incentive to borrow at a higher-than-optimal rate from public pen-sion plans through increasing benefits or lowering contributions. This means that state DB pensiondeficits should be higher in election years relative to non-election years.While the stylized model only includes one period before the election, it is easy to extrapolatebackwards to show that the incentive to manipulate election results would not extend backwards ifone were to include additional periods prior to the election period. This is due to the assumptionthat any potential information asymmetry between the incumbent and voters is resolved by theend of the period. Therefore, any opportunistic borrowing conducted through the pension planduring non-election years would be revealed by the time that the election occurs. This assumptionis motivated by the one year gap between when pension benefit and contributions policies are setand when their impact on pension funding levels are disclosed to the public, which is explainedin Section 3.2 from Chapter 3. An electoral year spike in pension deficits can also result if votersput more weight on the most recent performance during election time. This may arise from anirrational recency bias on the part of voters, or if voters are rational and understand that the mostrecent performance is more predictive of future performance.92On the other hand, one can also imagine the existence of a gradual electoral cycle pattern inwhich the political incentive to manipulate pension deficits increases as election time nears. Inthe context of the model, such gradual cycles can arise if the probability that the representativevoter discovers the actions taken by the incumbent is increasing in the amount of time between the92 For example, Persson and Tabellini (2002), Alt and Lassen (2006), and Shi and Svensson (2006) present modelsof electoral cycles in budget deficits based on the assumption that the politician’s innate ability follows an MA(1)process. This means that only election-year activities are informative about the incumbent future performance, andvoters rationally discard pre-election performance.102incumbent’s action and the next election. However, in empirical results presented in Chapter 3, Ifind that there is a sharp election year decrease in pension contributions.While Proposition 1 predicts an election year increase in pension benefits and Proposition 2predicts an election year decrease in pension contribution, I find empirical evidence for the latterbut not the former in results presented in Chapter 3. As discussed in Section 3.2 of Chapter 3, thisis consistent with the institutional realities of the public pension system, in which the Governorhas significant discretion to change contribution policy but not benefit policy on a yearly basis.Proposition 1(a) and Proposition 1(b) also imply that election year spikes in pension deficitsshould be larger for state DB pension plans that are more opaque relative to state DB pensionplans that are more transparent, following the insight that information asymmetry is a necessaryingredient in creating distortionary incentives. Indeed, empirical findings presented in Section 3.5from Chapter 3 indicate electoral cycles are more pronounced for states with public pension systemsthat are more opaque, using empirical proxies for the opacity of state pension systems.Next, Proposition 2(c) implies that the incentive to raise election year pension borrowing de-pends on employees enjoying a certain degree of protection over their benefits. The intuition is thatcontribution cutbacks are self-defeating as a means to reduce pre-election expenditures if they areoffset by employees demanding higher wages in exchange for future losses from unfunded benefits.The empirical findings from Section 3.5 from Chapter 3 support this interpretation, as electionyear reductions in pension contributions are signficantly more pronounced in states that providestronger legal protection over public pension benefits.Lastly, Proposition 1(b) and Proposition 2(b) imply that election year spikes in pension deficitsare larger for elections that are more closely contested, based on the idea that there is a greaterincentive to manipulate elections that are close to a “tipping point” than in manipulating electionsin which one candidate has a large lead and the election result is a foregone conclusion. Thisprediction receives robust empirical support from evidence presented in Section 3.5 from Chapter 3.4.5. ConclusionIn this essay, I construct a stylized model to explain how reelection incentives distort policymakers’decisions over policies relating to public sector defined benefit plans. The model shows that, whenvoters are imperfectly informed about public policy, incumbent politicians can realize short-termsavings by promising higher defined benefits to public sector employees or by cutting back on publicpension contributions. In equilibrium, voters are not fooled by these actions, but the incumbentstill has the incentive to follow through in order to protect his reputation.The model generates several predictions of electoral cycles in public pension policies. In partic-ular, one should expect benefits to spike and contributions to dip right before an election, and theseeffects should be larger for states with more opaque pension systems, stronger legal protection overpension benefits, and for elections that are more closely contested. These predictions are supportedby the empirical findings presented in the essay from Chapter 3.103Figure 4.1: Model Timeline104Chapter 5ConclusionIn this thesis, I present three essays on the interrelated topics of finance, labour, and politicaleconomy. Chapter 2 forms the first essay, in which I examine how regulatory constraints on firms’abilities to hire skilled workers can inhibit corporate investment. To this end, I exploit a 2003reduction in the annual quota for H-1B visas, which are used by domestic firms in the U.S. to hireskilled foreign workers on a temporary basis. I find that the quota reduction resulted in relativedecreases in capital expenditures for firms that were ex-ante more reliant on H-1B workers, and thatthis effect persisted for several years and was more pronounced for firms hiring workers in traditionalindustrial occupations, such as scientists and engineering. My findings suggest that human capitalconstraints, much like financial capital constraints, can hinder corporate investment.The findings presented in Chapter 2 lay fertile groundwork for future research on related topics.One potential avenue would be to investigate whether the rents generated by barriers to hiringskilled workers are captured by firms or by domestic workers—i.e. whether more restrictive policiesresult in higher wages for domestic workers or whether the increased scarcity of foreign workersraises the values of existing workers and subsequently captured by the firm. Answering suchquestions will help clarify the welfare implications of labour market restrictions, as well as improveour understanding of the political economy surrounding immigration policy.Chapter 3 forms the second essay, in which I document an electoral cycle in how states fund theirpublic sector defined benefit pension plans. Specifically, systematic election year cuts to governmentcontributions result in election year increases in the rate at which the government effectively borrowsthrough public pension plans, and this pattern is more pronounced for states with less transparentpublic pension systems, for states that provide stronger benefit protection, and for elections thatare more closely contested. These findings indicate that incumbent Governors have the incentiveto undertake “hidden borrowing” in an attempt to inflate performance and secure reelection whenthe pension system is opaque and when employees bear the consequences of unfunded liabilities.Chapter 4 forms the final essay, in which I present a theoretical model to rationalize the empiricalfindings from Chapter 3. The model illuminates an agency conflict between the incumbent politicianand taxpayer, in which the incumbent borrows on behalf of taxpayers through the public pensionsystem at a higher rate than taxpayers would choose if they could directly choose for themselves.The underlying friction driving this conflict is information asymmetry relating to the inability forvoters to perfectly observe the incumbent’s pension policy, which allows the incumbent to attemptto inflate his performance through higher benefits or lower contributions, although in equilibriumvoters see through these actions.105Chapters 3 and 4 combine to illustrate how opaque borrowing channels provided by publicentities like state DB pension plans are vulnerable to opportunistic actions taken by incumbentpoliticians. However, more work is needed to improve our understanding of the welfare implicationsof public pension policy manipulations. For example, meddling with public employees’ pensionbenefits may have consequences for workers’ labour supply decisions. There is also the difficultquestions of whether large unfunded pension liabilities can act as a form of public debt overhangthat reduces economic growth. Answering these questions presents a research agenda left for futurework.106BibliographyAgrawal, Ashwini K, and David A Matsa, 2013, Labor unemployment risk and corporate financingdecisions, Journal of Financial Economics 108, 449–470.Alesina, Alberto, and Nouriel Roubini, 1992, Political cycles in OECD economies, The Review ofEconomic Studies 59, 663–688.Alesina, Alberto, Nouriel Roubini, and Gerald D Cohen, 1997, Political cycles and the macroecon-omy (MIT press).Almeida, Heitor, Murillo Campello, Bruno Laranjeira, Scott Weisbenner, et al., 2012, CorporateDebt Maturity and the Real Effects of the 2007 Credit Crisis, Critical Finance Review 1, 3–58.Almeida, Heitor, Murillo Campello, and Michael S Weisbach, 2004, The cash flow sensitivity ofcash, The Journal of Finance 59, 1777–1804.Alt, James, Ethan Bueno De Mesquita, and Shanna Rose, 2011, Disentangling accountability andcompetence in elections: evidence from US term limits, The Journal of Politics 73, 171–186.Alt, James E, and David Dreyer Lassen, 2006, Fiscal transparency, political parties, and debt inOECD countries, European Economic Review 50, 1403–1439.Ansolabehere, Stephen, and James M Snyder Jr, 2002, The incumbency advantage in US elections:an analysis of state and federal offices, 1942–2000, Election Law Journal 1, 315–338.Ashraf, Rasha, and Rina Ray, 2016, Human Capital, Skilled Immigrants, and In-novation, Working Paper. Available at SSRN: https://ssrn.com/abstract=2849089 orhttp://dx.doi.org/10.2139/ssrn.2849089.Atanassov, Julian, and E Kim, 2009, Labor and corporate governance: International evidence fromrestructuring decisions, The Journal of Finance 64, 341–374.Autor, David H, 2003, Outsourcing at will: The contribution of unjust dismissal doctrine to thegrowth of employment outsourcing, Journal of Labor Economics 21, 1–42.Bader, Lawrence N, and Jeremy Gold, 2007, The case against stock in public pension funds, Fi-nancial Analysts Journal 63, 55–62.107Baker, Malcolm, Jeremy C Stein, and Jeffrey Wurgler, 2003, When does the market matter? Stockprices and the investment of equity-dependent firms, The Quarterly Journal of Economics 118,969–1005.Bebchuk, Lucian Arye, and Lars A Stole, 1993, Do Short-Term Objectives Lead to Under-or Over-investment in Long-Term Projects?, The Journal of Finance 48, 719–729.Becker, Gary S, 1983, A theory of competition among pressure groups for political influence, TheQuarterly Journal of Economics 98, 371–400.Benmelech, Efraim, Nittai K Bergman, and Amit Seru, 2011, Financing Labor, NBER WorkingPaper No. 17144.Bergstresser, Daniel, Mihir Desai, and Joshua Rauh, 2006, Earnings manipulation, pension assump-tions, and managerial investment decisions, The Quarterly Journal of Economics 121, 157–195.Bertrand, Marianne, Esther Duflo, Sendhil Mullainathan, et al., 2004, How Much Should We TrustDifferences-In-Differences Estimates?, The Quarterly Journal of Economics 119, 249–275.Bertrand, Marianne, Francis Kramarz, Antoinette Schoar, and David Thesmar, 2007, Politicians,Firms and the Political Business Cycle: Evidence from France, Unpublished Working Paper,University of Chicago.Besley, Timothy, and Anne Case, 1995, Does Electoral Accountability Affect Economic PolicyChoices? Evidence from Gubernatorial Term Limits, The Quarterly Journal of Economics 110,769–98.Bloom, Nicholas, and John Van Reenen, 2007, Measuring and Explaining Management PracticesAcross Firms and Countries, The Quarterly Journal of Economics 122, 1351–1408.Bloom, Nick, Stephen Bond, and John Van Reenen, 2006, Uncertainty and investment dynamics,The Review of Economic Studies 74, 394–415.Borjas, George J, 2001, Does immigration grease the wheels of the labor market?, Brookings Paperson Economic Activity 2001, 69–133.Brender, Adi, and Allan Drazen, 2008, How do budget deficits and economic growth affect reelectionprospects? Evidence from a large panel of countries, The American Economic Review 98, 2203–2220.Brinkman, Jeffrey, Daniele Coen-Pirani, and Holger Sieg, 2016, The Political Economy of Under-funded Municipal Pension Plans, NBER Working Paper No. 22321.Brown, Craig O, and I Serdar Dinc, 2005, The politics of bank failures: Evidence from emergingmarkets, The Quarterly Journal of Economics 120, 1413–1444.108Brown, Jeffrey R, and David W Wilcox, 2009, Discounting state and local pension liabilities, TheAmerican Economic Review 99, 538–542.Cameron, A Colin, and Douglas L Miller, 2015, A practitioner’s guide to cluster-robust inference,Journal of Human Resources 50, 317–372.Card, David, 1992, Using regional variation in wages to measure the effects of the federal minimumwage, ILR Review 46, 22–37.Carlson, Murray, Adlai Fisher, and Ron Giammarino, 2004, Corporate investment and asset pricedynamics: implications for the cross-section of returns, The Journal of Finance 59, 2577–2603.Chaney, Barbara A, Paul A Copley, and Mary S Stone, 2003, The effect of fiscal stress and balancedbudget requirements on the funding and measurement of state pension obligations, Journal ofAccounting and Public Policy 21, 287–313.Chen, Huafeng Jason, Marcin Kacperczyk, and Hernan Ortiz-Molina, 2011, Labor unions, operatingflexibility, and the cost of equity, Journal of Financial and Quantitative Analysis 46, 25–58.Coate, Stephen, and Stephen Morris, 1995, On the Form of Transfers to Special Interests, TheJournal of Political Economy 103, 1210–1235.Cole, Shawn, 2009, Fixing market failures or fixing elections? Agricultural credit in India, AmericanEconomic Journal: Applied Economics 1, 219–250.Correia, Sergio, 2015, Singletons, cluster-robust standard errors and fixed effects: A bad mix,Unpublished paper.Cunha, Igor, Miguel A Ferreira, and Rui Silva, 2016, Can Credit Rating Agencies Affect Elec-tion Outcomes?, Working paper. Available at SSRN: https://ssrn.com/abstract=2748458 orhttp://dx.doi.org/10.2139/ssrn.2748458.DiNardo, John E, and Jorn-Steffen Pischke, 1997, The Returns to Computer Use Revisited: HavePencils Changed the Wage Structure Too?, The Quarterly Journal of Economics 112, 291–303.Doran, Kirk, Alexander Gelber, and Adam Isen, 2014, The Effects of High-Skilled ImmigrationPolicy on Firms: Evidence from H-1B Visa Lotteries, NBER Working Paper No. 20668.Duchin, Ran, Oguzhan Ozbas, and Berk A Sensoy, 2010, Costly external finance, corporate invest-ment, and the subprime mortgage credit crisis, Journal of Financial Economics 97, 418–435.Edmans, Alex, 2009, Blockholder trading, market efficiency, and managerial myopia, The Journalof Finance 64, 2481–2513.Eisfeldt, Andrea L, and Dimitris Papanikolaou, 2013, Organization Capital and the Cross-Sectionof Expected Returns, The Journal of Finance 68, 1365–1406.109Elder, Erick M, Gary A Wagner, et al., 2015, Political Effects on Pension Underfunding, Economicsand Politics 27, 1–27.Epple, Dennis, and Katherine Schipper, 1981, Municipal pension funding: A theory and someevidence, Public Choice 37, 141–178.Fazzari, Steven M, R Glenn Hubbard, and Bruce C Petersen, 1988, Financing Constraints andCorporate Investment, Brookings Papers on Economic Activity 1988, 141–206.Fos, Vyacheslav, Kai Li, and Margarita Tsoutsoura, 2016, Do Director Elections Matter?, ChicagoBooth Research Paper No. 17-08. Available at SSRN: https://ssrn.com/abstract=2609815 orhttp://dx.doi.org/10.2139/ssrn.2609815.Foucault, Thierry, and Laurent Fresard, 2014, Learning from peers’ stock prices and corporateinvestment, Journal of Financial Economics 111, 554–577.Ghosh, Anirban, Anna Maria Mayda, and Francesc Ortega, 2014, The impact of skilled foreignworkers on firms: an investigation of publicly traded US firms, IZA Discussion Paper No. 8684.Available at SSRN: https://ssrn.com/abstract=2534712.Giertz, J Fred, and Leslie E Papke, 2007, Public Pension Plans: Myths and Realities for StateBudgets, National Tax Journal 60, 305–23.Giroud, Xavier, 2013, Proximity and Investment: Evidence from Plant-Level Data, The QuarterlyJournal of Economics 128, 861–915.Glaeser, Edward L, and Giacomo AM Ponzetto, 2014, Shrouded costs of government: the politicaleconomy of state and local public pensions, Journal of Public Economics 116, 89–105.Goldin, Claudia, and Lawrence F Katz, 1998, The origins of technology-skill complementarity, TheQuarterly Journal of Economics 113, 693–732.Griliches, Zvi, 1969, Capital-Skill Complementarity, The Review of Economics and Statistics 51,465–68.Gulen, Huseyin, and Mihai Ion, 2015, Policy uncertainty and corporate investment, The Review ofFinancial Studies 29, 523–564.Harris, Milton, and Artur Raviv, 2005, Allocation of decision-making authority, Review of Finance9, 353–383.Haselmann, Rainer, Thomas Kick, Markus Behn, and Vikrant Vig, 2015, The Political Economyof Bank Bailouts, IMFS Working Paper Series 86.Holmstro¨m, Bengt, 1999, Managerial incentive problems: A dynamic perspective, The Review ofEconomic Studies 66, 169–182.110Inman, Robert P, 1982, Public employee pensions and the local labor budget, Journal of PublicEconomics 19, 49–71.Julio, Brandon, and Youngsuk Yook, 2012, Political uncertainty and corporate investment cycles,The Journal of Finance 67, 45–83.Kahle, Kathleen M, and Ralph A Walkling, 1996, The impact of industry classifications on financialresearch, Journal of Financial and Quantitative Analysis 31, 309–335.Kaplan, Steven N, and Luigi Zingales, 1997, Do Investment-Cash Flow Sensitivities Provide UsefulMeasures of Financing Constraints?, The Quarterly Journal of Economics 112, 169–215.Kato, Takao, and Chad Sparber, 2013, Quotas and quality: The effect of H-1B visa restrictions onthe pool of prospective undergraduate students from abroad, Review of Economics and Statistics95, 109–126.Kelley, Dashle G, 2014, The political economy of unfunded public pension liabilities, Public Choice158, 21–38.Kerr, Sari Pekkala, William R Kerr, and William F Lincoln, 2015a, Firms and the Economics ofSkilled Immigration, Innovation Policy and the Economy 15, 115–152.Kerr, Sari Pekkala, William R Kerr, and William F Lincoln, 2015b, Skilled immigration and theemployment structures of US firms, Journal of Labor Economics 33, S147–S186.Kerr, William R, and William F Lincoln, 2010, The supply side of innovation: H-1B visa reformsand US ethnic invention, Journal of Labor Economics 28, 473–508.Kerr, William R, William F Lincoln, and Prachi Mishra, 2014, The dynamics of firm lobbying,American Economic Journal: Economic Policy 6, 343–379.Kido, Nolan, Reining Petacchi, and Joseph Weber, 2012, The influence of elections on the account-ing choices of governmental entities, Journal of Accounting Research 50, 443–476.Lemmon, Michael, and Michael R Roberts, 2010, The Response of Corporate Financing and In-vestment to Changes in the Supply of Credit, Journal of Financial and Quantitative Analysis45, 555–587.Levitt, Steven D, et al., 1997, Using Electoral Cycles in Police Hiring to Estimate the Effect ofPolice on Crime, American Economic Review 87, 270–90.Lewis, Ethan, 2011, Immigration, skill mix, and capital skill complementarity, The Quarterly Jour-nal of Economics 126, 1029–1069.Lewis, Ethan, Giovanni Peri, et al., 2015, Immigration and the Economy of Cities and Regions,Handbook of Regional and Urban Economics 5, 625–685.111Liu, Wai-Man, and Phong TH Ngo, 2014, Elections, political competition and bank failure, Journalof Financial Economics 112, 251–268.Lofstrom, Magnus, and Joseph J Hayes, 2011, H-1Bs: How Do They Stack Up to US Born Workers?,IZA Discussion Paper No. 6259. Available at SSRN: https://ssrn.com/abstract=1981215.Malmendier, Ulrike, and Geoffrey Tate, 2005, CEO overconfidence and corporate investment, TheJournal of Finance 60, 2661–2700.Martin, RL, and MC Moldoveanu, 2003, Capital versus talent. The battle that’s reshaping business.,Harvard Business Review 81, 36–41.Matsa, David A, 2010, Capital structure as a strategic variable: Evidence from collective bargaining,The Journal of Finance 65, 1197–1232.Meulbroek, Lisa K, Mark L Mitchell, J Harold Mulherin, Jeffry M Netter, and Annette B Poulsen,1990, Shark repellents and managerial myopia: An empirical test, Journal of Political Economy98, 1108–1117.Mian, Atif, and Amir Sufi, 2014, What Explains the 2007–2009 Drop in Employment?, Economet-rica 82, 2197–2223.Mitchell, Olivia S, and Robert S Smith, 1994, Pension Funding in the Public Sector, The Reviewof Economics and Statistics 76, 278–290.Molloy, Raven, Christopher L Smith, and Abigail Wozniak, 2014, Declining Migration within theUS: the Role of the Labor Market, NBER Working Paper No. 20065.Munnell, Alicia H, Jean-Pierre Aubry, Josh Hurwitz, and Laura Quinby, 2011a, Comparing com-pensation: State-local versus private sector workers, Center for Retirement Research Issue inBrief 20.Munnell, Alicia H, Jean-Pierre Aubry, and Laura Quinby, 2011b, Public pension funding in practice,Journal of Pension Economics and Finance 10, 247–268.Munnell, Alicia H, and Laura Quinby, 2012, Legal constraints on changes in state and local Pensions,Center for Retirement Research Issue in Brief 25.Nagarajan, Nandu J, K Sivaramakrishnan, and Sri S Sridhar, 1995, Managerial entrenchment,reputation and corporate investment myopia, Journal of Accounting, Auditing & Finance 10,565–585.Narayanan, MP, 1985, Managerial incentives for short-term results, The Journal of Finance 40,1469–1484.Nordhaus, William D, 1975, The Political Business Cycle, The Review of Economic Studies 42,169–190.112Nordhaus, William D, 2002, Productivity growth and the new economy, Brookings Papers onEconomic Activity 211–267.Novy-Marx, Robert, and Joshua Rauh, 2011, Public pension promises: how big are they and whatare they worth?, The Journal of Finance 66, 1211–1249.Novy-Marx, Robert, and Joshua Rauh, 2014, The revenue demands of public employee pensionpromises, American Economic Journal: Economic Policy 6, 193–229.Panousi, Vasia, and Dimitris Papanikolaou, 2012, Investment, idiosyncratic risk, and ownership,The Journal of Finance 67, 1113–1148.Peltzman, Sam, 1992, Voters as fiscal conservatives, The Quarterly Journal of Economics 107,327–361.Peri, Giovanni, Kevin Yang Shih, and Chad Sparber, 2014, Foreign Stem Workers and NativeWages and Employment in US Cities, NBER Working Paper No. 20093.Pe´rignon, Christophe, and Boris Valle´e, 2017, The political economy of financial innovation: evi-dence from local governments, The Review of Financial Studies 30, 1903–1934.Persson, Torsten, and Guido Enrico Tabellini, 2002, Political economics: explaining economic policy(MIT press).Poterba, James M, 1994, State Responses to Fiscal Crises: The Effects of Budgetary Institutionsand Politics, Journal of Political Economy 799–821.Rauh, Joshua D, 2006, Investment and financing constraints: Evidence from the funding of corpo-rate pension plans, The Journal of Finance 61, 33–71.Rauh, Joshua D, et al., 2010, Are State Public Pensions Sustainable? Why The Federal GovernmentShould Worry About State Pension Liabilities, National Tax Journal 63, 585–601.Reynolds, C Lockwood, 2014, State politics, tuition, and the dynamics of a political budget cycle,Empirical Economics 46, 1241–1270.Rogoff, Kenneth, 1990, Equilibrium Political Budget Cycles, American Economic Review 80, 21–36.Rogoff, Kenneth, and Carmen Reinhart, 2010, Growth in a Time of Debt, American EconomicReview 100, 573–8.Rogoff, Kenneth, and Anne Sibert, 1988, Elections and macroeconomic policy cycles, The Reviewof Economic Studies 55, 1–16.Rose, Shanna, 2006, Do fiscal rules dampen the political business cycle?, Public choice 128, 407–431.113Shi, Min, and Jakob Svensson, 2006, Political budget cycles: Do they differ across countries andwhy?, Journal of Public Economics 90, 1367–1389.Simintzi, E, V Vig, and P Volpin, 2015, Labor Protection and Leverage, The Review of FinancialStudies 28, 561–591.Splinter, David, 2015, State pension contributions and fiscal stress, Journal of Pension Economicsand Finance 16, 65–80.Stefanescu, Irina, Kangzhen Xie, and Jun Yang, 2015, Pay me now (and later):bonus boosts before pension freezes and executive departures, Kelley School of Busi-ness Research Paper No. 15-5. Available at SSRN: https://ssrn.com/abstract=2533459 orhttp://dx.doi.org/10.2139/ssrn.2533459.Stein, Jeremy C, 1988, Takeover threats and managerial myopia, Journal of Political Economy 96,61–80.Stein, Jeremy C, 1989, Efficient capital markets, inefficient firms: A model of myopic corporatebehavior, The Quarterly Journal of Economics 104, 655–669.Tirole, Jean, 1994, The Internal Organization of Government, Oxford Economic Papers 46, 1–29.Wahal, Sunil, and John J McConnell, 2000, Do institutional investors exacerbate managerial my-opia?, Journal of Corporate Finance 6, 307–329.Zingales, Luigi, 2000, In search of new foundations, The Journal of Finance 55, 1623–1653.114Appendix AProofsA.1. Proof of Lemma 1Proof. In order to express the partial derivative of Ω with respect to w, it is useful to first re-framethe voter’s election decision in terms of m∆2 , which we define as m∆2 = mI2 −mC2. Substituting4.2.7 and 4.2.8 into 4.2.10 and 4.2.11, we obtainθ =1 if m∆2 ≥ 00 if otherwise, (A.1.1)if the representative voter observes d before the election andθ =1 if m∆2 ≥ µ(w − w¯)0 if otherwise, (A.1.2)if the representative voter does not observe w before the election.At the beginning of period 1, m∆2 is a random variable that follows the distributionm∆2 |Ψ1 ∼ N(m∆1 ,µhI1), (A.1.3)where Ψ1 denotes the incumbent’s information set at the beginning of period 1, and we getV ar(m∆2 |Ψ1) = µhI1 from the fact thatV ar(m∆2 |Ψ1) = V ar((1− µ)mI1 + µz −mC2|Ψ1)= µ2V ar(ηI + I)=µhI1.Therefore, we can express Ω asΩ = E1[θ(ηI − ηC + x)]= E1[θ(mI2 −mC2 + x)]= E1[θ(m∆2 + x)],where the second line follows from applying the law of iterated expectations, and the third line115follows the definition of m∆2 . Next, we use the definition of the expectation function as an integral,and apply A.1.1, A.1.2, and A.1.3 to obtainΩ = ρ∞∫µ(w−w¯)(m∆2 + x)φ(m∆2 ;m∆1 ,µhI1)dm∆2 + (1− ρ)∞∫0(m∆2 + x)φ(m∆2 ;m∆1 ,µhI1)dm∆2 ,which follows from the fact that the representative voter’s decision follows A.1.2 with probabilityρ, and follows A.1.1 with probability 1− ρ.Differentiating both sides with respect to w and applying the fundamental theorem of calculus,we obtain the required result∂Ω∂w= −ρµφ(µ(w − w¯);m∆1 ,µhI1)(x+ µ(w − w¯)).A.2. Proof of Proposition 1Proof. To solve the optimization problem according to 4.2.12, we take the first order necessaryconditions of the LagrangianL = E1[ηI − w + 1 + ηC + θ(ηI − ηC + x)− b+ 2] + λ[u(w) + u(b)− u¯],to obtainω∗ + λu′(w) = 1 (A.2.1)λu′(b) = 1 (A.2.2)u(w) + u(b) = u¯ (A.2.3)λ > 0, (A.2.4)where ω∗ ≡ ∂Ω∂w∣∣∣∣w=w¯represents the equilibrium equilibrium election manipulation incentive.It is immediately clear from A.2.1 and A.2.2 that w = b under full transparency (i.e. whenω∗ = 0). To show how b varies with ρ, we differentiate both sides of A.2.1, A.2.2, and A.2.3 withrespect to ρ to obtain∂ω∗∂ρ+∂λ∂ρu′(w) + λu′′(w)∂w∂ρ= 0 (A.2.5)∂λ∂ρu′(b) + λu′′(b)∂b∂ρ= 0 (A.2.6)u′′(w)∂w∂ρ+ u′′(b)∂b∂ρ= 0. (A.2.7)116Solving for ∂b∂ρ , we obtain∂b∂ρ=u′(b)u′(w)λ(u′′(b)u′(w)2 + u′′(w)u′(b)2)∂ω∗∂ρ. (A.2.8)Since u′(·) > 0 and u′′(·) < 0, and λ > 0, it follows that u′(b)u′(w)λ(u′′(b)u′(w)2+u′′(w)u′(b)2) is negative,which implies that ∂b∂ρ has the opposite sign as∂ω∗∂ρ . But we know from 4.2.14 that∂ω∗∂ρ < 0, whichmeans that ∂b∂ρ > 0. This completes the proof for part (a) of the Proposition.Following a similar path, we differentiate both sides of A.2.1, A.2.2, and A.2.3 with respect tom∆1 and solve for∂b∂m∆1, we obtain∂b∂m∆1=u′(b)u′(w)λ(u′′(b)u′(w)2 + u′′(w)u′(b)2)∂ω∗∂m∆1. (A.2.9)Since we have already established u′(b)u′(w)λ(u′′(b)u′(w)2+u′′(w)u′(b)2) is negative, A.2.9 implies that∂b∂m∆1has the opposite sign as ∂ω∗∂m∆1. Using the definition of ω∗ from 4.2.14 and by the properties of thenormal distribution function, it follows that∂ω∗∂m∆1> 0 if m∆1 > 0< 0 if m∆1 < 0= 0 if m∆1 = 0,which means that∂b∂m∆1< 0 if m∆1 > 0> 0 if m∆1 < 0= 0 if m∆1 = 0,which completes the proof for part (b) and (c) of the Proposition.A.3. Proof of Lemma 2Proof. Following the same logic as the first part of the proof from A.1, we can express Ω as follows:Ω = ρ∞∫µ(w−w¯+k−k¯)(m∆2 + x)φ(m∆2 ;m∆1 ,µhI1)dm∆2 + (1− ρ)∞∫0(m∆2 + x)φ(m∆2 ;m∆1 ,µhI1)dm∆2which we apply the fundamental theorem of calculus to differentiate with respect to w and k,117respectively, to obtain the required results∂Ω∂w= −ρµφ(µ(w − w¯ + k − k¯);m∆1 ,µhI1)(x+ µ(w − w¯ + k − k¯)),and∂Ω∂k= −ρµφ(µ(w − w¯ + k − k¯);m∆1 ,µhI1)(x+ µ(w − w¯ + k − k¯)).A.4. Proof of Proposition 2Proof. The Lagrangian associated with the optimization problem according to 4.3.5 isL = E1[ηI − w − k + 1 + ηC + θ(ηI − ηC + x)− pi(b− k) + 2] + λ[u(w) + u(k + pi(b− k))− u¯]which yields the first order necessary conditionsω∗ + λu′(w) = 1 (A.4.1)ω∗ + (1− pi)λu′(s) = 1− pi (A.4.2)u(w) + u(s) = u¯ (A.4.3)λ > 0, (A.4.4)where ω∗ ≡ ∂Ω∂w∣∣∣∣w=w¯,k=k¯= ∂Ω∂k∣∣∣∣w=w¯,k=k¯represents the equilibrium equilibrium election manipulationincentive and s = k + pi(b− k) represents employees’ period 2 consumption.It is immediately clear from A.4.1 and A.4.2 that w = s under full transparency (i.e. whenω∗ = 0). To show how k varies with ρ, we differentiate both sides of A.4.1, A.4.2, and A.4.3 withrespect to ρ to obtain∂ω∗∂ρ+∂λ∂ρu′(w) + λu′′(w)∂w∂ρ= 0 (A.4.5)∂ω∗∂ρ+ (1− pi)(∂λ∂ρu′(s) + (1− pi)λu′′(s)∂k∂ρ) = 0 (A.4.6)u′′(w)∂w∂ρ+ (1− pi)u′′(s)∂k∂ρ= 0. (A.4.7)Solving for ∂k∂ρ , we obtain∂k∂ρ=−piu′(w)(1− pi)λ2(u′′(w)u′(s)2 + u′′(s)u′(w)2)∂ω∗∂ρ. (A.4.8)118Since u′(·) > 0 and u′′(·) < 0, and λ > 0, it follows that−piu′(w)(1− pi)λ2(u′′(w)u′(s)2 + u′′(s)u′(w)2)> 0 if pi > 0= 0 if pi = 0,which implies that ∂k∂ρ has the same sign as∂ω∗∂ρ if pi > 0, and is zero otherwise. But we know from4.3.7 that ∂ω∗∂ρ < 0, which means that∂k∂ρ < 0 if pi > 0. This completes the proof for part (a) of theProposition.Following a similar path, we differentiate both sides of A.4.1, A.4.2, and A.4.3 with respect tom∆1 and solve for∂k∂m∆1, we obtain∂k∂m∆1=−piu′(w)(1− pi)λ2(u′′(w)u′(s)2 + u′′(s)u′(w)2)∂ω∗∂m∆1. (A.4.9)Since we have already established −piu′(w)(1−pi)λ2(u′′(w)u′(s)2+u′′(s)u′(w)2) is nonnegative, A.4.9 impliesthat ∂k∂m∆1has the same sign as ∂ω∗∂m∆1if pi > 0, and is equal to zero otherwise. Using the definitionof ω∗ from 4.3.7 and by the properties of the normal distribution function, it follows that∂k∂m∆1< 0 if m∆1 < 0> 0 if m∆1 > 0= 0 if m∆1 = 0,if pi > 0, and ∂k∂m∆1= 0 if pi = 0. This completes the proof for part (b) and (c) of the Proposition.119Appendix BVariable DefinitionsB.1. Variable Definitions for Chapter 2CapEx : Quarterly capital expenditures (capxy) scaled by lagged total assets (atq). Note Compustat variable capxyis year-to-date cumulate, so for fiscal quarter 2, 3, 4, the lagged capxy is subtracted from current capxy (Source:Compustat).Tobin ′s Q : Market value of quarter-end total assets (atq+prccq×cshoq−ceqq− txditcq) scaled by quarter-end bookvalue of total assets (atq) (Source: Compustat).ln(Size): Natural log of quarter-end total assets (atq) (Source: Compustat).Cash Flow : Quarterly income before extraordinary items and depreciation (ibq + dpq) scaled by quarter-end totalassets (atq) (Source: Compustat).Cash Holdings: Quarter-end cash holdings (cheq) scaled by total assets (atq). (Source: Compustat).Leverage: Quarter-end long-term book value of debt (dlttq) scaled by quarter-end total assets (atq). (atq) (Source:Compustat).H1B use: The total number of H-1B initial petitions submitted to the USCIS during the 2001 calendar year, scaledby the average number of workers employed by the firm in 2001 (Source: USCIS petitions, Compustat).ln(H1B use): The natural log of H1B use as defined above (Source: USCIS petitions, Compustat).High H1B use: A dummy variable that takes a value of one if H1B use, as defined above, is above the sample median,and zero otherwise (Source: USCIS petitions, Compustat).H1B usej : The total number of H-1B initial petitions submitted to the USCIS during the 2001 calendar year forworkers in occupational category j, scaled by the average number of workers employed by the firm in 2001 (Source:USCIS petitions, Compustat).H1B : A dummy variable that takes on a value of one if the firm filed at least one H-1B initial petition to the USCISduring 2001, and zero otherwise (Source: USCIS petitions).H1B wage: The sum of wages listed across H-1B initial petitions submitted to the USCIS during the 2001 calendaryear, scaled by the product of the average number of workers employed by the firm in 2001 and the national industryaverage wage at the 3-digit NAICS level (Source: USCIS petitions, Compustat, BLS QCEW files).ln(H1B wage): The natural log of H1B wage as defined above (Source: USCIS petitions, Compustat, BLS QCEWfiles).120High H1B wage: A dummy variable that takes a value of one if H1B wage, as defined above, is above the samplemedian, and zero otherwise (Source: USCIS petitions, Compustat, BLS QCEW files).Wage: The wage listed for the position for the prospective H-1B worker on the H-1B petition (Source: USCIS peti-tions).High Wage: A dummy variable that takes a value of one if the firm average Wage, as defined above, for initialpetitions submitted in 2001 is above the sample median, and zero otherwise (Source: USCIS petitions).Occ Wage: The wage from BLS Occupational Employment Statistics corresponding to the DOT occupational code(cross-referenced with SOC codes) for the H-1B worker (Source: USCIS petitions, BLS OES files).Age: The age of the prospective H-1B worker listed on the H-1B petition (Source: USCIS petitions).High Age: A dummy variable that takes a value of one if the firm average Age, as defined above, for initial petitionssubmitted in 2001 is above the sample median, and zero otherwise (Source: USCIS petitions).Grad : A dummy variable that takes a value of one if the prospective H-1B worker is listed on the H-1B petition aspossessing a Master’s or PhD degree, and zero otherwise (Source: USCIS petitions).Grad : A dummy variable that takes a value of one if the firm average Grad , as defined above, for initial petitionssubmitted in 2001 is above the sample median, and zero otherwise (Source: USCIS petitions).HQ State: A dummy variable that takes a value of one if the prospective H-1B worker is listed to be in the samestate as the location of firm headquarters as reported in Compustat, and zero otherwise (Source: DOL LCA files,Compustat).NearHQ: A dummy variable that takes a value of one if the firm average HQ State, as defined above, for initialpetitions submitted in 2001 is above the sample median, and zero otherwise (Source: DOL LCA files, Compustat).Manufacturing : A dummy variable that takes on a value of one if the firm SIC classification is between 2000 and3999, and zero otherwise (Source: Compustat).Services: A dummy variable that takes on a value of one if the firm SIC classification is between 7000 and 8999, andzero otherwise (Source: Compustat).IT : A dummy variable that takes on a value of one if the firm SIC classification is 3341, 3342, 3343, 3344, 3345,3346, 5111, 5112, 5161, 5181, 5182, 5191, or 5415, and zero otherwise (Source: Compustat).New Econ: A dummy variable that takes on a value of one if the firm SIC classification is between 35, 36, 48 (2-digit),or 873 (3-digit), and zero otherwise (Source: Compustat).High TQ: A dummy variable that takes a value of one if the pre-treatment (2002) industry average Tobin ′s Q atthe 2-digit SIC level is above the sample median, and zero otherwise (Source: Compustat).High Size: A dummy variable that takes a value of one if the pre-treatment (2002) industry average ln(Size) at the2-digit SIC level is above the sample median, and zero otherwise (Source: Compustat).121High RD: A dummy variable that takes a value of one if the pre-treatment (2002) industry average R&D expendi-tures scaled by assets at the 2-digit SIC level is above the sample median, and zero otherwise (Source: Compustat).Notes: Compustat data comes from the Compustat Fundamentals Annual (annual) and Compustat FundamentalsQuarterly files (quarterly). USCIS data comes from the United States Citizenship and Immigration Services via aFreedom of Information Act (FOIA) request. DOL LCA data comes from the Department of Labor’s website atwww.foreignlaborcert.doleta.gov/performancedata.cfm. BLS QCEW data comes from www.bls.gov/cew/. BLS OESdata comes from www.bls.gov/oes/.B.2. Variable Definitions for Chapter 3Electionit: Indicator variable that takes on a value of one if a gubernatorial election occurs before the end of thefiscal year for plan i’s state in fiscal year t, and zero otherwise (Source: Klarnerpolitics.com, The Book of the States).ContribGov it: Total employer contributions (contrib ER regular + contrib ER state) divided by total pensionableearnings of plan participants (payroll) (source: CRR Public Plans Database).ContribMbrsit: Total employee contributions (contrib EE regular+contrib ER other+contrib EE PurchaseService)divided by total pensionable earnings of plan participants (payroll) (source: CRR Public Plans Database).Contribit: The sum of ContribGov it and ContribMbrsit (source: CRR Public Plans Database).AccGov it: The employer’s share of the normal cost rate (NormCostRate ER) (source: CRR Public Plans Database).AccMbrsit: The employee’s share of the normal cost rate (NormCostRate EE) (source: CRR Public Plans Database).Accit: The sum of AccGov it and AccMbrsit (source: CRR Public Plans Database).PenDef it: The difference between Accit and Contribit (source: CRR Public Plans Database).PenDefGov it: The difference between AccGov it and ContribGov it (source: CRR Public Plans Database).PenDefMbrsit: The difference between AccMbrsit and ContribMbrsit (source: CRR Public Plans Database).ln(Payroll)it: The natural log of total pensionable earnings of plan participants (payroll) (source: CRR Public PlansDatabase).ln(Avg Salary)it: The natural log of the average salary among active participants (ActiveSalary avg) (source: CRRPublic Plans Database).Incomeit: The difference between total income (income net) and total contributions (contrib tot), divided by totalpensionable earnings of plan participants (payroll) (source: CRR Public Plans Database).Discount Rateit: The assumed return on investments used to discount plan liabilities reported under GASB require-ments (InvestmentReturnAssumption GASB) (source: CRR Public Plans Database).122Inflation Rateit: The assumed inflation rate (InflationAssumption GASB) (source: CRR Public Plans Database).CostMthd EAN it: An indicator variable that takes on a value of one if the plan uses the Entry Age Normal costmethod in order to evaluate pension liabilities, and zero otherwise (source: CRR Public Plans Database).Deficit Shock it: Per capita unexpected budget deficit—i.e. (expenditure shock − revenue shock)/state population,where expenditure shock = actual expenditures− projected expenditures− enacted expenditure adjustments andrevenue shock = actual revenue− projected revenue− enacted expenditure revenue (see Poterba (1994)) (source:National Association of State Budget Officers (NASBO) Fiscal Survey of States).State Unempit: State unemployment rate (source: Bureau of Labor Statistics).Pub Union Mbrshpit: Proportion of state public-sector workers that are members of a labour union (source: Union-stats.com (Hirsch and Macpherson)).∆UnfundedLiabi: The plan-level time series average for ∆Unfunded LiabitPayrollit, where ∆ indicates the first difference opera-tor, and Unfunded Liabilityit is the unfunded actuarial accrued liability (UAAL GASB) (source: CRR Public PlansDatabase).PenDefCyci: E¯i[PenDef it|Electionit = 1] − E¯i[PenDef it|Electionit = 0], where E¯i[X|Y ] denotes the time-seriesaverage, for plan i, of X conditional on Y (source: CRR Public Plans Database).PenDefCycDi: E¯i[PenDefDit|Electionit = 1] − E¯i[PenDefDit|Electionit = 0], where E¯i[X|Y ] denotes the time-series average, for plan i, of X conditional on Y and PenDefDit represents the residual term from estimatingPenDef it = α+ δ · t+ it (source: CRR Public Plans Database).PenDefCycRi: E¯i[PenDefRit|Electionit = 1] − E¯i[PenDefRit|Electionit = 0], where E¯i[X|Y ] denotes the time-series average, for plan i, of X conditional on Y and PenDefRit represents the residual term from estimatingPenDef it = α+ κi + λt +Xitβ + it (source: CRR Public Plans Database).Budget Year it: An indicator variable that takes on a value of one if the plan i is located in a state that passed abudget in year t (source: Klarnerpolitics.com).LegisExpit: An indicator variable that takes on a value of one if the Governor has prior experience in the statelegislature (source: Klarnerpolitics.com).Opaque Pensionsit: An indicator variable that takes a value of one if plan i is in a state that is in the bottom decilein terms of the SII transparency indicator for state pension fund management, and zero otherwise (source: Centerfor Public Integrity State Integrity Investigation).Transparent Pensionsit: An indicator variable that takes a value of one if plan i is in a state that is in the top decilein terms of the SII transparency indicator for state pension fund management, and zero otherwise (source: Centerfor Public Integrity State Integrity Investigation).Opaque Budget it: An indicator variable that takes a value of one if plan i is in a state that is in the bottom decilein terms of the SII transparency indicator for state budget process, and zero otherwise (source: Center for PublicIntegrity State Integrity Investigation).123Transparent Budget it: An indicator variable that takes a value of one if plan i is in a state that is in the top decilein terms of the SII transparency indicator for state budget process, and zero otherwise (source: Center for PublicIntegrity State Integrity Investigation).Transparent Budget it: An indicator variable that takes a value of one if plan i is in a state that is in the top decilein terms of the SII transparency indicator for state budget process, and zero otherwise (source: Center for PublicIntegrity State Integrity Investigation).VicMarginit: The margin of victory (as a fraction of 1) between the winning candidate and the runner up given agubernatorial election occurs in year t, and zero otherwise (source: Klarnerpolitics.com).IncumbLosesit: An indicator variable that takes on a value of one if the incumbent Governor loses an election in yeart, and zero otherwise. (source: Klarnerpolitics.com).Lame Duck it: An indicator variable that takes on a value of one if the plan i is located in a state a Governor facingbinding term limits in year t (source: Klarnerpolitics.com).Republicanit: An indicator variable that takes on a value of one if the incumbent Governor belongs to the Republicanparty. (source: Klarnerpolitics.com).Strong Protect i: An indicator variable that takes on a value of one if the plan i is located in a state that offersconstitutional protection of state DB pension benefits (source: Munnell and Quinby (2012)).Weak Protect i: An indicator variable that takes on a value of one if the plan i is located in a state that offersprotection of state DB pension benefits under the gratuity principle (source: Munnell and Quinby (2012)).Unconditional Protect i: An indicator variable that takes on a value of one if the plan i is located in a state that offersunconditional protection of state DB pension benefits (source: Munnell and Quinby (2012)).Gov Changeit: An indicator variable that takes a value of one if plan i is in a state where there was an unexpectedGovernor change due to death, resignation, or impeachment in year t and zero otherwise (source: Klarnerpolitics.com,The Book of the States).ln(GDP Growth)j : The time-series mean in the annual log growth rate of real GDP for state j over the 2001-2015sample period (source: Bureau of Economic Analysis).ln(HPI Growth)j : The time-series mean in the quarterly log growth rate of seasonally-adjusted house price indexvalues (based on purchases only) for state j over the 2001-2015 sample period (source: Federal Housing FinanceAgency).124Appendix CMiscellaneousC.1. Occupation DefinitionsArchitecture, Engineering, And Surveying (Engineering)001 Architectural Occupations002 Aeronautical Engineering Occupations003 Electrical/Electronics Engineering Occupations005 Civil Engineering Occupations006 Ceramic Engineering Occupations007 Mechanical Engineering Occupations008 Chemical Engineering Occupations010 Mining And Petroleum Engineering Occupations011 Metallurgy And Metallurgical Engineering Occupations012 Industrial Engineering Occupations013 Agricultural Engineering Occupations014 Marine Engineering Occupations015 Nuclear Engineering Occupations017 Drafters, N.E.C.018 Surveying/Cartographic Occupations019 Occupations In Architecture, Engineering, And Surveying, N.E.C.Mathematics And Physical Sciences (Combined under Sciences)020 Occupations In Mathematics021 Occupations In Astronomy022 Occupations In Chemistry023 Occupations In Physics024 Occupations In Geology025 Occupations In Meteorology029 Occupations In Mathematics And Physical Sciences, N.E.C.Computer-Related Occupations (Computers)030 Occupations In Systems Analysis And Programming031 Occupations In Data Communications And Networks032 Occupations In Computer Systems User Support033 Occupations In Computer Systems Technical Support039 Computer-Related Occupations, N.E.C.Life Sciences (Combined under Sciences)040 Occupations In Agricultural Sciences041 Occupations In Biological Sciences045 Occupations In Psychology125049 Occupations In Life Sciences, N.E.C.Administrative Specializations (Admin)160 Accountants, Auditors, And Related Occupations161 Budget And Management Systems Analysis Occupations162 Purchasing Management Occupations163 Sales And Distribution Management Occupations164 Advertising Management Occupations165 Public Relations Management Occupations166 Personnel Administration Occupations168 Inspectors And Investigators, Managerial And Public Service169 Occupations In Administrative Specializations, N.E.C.Managers And Officials, N.E.C. (Management)180 Agriculture, Forestry, And Fishing Industry Managers And Officials181 Mining Industry Managers And Officials182 Construction Industry Managers And Officials183 Manufacturing Industry Managers And Officials184 Transportation, Communication, And Utilities Industry Managers And Officials185 Wholesale And Retail Trade Managers And Officials186 Finance, Insurance, And Real Estate Managers And Officials187 Service Industry Managers And Officials188 Public Administration Managers And Officials189 Miscellaneous Managers And Officials, N.E.C.126C.2. Actuarial Valuations MethodsThe information provided here is a brief summary of the much fuller description, including detailedformulas, found in Section II of Novy-Marx and Rauh (2011). We begin with the concept of theAccumulated Benefit Obligation (ABO), which reflects the terminal value of a plan’s liabilities if allbenefits were permanently frozen at its current level. Calculating the ABO requires assumptionsabout mortality rates and future inflation, and these assumptions are applied to the current benefitformula, wages, and employees’ accumulated years of service to arrive at a discounted present value.In essence, the ABO captures benefits that have already been promised and accrued.A broader concept of pension liabilities is the Projected Value of Benefits (PVB), which accountsfor expected future years of service and wage growth for current employees. Estimating the PVBrequires additional actuarial assumptions about salary growth rates and job separation rates. ThePVB method is a significantly more conservative estimate of pension liabilities relative to the ABO,as it operates under the implicit assumption that the plan sponsor cannot curtail future benefitaccruals for current employees.Almost all state plans apply one of two liability measures—the Projected Benefit (PBO) andthe Entry Age Normal (EAN)—both of which fall in between ABO and the PVB in terms ofconservatism. The PBO takes the PVB and prorates it by current years of accrued service, whichimplies recognition of projected wage growth but not future years of service. The EAN takes thePVB and amortizes it into a series of annual accruals such that each accrual is a constant percentageof projected salary. Assuming that the wage growth rate is lower than the discount rate, the EANis more conservative than the PBO, and is interpreted to account for some future service in additionto wage growth.127

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}]}"
                            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.24.1-0349034/manifest

Comment

Related Items