Taxpayer Responses over the Cycle: Evidence from Irish Notches Enda Patrick Hargaden Department of Economics University of Michigan IFS Labour Market Conference March 1, 2016 Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 1 / 51 Introduction and Motivation Why would cyclical variation be interesting? Standard theory shows that the behavioral response optimally determines: 1 Tax rates 2 Extent of redistribution 3 Enforcement levels Should these vary over the cycle? Some have argued a Keynesian cyclicality story on the expenditure side. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 2 / 51 Introduction and Motivation Why would cyclical variation be interesting? Standard theory shows that the behavioral response optimally determines: 1 Tax rates 2 Extent of redistribution 3 Enforcement levels Should these vary over the cycle? Some have argued a Keynesian cyclicality story on the expenditure side. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 2 / 51 Introduction and Motivation Which way does cyclical variation swing? Arguments for increase in responsiveness during a recession: If firms are struggling to survive, they may search harder for cost savings With income effects, employees may value a euro more during a recession Arguments for decrease: There may be fewer opportunities to hide income when GDP is smaller It is harder to adjust hours during a recession, by e.g. switching jobs Wage rigidity could be a factor in downturns Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 3 / 51 Introduction and Motivation Which way does cyclical variation swing? Arguments for increase in responsiveness during a recession: If firms are struggling to survive, they may search harder for cost savings With income effects, employees may value a euro more during a recession Arguments for decrease: There may be fewer opportunities to hide income when GDP is smaller It is harder to adjust hours during a recession, by e.g. switching jobs Wage rigidity could be a factor in downturns Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 3 / 51 Introduction and Motivation What does this paper have? Administrative employee tax return data, and a set of notches Unprecedentedly volatile developed economy Insight into “frictions”, i.e. the determinants of reporting tax disadvantaged incomes Focus on the inter-temporal component of taxpayer response A new method (bunching in differences) to investigate the responsiveness Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 4 / 51 Introduction and Motivation Preview of Results The evidence of taxpayer response is much weaker during the recession. Pre-2008 ‘treatment effects’ are about three times as large as post-2008. Much of the difference reflects non-random attrition from labor force. I Reduced employment in sectors (e.g. construction) that exhibit above-average ability to report tax-advantaged incomes creates composition effect/composition bias. However, it’s more than just a composition effect. I Holding the labor force constant, people are less responsive to the incentives. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 5 / 51 Introduction and Motivation Overview of Talk 1 The theory of notches 2 Taxpayer responses and the elasticity of taxable income 3 Data description and introduction to methods 4 Empirical evidence of cyclical variation 5 Determinants of tax avoidance 6 Inverse probability weighting estimates? Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 6 / 51 Theory of Notches Question: What exactly are notches? Post-tax income Y1 Hargaden (University of Michigan) Y2 Irish Tax Notches Pre-tax income March 1, 2016 7 / 51 Theory of Notches Question: What exactly are notches? Post-tax income Y1 Hargaden (University of Michigan) Y2 Irish Tax Notches Pre-tax income March 1, 2016 7 / 51 Theory of Notches Question: What exactly are notches? Post-tax income Y1 Hargaden (University of Michigan) Y2 Irish Tax Notches Pre-tax income March 1, 2016 7 / 51 Theory of Notches Question: What exactly are notches? Post-tax income Y1 Hargaden (University of Michigan) Yl Irish Tax Notches Y2 Yu Pre-tax income March 1, 2016 7 / 51 Theory of Notches Question: What exactly are notches? Post-tax income Strictly dominated region Y1 Hargaden (University of Michigan) Yl Irish Tax Notches Y2 Yu Pre-tax income March 1, 2016 7 / 51 Understanding Taxpayer Responses Understanding Taxpayer Responses Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 8 / 51 Understanding Taxpayer Responses Taxpayer Response and Taxable Income Debate over which margin of adjustment: hours, gross income, evasion/avoidance? Feldstein (1995) shifted the focus to taxable income (ETI). Many margins of adjustment, all captured by taxable income. Under certain circumstances, this measures the welfare costs of taxation. Readily available empirical source. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 9 / 51 Understanding Taxpayer Responses Bunching at notches to identify taxpayer response Notches provide very strong incentives, and clean tests of ability to avoid taxes. Extent of level-bunching can identify ETI (Saez, 2010). The larger the bunching, the larger the ETI. Chetty et al (2011) and Kleven & Waseem (2013) bunching in Denmark and Pakistan. Almunia & Lopez-Rodriguez (2014), Auten & Carroll (1999), Bastani & Selin (2014), Gruber & Saez (2002), Jenderny (2015), Kopczuk & Munroe (2015), Marx (2012), Onji (2009), Ramnath (2013), Sallee & Slemrod (2012), Slemrod et al (2015), Yelowitz (1995). Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 10 / 51 Understanding Taxpayer Responses Slemrod & Kopczuk: ETI as a policy choice ETI is not an immutable parameter. For a cost, we can influence it. Dependence on tax system: compare Denmark and Greece. Increased enforcement will lower responsiveness — ETI can be lowered (at a cost). Subtle point: I find a very low responsiveness for Ireland. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 11 / 51 Understanding Taxpayer Responses Policy and academic implications of the paper Time-varying responsiveness affects: 1 Optimal enforcement varies with the phase of the cycle: fund HMRC in good times. 2 Stimulus package identification bias. Small aside: tax cuts likely less stimulating than hoped. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 12 / 51 Understanding Taxpayer Responses Policy and academic implications of the paper Time-varying responsiveness affects: 1 Optimal enforcement varies with the phase of the cycle: fund HMRC in good times. 2 Stimulus package identification bias. Small aside: tax cuts likely less stimulating than hoped. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 12 / 51 Data Description Data and Empirics Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 13 / 51 Data Description Sample from Universe of Irish Tax Returns I have access to an administrative panel of employee tax returns with demographics and details on their employer. Employers must file an income statement for each employee Dataset is a panel (2006–2013) of 10% of all employees, representative of full sample Demographic information on employee (age, sex, nationality) Some business information on employer (legal form, number of employees, detailed sector) Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 14 / 51 Data Description Trends in the data Earnings distribution in Ireland, 2006-2013 40,000 Frequency 30,000 20,000 10,000 0 Hargaden (University of Michigan) 20,000 40,000 60,000 Total Annual Pay (€) Irish Tax Notches 80,000 100,000 March 1, 2016 15 / 51 Data Description Trends in the data 100 Real Output (2005 Q1=100) 105 110 115 120 Irish and US Business Cycles 2006 2008 2010 Year Ireland 2012 2014 USA Sources: Central Statistics Office, FRED Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 16 / 51 Data Description Trends in the data 12.5 10 7.5 5 2.5 Unemployment Rate (seasonally adjusted) 15 Irish and US Unemployment Rates 2005 2006 2007 2008 2009 2010 Ireland 2011 2012 2013 2014 USA Sources: Central Statistics Office, BLS Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 17 / 51 Data Description Trends in the data 100 80 60 40 Employment by Sector (2006 Q2 = 100) 120 Employment in Ireland 2007 2008 Total 2009 2010 2011 Year Non-construction 2012 2013 2014 Construction Source: Central Statistics Office/QNHS Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 18 / 51 Data Description Trends in the data Limitation of the data The notches are not applicable to people who are (i) married; (ii) jointly filing; and (iii) both spouses working. Will attenuate results. Likely becomes less of a problem over time. Marriage rate Employment rate Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 19 / 51 Data Description Experimental Design A Tale of Three Notches Health Levy — earmarked for contribution to public health service, enacted in 1979 Income Levy — enacted by the Financial Emergency Measures in the Public Interest Act, 2009 Universal Social Charge (USC) — enacted in 2011 to replace both the Health and Income Levies. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 20 / 51 Data Description Experimental Design A Tale of Three Notches Health Levy — earmarked for contribution to public health service, enacted in 1979 Income Levy — enacted by the Financial Emergency Measures in the Public Interest Act, 2009 Universal Social Charge (USC) — enacted in 2011 to replace both the Health and Income Levies. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 20 / 51 Data Description Experimental Design A Tale of Three Notches Health Levy — earmarked for contribution to public health service, enacted in 1979 Income Levy — enacted by the Financial Emergency Measures in the Public Interest Act, 2009 Universal Social Charge (USC) — enacted in 2011 to replace both the Health and Income Levies. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 20 / 51 Data Description Experimental Design A Tale of Three Notches Health Levy — earmarked for contribution to public health service, enacted in 1979 Income Levy — enacted by the Financial Emergency Measures in the Public Interest Act, 2009 Universal Social Charge (USC) — enacted in 2011 to replace both the Health and Income Levies. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 20 / 51 Data Description Experimental Design List of Quasi-Experimental Variation Year Health Levy 2005 2006 2007 2008 2009 2010 2011 2012 2013 20,800 22,880 24,960 26,000 26,000 26,000 Hargaden (University of Michigan) Income Levy USC 15,028 15,028 4,004 10,036 10,036 Irish Tax Notches Tax Penalty (416) (458) (499) (520) (867) (1,040) (80) (201) (201) (200) (301) March 1, 2016 21 / 51 Data Description Experimental Design Order of the Game 1 December of Year T − 1: government announces annual budget. 2 January 1 of Year T : tax changes take effect. Wages are paid and 100% of taxes are deducted by employers at source. 3 Early in Year T + 1: employers send tax forms to Revenue detailing total annual pay. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 22 / 51 Bunching Estimates Empirical Design Estimation The literature’s standard estimation technique measures excess mass (“bunching”) near the threshold. 1 Plot histogram of earnings near the threshold. 2 Fit high-order polynomial to these data. 3 Compare actual mass to the predicted mass. Identification requires the level of the income distribution to be flat at the notch threshold. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 23 / 51 Bunching Estimates An Illustrative Example Bunching (Levels) in 2008 200 0 100 Frequency 300 400 Extent of excess bunching b= 7.5 (0.6) 25,490 25,690 25,890 26,090 26,290 26,490 Income 2005 2006 2007 Hargaden (University of Michigan) 2008 2009 2010 Irish Tax Notches 2009 IL 2010 IL 2012 March 1, 2016 2013 24 / 51 Bunching Estimates Round-number problem: spurious bunching 600 500 300 400 Frequency 700 800 Earnings Distribution, 2006-2007 20,000 21,000 22,000 23,000 24,000 25,000 26,000 27,000 28,000 29,000 30,000 Annual Income Clear round-number bunching for multiples of e1,000. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 25 / 51 Bunching Estimates Changes in bunching as an alternative strategy Rather than look at the bunching in levels, we can look at the bunching in differences. Identification is valid if the change in the income distribution is flat at the notch threshold. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 26 / 51 Bunching Estimates Difference Bunching Bunching Clear in Early Years: 2006 Bunching (Differences) in 2006 100 0 50 Frequency 150 200 Extent of excess bunching b=72.3 (21.7) 22,370 22,570 22,770 22,970 23,170 23,370 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 27 / 51 Bunching Estimates Difference Bunching Bunching Clear in Early Years: 2007 Bunching (Differences) in 2007 100 0 50 Frequency 150 200 Extent of excess bunching b=146.5 (24.4) 24,450 24,650 24,850 25,050 25,250 25,450 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 28 / 51 Bunching Estimates Difference Bunching Bunching Clear in Early Years: 2008 Bunching (Differences) in 2008 50 -50 0 Frequency 100 150 Extent of excess bunching b=153.6 (18.8) 25,490 25,690 25,890 26,090 26,290 26,490 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 29 / 51 Bunching Estimates Difference Bunching No Bunching: 2009 Bunching (Differences) in 2009 50 -50 0 Frequency 100 150 Extent of excess bunching b=15.6 (10.2) 14,520 14,720 14,920 15,120 15,320 15,520 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 30 / 51 Bunching Estimates Difference Bunching No Bunching: 2009 Bunching (Differences) in 2009 50 -50 0 Frequency 100 150 Extent of excess bunching b=-20.3 (10.0) 25,490 25,690 25,890 26,090 26,290 26,490 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 31 / 51 Bunching Estimates Difference Bunching No Bunching: 2010 Bunching (Differences) in 2010 50 -50 0 Frequency 100 150 Extent of excess bunching b=-2.0 (9.5) 14,520 14,720 14,920 15,120 15,320 15,520 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 32 / 51 Bunching Estimates Difference Bunching No Bunching: 2012 Bunching (Differences) in 2012 50 -50 0 Frequency 100 150 Extent of excess bunching b= 9.4 (8.3) 9,530 9,730 9,930 10,130 10,330 10,530 Income Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 33 / 51 Declining Responsiveness Salience? Could this be a salience story? One plausible explanation is that people knew about the tax in the initial years and, in the midst of a recession, lost sight of their personal finances. This seems quite unlikely. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 34 / 51 Declining Responsiveness Salience? Could this be a salience story? One plausible explanation is that people knew about the tax in the initial years and, in the midst of a recession, lost sight of their personal finances. This seems quite unlikely. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 34 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Declining Responsiveness Salience? Public attention of USC Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 35 / 51 Determinants of Bunching Who bunches? Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 36 / 51 Determinants of Bunching Who bunches? If it’s not a salience story, what sort of people place themselves in the tax-advantaged portion of the income distribution? To address this, consider ( a dummy variable: 1 if income in dominated region Dominated Region = 0 if income ∈ (Threshold-e100, Threshold] This permits estimating probits on the determinants of tax-reporting behavior. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 37 / 51 Determinants of Bunching Regression Results Self-employed construction workers avoid taxes Table: Selected determinants of reporting tax-disadvantaged income Avg. Marginal Effects Personal Characteristics Firm Characteristics Industry Hargaden (University of Michigan) Number of Employers EU 2004 Fifty-two weeks Self-employment Partnership Private Limited Foreign Branch Public Limited Firm-size Construction Retail Hotel & restaurants Public admin Utilities Irish Tax Notches 1.9??? 2.0?? -3.7??? -16.0??? 3.7?? 5.4??? 10.0??? 10.2??? 0.02??? -3.0??? -2.0?? -3.2?? 3.2?? 4.2 March 1, 2016 38 / 51 Determinants of Bunching Regression Results Discussion of regression results Bunchers share traits. They are younger, are employed primarily in small enterprises, typically for 52 weeks of the year, in certain industries. Perhaps these covariates are changing over time, and this is affecting bunching. What happens when we reweight the pre- and post-recession samples to look like each other? Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 39 / 51 Counterfactual Estimation VAT Notch in Japan Figure: Onji (JPubEc, 2009) Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 40 / 51 Counterfactual Estimation DiNardo, Fortin, Lemieux approach DFL Density Estimates 2006-2007 .00055 Density .0005 .00045 .0004 24,500 25,000 Annual Pay (€) Actual (pre) Actual (post) 25,500 Counterfactual Controls: male age irish emp_size legal_form sector fiftytwo Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 41 / 51 Counterfactual Estimation DiNardo, Fortin, Lemieux approach DFL Density Estimates 2007-2008 .0005 Density .00048 .00046 .00044 .00042 25,500 26,000 Annual Pay (€) Actual (pre) Actual (post) 26,500 Counterfactual Controls: male age irish emp_size legal_form sector fiftytwo Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 42 / 51 Counterfactual Estimation DiNardo, Fortin, Lemieux approach DFL Density Estimates 2008-2009 .00055 Density .0005 .00045 .0004 25,500 26,000 Annual Pay (€) Actual (pre) Actual (post) 26,500 Counterfactual Controls: male age irish emp_size legal_form sector fiftytwo Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 43 / 51 Counterfactual Estimation DiNardo, Fortin, Lemieux approach DFL Density Estimates 2009-2010 .00046 Density .00044 .00042 .0004 .00038 14,500 15,000 Annual Pay (€) Actual (pre) Actual (post) 15,500 Counterfactual Controls: male age irish emp_size legal_form sector fiftytwo Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 44 / 51 Controlling for composition It’s not just composition bias Table: Changes in wage dynamics for job-stayers Year 2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12 2012-13 Proportion Freezes Proportion Cuts Proportion Increases 2.5% 2.5% 2.8% 3.3% 4.4% 6.8% 10.3% 10.0% 17.2% 17.6% 22.9% 52.7% 55.2% 39.3% 34.2% 34.3% 80.4% 79.9% 74.2% 44.0% 40.3% 53.9% 55.5% 55.7% Job-stayers are defined as employees who remain with the same employer over multiple years. Table shows the proportion of job-stayers whose pay was frozen within ±0.1% of previous year’s pay, cut, or increased. Source: Doris et al. (2013) Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 45 / 51 Controlling for composition Inverse Probability Weighting I use weights to match the pre- and post-recession samples One way to disentangle composition bias from additional constraints on behavior is to use inverse probability weighting (IPW). This calculates the determinants of bunching “as if” the composition of the labor force were unchanged from previous years. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 46 / 51 Controlling for composition Inverse Probability Weighting Post-matching magnitudes vary Table: Matched sample determinants Fifty-two weeks Firm Size (’000s) Self-employment Number of employers Irish Male 2006 2010 (match) -3.18? -0.01 -12.66??? 1.03 1.72 -0.27 -7.58??? 0.05?? -21.33??? 2.14 0.50 2.58? Table shows selected mean marginal effects from probit regression on the determinants of bunching. The first column shows the results for the notch at e22,880 in 2006. The second column shows the results for the 2010 notch, but with the results weighted so the covariates match those of the 2006 sample. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 47 / 51 Controlling for composition Inverse Probability Weighting Interpretation of recession in potential outcome framework One interpretation of the results is to consider the recession as a treatment with non-random assignment (e.g. construction). If a recession imposes no additional constraints on behavior, then selection into the treatment should explain the change in outcomes. This is rejected by the data. Regardless of the specification, the change in tax-reporting behavior is not explained by the covariates. People are 4–5pp less likely to respond to these tax incentives post-2008. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 48 / 51 Conclusion Conclusions and contributions This paper has three things really going for it: 1 Great variation: several changes in location and severity of notch I Motivation for the difference bunching strategy 2 Great data: third party-reported, administrative panel 3 Huge business cycle volatility Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 49 / 51 Conclusion Conclusions and contributions Developed a new bunching strategy with more compelling identification than existing approaches. Tested the response to the ‘treatment’ of notch introduction. Contributed to the literature on what covariates are consistent with avoiding taxes. Decomposed the total response into a labor composition effect, and a change in behavior effect. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 50 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 Conclusion Conclusions and contributions Clear, unambiguous results in the initial period: people avoid taxes. That evidence disappears. The employer and employee characteristics are determinants of reporting behavior. Changes in the composition of the labor force can explain much of the changes in reporting. . . . . . but composition alone cannot explain the findings. In a model of cyclical responsiveness, optimal tax rate and enforcement vary over the cycle. People are less able to avoid taxes during recessions: fund the tax authorities in good times. Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 51 / 51 95 Employment (2005 Q1=100) 105 100 110 115 Irish and US Employment Levels 2006 2008 2010 Year Ireland 2012 2014 USA Sources: Central Statistics Office, BLS Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 1 / 12 0 20 Weddings (2005=100) 40 60 80 100 120 Weddings Registered in Ireland 2005 2006 2007 2008 2009 2010 Year 2011 2012 2013 2014 Source: Central Statistics Office Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 2 / 12 Bunching (Levels) in 2005 150 100 0 50 Frequency 200 250 Extent of excess bunching b= 3.8 (1.2) 20,290 20,490 20,690 20,890 21,090 21,290 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 3 / 12 Bunching (Levels) in 2006 150 100 50 Frequency 200 250 Extent of excess bunching b= 1.9 (0.8) 22,370 22,570 22,770 22,970 23,170 23,370 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 4 / 12 Bunching (Levels) in 2007 0 100 Frequency 200 300 Extent of excess bunching b= 3.5 (1.0) 24,450 24,650 24,850 25,050 25,250 25,450 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 5 / 12 Bunching (Levels) in 2008 200 0 100 Frequency 300 400 Extent of excess bunching b= 7.5 (0.6) 25,490 25,690 25,890 26,090 26,290 26,490 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 6 / 12 Bunching (Levels) in 2009 100 0 50 Frequency 150 Extent of excess bunching b= 4.8 (0.9) 14,520 14,720 14,920 15,120 15,320 15,520 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 7 / 12 Bunching (Levels) in 2009 200 0 100 Frequency 300 Extent of excess bunching b= 8.2 (1.0) 25,490 25,690 25,890 26,090 26,290 26,490 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 8 / 12 Bunching (Levels) in 2010 100 0 50 Frequency 150 200 Extent of excess bunching b= 4.4 (0.9) 14,520 14,720 14,920 15,120 15,320 15,520 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 9 / 12 Bunching (Levels) in 2010 200 0 100 Frequency 300 400 Extent of excess bunching b=10.0 (0.9) 25,490 25,690 25,890 26,090 26,290 26,490 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 10 / 12 Bunching (Levels) in 2012 100 0 50 Frequency 150 200 Extent of excess bunching b= 3.3 (3.7) 9,530 9,730 9,930 10,130 10,330 10,530 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 11 / 12 Bunching (Levels) in 2013 100 0 50 Frequency 150 200 Extent of excess bunching b= 3.2 (5.4) 9,530 9,730 9,930 10,130 10,330 10,530 Income Back Hargaden (University of Michigan) Irish Tax Notches March 1, 2016 12 / 12