Does Raising Contribution Limits Lead to More Saving? Evidence from the Catch-up Limit Reform

Transcription

1 Does Raising Contribution Limits Lead to More Saving? Evidence from the Catch-up Limit Reform Adam M. Lavecchia University of Toronto National Tax Association 107 th Annual Conference on Taxation Adam M. Lavecchia (UofT) 1 / 1

2 Research questions Does raising IRA and 401(k) contribution limits increase retirement contributions? If so, for whom? The effectiveness of raising contribution limits has been widely debated with no clear consensus (Venti and Wise, 1990a,1990b; Gale and Scholz, 1994) Do contributions to IRAs and 401(k)s represent new saving? Or are contributions achieved by displacing saving in taxable accounts? Important for understanding how to address the perceived retirement savings shortfall (Poterba, 2014) Mixed findings in previous studies may be due to limitations with research designs (Bernheim, 2002; Chetty et al., 2014) Literature review Adam M. Lavecchia (UofT) 2 / 1

3 This paper I estimate the causal policy effect of raising IRA and 401(k) contribution limits on retirement saving Do these policies only affect agents whose limit is binding? Real-world estimates are scare because statutory limit changes are infrequent Exogenous variation in limits due to eligibility for catch-up limits, an IRA and 401(k) contribution rule that allows individuals who are 50 or older to contribute more than those under the age of 50 Policy variation arguably provides a cleaner and more convincing estimate of the savings effect of these programs in the U.S. Adam M. Lavecchia (UofT) 3 / 1

4 What are catch-up limits? Introduced in 2002 as part of the Economic Growth and Tax Relief Reconciliation Act (EGTRRA) EGTRRA also legislated universal IRA and 401(k) limit increases The previous IRA limit was $2,000 from 1981 to 2001 Allows those who will turn age 50 or older by the end of the calendar year to contribute more to their IRA and 401(k) than those under 50 Catch-up limits are intended to give older workers a way to catch-up on their retirement saving Like regular IRA and 401(k) limits, unused catch-up limits cannot be carried forward Only provision in EGTRRA to affect those over age 50 differentially Adam M. Lavecchia (UofT) 4 / 1

5 (Nominal) IRA contribution limits before and after EGTRRA Year Regular IRA limit Catch-up limit Regular + catch-up limit 2001 $2, $3,000 $500 $3, $3,000 $500 $3, $3,000 $500 $3, $4,000 $500 $4, $4,000 $1,000 $5,000 The 2014 regular IRA limit is $5,500 and the regular + catch-up limit is $6,500. EGTRRA implemented similar increases for 401(k) limits, though initial limits were much higher. Adam M. Lavecchia (UofT) 5 / 1

6 Who benefits from eligibility for catch-up limits? The proposal would have virtually no effect, however, on families and individuals who do not make any deposits to IRAs under the current law or who deposit less than the current $2,000 limit. This proposal would directly benefit only those already making the $2,000 maximum contribution...those at the limit are almost certainly among the most affluent of taxpayers eligible for IRAs. (Orszag and Orszag, Center on Budget and Policy Priorities, 2001) Adam M. Lavecchia (UofT) 6 / 1

7 Who benefits from catch-up limits? Many in the media and financial planning community praised the reform as a way to help procrastinators Descriptive evidence from industry publications suggests that banks 1 viewed catch-up limits as an opportunity to increase assets under management (Ellens, 2004) 2 directed their IRA advertising to those over 50 For example, send a direct mail piece explaining the new catch-up contributions to your IRA owners who are age 50 or older...not only will you educate members about the IRA changes, you will create member awareness of the products and services your credit union offers. (Zuehlke, 2001) Adam M. Lavecchia (UofT) 7 / 1

8 Data Repeated cross-sections from the Survey of Income and Program Participation (SIPP) spanning 2002 and (post-egtrra years) Detailed individual-level information on IRA and 401(k) ownership, account balance and tax-deductible contributions Respondents asked about IRA and 401(k) contributions during tax-preparation season (January-April) for the preceding year Running variable: age it = t YOB i t denotes calendar year and YOB i is i s year of birth reported in the SIPP Eligibility for catch-up limits is a deterministic function of an individual s year of birth Adam M. Lavecchia (UofT) 8 / 1

9 Empirical Strategy R it = α + βover50 it + P φ page p it + P ψ pover50 it age p it + Z it γ + δ t + u it (1) p=1 p=1 R it : IRA or 401(k) contribution (or participation dummy) for individual i in year t over50 it : catch-up limit eligibility indicator α: estimate of the counterfactual level of savings at age 50 β: causal effect of eligibility for catch-up limits RDD identification assumption Counterfactual IRA and 401(k) savings function is continuous in age at age 50 The counterfactual is a non-differential change in contribution limits Adam M. Lavecchia (UofT) 9 / 1

10 Catch-up Limits and IRA Contributions Pre-Reform: β = (7.05) Mean IRA Contributions by Normalized Age: Pre Reform Post-Reform: β = (7.27)*** Mean IRA Contributions by Normalized Age: Post Reform IRA Contributions (1996 $) Age - 50 IRA Contributions (1996 $) Age - 50 Can t reject continuous IRA contributions at age 50 before EGTRRA Estimate for β suggests that eligibility for catch-up limits increased average IRA contributions by 28% Adam M. Lavecchia (UofT) 10 / 1

11 Catch-up Limits and IRA Participation Rates Pre-Reform: β = (0.003) Mean IRA Participation Rate by Normalized Age: Pre Reform Post-Reform: β = (0.004)** Mean IRA Participation Rate by Normalized Age: Post Reform IRA Participation Rate Age - 50 IRA Participation Rate Age - 50 Can t reject continuous IRA participation rates at age 50 before EGTRRA Estimate for β suggests that eligibility for catch-up limits increases the likelihood of making a positive IRA contribution by 25% Adam M. Lavecchia (UofT) 11 / 1

12 What can explain these findings? In the standard life-cycle model, a changing contribution limit will only affect the choices of agents for whom the limit is binding With fixed costs of contributing, some non-contributors may be induced to make a contribution following a limit change New contributions for these agents must be greater than the pre-reform limit otherwise contributing under the old limit would have been optimal However, I find that the contributions of those induced to participate are relatively small (and below the previous $2,000 limit) Those just over 50 who report saving in an IRA for less than one year contribute $1,694 on average A bounding procedure also rules out contributions larger than $2,000 Adam M. Lavecchia (UofT) 12 / 1

13 Conclusion This paper studies the effect of increasing contribution limits on IRA and 401(k) saving Previous studies estimate structural models with simulations I recover a causal estimate of the effect of raising limits in a real-world setting Results suggest that the initial response to eligibility for catch-up limits was not limited to constrained savers Large effects on both IRA contributions and the likelihood of making a deductible contribution Relatively small average contributions for those induced to participate No significant effect on 401(k) contributions Descriptive evidence is consistent with targeted advertising by financial institutions explaining the increase in IRA participation rates This mechanism has long been hypothesized but difficult to test directly (Poterba, Venti and Wise, 1995; Bernheim, 2002) Recent research suggests that advertising may be important for explaining the demand for retail banking products (Gurun et al., 2013; Hastings et al., 2013; Honka et al., 2014) Adam M. Lavecchia (UofT) 13 / 1

14 EXTRA ( BACK-UP ) SLIDES Adam M. Lavecchia (UofT) 14 / 1

15 Literature Review More than 20 years of research on whether (and how) IRAs and 401(k)s increase saving has yielded mixed results. Papers can be grouped into two broad categories: IRA and 401(k) contributions represent new saving: Feenberg and Skinner (1989); Venti and Wise (1990a,1990b); Poterba, Venti and Wise (1995); Gelber (2011) 1 IRA and 401(k) contributions crowd-out taxable saving: Gale and Scholz (1994); Attanasio and DeLeire (2002); Benjamin (2003); Engelhardt and Kumar (2007); Chetty at al., (2014) more recently for IRAs in Denmark Differing results due to: Different research designs and identification assumptions (mostly strong CIAs) The identification assumption of a continuous counterfactual savings function at age 50 is arguably much weaker than most of the previous literature Heterogeneous effects of tax-preferred accounts on total saving Research questions 1 The crowd-out parameters estimated in Gelber (2011) are imprecisely estimated and he concludes that becoming eligible to participate in your firm s 401(k) may slightly increase or decrease total personal saving. Adam M. Lavecchia (UofT) 15 / 1

16 Table: Summary Statistics: Post-Reform (pooled) Saver Type Variable All IRA Owners 401(k) Owners Own IRA Made IRA Contribution IRA Contribution Own 401(k) (k) Contribution , , Bank Saving Other Equity Saving Stock Saving -6, , , Unsecured Debt Change Age Female White Black Hispanic HH Total Income 55, , , Household Income 49, , , Personal Earned Income 22, , , Number of Kids < High School Some College College Degree N 104,021 20,140 26,182 Notes: All dollar amounts are deflated to 1996 dollars using the Bureau of Labor Statistics CPI Inflation Calculator. The sample in column 1 is respondents from the 2001 and 2004 SIPP panels between the ages of 18 and 65. The sample in column 2 is restricted to IRA owners. The Bank Saving, Other Equity Saving, Stock Saving and Unsecured Debt Change variables for any year t are imputed by subtracting the reported assets for that variable at year t - 1 from the balance at year t.. Adam M. Lavecchia (UofT) 16 / 1

17 Figure: Distribution of IRA Contributions (a) Pre-Reform Ages (b) Pre-Reform Ages Frequency IRA Contributions (nominal $) Frequency IRA Contributions (nominal $) (c) Post-Reform Ages (d) Post-Reform Ages Frequency Frequency IRA Contributions (nominal $) IRA Contributions (nominal $) Adam M. Lavecchia (UofT) 17 / 1

18 Regression RDD Results: IRAs Table: IRA Results (1) (2) (3) (4) Contributions Participation Rates Full Sample: N = 51,725 Control Mean (se) (8.18)*** (13.78)*** (0.005)*** (0.006)*** over50 it (se) (7.27)*** (6.95)*** (0.004)*** (0.004)** Covariates N Y N Y IRA Owners: N = 12,261 Control Mean (se) (33.57)*** (103.54)*** (0.019)*** (0.045)*** over50 it (se) (28.99)** (27.83)** (0.017)* (0.017)* Covariates N Y N Y Positive Contributors: N = 2,501 Control Mean 2,019 1, (se) (61.20)*** (271.42)*** (0.000)*** (0.000)*** over50 it (se) (75.33) (75.88) (0.000) (0.000) Covariates N Y N Y Standard errors are clustered at the age level.*** p < 0.01,** p < 0.05, * p < All columns present estimates for linear spline regressions with a bandwidth of 10 and year FE. The covariates are: female, white, black, hispanic, married, personal earned income, number of kids under age 18 and 3 education dummies. Adam M. Lavecchia (UofT) 18 / 1

19 Robustness 1 Functional form assumptions: Main results use a polynomial of degree 1 Results robust to polynomial of degree 2 Visual inspection of the data and goodness of fit tests suggests that a polynomial of degree 1 is appropriate 2 Varying the age window : Main results restrict sample to year olds ( bandwidth of 10) Results robust to varying the age window from to year olds Results robust to using the IK or CCT optimal bandwidths 3 Pooling pre- and post-reform data and using a differences-in-differences design 4 Placebo cutoffs 5 Standard RDD validity tests 6 Various estimates for standard errors (cluster at HH level) Adam M. Lavecchia (UofT) 19 / 1

20 Robustness: Bandwidth specification (a) IRA Contributions (b) IRA Participation Rates Estimted Treatment Effect (1996 $) Estimated Treatment Effect (percentage points) Bandwidth Bandwidth 95% Confidence Interval Point Estimate CI Point Estimate Adam M. Lavecchia (UofT) 20 / 1

21 Robustness: Placebo Cutoffs (a) IRA Contributions (b) IRA Participation Rates Frequency Age 50 Estimated Treatment Effect: $ Frequency Age 50 Treatment Effect: Estimated Treatment Effect (1996$) Estimated Treatment Effect (percentage points) Adam M. Lavecchia (UofT) 21 / 1

22 Catch-up Limits and 401(k) Contributions Pre-Reform: β = (41.37)* Post-Reform: β = 5.29 (57.41) 401k Contributions (1996 $) Mean 401k Contributions by Normalized Age: Pre Reform Age k Contributions (1996 $) Mean 401k Contributions by Normalized Age: Post Reform Age - 50 Can t reject continuous 401(k) contributions at age 50 during the pre-reform years Estimate for β suggests that eligibility for catch-up limits has no effect on 401(k) contributions Adam M. Lavecchia (UofT) 22 / 1

23 Catch-up Limits and 401(k) Participation Rates Pre-Reform: β = (0.011) 401k Participation Rate by Age: Pre Reform Post-Reform: β = (0.013) 401k Participation Rate by Age: Post Reform 401k Participation Rate k Participation Rate Age Age - 50 Can t reject continuous 401(k) participation rates at age 50 during the pre-reform years Estimate for β suggests that eligibility for catch-up limits has no effect on the likelihood of making a positive 401(k) contribution Adam M. Lavecchia (UofT) 23 / 1

24 Regression RDD Results: 401(k)s Table: Post Reform: 401(k) Contributions (1) (2) (3) (4) Contributions Participation Rates Full Sample: N = 29,365 Control Mean (se) (47.49)*** (83.88)*** (0.011)*** (0.020)** over50 it (se) (56.78) (46.14) (0.013) (0.010)* Covariates N Y N Y Positive Contributors: N = 9,037 Control Mean 3, (se) (81.89)*** (254.39)*** (0.000)*** (0.000)*** over50 it (se) (81.66)** (80.05) (0.000) (0.000) Covariates N Y N Y Standard errors are clustered at the age level.*** p < 0.01,** p < 0.05, * p < All columns present estimates for linear spline regressions with a bandwidth of 10 and year FE. The covariates are: female, white, black, hispanic, married, personal earned income, number of kids under age 18 and 3 education dummies. Adam M. Lavecchia (UofT) 24 / 1

25 Did increases in IRA contributions lead to new saving? To estimate whether increases in IRA contributions (due to eligibility for catch-up limits ) are achieved by crowding-out taxable savings I estimate N it = μ 0 + μ 1 R it + P p=1 N it : non-ira (non-401(k)) taxable savings for individual i in year t μ 2p age p it + Z it μ 3 + e it (2) Taxable savings variables: bank saving, stock/mutual fund saving, unsecured debt, car saving R it : IRA contribution for individual i in year t R it instrumented with over50 it, the catch-up limit eligibility indicator Z it : vector of baseline covariates μ 1 : crowd-out parameter of interest; can be interpreted as a LATE IV Identification Assumptions over50 it : (i) predicts R it and (ii) is uncorrelated with e it If the RDD identification assumption holds and RDD estimates are significantly different from zero then (i) and (ii) hold Adam M. Lavecchia (UofT) 25 / 1

26 Crowd-out Estimates (IRA Owners) Table: Crowd-out Estimates: IRA Owners (1) (2) (3) (4) Bank Stocks U. Debt Cars OLS Estimates: N = 6,822 μ (se) (0.005) (0.003)** (0.002) (0.002) 2SLS Estimates: N = 6,822 μ (se) (0.337) (0.508) (0.242) (0.110) First-Stage F-statistic: 1.38 Standard errors are clustered at the age level.*** p < 0.01,** p < 0.05, * p < All columns present IV regression results for IRA owners between 40 and 59 years of age and year FE. The level of IRA contributions is instrumented with an catch-up limit provision eligibility indicator. Adam M. Lavecchia (UofT) 26 / 1