The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
TAX POLICY, LUMP-SUM PENSION DISTRIBUTIONS, AND HOUSEHOLD SAVING by Angela E. Chang Federal Reserve Bank of New York Research Paper No. 9507 April 1995 This paper is being circulated for purposes of discussion and comment only. The contents should be regarded as preliminary and not for citation or quotation without permission of the author. The views expressed are those of the author and do not necessarily reflect those of the Federal Reserve Bank of New York or the Federal Reserve System. Single copies are available on request to: Public Information Department Federal Reserve Bank of New York New York, NY 10045 Abstrac t As of May 1988, over 8 million workers had received their pension benefits as lump-sum distribut ions (LSDs) when they changed jobs. In 1986 Congress imposed a 10% tax penalty on the amount of LSDs not rolled over into tax-defer red instrumen ts. This paper examines the effects of this tax penalty on the rollover decisions of LSD recipient s. The penalty increases the probabili ty of rollover among higher-inc ome recipient s; an increase of 1 percentag e point in the penalty is estimated to increase the probabili ty of rollover by 1.1 percentag e point. However, the. penalty has not affected the rollover decisions of lower-inco me recipient s, who are more likely to be liquidity- constrain ed. I would like to thank Peter Diamond, Jonathan Gruber, James Poterba, and participa nts at the Harvard-M IT Public Economics Seminar for helpful comments. I would also like to thank.Dan iel Feenberg for tabulatio ns from the U.S. Treasury Individua l Tax Model File (1984-89) . Financial support from the National Science Foundatio n, Bradley Foundatio n, and MIT World Economy Laborator y while at MIT is gratefull y acknowled ged. 1. Introduct ion As of May 1988, over 8 million workers had received a total of $42 billion (1988 dollars) in pre-retire ment lump-sum distribut ions (LSDs) from their pension plans when they changed jobs. Most of these workers did not roll over their LSDs into tax-defer red instrumen ts (e.g., IRAs) which are close substitut es of pension plans. Instead, they saved their LSDs in non tax-defer red instrumen ts, such as savings accounts, or spent them (Fernande z 1992). Congress is concerned about the low rollover rate, because the spending of LSDs may reduce workers' accumulat ed retiremen t savings. Also, the fact that most LSD recipient s choose not to save their LSDs for retiremen t raises the question of whether the recipient s ever intended to save their pension money for retiremen t (Fernande z 1992). pension plans as tax shelters. Instead, they may have used the To encourage the rollover of LSDs, Congress introduce d in the Tax Reform Act of 1986 a 10% tax penalty on the amount of LSDs not rolled over into tax-defer red instrumen ts, like the 10% penalty on premature withdrawa ls from Individua l Retiremen t Accounts (Joint Committee on Taxation 1987). This paper examines the impact of the 10% tax penalty on workers' decisions to roll over their LSDs. 1 Past studies of individua ls' use of LSDs have not examined the impact of the tax penalty on rollovers (Andrews 1991, Piacentin i 1990). Estimates of the tax sensitivi ty of workers' rollover decisions may provide insight into whether a penalty on consumpti on actually 1This paper does not examine the rollovers of lump-sum distribut ions received after age 59½, which are called normal or retiremen t distribut ions. In 1990, pre-retire ment LSDs (i.e., received before the age 59½) accounted for about 60% of all LSDs and 38% of the total amount of LSDs. Normal distribut ions accounted for about 25% of all LSDs and 34% of the total amount of LSDs in 1990 (Yakobosk i ·1994). 2 increase s saving. This study may also contribu te to the empiric al literatu re on individu als' saving decision . The paper proceeds as follows. Section 2 provides backgrou nd informa tion about when workers receive LSDs and how the tax treatmen t differs accordin g to the particu lar use of LSDs. It also details changes in the tax treatmen t of LSDs that occurred in 1986. Section 3 describe s the May 1988 Current Populat ion Survey (CPS), which is the primary dataset. Section 4 describe s the econome tric framewo rk, data limitati ons, and econome tric issues. The results are presente d in Section 5. The results indicate that the tax penalty signific antly increase s the probabi lity of rollover among higherincome recipien ts. In contras t, the tax penalty does not signific antly affect the rollove r decision s of lower-in come recipien ts. Since lower-in come recipien ts comprise the majority of LSD recipien ts, the tax penalty has raised conside rably more tax revenue than Congress anticipa ted. tax penalty raised over $1.9 billion in tax revenue; 2 $547 million (U.S. Joint Tax Committe e 1987). 3 During 1987-89 , the Congress anticipa ted Section 6 summari zes the main findings and conclude s the paper with a discussi on of the policy implica tions of the results. 2. Backgrou nd Workers can receive their pension benefit s as LSDs when they change jobs 2 This figure is an estimate from the U.S. Treasury Individu al Tax Model dataset for 1984-89 and from the Interna l Revenue Service Individu al Income Tax Returns for 1986-88. 3 The revenue potenti al of the tax penalty may have motivate d a recent change in the adminis tration of the tax penalty . As of January 1993, workers who choose to receive their LSDs are subject to a 20% withhold ing tax [IRS Taxpaye r Service s, Janssen (1992), Schultz (1992)]. This withhold ing tax was included in the July 1992 bill that extended unemploy ment benefit s; it was expected to pay for part of the cost of the extended unemploy ment benefit s. 3 under two circum stance s. First, certai n pensio n plans give worke rs the option to take their pensio n benef its as LSDs upon separa tion. Worke rs in such plans can exerci se the LSD option and thereb y receiv e LSDs when they change jobs. Define d contri bution (DG) pensio n plans typica lly give worke rs the LSD option ; where as, define d benef it (DB) pensio n plans typica lly do not. 4 DB plans typica lly pay worke rs their pensio n benef its at retirem ent, even to those who leave the emplo yers before retirem ent. Second , worke rs may receiv e LSDs as a result of unilat eral cashou ts by their emplo yers. The Employ ee Retire ment Income Secur ity Act of 1974 (ERISA) allowe d emplo yers to cash out employ ees with small accrue d benef its unilat erally . In other words, emplo yers can give LSDs to .worke rs upon separa tion regard less of whethe r the worke rs want to receiv e them or not, provid ed that the amount of accrue d benef its is less than the legal maximum unilat eral cashou t limit. The curren t maximum unilat eral cashou t limit is $3,500 (Ferna ndez 1992). Worke rs can alloca te their LSDs in three ways: (1) leave the money with the firms they are leavin g for future withdr awal, (2) receiv e checks for the amoun t of the LSDs and save or spend as they please , or (3) arrang e to have the money rolled over into their new emplo yers' DG pensio n plans. 5 Taxes may affect indivi duals' choice s among these option s. The tax treatm ent of LSDs has change d over the years. Prior to 1987, indivi duals paid income taxes on the amoun t of LSDs that they did not invest in tax-de ferred instru ments such as Indivi dual Retire ment Accou nts (IRAs) , About 10% of DB plans and betwee n 80-90% of DG plans give worke rs the option of taking LSDs upon separa tion (Ferna ndez 1992). 4 5Few pensio n plans offer worke rs the option of transf erring money from former emplo yers' pensio n plans. 4 insurance annuities, or their new employers' pension plans. LSDs in such instruments is called rollovers. Investment of Not investing LSDs in tax- deferred instruments will be referred to as non-rollovers. The Tax Reform Act of 1986 (TRA) changed the tax price of non-rollovers in two ways. First, it changed the structure of federal personal income taxes, generally lowering the marginal tax rates 6 and reducing the number of tax brackets. Second, it imposed a 10% penalty in addition to income taxes on the amount of LSDs not rolled over within 60 days of receipt (Joint Committee on Taxation 1987). Certain workers are exempt from the 10% penalty on non-rollovers of LSDs. For example, workers who receive LSDs because they are totally or permanently disabled are not subject to the penalty. Workers who receive LSDs due to medical expenses which are deductible are also not subject to the penalty. For the purposes of this paper, the exception of interest is that workers who receive LSDs during or after the year they turn 55 years of age are not subject to the 10% penalty (Joint Committee on Taxation 1987). Thus, the penalty raised the tax price of non-rollovers for only younger recipients (under the age of 55). The following discussion will focus on married recipients filing joint returns. Single recipients experienced similar changes in the tax price of non-rollovers. Figure 1 illustrates the marginal tax price of non-rollovers for married recipients before and after TRA and across age groups (younger and older than 55),. Figure 1 shows that in general older recipients faced a lower 6 For example, in 1986 the top marginal tax rate for married taxpayers filing joint returns was 50%; in 1987, 38.5%; and in 1988, 33% (Coopers and Lybrand 1986, Statistical Abstract of the United States 1982-90 editions). 5 margin al tax price of non-ro llover s in 1987 than in 1986. For older recipi ents, the change in the struct ure of the federa l income tax rates was the sole source of variat ion in the tax price of non-ro llover s before and after 1986, since they were not subjec t to the 10% penalt y. Like older recipi ents, younge r recipi ents faced a margi nal tax price of non-ro llover s equal to the federa l income tax rates in 1986. After 1986, young er recipi ents were subjec t to the 10% penalt y, such that their margin al tax price of non-ro llover s equale d the federa l income tax rates plus 10 percen tage points . For most levels of taxabl e income , the net effect of lower margin al tax rates and the 10% penalt y has been an increa se in the margin al tax price of non-ro llover s for younge r recipi ents (Figur e 1). In fact, the margin al tax price of non-ro llover s increa sed more for young er recipi ents with lower taxabl e income than for those at the high end of taxabl e income . 3. Data The analys is in this paper is based prima rily on data from the May 1988 Curren t Popula tion Survey (CPS), which includ es a supple ment on employ ee benef its. The supple ment questi ons were asked of adults in one-h alf of the basic CPS sample who were employ ed for pay at the time of the interv iew. Out of 109,19 2 adults in the basic CPS sample 27,701 worke rs answe red the supple ment questi ons. Like the basic CPS data, the supple ment data provid e weigh ts that allow adjust ing the sample to be repres entati ve of the U.S. popul ation. The supple ment includ es a set of questi ons about the receip t of LSDs from pensio n plans on prior jobs. about their most recent LSDs. LSD recipi ents answe red questi ons only They were asked to report the year they receiv ed their most rece~ t LSDs and the amoun t. 6 About 8% of the supplement sample, or 2,162 workers, reported having received at least one LSD from a prior job. This translates into a weighted figure of approximately 8.5 million workers who received LSDs. a total of approximately $42 billion. They received This total is in 1988 dollars, as are all dollar figures in subsequent tables. The distribution of LSD recipiency is not uniform across age or family income (Table 1). Over half of the recipients were under the age of 35 when they received their most recent LSDs. About 45% of individuals who received LSDs had a family income less than $30,000 in 1987. Since younger and lower- income workers are more likely to be liquidity-constrained, Table 1 suggests that liquidity constraints may affect the rollover decisions of most LSD recipients. The last two columns of Table 1 indicate that the LSDs of younger recipients were small. Recipients under 35 comprised 61% of the recipients yet received only 37% of the total amount of LSDs, a disproportionately low percentage of the total amount. Similarly, individuals with family income less than $30,000 in 1987 received a disproportionately low fraction of the total amount of LSDs. The distribution of the size of LSDs also is not uniform across age or family income (Table 2). The average amount of LSDs rises with the age at the time of receipt and with the recipients' 1987 family income. average LSD was $5,989, the median was $2,451. While the This discrepancy between the average and the median LSD is due to the fact that most LSDs were small; about 40% of the LSDs were less than $1,000. The dataset contains yes/no responses to whether individuals allocated any of their LSDs to various uses. An overwhelming majority of recipients 7 (86.5%) did not roll over any of their LSDs into tax-deferred instruments (Table 3). In fact, approximately 41% of the recipients did not save any of their LSDs. 7 Weighted by dollars, roughly three-fourths of the total amount of LSDs were not allocated to tax-deferred instruments.• Table 3 shows that about 16% of the recipients allocated at least some of their LSDs toward investments such as starting or buying a business, buying a house, 9 or paying educational expenses. About 7% of the recipients reported using some of their LSDs to pay expenses incurred during unemployment. Lastly, slightly over a quarter of the recipients used some of their LSDs for uses not specified in the CPS questionnaire. Table 3 also shows that the percentage of recipients who save at least part of their LSDs in non tax-deferred instruments (e.g., savings accounts) is substantially higher than the percentage who rolled over their LSDs. Approximately 14% of the recipients invested at least some of their LSDs in tax-deferred instruments; whereas, 24% invested at least some of their LSDs in non tax-deferred instruments. In terms of dollars, the percentage of the 7 I defined saving as allocating at least part of an LSD to financial instruments or paying off debt. 8 The May 1988 CPS supplement does not provide the exact amount of LSDs allocated to various uses. A range of the amount of rollovers was inferred based on the number of uses reported. If a recipient reported more than one use for the LSD, then he/she was assumed to have allocated a minimum of $1 and a maximum of the entire amount to each reported use. Summing across these minimum and maximum figures across recipients produces a range of the total amount rolled over. The median value of the range of rollovers is $9.4 billion or about 22% of the total amount of LSDs. 9 Buying a house can be considered a form of saving for retirement, since individuals can use the value of the house to obtain financial resources via instruments such as home equity loans and reverse mortgages; however, Venti and Wise (1989) show that the elderly are reluctant to reduce housing equity. This suggests that financing consumption during retirement may not be the main motivation for buying a house. 8 total amount allocat ed to non tax-def erred instrum ents is around 23%, which is about the same as the percent age allocat ed to tax-de ferred instrum ents. 4. Econom etric Framework and Issues 4.1. Econom etric Framework The estima tion procedu re is based on the probit model. compel led the use of a discret e depend ent variab le. Data limitat ions The ideal depend ent variab le is the amount of the LSD rolled over into tax-de ferred instrum ents, such that the effect of the penalty on the amount rolled over can be estimat ed. The May 1988 CPS does not provide inform ation on the amount rolled over but yes/no respons es to whethe r any of the LSD was allocat ed to particu lar uses. The datase t allows inferen ce about whethe r recipie nts rolled over none, some, or all of their LSDs. Within the contex t of the probit model, the observe d discret e choice of recipie nt i to roll over at least some of his/he r LSD is assumed to be the outcome of an unobse rvable proces s: if S-= [ 1 i 0 if s/>O s/~O (1) where Si denotes the observe d rollove r decisio n of the recipie nt; and si", the unobse rvable desired amount of LSD rollove r. Si has been defined as equal to 1 if a recipie nts rolls over at least some of his/he r LSD and O otherw ise. 10 Based on the probit model, the followi ng equatio n is estima ted: si can also be defined as equal to 1 for recipie nts who roll over all of their LSD and O otherw ise. The results using this measure of Si are simila r to those presen ted in the paper. I also estima ted ordered probit equatio ns, where Si equals 2 for rollove r of the entire LSD, 1 for partia l rollove r, and O for no rollove r. The simila rity in the results is not surpris ing, ·since 79% of recipie nts who rolled over any of their LSDs actuall y rolled over their entire LSDs. 10 9 (2) wher e~(.) is the cumu lative distr ibuti on funct ion for a stand ard norm al distr ibuti on; and x 1 , a vecto r of obser vable char acter istic s of the recip ient and of the LSD. The follo wing facto rs may influ ence the rollo ver decis ion and have been inclu ded in the vecto r x 1 : (1) the amount of the LSD in 1988 dolla rs, (2) the feder al income tax rates (MTRs), (3) the 10% tax pena lty, 11 (4) age at the time of recei pt of the LSD, (5) famil y incom e, (6) years of schoo ling comp leted , (7) gend er, and (8) mari tal statu s. The means and stand ard devia tions of the varia bles used in the prob it estim ation are in Table 4. I allow ed the MTRs and the tax pena lty to have separ ate effec ts on the prob abili ty of rollo ver, becau se their effec t on rollo vers may diffe r accor ding to the incom e of the recip ients . For exam ple, Figur e 1 shows that there are fewer tax brack ets at the highe r end of taxab le income than at the lower end. This impl ies that the 10% pena lty is the major sourc e of varia tion in the marg inal tax price of non- rollo vers for highe r-inco me recip ients . This is espe ciall y true for the recip ients in the top tax brack et, for whom the 10% pena lty is the only sourc e of varia tion in the marg inal tax price of nonrollo vers. For lower -inco me recip ients , both the MTRs and the pena lty gene rate varia tions in the marg inal tax price of non- rollo vers. Chang es in the tax price of non- rollo vers impos ed by the Tax Refor m Act of 1986 (TRA) make poss ible the iden tific ation of the effec t of MTRs on rollo ver decis ions. Beca use the MTRs are a nonli near funct ion of famil y incom e, witho ut the 10% pena lty the effec t of MTRs on rollo vers canno t be 1 1The tax pena lty varia ble equa ls 0.1 for youn ger recip ients after 1986 and O for all other recip ients . 10 identifi ed from the effect of family income on rollover s. 12 The impositi on of the tax penalty on younger recipien ts provides a natural experim ent, whereby recipien ts with similar charact eristics faced differen t tax prices dependin g on their age and whether they received their LSDs before or after TRA. The older recipien ts serve as the "control " group, since they were not subject to the 10% penalty. The younger recipien ts were subject to the 10% penalty after 1986 and thus are the "treatme nt" group . 13 Differe nce-in-d ifferenc e probit equation s can be used to test whether the penalty had any effect on the "treatme nt" group relative to the "control " group. Rollove rs should have increase d more among younger recipien ts than among older recipien ts after 1986, all else being equal. The sample for the probit equation s has been limited to workers who received LSDs between 1984-88. Not includin g years earlier than 1984 is an attempt to control for aggrega te conditio ns. Ideally, the sample would include the two years before and two years after 1986, the year the Tax Reform Act of 1986 was passed. Given that the CPS was taken in May 1988, the sample includes data for only about l½ years after the passage of TRA. 4.2. Data Limitati ons The advantag e of using the May 1988 CPS dataset is that it provides detailed informa tion about pre-reti rement LSDs for a represe ntative sample. From the CPS, workers who were subject to the 10% tax penalty can be accurate ly identifi ed. However , the CPS does present several data limitati ons for the computin g the federal income tax rates. 12 Feenberg 13 For (1992). The most severe limitati on is (1987) gives a clear discussi on of this identifi cation issue. a discussi on of differen ce-in-di fference estimati on, see: Gruber 11 ·the availabil ity of family income only for 1987. To estimate the recipient s' family income and their federal marginal tax rates (hencefor th, MTRs) at the time of receipt of the LSDs, I assumed that real family income at the time of receipt was highly correlate d with family income in 1987. This assumptio n will be too strong if most workers receive their LSDs as a result of being laid off and experienc e large unexpecte d drops in income in the year they received their LSDs. Tabulatio ns from the CPS data suggest that only about 10% of workers who received LSDs as a result of layoffs in 1987. 14 Furthermo re, only 7% of recipient s used their LSDs to pay expenses while unemploye d (Table 3). These tabulatio ns suggest that most workers do not experienc e large unexpecte d drops in their income when they receive their LSDs. Other considera tions pertain to the validity of the assumptio n. First, MTRs apply to brackets of income; therefore , the MTRs computed from the true income in the year of LSD receipt may not differ much from the MTRs computed from income in 1987. Second, the analysis is limited to the period 1984-1988 . Thus, the differenc e in time between the time of LSD receipt and 1987 is at most three years. These considera tions suggest that the assumptio n may not be too unreasona ble. The second data limitatio n is the availabil ity of the 1987 family income in brackets. To compute MTRs, I assumed the recipient s had 1987 family income equal to the median value of the income brackets. In other words, if a 14 This figure is based on workers who received their LSDs in 1987 or 1988. For this group of workers, the May 1988 CPS provides enough informati on to infer whether they received their LSDs as a result of quits or layoffs. Results from probit equations estimated for this group indicate that whether a worker was laid off in 1987 did not have a significa nt effect on the rollover decision. 12 recipient reported having a family income in 1987 between $30,000-3 4,999, he/she was assumed to have had family income equal to $32,500 in 1987. Other assumptio ns were required to compute MTRs. First, the marital status of the recipient s as reported in 1987 was assumed to be the same in the year they received the LSDs. joint returns. Second, married recipient s were assumed to file Third, recipient s were assumed to take the standard deduction s and not to itemize deduction s. Lastly, single recipient s were assumed to take one personal exemption ; and married recipient s, two personal exemption s. The number of dependent exemption s was based on the assumptio n that the number of children in the year of LSD receipt was the same as the number of children reported in the May 1988 CPS. 15 4.3. Self-Sele ction Bias Using the CPS data to estimate the effect of the tax penalty on the rollover of LSDs raises the issue of self-sele ction bias. A self-sele ction bias may be present, because the sample consists of only workers who received LSDs and does not include all workers who had the LSD option. The total effect of the tax penalty on rollovers consists of two componen ts: (1) the effect on the decision to exercise the LSD option and (2) the effect on the decision to roll over an LSD that has been received. With a sample of LSD recipient s, the second effect is estimated without taking into account the first effect; this will produce a self-sele ction bias. Because individua l-level data are not available with which to separate workers with and without the LSD option (i.e., belong to pension plans that allow 15Results from probit equations with and without personal and dependent exemption s are not substanti vely different . 13 LSDs) , 16 econom etric "correc tions" for self-se lection bias (e.g., Heckma n's two- step proced ure) cannot be utilize d. 17 Althoug h I cannot correc t for the self-se lection bias, the directi on of the self-se lection bias makes the estima ted effect of taxes in Section 5 more plausib le. If the tax penalty induced many workers to leave their pension money with their former employ ers, then the estimat ed effect of the tax penalty on the rollove r decisio ns of worker s who chose to receive LSDs will be lower than the true effect for all worker s who had the LSD option. In other words, since leaving the money with the employ ers is equiva lent to rollove rs, my sample underr epresen ts the number of rollove rs due to the penalty . Thus, if a self-se lection bias is presen t, then the estima ted effects of the MTRs and the tax penalty in Section 5 are lower bounds on their true effects for ali' workers eligibl e for LSDs. 5. Results 5.1 Aggreg ate Eviden ce Aggreg ate data indicat e that the 10% tax penalty has not increas ed rollove rs to the extent that Congre ss expecte d. revenue than Congre ss anticip ated. Hence, it has raised more tax Congre ss anticip ated raising $547 million in revenue from the tax penalty during 1987-8 9. The next table shows that the 16A Gallup survey conduc ted during Februa ry-June 1991 provide s the only availab le figures on the number. of worker s who left their pension money with their former employ ers. It indica tes that only 17% of individ uals between the ages 18 and 54 who had the LSD option chose not to exercis e the option. The sample consist ed of 327 individ uals. David Wray at the Profit Sharing Coun~i l kindly provide d the survey results . 17 For a descrip tion of Heckma n's two-ste p proced ure, see Chapte rs 8 and 9 in Maddal a (1983). 14 tax penalty actuall y raised over $1.9 billion in revenue during 1987-89 8 .1 Tax Revenu e from the Tax Penalty Year Number Paid Penalty (thousa nd) Collec ted Revenue (millio n) Anticip ated Revenu e (millio n) 1987 876 $364 $97 1988 1,524 711 209 1989 1,653 833 241 Source: IRS U.S. Treasu ry Individ ual Tax Model File and SOI publica tions (numbe r paid penalty , collect ed revenu e); U.S. Joint Tax Commit tee, p. 719 (antici pated revenue ) The number of recipie nts who paid the 10% LSD penalty and the collect ed revenue from the penalty are estima ted from the U.S. Treasu ry Individ ual Tax Model and SOI public ations. The IRS aggreg ates figures for the 10% penalty on premat ure IRA withdra wals with those for the 10% penalty on non-ro llover of LSDs. To estima te the figures for the 10% LSD penalty , I assumed that for 1987-89 the number of taxpay ers who paid the 10% IRA penalty and the revenue from the IRA penalty were the same as in 1986. 19 Tabula tions from the May 1988 CPS also sugges t that the impact of the penalty on the overal l group of LSD recipie nts has been small. The followi ng 18Figures from the May 1988 CPS are similar to those in the table. Based on the CPS data, about 926,000 LSD recipie nts paid an estima ted $360 million in tax penalty in 1987. Figures for 1988 cannot be tabulat ed from the CPS, since it does not cover the entire year of 1988. 19How this assump tion biases the figures for the 10% LSD penalty is ambigu ous. If the upward trend in the revenue from the IRA penalty during 1984-86 continu ed during 1987-89 or if more taxpay ers made prematu re IRA withdra wals in 1987-89 to make advanta ge of the lower tax rates, then my estima tes of the 1987-89 revenue from the LSD penalty are too high. On the other hand, IRA partici pation decline d dramat ically after 1986 (Engen et al. 1994), such that the number of premat ure IRA withdra wals may have decline d after 1986·. This would sugges t my estima tes of the revenue from the LSD penalty are too low. 15 table shows that rollover s have increase d over time; thus, the increase in rollove rs after 1986 may be due to a time trend or changes in the aggrega te conditio ns of the U.S. economy and not necessa rily due to the tax penalty . Rollove r by Year of Receipt Year Percenta ge of Recipien ts Before 1980 Percenta ge of Amount 4.0% 7.6% 1980-84 14.9 29.0 1985-86 17.5 23.8 1987-88 20.6 30.7 Source: May 1988 CPS A cross tabulati on by year of receipt and age group is more meaning ful, since only younger recipien ts (under 55) became subject to the tax penalty after 1986. The table below shows that rollover s increase d among younger recipien ts after 1986 but remained fairly steady among older recipien ts. Rollove r by Year of Receipt and Age Group Year Percenta ge of Younger Recipie nts Before 1980 3.6% Percenta ge of Older Recipie nts 27.0% 1980-84 13.3 44. 7 1985-86 15.9 28.0 1987-88 20.3 27.4 Source: May 1988 CPS The increase in rollove rs among younger recipien ts after 1986 but not among older recipien ts suggests that the tax penalty imposed on younger recipien ts may have increase d rollove rs. Since there has been an upward trend in rollove rs among younger recipien ts througho ut the 1980s, the increase in 16 .rollo vers after 1986 canno t be attrib uted solely to the tax penalt y. A time trend or change s in the demog raphic chara cteris tics of the recipi ents may explai n the increa se in rollov ers. 5.2 Probi t Resul ts A natura l extens ion of the cross tabula tions above is the differ ence- indiffer ence probit equati ons that includ e demog raphic variab les but not the tax variab les. They provid e insigh t into whethe r differ ent trends for rollov ers are observ ed for young er and older recipi ents, withou t relyin g on tax variab les which may be noisy (e.g., MTRs as compu ted from the CPS). Result s from probi t equati ons that includ e the tax variab les are presen ted later in this sectio n. The depend ent variab le in the differ ence-i n-diff erence equati ons equals 1 for recipi ents who rolled over at least some of their LSDs and O for recipi ents who did not roll over any of their LSDs. The indepe ndent variab les of intere st are as follow s: • LESS55: 0 for recipi ents 55 or older, 1 for recipi ents under 55; • POST86: 0 for LSD receip t before 1986, 1 for after 1986; and • (LESS55)*(POST86): an intera ction term of the two variab les. The differ ence- in-dif ferenc e probit equati ons take the follow ing form: PI(S1 =1)= ~[p 0 +P 1 (LESS55)+P 2 (POST86)+P 3 (LESS 55)*(P OST86 )+y 1z 1] (3) where z; is a vector of demog raphic variab les and the real amount of the LSD. The hypoth esis that the tax change s affect ed only younge r recipi ents after 1986 transl ates into a test that the coeffi cient P >0. 3 This test is based on the assum ption that there are no unobse rvable variab les that 17 increase d rollover s among younger recipien ts but not among older recipien ts (i.e., age-spe cific time trends). The estimate d coeffici ents for the interact ion term (LESS55)*(POST86) are positive but not statisti cally signific ant (Table 5). The P-values for the one-side d test (Ho: P3-0, H1 : p 3>0) reflect the statisti cal insignif icance of the coeffici ent Pa- The P-values are 0.20 for both equation s; the null hypothe sis cannot be rejected at the convent ional 5% level. A measure similar to a tax elastici ty can be computed from the differen ce-in-d ifferenc e probit results as follows: (4) where M denotes the change in the predicte d probabi lity of rollover between 1984-86 and 1987-88; and ~t, the change in the margina l tax price of nonrollove rs. The supersc ripts denote the age groups: Y for younger recipien ts and O for older recipien ts. Several assumpt ions were made to compute the predicte d probab ilities for the "average " younger and older recipien ts. married males with 14 years of schoolin g. The recipien ts were assumed to be For younger recipien ts, the real amount of LSD was assumed to equal $6,000; and for older recipien ts, $16,000 . The real amount of LSD and the persona l charact eristics were chosen to resemble the means for the two age groups. Based on these predicte d probabi lities, the estimate d "tax elastici ty" is 0.8 percenta ge point. In other words, a 1 percenta ge point increase in the margina l tax price of non-roll overs for younger recipien ts relative to older recipien ts raises the probabi lity of rollove r by 0.8 percenta ge point more among younger recipien ts relative to older recipien ts, control ling for 18 demogra phic variable s. This estimate of the "tax elastici ty" is similar to those from the probit equation s with the tax penalty variable , which are presente d below. The estimate s from the probit equation s that include the tax variabl es are in Table 6. The depende nt variable in these equation s equals 1 for recipien ts who rolled over at least part of their LSDs into tax-defe rred instrume nts and O for recipien ts who did not roll over any of their LSDs. Table 6 shows that MTRs do not signific antly affect the probabi lity of rollover . income. One explana tion may be the high correlat ion between MTRs and family Multico llineari ty would also explain why MTRs are signific ant in probit equation s that do not include family income. In contras t, the coeffici ent for the tax penalty fs statisti cally signific ant and positive . This indicate s that younger recipien ts who are subject to the tax penalty are more likely to roll over their LSDs. Table 7 shows that the estimate d margina l effect of the tax penalty on the probabi lity of rollover is 0.6 percenta ge point. In other words, increasi ng the tax penalty by 1 percenta ge point would raise the probabi lity of rollover by 0.6 percenta ge point. This implies a rough approxim ation that the impositi on of the tax penalty (i.e., an increase in the penalty from Oto 10%) raised the probabi lity of rollove r by 6 percenta ge points. 20 This is not a negligib le effect, since the probabi lity of rollover among younger recipien ts prior to the impositi on of the tax penalty was only about 16%. The probabi lity of rollover also increase s with the real amount of LSD, 20 This is a rough approxim ation, because the effect of the tax penalty appears to be nonline ar; its effect diminish es as the tax penalty increase s. 19 ·age, and family income (Table 6). 21 These findings challenge the predictions of the standard Life-Cycle Hypothesis. The standard Life-Cycle Hypothesis predicts that taxes, the real amount of LSDs, and the age at the time of receipt should have little effect on the probability of rollover, so long as the receipt of an LSD is basically a change in the timing of income receipt. On the other hand, the findings are consistent with an extended version of the Life-Cycle Hypothesis, in which consumption decisions are made under uncertainty about future income. Given uncertainty about future income, workers may prefer to keep some liquid assets. Since tax-deferred instruments such as IRAs are less liquid than non-tax deferred instruments, 22 as uncertainty increases workers may prefer non tax-deferred instruments. In particular, if job separations increase uncertainty about future income, then workers may prefer not to roll over their LSDs until they are more certain of their future prospects. Within this framework, if workers desire a fixed amount of liquid assets as protection against unexpected drops in income, then they would be more likely to roll over larger LSDs; they would roll over the portion of the LSDs that exceeds the desired amount of liquid assets. If younger workers and lower-income workers face more uncertainty about their future income or are less likely to have accumulated the desired amount of liquid assets, then the probability of rollover would increase with age and family income. The results are also consistent with two other models of saving 21 In the probit equations, the base age group is 35-44 and the base income group is $30,000-39,999. 22 Engen et al. (1994) analyze how individuals may not view tax-deferred and non tax-deferred instruments as close substitutes and may want to invest in both. 20 behavior. The first is the model of saving under liquidity constrain ts. Liquidity constrain ts are especiall y relevant within the context of preretiremen t LSDs, since most recipient s are younger and lower-inco me workers (Table l) who are more likely to be liquidity -constrai ned. Liquidity constrain ts may thus explain the increase in the probabili ty of rollover with the age and family income. Liquidity constrain ts may also explain the positive relations hip between LSD size and the probabili ty of rollovers . Larger LSDs are more likely to exceed the desired level of consumpti on of liquidity -constrai ned recipient s; liquidity -constrai ned recipient s would roll over the amount of the LSDs that exceeds what is necessary to attain the desired level of consumpti on. The second is the behaviora l model of saving that emphasize s individua ls' categoriz ing bundles of money into three different "mental accounts" : (1) current income, (2) assets, and (3) future income. 23 The model assumes that individua ls have the lowest marginal propensit y to save bundles of money that they have categoriz ed as current income, and the highest marginal propensit y to save money categoriz ed as future income. This behaviora l model provides an explanati on for the positive effect of the amount of LSD and age on the probabili ty of rollover. Recipient s are more likely to classify larger LSDs as assets or future income than as current income; this would explain why larger LSDs are more likely to be rolled over. Similarly , older recipient s may be more concerned about retiremen t income and more likely to classify LSDs as assets or future income than younger recipient s. 23 For The behaviora l model, however, does not provide a good a detailed discussio n of behaviora l models of saving behavior, see: Shefrin and Thaler (1988) and Thaler (1991). 21 explanation for the positive relationship between family income and the probability of rollover. To explore further the differential effect of the tax penalty by family income, I estimated separate probit equations for lower-income and higherincome recipients. Lower-income recipients are defined as those with less than $40,000 in family income in 1987. Higher-income recipients are defined as those with 1987 family income greater than $30,000. Table 7 presents the estimated marginal effect of the tax penalty for the two income groups. The marginal effect of the tax penalty is larger for higher-income recipients than for lower-income recipients. An increase of 1 percentage point in the tax penalty raises the probability of rollover among the higher-income group by 1.1 percentage point. In contrast, the marginal effect for the lower-income group is 0.4 percentage point. This contrast is noteworthy, since the tax penalty raised the marginal tax price of nonrollovers more for lower-income recipients than for higher-income recipients. The results cast doubt on the idea of including one tax price variable that equals the sum of MTRs and the tax penalty. The results show that the MTRs do not have a statistically effect on the probability of rollover. In contrast, the coefficient for the tax penalty is significantly positive. Nonetheless, I have done likelihood ratio tests to determine if the coefficients on the MTRs and the tax penalty are equal (i.e., the MTRs and the tax penalty can be summed into one variable). The results from the likelihood ratio tests indicate that for the whole group of LSD recipients and for the lower-income group, the null hypothesis that the coefficients are equal cannot be rejected at the conventional 5% level. Since most LSD recipients are in the lower-income group, it is not 22 .surpris ing that the results from the likeliho od ratio tests for the whole group and for the lower-in come group lead to the same conclusi on. On the other hand, the null hypothe sis is rejected at the 1% level for the higherincome group. The MTRs and the tax penalty cannot be summed for the higher- income group. The results from the likeliho od ratio tests are consiste nt with the discussi on in Section 4.1. The 10% penalty is the major source of variatio n in the margina l tax price of non-roll overs for higher-i ncome recipien ts. In contras t, for lower-in come recipien ts, both the MTRs and the penalty generate variatio ns in the margina l tax price of non-roll overs. Thus, the effect of the tax penalty differs from the effect of the MTRs more among the higherincome group than among the lower-in come group. The presenc e of a self-sel ection bias, as discusse d in Section 4.3, would imply that the estimate s in Tables 6 and 7 underes timate the true effect of the tax penalty on the entire group of workers who had the LSD option. While more precise estimate s may be possible in the future with better data, the policy interes t in the rollove r decision s of LSD recipien ts warrant s the present endeavo r to estimate the effect of the tax penalty. 6. Conclus ion Congres s has often attempte d to raise persona l saving with tax penaltie s on consump tion. They imposed a 10% tax penalty on early withdraw als from IRAs since 1982, when IRAs became availab le to all workers (Ozanne 1992). In 1986, Congres s imposed a 10% tax penalty on the amount of lump-sum distribu tions (LSDs) not rolled over into tax-defe rred instrume nts. This paper presents the first estimate s of the effect of the 10% tax penalty and federal income taxes on individu als' use of LSDs. Aggrega te data 23 show that rollover s increase d after the tax penalty went into effect but not to the degree expected by Congres s. Consequ ently, the penalty has raised more tax revenue than Congress anticipa ted. An analysis of individu al-level data from the May 1988 CPS indicate s that the tax penalty signific antly increase s the probabi lity of rollove r among higher-i ncome recipien ts (i.e., recipien ts with income above $30,000 ). A 1 percenta ge point increase in the tax penalty produces a 1.1 percenta ge point increase in the probabi lity of rollover among higher-i ncome recipien ts. In contras t, the tax penalty does not signific antly affect the rollove r decision s of lower-in come recipien ts. One explana tion for the differen tial impact of the tax penalty by income is that the lower-in come recipien ts are more likely to be liquidit yconstrai ned; therefor e, their rollover decision s are less sensitiv e to tax concern s. The importan ce of liquidit y constra ints in the rollove r decision s of LSD recipien ts cannot be examined with the currentl y availab le datasets . Once adequate data become availab le, an examina tion of the importan ce of liquidit y constra ints in the rollove r decision may provide importa nt insights into the tax sensitiv ity of rollove rs. As the trend continue s toward defined contribu tion pension plans, more workers will receive LSDs. Most of the recipien ts are young, lower-in come workers who use their LSDs to finance current consump tion. Thus, for most recipien ts, the receipt of LSDs is associat ed with a reductio n in their accumul ated retireme nt savings. The combina tion of increasi ng prevalen ce of LSDs, the spending of most LSDs, and the insignif icant effect of the tax penalty on the rollove r decision s of lower-in come recipien ts suggests that the non-rol lover of LSDs will continue to be a policy concern in the near future. 24 Figure 1. (I) (.) ·c: a. Marginal Tax Price of Non-Rollovers: Younger and Older Married Recipients 0.6 X ~ <ti 0.5 1987: Younger C: ·e <ti ~ 0.4 1986 1987: Older 0.3 0.2 0.1 0.0 ..._____ ___,___________._...._ ...._.... .,_......_..... ,_ __.__ 0 20 40 60 80 100 120 140 160 180 200 Taxable Income (Thousands) Sourc e: Coope rs & Lybra nd (1986 )[p.2, 166], ACIR (1986 )[p.22 ], ACIR (1987 )[p.20 ]. 25 Table 1. Distrib ution of Recipie ncy of Lump-Sum Distrib utions Charac teristic s Total Age at time of receip t: Under 35 35-44 45-54 55+ 1987 Family Income: Less than $10,000 $10,000 -19,999 $20,000 -29,999 $30,000 -39,999 $40,000 -49,999 $50,000 -74,999 $75,000 + Recipie nt-Wei ghted thousan ds percen tage Dollar- Weight ed billion s percen tage 8,478 100% $42.0 100% 5,201 2,042 850 385 61.4 37.4 24.1 10.0 4.5 15.7 15.3 6.5 4.5 344 1,610 1,710 2,042 1,002 682 818 4.2 19.6 20.8 24.9 12.2 8.3 10.0 1.8 6.6 6.6 9.9 4.9 3.8 7.6 36.4 15.5 10.7 4.3 15.7 15. 7 23.6 11.7 9.0 18.l All figures have been compute d using weight s provide d in the May 1988 CPS supplem ent. The sample consis ts of 2,162 worker s who reporte d receivi ng at least one pre-ret iremen t lump-sum distrib ution from a prior job. 26 Table 2. Characteristics Distribution of Size of Lump-Sum Distributions Average Amount of Lump-Sum Distribution $1$1,000 $3,000 $5,000 $10,000 999 -2,999 -4,999 -9,999 -19,999 $20,000+ Age at time of receipt: Under 35 35-44 45-54 55+ $3,673 8,546 9,789 15,449 46.4% 29.6 35.3 30.6 25.4% 19.1 16.1 14.1 10.7% 11. 7 11.4 10.5 10.4% 17.0 18.8 13.3 5.2% 13.5 8.8 12.9 1. 9% 9.1 9.5 18.6 1987 Family Income: Less than $10,000 $10,000-19,999 $20,000-29,999 $30,000-39,999 $40,000-49,999 $50,000-74,999 $75,000+ $6,543 4,836 4,751 5,534 6,059 6,786 11,739 45.0% 44.5 43.5 38.5 37.8 32.9 32.2 21. 7% 22.9 24.9 24.9 19.5 19.6 19.0 8.2% 12.7 9.3 12.2 9.5 16.5 8.5 10.2% 8.5 12.4 12.9 17.3 18.0 16.7 8.7% 2.4 6.5 7.8 12.2 6.2 8.8 6.2% 4.0 3.5 3.9 3.7 $5,989· 2,451b 40.5% 2.2 22.4% 8.3 11.0% 8.6 13.0% 10.9 7.9% Recipient-Weighted Dollar-Weighted •Mean amount of lump-sum distributions received bMedian amount of lump-sum distributions received 21. 7 6.9 14.8 5.2% 40.6 27 Table 3. Uses Uses of Lump-Sum Distribut ions Recipient -Weighted thousands percentag e Tax-defer red instrumen ts: Retiremen t programs 1,142 Insurance annuities Savings accounts 198 1,478 Other financial instrumen ts 944 532 13 .5% Dollar-We ighted billions percentag e 11.1 2.3 17.4 6.3 $9.4 7.7 1. 7 5.1 4.5 22.3% 18.4 4.1 12.0 10.6 2.5 4.8 1.4 4.9 1.6 5.9 11.5 3.4 11. 7 3.7 1.8 7.7 4.3 18.4 Start or buy a business Buy a house Buy a car Pay off debt Pay education al expenses Pay expenses during 233 756 310 1,843 355 2.7 8.9 3.7 21. 7 4.2 unemploym ent Other 552 2,250 6.5 26.5 All figures have been computed using weights provided in the May 1988 CPS supplemen t. The sample consists of 2,162 workers who reported having received at least one pre-retire ment lump-sum distribut ion from a prior job. The May 1988 CPS supplemen t does not provide the exact amount of lump-sum distribut ions allocated to various uses; therefore , only a range of the amount allocated to each use can be inferred. If a recipient reported more than one use, then he/she was assumed to have allocated a minimum of $1 and a maximum of the entire amount of the lump-sum distribut ion to each reported use. Summing across these minimum and maximum figures across recipient s yields a range of total amount allocated to each use. The figures in the fourth and fifth columns are the median values of these ranges. 28 Table 4. Means: By Age Group and Time Period Variables Younger Recipient s Older Recipient s 1984-86 1987-88 1984-86 1987-88 % rolled over any of lump-sum 16.0 [36.7] 19.2 [39.5] 35.7 (48.5] % 12.8 (33.4] 15.6 [ 36. 3 J 28.6 [45.7] 27.3 [ 45. 6 J 13.6 [35.1] 5,916 [10,301] 5,792 [11,542] 16,373 [20,669] 11,785 [11,910] 0.30 [0.10] 0.34 [0.09] 0.29 [0.12] 0.27 [0.10] 34.8 (8.3] 34.5 [7.9] 59.2 [ 3. 3 J 60.1 (4.4] 35,396 [20,350] 30,609 [17,649] 28,748 [18,176] 31,350 [20,507] 14.2 [ 2. 4 J 14.2 [2.3] 13.2 [3 .2J 13.4 (3.1] distribut ion (LSD) rolled over all of LSD Amount of LSD in 1988 dollars Marginal tax price of non-rollo vers Age at time of LSD receipt Family income in 1987 Years of schooling completed % female 46.9 47.0 (SO.OJ [SO.OJ 23.8 [43.1] 45.5 (51.0] % married 79.2 [40.6] 67.3 (47.0] 88.1 [32.8] 68.2 (47.7] % owned home in 1987 73.1 (44.4] 66.2 (47.4] 90.5 (29.7] 81. 8 [39.5] 17.6 (38.1] 19.7 (39.9] 35.7 (48.5] 27.3 [45.6] 501 385 42 22 % contribut ed to IRA in 1987 Number of observati ons The figures above are unweighte d. The standard deviation s are listed in brackets. The marginal tax price of non-rollo vers incorpora tes the 10% penalty that was imposed on younger recipient s after 1986. The sample consists of 950 workers who received pre-retire ment lump-sum distributi ons between 1984-88. 29 Table 5. Differen ce-in-D ifferenc e Probit Results Explana tory Variable s Real amount of LSD If less than 55 years of age: LESS55 If received LSD after 1986: POST86 (Ps)(LESS55)*(POST86) Years of schoolin g Coeffic ients 1. 7x10- 5 •• 1. 6x10- 5•• (4. Oxl0- 6 ) (4.0xl0- 6 ) -0.46** -0.53''* (0.24) (0.24) -0.18 -0.17 (0.38) (0.38) 0.34 0.33 (0.40) (0.40) --- 0. 05** (0.02) If female -- - 0.02 (0.10) If married - -- 0.04 (0.12) Constan t • -0. 67** -1.37** (0.23) (0.39) -Log likeliho od 391.45 388.51 Test: Ps>O [p values] 0.34 0.33 (0.20] [o. 20 J Statisti cally signific ant at the 10% level ** Statist ically signific ant at the 5% level The depende nt variable is binary and represen ts whether a recipie nt rolled over any of his/her lump-sum distribu tion (LSD) into tax-defe rred instrum ents. Standard errors are in parenth eses. The sample size is 855. 30 Table 6. Probit Results Exnlanato""'' Variables Real amount of LSD Federal income tax rates (MTRs) Tax penalty Age when received LSD: Under 35 45-54 55+ Family income in 1987: Less than $10,000 $10,000-19,999 $20,000-29,999 $40,000-49,999 $50,000-74,999 $75,000+ Years of schooling If female If married Constant - Loo: likelihood I Coefficients 9. 5x10-s• (4. 9xl0- 6 ) 0. 77 (0.89) 2.34* (1. 25) I -0. 23* (0.12) 0. 34•• (0.16) 0. 51 •• (0.22) -0.14 (0.32) -0. 37•• (0.17) -0.11 (0.15) 0.05 (0.19) 0.03 (0.24) 0.20 (0.22) 0.03 (0.02) 0.02 (0 .11) -0.04 (0.14) -1. 60·· /0.49) 373.48 • Statistically significant at the 10% level •• Statistically significant at the 5% level The dependent variable denotes whether a recipient rolled over any of his/her LSD. Standard errors are in parentheses. The sample size is 855. 31 Table 7. Marginal Effect of.the Tax Penalty on Non-Rollovers: By Income Group Sample Full Sam12le: coefficient for tax penalty ll(prob)/ll(p) Estimates 2. 34* (1. 25) 0.6% Lower Income: a coefficient for tax penalty fl (prob)/ /l(p) 1. 74 (1. 67) 0.4% Higher Income:• coefficient for tax penalty 1.1% fl(prob)/ll(p) • •• a 3. 74•• (1.61) Statistically significant at the 10% level Statistically significant at the 5% level The base family income bracket is $30,000-39,999. Recipients in this bracket are in both income groups. ~ This table summarizes estimates from three probit equations. The standard errors are in parentheses. The MTRs in these probits are the federal income tax rates. The probit equations also include the following variables: real amount of LSD, family income brackets, marital status, years of schooling completed, gender, and age brackets. The estimates for ll(prob)/ll(p) represent the change in the probability of rollover per 1 percentage point change tax penalty.