How To Find Out How A Person'S Consumption Is Affected By A Social Security Card

Transcription

1 Te Journal of Risk and Uncertainty, 25:1; 5 20, 2002 c 2002 Kluwer Academic Publisers. Manufactured in Te Neterlands. Explicit Versus Implicit Income Insurance THOMAS J. KNIESNER Center for Policy Researc, Syracuse University, Syracuse, NY , USA TKniesne@Maxwell.Syr.Edu JAMES P. ZILIAK Department of Economics, University of Kentucky, Lexington, KY , USA Abstract By supplementing income explicitly troug payments or implicitly troug taxes collected, income-based taxes and transfers make disposable income less variable. Because disposable income determines consumption, policies tat smoot disposable income also create welfare improving consumption insurance. Wit data from te Panel Study of Income Dynamics we find tat annual consumption variation is reduced by almost 20 percent due to explicit and implicit income smooting. Consumption insurance is as important economically as private ealt or automobile insurance. Altoug taxes ave become an increasingly important source of consumption insurance, te 2001 income-tax reform legislation sould ave little effect on implicit consumption insurance. Keywords: consumption, implicit insurance, income taxes, transfer payments, PSID JEL Category: H21 Everyone agrees tat social insurance is an important mecanism used to stabilize income, and in turn consumption. In practice, te most important governmentally provided income insurance may not be explicit social insurance programs but rater income taxes. Wit income-based taxation wen before-tax income falls te ouseold s tax burden also falls so tat after-tax spendable income drops by less tan te drop in pre-tax income. We examine bot implicit and explicit income insurance and find tat te amount of implicit income insurance occurring troug te structure of income taxes is actually comparable to or of greater magnitude tan te muc-explored explicit social insurance. Income insurance is important because a central goal of economic policy is to stabilize ouseold consumption in te presence of adverse economic events. Wen tere is an economy-wide income sock, suc as a recession, te Federal Reserve or Congress implement counter-cyclical monetary or fiscal policy to support te employment and incomes of ouseolds generally. Wen an income sock is idiosyncratic to te ouseold, peraps due to poor ealt of a primary earner, a cange in family structure, or a job loss caused by an industry or occupation sakeout, relevant public policy is te system of explicit and implicit income insurance programs. Te most well known examples of explicit income insurance are Social Security (OASI), Unemployment Insurance (UI), Medicare, Workers To wom correspondence sould be addressed. Ester Gray and Ann Wicks ceerfully and skillfully elped wit manuscript preparation. Marta Bonney provided er usual expert referencing elp.

2 6 KNIESNER AND ZILIAK Compensation Insurance (WC), and means-tested transfers suc as te Food Stamp Program, Medicaid, Supplemental Security Income (SSI), and Temporary Assistance to Needy Families (TANF). Economically more subtle is te income insurance embedded in te federal and state income tax system, including te payroll tax (FICA) and te Earned Income Tax Credit (EITC). Te main focus of our researc is to assess empirically te relative contributions of explicit versus implicit income insurance in reducing consumption volatility in te United States. It is instructive to establis some basic relative magnitudes of explicit and implicit insurance. Table 1 presents expenditures on social insurance programs, means-tested transfer payments, and income-based taxes and tax credits for 1979, 1989, and Social security and Medicare smoot consumption during retirement, and UI, DI, and WC buffer income and consumption during te working life. Overall real expenditures on social insurance almost doubled in te last 20 years due largely to notable growt in social security and disability insurance coupled wit uge increases in Medicare expenditures. Adding to enanced income and consumption stabilization since 1979 are real outlays on means-tested transfers, wic increased nearly 140 percent. Transfer payment increases in te United States are from expansions in Medicaid, SSI, and ousing assistance programs. 1 Altoug social insurance as a fraction of real GDP remained rougly constant over time, means-tested transfers as a fraction of real GDP increased by 30 percent. Real growt in individual income tax collections during is similar to te real growt in social insurance and te sare of real GDP paid as individual income taxes also remained relatively constant. Because of te relative growt in te payroll and state income taxes and a six-fold increase in te EITC, wic reduced tax collections, te sare of federal income tax receipts in total tax collections as fallen since State and local income-based taxation as become relatively more important as implicit income insurance. Table 2 igligts some of te economically significant canges to tax parameters in te federal income tax code, te payroll tax, and te EITC. Te Economic Recovery Tax Act of 1981 (ERTA) and te Tax Reform Act of 1986 (TRA86) broadened te tax base and reduced te number of federal income tax brackets from 16 to four. Te marginal tax rate on te igest income earners dropped from 70 percent in 1979 to 28 percent in 1989 and ten rose to 39.6 percent following te Omnibus Budget Reconciliation Act of Altoug te tax reforms of te 1980s removed several million ouseolds from te federal tax rolls, substantial expansions in te payroll tax base caused a sift in tax burdens from income to payroll. Te fraction of families wit relatively iger payroll tax burdens increased from 44 percent in 1979 to nearly 67 percent in 1999 (Mitrusi and Poterba, 2000). Troug creating iger pase-in (subsidy) rates, iger income cutoffs, and differential benefits based on te ouseold s number of qualifying cildren te Omnibus Budget Reconciliation Acts of 1990 and 1993 increased EITC generosity. A major part of te secular cange in source of ouseold tax liability in te United States is te expansion in te EITC during te 1990s. 2 Access to te explicit insurance described in Table 1 is restricted based on age, ealt status, income status, asset status, or industry (weter employment is covered by UI). Because access to te federal and state income tax code is automatic it migt be te case tat income-based taxation is a readier cannel of income insurance and subsequent consumption stability for many ouseolds. Te implicit insurance tat income taxes provide is peraps

3 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 7 Table 1. Canges in selected sources of income insurance ( ) (billions of $1999) Social insurance (Percent of real GDP) (6.9) (7.3) (7.4) OASI Medicare Unemployment insurance Workers compensation a Disability insurance Means-tested transfers (Percent of real GDP) (2.3) (2.3) (3.0) Medicaid Supplement security AFDC/TANF Food stamps Housing assistance Individual income taxes (Percent of real GDP) (17.4) (17.4) (17.9) Federal State Payroll (FICA) Earned income tax credit Notes: Data on social insurance, means-tested transfers, and te EITC are from Scolz and Levine (2002). Data on federal income taxes, payroll taxes, and real GDP are from te 2001 Economic Report of te President. Data on state income taxes are from te Department of Commerce s State Government Tax Collections in 1979, 1989, and a Due to missing data, te 1980 value appears for 1979; Statistical Abstract of te United States, 1998, p enanced furter by te substantial canges in te tax code over te past two decades. In particular, Kniesner and Ziliak (2002) sow tat a married couple wit median income wo suffer a 30 percent income loss experience just over a 20 percent consumption drop during te late 1980s but only a 14 percent consumption drop during te late 1990s. Te reason for te increased implicit insurance recently can be seen in Table 2, were we see tat te median family not only faces a iger payroll tax rate in te late 1990s but also faces a relatively steep pase-out rate of 21 percent in te EITC. We use data on U.S. ouseolds from te Panel Study of Income Dynamics (PSID) for to examine te impact of explicit and implicit insurance on income and consumption volatility. Based on te work of Kniesner and Ziliak (2002) we specify a model were consumption volatility depends on estimated consumption function parameters, income variances, and covariances among gross income, transfer income, and tax payments. Our project first examines te connection between income and consumption volatility across

4 8 KNIESNER AND ZILIAK Table 2. Canges in selected federal tax parameters, a MTR (Percents) Range ($1,000) MTR (Percents) Range ($1,000) MTR (Percents) Range ($1,000) Income tax b Payroll tax c (Range in $1,000) 1989 (Range in $1,000) 1999 (Range in $1,000) Pase-in rate Pase-out rate Pase-in rate Pase-out rate Pase-in rate Pase-out rate Earned income tax credit d (0 5.0) ( ) (0 6.5) ( ) (0 4.53) ( ) (0 6.8) ( ) (0 9.54) ( ) a Data on te federal income tax and payroll tax parameters are from te Commerce Clearing House U.S. Master Tax Guide for 1980 and 1990, and from te 1999 Forms and Publications link on te IRS WebPages <ttp:// pubs/formpub99.tml>. Data on te EITC parameters are from Ventry (2000). b Federal income tax rates and ranges are for a married couple filing jointly. Te Tax Reform Act of 1986 altered te definition of taxable income to eliminate te so-called zero bracket amount. c Beginning in 1991 separate bases applied to te retirement and ealt-insurance portions of te payroll tax. Te tax base is labor market earnings. d Beginning in 1991, separate EITC parameters applied to families wit one qualifying cild versus more tan one qualifying cild. Beginning in 1994 te EITC extends to families wit no dependents. Te tax base is labor market earnings.

5 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 9 families generally. We ten consider low versus moderate versus ig average income families to identify te part of te long-term income distribution most affected by explicit versus implicit insurance. We find tat explicit insurance reduces overall average consumption volatility by 8.5 percent wile income taxes reduce consumption volatility by an additional 10 percent. Implicit consumpiton insurance increases as one moves up te income distribution, and it increased across te board by te early 1990s compared to te early 1980s. Tax reforms enacted in 2001 sould do little to cange consumption insurance. We calculate tat te consumption insurance present in te U.S. system of taxes and transfers is as economically important as automobile insurance or private ealt insurance. 1. Disposable income insurance and consumption volatility Te workorse of te recent empirical consumption insurance literature is te Euler equation for te relationsip among canges in per person consumption (c/n), disposable income (y d ), and aggregate resources as metered by total consumption (C), ln ( c t / n t ) = α ln(ct ) + β ln ( y dt) + ε t, (1) were indexes ouseolds. 3 Government policy stabilizes disposable income and consumption troug two avenues because disposable income is ydt = yt (1 + g t τt ), were gt = Gt /y t is te average transfer rate and τt = Tt /yt is te average tax rate. Because disposable income depends on gross income (yt ) and te average tax and transfer rates, wic also depend on gross income, te effect of canges in gross income on disposable income and consumption is dampened by coincident canges in te average and marginal tax rates and te transfer rates. Not always fully appreciated is tat even a flat-rate income tax stabilizes consumption. To obtain some additional intuition consider te case of α = 0 = εt, wic nets out any effects of group insurance and random socks or measurement errors. 4 In te simple case of no extra-family income effects te variance of consumption growt is Var ( ln ( c t / n t )) = β 2 Var ( ln y dt). (2) Given an estimate of β, Eq. (2) fleses out ow te income tax and transfer systems reduce te variability of consumption canges once we substitute for te connection between te variation in disposable income and policy. Taking te natural log of disposable income and differencing transforms (2) into Var ( ln ( c t / )) n t = β 2 Var ( ln yt + ln ( 1 + gt τt )). (3) Because te log of 1 plus and minus two small numbers is approximately te difference in te two small numbers we can rewrite te variance decomposition in (3) as Var ( ln ( c t / )) n t β 2 Var ( ln yt + gt τt ). (4)

6 10 KNIESNER AND ZILIAK Te complete expression for te variance of consumption growt in ligt of te components of disposable income and teir variances and covariances is ten Var ( ln ( ct / )) n t β 2 { Var ( ln yt ) ( ) ( ) + Var g t + Var τ t + 2Cov ( ln yt ) (, g t 2Cov ln y t, τt ) 2Cov ( gt )}, τ t. (5) To implement te decomposition in (5) we use estimates of ˆβ from Kniesner and Ziliak (2002) along wit te squared residuals from Mincer (1974)-type equations for gross income, average transfer rates, and average tax rates as estimates of Var( ln yt ), Var( g t ), and Var( τt ).5 For example, te dependent variable in te first Mincer-type regression equation is te proportionate cange in gross income from year to year and te independent variables are a quadratic in age, family size, self-employment, union membersip, state unemployment rate, and year dummies. A fixed effect sweeps out all time-invariant regressors, suc as race and education. Te dependent variables in te two oter Mincer-type equations are te annual canges in te average tax rate and te annual canges in te average transfer rate. Our measures of uncertainty are ten computed as follows. We save te N(T 1) 1 vectors of residuals from te tree Mincer equations, square te residuals, and ten calculate te average squared residual for eac year, wic tracks income, tax-rate, and transfer-rate uncertainty over time. 6 We construct te needed covariances directly using per-period residuals from te tree regressions for canges in gross income and te tax and transfer rates just described. Te covariance of income and te tax rate is Cov ( ln yt ) 1, τ t = H H ln yt =1 τ t ln y t τ t, for example. We identify te importance of implicit versus explicit programmatic insurance of consumption by examining te decomposition in (5) wit and witout taxes or transfers. 2. Data Our data are from te Panel Study of Income Dynamics (PSID) for te interview years of , wic is wen te PSID as its most accurate ouseold tax information. Te primary attraction of te PSID for our researc is its detailed information on ouseold income and composition and te lengt and timing of te panel, wic encompasses several reforms concerning te taxes and transfers tat provide income and consumption insurance in te United States. Altoug te PSID contains information for years before 1980 and after 1991, our sample begins in 1980 because it is wen te PSID started including its most accurate, computer-generated, income tax data, and our sample ends in 1991 because it is wen te PSID stopped collecting tax information. Because we ave no reason to suspect a connection between consumption expenditure dynamics and participation in te PSID our data are an unbalanced panel were we treat missing person years as statistically ignorable (exogenous). Persons included in our sample

7 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 11 are ouseold eads tat (1) are at least 25 years old in 1980 and less tan 65 years old in 1991, (2) finised wit scool by 1980, (3) were not permanently disabled or institutionalized, and (4) did not cange marital status during te sample period. 7 To downplay te influence of outliers we omit person years wit over a 300 percent increase in disposable income or more tan a 75 percent decrease in income from te previous year. As a final data cleaning procedure we only consider a person year wit disposable income of at least $1, Te selection criteria we use create a sample of 12,341 person years for 1,298 ouseolds. Te focal variables in te variance decomposition model in Eq. (5) are gross income, average tax rates, and average transfer rates. Gross income is labor earnings plus income from rent, interest, and dividends. Transfers include social insurance (Social Security, SSI, AFDC, food stamps, and veteran s benefits) as well as private transfers (cild support, alimony, and gifts from relatives). Because te main sources of transfers are income conditioned, transfer payments are one of two main components of insurance supporting disposable income and in turn consumption. Te oter component of consumption insurance, income taxation, needs some discussion about its calculation and accuracy. Until 1992 te PSID used te ouseold s exemptions, based on reported information on dependents, and te PSID used deductions, based on te larger of te standard deduction and typical itemized deductions for te family of interest, to calculate te ouseolds federal taxes and tax rates. Te panel also contains an estimate of te family s potential Earned Income Tax Credit, wic we incorporate. Finally, we also include FICA taxes and te relevant state income tax payments, wic for tractability we take as a proportional tax on income wit te tax rate determined by te average income tax rate in te state (State Government Tax Collections, Annual, Tax Years). 3. Results We present our empirical results in two stages. First we examine te relative importance of unexpected canges in disposable income variability across ouseolds and over time. Across ouseolds we study te uncertainty component of incomes by permanent income groupings and note ow income uncertainty is affected by taxes versus transfers across income groups. Over time we study ow te various tax reforms ave affected te amount of income uncertainty by smooting disposable income. Te second step of our empirical researc is to use Eq. (5) to connect te degree of income variability to consumption variability, wic is a negative component of ouseold economic well being. As we saw in te last section, if consumption is simply proportional to disposable income ten consumption variability is just a fractional multiple of disposable income variability. Te situation is more complex if te effect of disposable income on consumption varies over time or across income groups, wic we also consider Estimated income uncertainty To maximize understanding and te robustness of our inferences we examine income in a sequence: (1) labor and capital incomes, (2) labor and capital incomes plus transfers,

8 12 KNIESNER AND ZILIAK (3) labor and capital incomes less taxes, and (4) labor and capital incomes plus transfers less taxes. Te calculations in (2) and (3) inform us about te relative reduction in income volatility due to taxes alone and transfers alone, wile te calculation in (4) also incorporates a covariance between taxes and transfers. Recall tat transfers include social insurance as well as private transfers, wile taxes include federal income taxes, FICA, average state income taxes, and Earned Income Tax Credits received. Netting out predictable canges in components of disposable income is te next step in examining ow income uncertainty varies across ouseolds inclusive of taxes and transfers. Based on Eq. (5), te top line in Figure 1 plots te annual average squared residual measure of income uncertainty in te absence of taxes and transfers. Te line connecting te points denoted by s sows te reduction in income variability wen taxes are netted out, te line connecting te points denoted by s adds in transfer payments to gross income, and te bottom line sows te full reduction in variability due to transfers and taxes. Figure 1 sows tat overall taxes and transfers reduce disposable income variability by rougly te same amounts (about 14 percent eac). 9 Also apparent is tat te policy canges of te 1980s led to a greater income variability over ouseolds in te United States tat seems to be reversing in te 1990s. Figures 2 4 are similar to Figure 1 but wit te extra feature tat te data are organized by te ouseold s location in te distribution of permanent income. To elaborate, we first compute te 12-year average income for eac ouseold, call tat permanent income, order permanent incomes, and find median permanent income. We label as low income te ouseolds wit permanent incomes less tan 50 percent of te median, label as moderate income te ouseolds wose permanent incomes range percent of te median, and label as ig income tose ouseolds wit permanent incomes greater tan 150 percent of te median. Figure 1. Income uncertainty for all families, Labor & capital income, plus transfers, less taxes, and plus transfers less taxes.

9 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 13 Figure 2. Income uncertainty for low-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. Figure 3. Income uncertainty for moderate-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. Low and ig permanent income families ave patterns distinct from te average family and distinct from eac oter. For low-income ouseolds te temporal pattern is similar to te overall picture, but transfers are a more important disposable income stabilizer tan taxes. Te situation is distinctly different for ig-income families, wo get virtually no transfer payments so tat teir disposable incomes are stabilized mainly by income-based taxes. Tax rate reductions after TRA86 sarply increased te disposable income uncertainty

10 14 KNIESNER AND ZILIAK Figure 4. Income uncertainty for ig-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. for te igest income ouseolds, wic as not totally reversed during te 1990s as it as for lower income groups Estimated consumption volatility To operationalize te consumption variance decomposition summarized in (5) we again net out predictable components of te data series by using te person-year squared residuals from fixed-effect income growt regressions and parallel regressions for te annual cange in te average transfer rate and te annual cange in te average tax rate. Te error variances and covariance are te components of (5) tat track te relative contributions to te ouseold s consumption variability and subsequent welfare loss. For our calculations across all families we use our preferred estimate of ˆβ = 0.78 from Kniesner and Ziliak (2002). 10 For calculations stratifying by te tree permanent income groups we use our estimates, ˆβ 1 = 0.81, ˆβ 2 = 0.70, and ˆβ 3 = 0.63, wic sow te effect of disposable income canges on consumption canges declining wit permanent income level. 11 Table 3 summarizes te outcomes of interest. Again we igligt te contributions of policy by displaying te variability of consumption first in te absence of policy ten in te presence of transfers separately, taxes separately, and taxes and transfers jointly as income components. Across all families in our data, adding transfers reduces consumption volatility by about 8.5 percent on average (4.481/4.895 = 0.915). Netting out taxes reduces consumption fluctuations an additional 10 percent (4.018/4.895). Across all families te tax code provides more consumption insurance, 10.5 percent, tan social insurance and public and private transfers. For low-income families, transfers stabilize consumption by about 10 percent, and taxes combined wit transfers stabilize consumption by 18 percent. For moderate-income families, transfers stabilize consumption by about 10 percent and taxes

11 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 15 Table 3. Consumption volatility, transfers, and taxes a Means Standard deviations Total sample Gross labor and capital income Plus transfers Less taxes Plus transfers less taxes Low income sub-sample Gross labor and capital income Plus transfers Less taxes Plus transfers less taxes Moderate income sub-sample Gross labor and capital income Plus transfers Less taxes Plus transfers less taxes Hig income sub-sample Gross labor and capital income Plus transfers Less taxes Plus transfers less taxes a Te low-income ouseolds ave 12-year average incomes below one-alf te median income; moderate-income ouseolds ave average incomes between one-alf and one and one-alf te median income; ig-income ouseolds ave average incomes above one and one-alf te median income. plus transfers stabilize consumption by 21 percent. For te igest income families, transfers stabilize consumption by only about two percent, but taxes alone reduce consumption volatility by 10 percent. Table 3 empasizes ow across income substrata taxes are a more important stabilizer of consumption tan transfers wit te largest consumption stabilizing impact of taxes coming at te igest levels of permanent income. Figures 5 8 illustrate te year-to-year patterns tat underlie te overall averages in Table 3. Most evident is ow taxes and transfers mitigate te aggregate level of consumption variability (Figure 5) wit te importance of taxes and transfers equally split during te early 1980s except for te ig-income families (Figures 6 8). More recently, taxes ave gained in importance across te board as a source of income and consumption stabilization. Particularly striking in Figures 5 8 is ow consumption insurance as evolved during for low and moderate-income families. In te early 1980s low and moderate income families received a relatively large proportion of teir consumption stabilization from transfer payments. After TRA86 te tide turned away from transfers and toward taxes as automatic

12 16 KNIESNER AND ZILIAK Figure 5. Consumption volatility for all families, Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. Figure 6. Consumption volatility for low-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. stabilizers. Te cange in focus of U.S. policies concerning income support and redistribution continues in te middle 1990s wit te coupling of welfare reform wit EITC expansions. As te result of policy since te early 1980s, comparatively little consumption smooting now emanates from transfer payments compared to te income tax code. Interestingly, wen tere is decreasing absolute risk aversion in permanent income, as in our case ere, ten insurance is an inferior good (Mossin, 1968; Arrow, 1971, c. 3). Unlike te typical case were efficiency and equity considerations are offsetting te reductions in

13 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 17 Figure 7. Consumption volatility for moderate-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. Figure 8. Consumption volatility for ig-income families. Labor & capital income, plus transfers, less taxes, and plus transfers less taxes. income insurance wit permanent income level we identify are not only equitable but are also economically efficient Te Economic Growt and Tax Relief Reconciliation Act of 2001 Recently, Congress passed and President Bus signed into law Te Economic Growt and Tax Relief Reconciliation Act of 2001 (EGTRRA). Among oter tings, te law lowers

14 18 KNIESNER AND ZILIAK te rate in te igest federal income tax bracket and creates a new (10 percent) bracket for some persons formerly in te 15 percent federal income tax bracket. By lowering te tax rates for te igest income families and te lowest income families wit taxes owed, EGTRRA reduces implicit insurance. An offsetting factor is tat te structure of tax rates becomes more progressive tan before as te ratio of te igest to te lowest positive tax rate is now greater (over 3 : 1 versus 2.5 : 1). An insurance neutral aspect of te Act is tat a large concentration of taxpayers (about 50 percent of returns) remains in te 15 percent income tax bracket despite its narrowing. Because some components of EGTRRA add to disposable income smooting, some components reduce disposable income smooting, and oter components leave te amount of smooting more or less uncanged we expect little effect on economic well being stemming from canges in implicit insurance in te most recent round of tax reform. 4. Discussion Te welfare-enancing effect of reduced consumption variability is an under-appreciated dimension of income-based taxes. Annual consumption variation is reduced by about 10 percent due to implicit income smooting. In te United States taxes do as muc to stabilize income and consumption implicitly as do social insurance programs explicitly. Over time te stabilizing effect of taxes as also been growing relative to te stabilizing effect of transfers. An instructive way to frame te importance of consumption insurance implicit in te income tax code is to compare it to formal market purcased insurance. In Kniesner and Ziliak (2002) we find tat te current set of income-based taxes in te United States enances te typical ouseold s economic well being by an amount equivalent to one to two percent more total consumption. 12 Expressed in 2001 dollars te aggregate welfare gain from income tax based implicit consumption insurance in te U.S. is in te range $ billion. Implicit consumption insurance is ten of a magnitude rougly similar to total private ealt insurance premiums collected, $155 billion, or total automobile insurance premiums collected, $147 billion. 13 A notable implication of te welfare enancing reductions in consumption variability from income-based taxation we find concerns te future of researc into and practical design of an economically optimal income tax. Up to now most of te optimal income tax researc as considered two dimensions of social welfare, te deadweigt welfare loss because te income tax distorts labor supply (Ziliak and Kniesner, 1999) and te welfare gains because a nonlinear income tax produces greater equality of spendable income and consumption (Auerbac and Hines, 2001). Our results demonstrating te welfare-enancing importance of an income-based tax troug its effects on consumption dynamics imply tat future optimal tax researc need consider a tird dimension, wic is ow an optimal tax creates beneficial dampening of oterwise unintended fluctuations in spendable income and consumption. Te interesting complexity of tree-dimensional optimal tax researc will be tat altoug a flatter tax structure as welfare-increasing effects because it reduces labor supply distortions, a flatter tax structure also as welfare reducing effects because it may not only promote less equal after-tax incomes and consumption levels but it also lowers implicit insurance of incomes and consumption.

15 EXPLICIT VERSUS IMPLICIT INCOME INSURANCE 19 Notes 1. An extensive discussion of social insurance and means-tested transfer programs is in Scolz and Levine (2002). 2. Some useful empirical studies of recent income tax reforms are Auerbac (1996), Auerbac and Feenberg (2000), Auerbac and Slemrod (1997), Burman, Gale, and Weiner (1998), Engen and Gale (1996), Kasten, Sammartino, and Toder (1994), and Pecman (1985). 3. A sort reading list is Cocran (1991), Gruber (1997), Ham and Jacobs (2000), Hayasi, Altonji, and Kotlikoff (1996), Kniesner and Ziliak (2002), Mace (1991), Morduc (1995), Nelson (1994), and Townsend (1994). Muc of te literature is concerned wit testing for complete insurance, or weter ˆβ = By setting α = 0 we simply adjust te scale of insurance but do not alter te relative contributions of taxes and transfers across ouseolds or over time. 5. In Kniesner and Ziliak (2002) we produce ˆβ wit a forward-filter instrumental variables estimator applied to Eq. (1). Additional control variables include te number of cildren, te age of te youngest cild, as well as te race, education, and five-year birt coort of te ouseold ead. 6. For additional discussion of measuring income volatility see Dynarski and Gruber (1997) and Gottscalk and Moffitt (1994). 7. Te marital status screen means tat te persons in our data keep te same tax table, wic simplifies understanding ow tax reforms tat interact wit income-splitting provisions affect insurance implicit in income taxes. 8. Tere are a small number of person-years wit average tax rates or average transfer rates tat exceed 100 percent. Close examination of te observations wit extreme tax or transfer rates reveals tat tey are outliers, wic we remove from our calculations. 9. Additional useful references on income and consumption stabilization in te United States include Auerbac and Feenberg (2000), Coen and Follette (2000), Gruber (1997, 2000), and Hamermes (1982). 10. Estimated from models of total consumption, defined as disposable income less saving. 11. We do not reject te null ypotesis tat ˆβ is uncanging over time, owever. 12. Te basis of comparison is a lump-sum tax tat collects te same revenue but provides no implicit income and consumption insurance. 13. Statistical Abstract of te United States 2000, pp expressed in 2001 dollars using te CPI. References Arrow, Kennet J. (1971). Essays in te Teory of Risk-Bearing. Cicago: Markam Publising Co. Auerbac, Alan. (1996). Tax Reform, Capital Allocation, Efficiency, and Growt. In Henry Aaron and William Gale (eds.), Economic Effects of Fundamental Tax Reform. Wasington, DC: Te Brookings Institution. Auerbac, Alan and Daniel Feenberg. (2000). Te Significance of Federal Taxes as Automatic Stabilizers, Journal of Economic Perspectives 14, Auerbac, Alan and James R. Hines, Jr. (2001). Taxation and Economic Efficiency, Working Paper No. 8181, National Bureau of Economic Researc, Cambridge, MA. Auerbac, Alan and Joel Slemrod. (1997). Te Economic Effects of te Tax Reform Act of 1986, Journal of Economic Literature 35, Burman, Leonard, William Gale, and David Weiner. (1998). Six Tax Laws Later: How Individuals Marginal Tax Rates Canged Between 1980 and 1995, National Tax Journal 51, Cocrane, Jon. (1991). A Simple Test of Consumption Insurance, Journal of Political Economy 99, Coen, Darrel and Glenn Follette. (2000). Te Automatic Fiscal Stabilizers: Quietly Doing Teir Ting, Federal Reserve Bank of New York Economic Policy Review 6, Commerce Clearing House. U.S. Master Tax Guide ( Tax Years), Cicago, IL. Dynarski, Susan and Jonatan Gruber. (1997). Can Families Smoot Variable Earnings? Brookings Papers on Economic Activity 1, Engen, Eric and William Gale. (1996). Te Effects of Fundamental Tax Reform on Saving. In Henry Aaron and William Gale (eds.), Economic Effects of Fundamental Tax Reform. Wasington, DC: Te Brookings Institution.

16 20 KNIESNER AND ZILIAK Gertler, Paul and Jonatan Gruber. (1997). Insuring Consumption Against Illness, Working Paper No. 6035, National Bureau of Economic Researc, Cambridge, MA. Gottscalk, Peter and Robert Moffitt. (1994). Te Growt of Earnings Instability in te U.S. Labor Market, Brookings Papers on Economic Activity 0, Gruber, Jonatan. (1997). Te Consumption Smooting Benefits of Unemployment Insurance, American Economic Review 87, Gruber, Jonatan. (2000). Cas Welfare as a Consumption Smooting Mecanism for Single Moters, Journal of Public Economics 75, Ham, Jon and Kris Jacobs. (2000). Testing for Full Insurance Using Exogenous Information, Journal of Business and Economic Statistics 18, Hamermes, Daniel. (1982). Social Insurance and Consumption: An Empirical Inquiry, American Economic Review 72, Hayasi, Fumio, Josep Altonji, and Laurence Kotlikoff. (1996). Risk Saring Between and Witin Families, Econometrica 64, Kasten, Ricard, Frank Sammartino, and Eric Toder. (1994). Trends in Federal Tax Progressivity, In Joel Slemrod (ed.), Tax Progressivity and Income Inequality. Cambridge: Cambridge University Press. Kniesner, Tomas J. and James P. Ziliak. (2002). Tax Reform and Automatic Stabilization, American Economic Review 92. Mace, Barbara. (1991). Full Insurance in te Presence of Aggregate Uncertainty, Journal of Political Economy 99, Mincer, Jacob. (1974). Scooling, Experience, and Earnings. New York: Columbia University Press. Mitrusi, Andrew and James Poterba. (2000). Te Distribution of Payroll and Income Tax Burdens, , National Tax Journal 53, Morduc, Jonatan. (1995). Income Smooting and Consumption Smooting, Journal of Economic Perspectives 9, Mossin, Jan. (1968). Aspects of Rational Insurance Purcasing, Journal of Political Economy 76, Nelson, Julie. (1994). On Testing Full Insurance Using Consumer Expenditure Survey Data, Journal of Political Economy 102, Pecman, Josep. (1985). Wo Paid te Taxes: ? Wasington, DC: Te Brookings Institution. Scolz, Jon Karl and Kara Levine. (2002). Te Evolution of Income Support Policy in Recent Decades. In Seldon Danziger and Robert Haveman (eds.), Understanding Poverty. Cambridge, MA: Harvard University Press, Capter 6. Townsend, Robert. (1994). Risk and Insurance in Village India, Econometrica 62, U.S. Census Bureau. (1998). Statistical Abstract of te United States: 1998 (118t edition). Wasington, DC. U.S. Census Bureau. (2000). Statistical Abstract of te United States: 2000 (120t edition). Wasington, DC. U.S. Department of Commerce. State Government Tax Collections (Annual, Tax Years). Wasington, DC. U.S. Government Printing Office. (2001). Economic Report of te President Wasington, DC. Ventry, Dennis. (2000). Te Collision of Tax and Welfare Politics: Te Political History of te Earned Income Tax Credit, , National Tax Journal 53, Ziliak, James P. and Tomas J. Kniesner. (1999). Estimating Life Cycle Labor Supply Tax Effects, Journal of Political Economy 107,