The Household Spending Response to the 2003 Tax Cut: Evidence from Survey Data *

Transcription

1 DRAFT DO NOT CITE The Household Spendng Response to the 2003 Tax Cut: Evdence from Survey Data * Jula Lynn Coronado Watson Wyatt Joseph P. Lupton Federal Reserve Board Louse M. Shener Federal Reserve Board February 25, 2004 * The vews expressed are those of the authors and not necessarly those of the Federal Reserve Board or ts staff.

2 Introducton By the end of 2002, the U.S. economy exhbted consderable slack. Despte the fact that the 2001 recesson offcally ended n November of that year, the followng thrteen months showed lttle sgn of a recovery: Roughly 1 mllon jobs were lost, capacty utlzaton n the manufacturng sector remaned at ts trough almost 8 percentage ponts below ts hstorcal average, real wages and salares edged down, and nflaton had moderated. In the context of a lackluster economc recovery, Congress enacted the Jobs and Growth Tax Relef Reconclaton Act (JGTRRA) n May 2003, an extenson of the Economc Growth and Tax Relef Reconclaton Act (EGTRRA), whch was enacted n September The EGTRRA and the JGTRRA were advertsed as textbook fscal stmulus. When asked on February 9, 2004 about the contrbuton of the EGTRRA and JGTRRA to economc growth, the charman of the Councl of Economc Advsers, N. Gregory Mankw, stated: I have no doubt that the economy s stronger today, there are more people workng today because of the Presdent s tax cuts. If we had left taxes exactly as they were when the Presdent took offce, many, many more people would be unemployed today. What I m sayng s standard textbook economcs. When you cut taxes: that expands aggregate demand. (Mankw, 2004) Whle t s lkely that the Bush tax cuts boosted aggregate demand, the magntude of the effect s unclear. Several studes have examned the mpact of the EGTRRA on personal consumpton expendtures usng household level survey data and the results are mxed (Shapro and Slemrod, 2003a, 2003b; Agarwal, Lu, and Souleles, 2004; Johnson, Parker and Souleles, 2004; Mchel and Rector, 2004). In contrast, the mpact of the JGTRRA on household spendng remans an unexamned queston. In ths paper, we provde the frst evdence based on household level survey data on the spendng response to the 2003 JGTRRA. The methodology we use s based on Shapro and Slemrod s semnal work usng survey data to explctly ask households how they responded to the 1992 temporary tax reducton and to the 2001 EGTRRA (1995, 2003a, 2003b). Our analyss yelds three prncpal fndngs. Frst, we show that households were remarkably aware of recevng both the advance chld rebate that was maled out n checks to households n the late summer of 2003 and the boost to dsposable ncome that resulted from decreased wthholdng. Second, we fnd no dfference between the spendng response out of the maled-out advance 1

3 chld-credt and the lowered wthholdng taxes a fndng that runs contrary to the dscusson at the tme on how to most effectvely stmulate spendng. Fnally, our results ndcate that roughly a quarter of the proceeds of the total tax cut were consumed wthn the frst two quarters of the enacted legslaton. The spendng response does not appear to vary sgnfcantly by household demographc characterstcs, although there s some evdence that households that own stocks and have hgher ncome spent a larger share of ther tax cut. However, the qualty of the avalable ncome data s suspect. These results are consstent wth the spendng response to the EGTRRA reported n Shapro and Slemrod (2003a, 2003b). We apply the estmated spendng response to aggregate data and fnd that household spendng was boosted by about $26 bllon n the thrd quarter of 2003 or roughly 1 percent of GDP at an annual rate. Tax Polcy and Household Spendng Tax cuts are often advertsed as a quck and effectve way to stmulate the economy n the near-term and reduce macroeconomc fluctuatons n the long run. However, the effectveness of tax polcy largely depends on the degree to whch household spendng responds to changes n taxes. Unfortunately, wth regard to the margnal propensty to consume out of a tax cut, economc theory runs the gamut from zero (Barro, 1982) to more than one-half (Campbell and Mankw, 1989). Consequently, the spendng response to tax cuts has largely been an emprcal matter. A textbook treatment of the permanent ncome hypothess (PIH) suggests that changes n household spendng are proportonate to the present dscounted value of news regardng changes n future dsposable labor ncome. One mplcaton of ths s that households adjust ther spendng upon learnng about a tax cut rather than at the tme of the actual ncrease n dsposable ncome. The lterature on the excess senstvty of consumpton to changes n ncome s legon. The evdence however s mxed due n large part to the dffculty n dentfyng exogenous changes to dsposable labor ncome and n determnng the permanence of these changes (Brownng and Lusard, 1996). In these two respects, changes to tax polcy provde an deal natural experment. Wlcox (1989) found that socal securty recpents ncreased ther spendng only at the tme they receved an ncrease n benefts rather than when the ncrease was announced, whle Parker (1999) showed that household spendng moves up wth dsposable ncome at the tme n 2

4 the calendar year when the cumulated annual payroll taxes reach ther mandatory cap. Both Wlcox (1990) and Souleles (1999) note that consumer spendng tends to jump upon the recept of pre-determned annual tax refunds. Souleles (2002) examned the tmng of the spendng response to the Economc Recovery Tax Act of 1981 and concluded that many households only ncreased ther spendng when the change n wthholdngs actually occurred n ther paychecks rather than at the tme of the enactment of the legslaton. More recently, Johnson, Parker and Souleles (2004) showed that households spent the rebate porton of the Economc Growth and Tax Recovery Act of 2001 only at the tme of recept. Although Souleles (2002) reports that lqudty-constrants do not explan excess senstvty to the 1981 tax cut, Johnson et al. (2004) suggest lqudty-constrants may partally explan the response to the 2001 tax cut. The obverse of excess senstvty s that spendng should respond to news about future ncome. Poterba (1988) examned several fscal polcy changes over the past few decades and found that aggregate consumer spendng dd not react to the news of these changes n polcy, contrary to the PIH. However, t s dffcult to nterpret estmates of the effect of fscal stmulus based on aggregate data due to the multtude of other factors that affect aggregate demand. The PIH also provdes a benchmark for the magntude of the response to a change n tax polcy. Indeed, the potental for both near- and long-term macroeconomc stablty depends more on the magntude of the spendng response to tax polcy and less on whether spendng s excessvely senstve. In ths respect, the success of tax polcy s usually measured by the extent to whch aggregate demand s stmulated. Accordng to the PIH, only the present dscounted value of the change n expected future ncome s relevant. Consequently, the magntude of the spendng response depends largely on the duraton of the change n tax polcy. Temporary tax cuts have a smaller effect on spendng than permanent tax cuts, and tax cuts that are explctly offset n the near future should have an nsgnfcant effect. 1 Prevous studes of households spendng and changes n tax polcy report results that are somewhat n lne wth the PIH n that the response to permanent tax cuts are larger than the response to temporary tax cuts, although there are some exceptons. Also, the responses to temporary tax cuts are somewhat larger than would be expected from the PIH. Blnder (1981) and Poterba (1988) fnd an ntal mpc n the range of 0.16 to 0.24 out of the one-tme tax rebate 1 More generally, fscal stmulus should be completely neffectve f government debt s treated as a part of household net worth (Barro, 1982). 3

5 n the 1975 Tax Reducton Act, and an mpc after several quarters far larger than what the PIH would predct. Shapro and Slemrod (1995) also fnd evdence of a large response to the to the 1992 tax cut. Ths result s partcularly surprsng because the 1992 tax cut was completely offset the followng year and ths was publcly known at the tme of the tax cut. Despte the relatvely large responses to temporary tax cuts, some evdence suggests an mpc out of the permanent tax cut n the Economc Recovery Tax Act (ERTA) of 1981 on the order of 0.6 to 0.9 (Souleles, 2002). In contrast to ths large response, Shapro and Slemrod (2003) conclude that the spendng response to the 2001 EGTRRA was surprsngly low despte ts explct purpose to boost aggregate demand. There are several reasons that could explan the strkng dfference between these two results. Frst, the spendng response to a change n tax polcy depends on the economc condtons at the tme. The 1981 ERTA was enacted one month nto the 1981/82 recesson, on the heels of the 1980 recesson, and at the end of an era of hgh nflaton and low growth. The 2001 EGTRRA was enacted sx months nto the 2001 recesson, whch followed an era of unprecedented growth. Fnancal condtons were far more accommodatve n 2001 than n 1981 and household wealth was more lqud. Moreover, despte the fall n value of corporate equtes n 2001, the rato of wealth to ncome remaned well above ts level n The result s that household dscretonary spendng was better supported at the tme of the 2001 tax cut than at the tme of the 1981 tax cut and so the margnal value of addtonal resources was also less. Second, the EGTRRA s not a permanent tax cut but s set to expre n 2010, dampng the potental spendng response. Estmatng the Spendng Response to the JGTRRA The JGTRRA was prmarly a pullng-forward of the provsons enacted n the EGTRRA. The legslaton had several provsons. Frst, the JGTRRA reduced most margnal tax rates above 15 percent by 2 percentage ponts and reduced the top margnal tax rate by 3.6 percentage ponts. These rate reductons had already been ncluded n EGTRRA, but had been scheduled to go nto effect gradually, wth a 1 percentage pont rate reducton n 2004 and the remander n Under the 2003 tax act, these rates reductons all occurred n 2003, and were made retroactve to January 1 of that year. As n the EGTRRA, these tax rate reductons returned to ther pre-egtrra levels n Second, the JGTRRA provded an ncrease n the 10 percent margnal ncome tax rate bracket and complete marrage penalty relef, whch the EGTRRA dd 4

6 not provde untl Both of these provsons were effectve only for 2003 and Fourth, the JGTRRA rased the chld tax credt from $600 per chld to $1000 per chld for 2003 and Under the EGTRRA, the chld credt was scheduled to ncrease gradually to $1000 by 2010 and then to revert to $500 n The JGTRRA rased t to $1000 for two years, but then had t revert back to ts prevous schedule. The 2003 porton of the ncrease ($400 per chld) was sent as advance refund checks to those who had clamed chld tax credts on ther 2002 tax returns. These checks were sent out n 3 batches the last week of July and the frst two weeks of August. 2 Ffth, the alternatve mnmum tax exempton was boosted by roughly $5,000 for sngle households and $10,000 for marred households n 2003 and 2004 only. In whole, the JGTRRA boosted real dsposable ncome by about $100 bllon at an annual rate n the thrd quarter of 2003 and by $45 bllon n the fourth quarter. Real DPI was boosted by about $100 bllon n 2004, $80 bllon n 2005 and $60 bllon n As ths descrpton makes clear, t s dffcult to know how these changes n tax law mght have affected taxpayer calculatons of permanent ncome. Dd taxpayers beleve that the temporary provsons would actually be allowed to expre? Had taxpayers already reacted to the scheduled future tax cuts, or were they unaware of them untl they were actually put n place? A casual examnaton of the aggregate data suggests a szable spendng response. Total real personal consumpton expendtures (PCE) surged 5.0 percent at an annual rate n the thrd quarter of If we were to assume the entre devaton from real PCE growth for 2003 as a whole (3.8 percent) were attrbutable to the JGTRRA, the ntal spendng response would be about $22 bllon mplyng an mpc of Unfortunately, the boost to dsposable ncome from the JGTRRA concded wth record fnancal ncentves for motor vehcles and a fall n real nterest rates to hstorcal lows. Indeed, although spendng on nondurables dd leap n 2003Q3, the jump n personal outlays for motor vehcles contrbuted 1.4 percentage ponts to total real PCE growth n that quarter. Although the JGTRRA lkely contrbuted to the surge n motor vehcles spendng, t s dffcult to dentfy the margnal spendng response attrbuted to the JGTRRA excludng the effect favorable fnancng condtons usng aggregate consumpton data alone. Moreover, offsettng these effects were the 2 The JGTRRA also set the tax rate on dvdend ncome equal to the rate at whch captal gans are taxed, and reduced the captal gans to 10 percent on longer-term gans and 5 percent on short term gans. The effects of these provsons are not examned n ths paper. 5

7 ongong concerns over a labor market that contnued to loss jobs. [MORE STRUCTURAL ESTIMATE.error correcton model.] Because the tax cut vared by ncome and the number of chldren, household level data on consumer spendng could be used to dentfy the ndependent effect of the JGTRRA, although t would stll be mportant to control for motor vehcle ncentves and fallng nterest rates. An alternatve strategy s to smply ask households how they responded to the JGTRRA, as was done n the Unversty of Mchgan s Survey of Consumers. The largest drawback to ths methodology s that households may not always do as they say (Bertrand and Mullanathan, 2001). However, Shapro and Slemrod (2003b) argue the data from Mchgan Survey provde a reasonable estmate of the spendng response to the EGTRRA when compared to the aggregate data. Moreover, the advantage of ths methodology s that t controls for the confoundng aggregate factors noted above, as well as much of the observed and unobserved household heterogenety that affects preferences over the lfe-cycle. The Mchgan Survey of Consumers Each month the Survey of Consumers asks a representatve sample of about 500 U.S. households a seres of questons about ther economc stuaton, as well as ther mpressons of the macroeconomy. The survey s conducted by telephone. In August, September, and October of 2003, supplemental questons were added to the survey that asked respondents what they dd wth ther tax cut. The questons were desgned n much the same way as a tax rebate module that was added to the Survey of Consumers n 2001 regardng the rebate checks that were sent out as part of the Economc Growth and Tax Relef Reconclaton Act of The questons dstngushed between the mal-out advance chld tax credt and the reducton n ncome tax wthholdng that resulted from reductons n margnal tax rates. In partcular, the key questons were as follows: Earler ths year a Federal law was passed that ncreased the chld tax credt and reduced tax rates. Those who qualfy for the ncreased chld credt wll receve rebate checks worth four hundred dollars for each chld. Startng on July frst, changes n wthholdng went nto effect resultng n an ncrease n take home pay for those who qualfy for reduced taxes. Thnkng about your (famly s) fnancal stuaton ths year, wll the [rebate 3 See Shapro and Slemrod (2003a) for a detaled descrpton of the survey desgn wth regard to the tax cuts. 6

8 check/ncrease n take home pay] lead you to mostly to ncrease savng, mostly to pay off debt, or are you not elgble for [the chld tax credt/ncrease n take home pay]? The queston tred to ask questons n everyday language that map nto well-defned economc concepts. Both the savng and payng off debt responses map nto savng. The respondents were then asked whether ther reacton would not change for at least a year. The survey s thus amed at measurng the overall propensty of households to spend or save the tax cut and does not capture any specfc tmng that would allow us to shed lght on the lterature concernng the excess senstvty of consumpton to ncome. All results are computed usng the survey weghts that provde a representatve sample of households. 4 Survey Valdaton We frst examned whether people n the Survey of Consumers were reasonably aware of recevng the tax cut. Usng data from the CPS and matchng t to observable characterstcs n the Mchgan survey, we asked whether the people who were lkely to have receved ether the advance chld credt rebate or a reducton n wthholdng reported beng elgble for these tax cuts. We use data from the March 2002 CPS to determne how the rebate checks and changes n wthholdng lkely were dstrbuted across dfferent populaton groups. The CPS contans nformaton on household adjusted gross ncome (AGI), the number of chldren, and the wages of each worker n a household. 5 Usng such nformaton, we calculated whether households would have receved a rebate check and whether they would have experenced a change n wthholdng. Usng the changes n the wthholdng tables, we also calculated how large the change n wthholdng was lkely to have been. These calculatons provde a reasonable analyss of how households were affected by the tax cut provsons. However, t s not perfect. For example, the calculatons wll not perfectly predct whch households receved a chld tax rebate, as chldren lvng wth one parent may actually be clamed as dependents on the other parent s tax return (so we would mpute a rebate check to the wrong household). Smlarly, the changes n wthholdng are calculated based on 4 The weghts had no substantve effect on the results. 5 AGI s top coded at $99,999 n the CPS. For those households wth AGI n excess of $99,999 we constructed an AGI equal to total famly ncome plus captal gans less non-taxed ncome lke workmen s compensaton, socal securty, and SSI benefts. 7

9 the prevous year s wages, whereas actual changes n wthholdng would have been based on wages earned at the tme of the survey. Table 1 compares these mputed changes n tax payments wth the results from the Mchgan Survey. The table shows a hgh degree of awareness of the advance chld credt rebate checks. Among all households, 26 percent of respondents n the Mchgan survey reported recevng or were expectng to receve advance chld credt rebate check compared to 33 percent of all households mputed n the CPS. Ths small dfference dsappears when restrctng the sample to households wth chldren. The awareness of changes n wthholdng was smaller. Whle only 48 percent of households reported a change n ther wthholdngs, the mputatons n the CPS suggest 63 percent of households had a reducton n wthholdngs, a share smlar to estmated provded by the Congressonal Budget Offce. Restrctng the sample to households less than age 62 does not elmnate the dfference, although both share ncrease by about 10 percentage ponts. Table 2 examnes the responses regardng the rebate check by ncome class. There are some dstnct dfferences between the Mchgan survey responses and the mputed responses from the CPS. In both the Mchgan survey and the CPS, households n the hghest and lowest ncome categores were less lkely to report recevng or expectng to receve a check. The two surveys are somewhat smlar across the largest groups of ncome, rangng from $15,000 to $150,000, wth the most smlarty among households wth ncome between $15,000 and $30,000. However, the relatonshp between ncome n the Mchgan survey and recept of the rebate check s much weaker than n the CPS. Ths s most true at the tals of the ncome dstrbuton where fewer households should have receved an advance chld credt rebate. One possble explanaton for these dfferences s that famles were not aware of the ncome lmts for the rebate checks, and that many lower or hgher ncome famles antcpated recevng a check when they were not enttled. However, an examnaton of the dfferences n responses across the August, September, and October surveys does not show the systematc pattern one mght expect n ths case that s, perceved elgblty for the rebate checks dd not declne over tme. Instead, a more lkely explanaton s that the household ncome response n the Mchgan survey s a less accurate measure of taxable ncome than n the CPS, where efforts are made to make sure ncome measures are accurate. In addton, measurement error s amplfed by the smaller cell szes for the hgh and low ncome groups. 8

10 Table 3 examnes the survey responses regardng the reducton n wthholdng taxes by ncome class. As wth the chld credt, there s a far weaker relatonshp n the Mchgan survey between ncome and a reducton n wthholdng taxes. The dfferences are smallest n the largest ncome groups and largest n the tals of the ncome dstrbuton, an ndcaton agan of potental measurement error. Stll, there s a 30 percentage pont underreportng n the Mchgan survey among households wth ncome between $30,000 and $75,000. Whle almost all households n the CPS n ths ncome class are mputed to have receved a reducton n wthholdng taxes, only 66 percent of the households n the Mchgan survey report recevng a reducton. The comparson between the Mchgan survey results and the CPS mputatons shows that households were qute aware of the rebate checks and were also aware, but less so, of the wthholdng changes. To the degree that ths comparson ndcates errors n awareness, they are concentrated n the tals of the ncome dstrbuton. The bulk of the populaton appeared to be aware of recevng the boost to dsposable ncome from the tax cut. Survey Results The spendng responses of households that reported recevng the chld tax rebate are shown n Table 4 and the responses for households that reported recevng an ncrease n take home pay are shown n Table 5. Overall, 27.0 percent of those who reported recevng the chld tax rebate sad they saved most of t, 49.0 percent sad they mostly pad down debt, and 24.0 percent reported mostly spendng the rebate. Among those who sad they had receved an ncrease n take home pay, 36.8 percent reported savng most of t, 42.5 percent sad they mostly pad off debt wth the ncreased dsposable ncome, and 20.7 percent sad they spent most of the proceeds. If the tax cuts were vewed as permanent, we would expect households to spend most of the ncrease n dsposable ncome. However, the JGTRRA was not a permanent tax cut but expred after seven years and the ncremental value of the JGTRRA relatve to the EGTRRA was smaller than the ntal effect of the EGTRRA on the present dscounted value of future dsposable ncome. As a result, we would expect the spendng response to the JGTRRA to have been smaller than the response to the EGTRRA. However, the spendng response out of both the chld credt and the reducton n wthholdngs was a bt larger than the 21.8 percent spendng response to the EGTRRA reported n Shapro and Slemrod (2003a). 9

11 The tables present the responses by varous categores of observable characterstcs. The fracton of respondents reportng spendng most of the proceeds of the tax cut rses wth ncome, stock ownershp, educaton, and age. These results are consstent wth the fndngs of Shapro and Slemrod (2003) and at frst blush mght appear to contradct our noton that people wth lowncome and low-educaton are more lkely to face lqudty constrants or follow rule of thumb behavor and thus consume all changes n ncome. However, these patterns reverse f we assume households that reported usng the tax cut to pay down debt actually boosted spendng quckly after payng down ther debt. Focusng only on the percentage of households that reported savng the tax cut n Table 4, there s a strong postve relatonshp between ncome and the share of households that report savng most of the advance chld credt rebate. As ndcated n Table 5, the relatonshp s weaker for the reducton n wthholdngs but stll postve. The ratonale for assumng households that pay down debt wll quckly ncrease spendng s based on a study of the effects of the 2001 tax cut on spendng and credt card debt by Agarwal, Lu and Souleles (2004). Agarwal et al fnd that households who used ther tax cut to pay down credt card debt dd so only temporarly and that after a one-year perod credt card debt was back to the pre-tax cut level, ndcatng that the tax cut was spent. The potental for ths behavor should be evdent n the share of households that report that they wll spend most of the tax cut wthn a year. These results are reported n the fnal columns of Table 4 and 5 and do not support the hypothess that most of the households that pay debt wll quckly ncrease spendng. Ths fndng s consstent wth the results n Shapro and Slemrod (2003b) who use the small panel component of the Mchgan survey to show that households dd not change ther spendng response roughly sx months later. Consstent wth the results n Shapro and Slemrod (2003a), households that own corporate equtes are more lkely to report spendng most of ther tax cuts. Ths apples to both the advance chld credt rebate (Table 4) and the reduced wthholdngs (Table 5). There also appears to be a slght postve relatonshp between educaton and the share of households that report spendng most of the tax cut. Fnally, consstent wth the lfe-cycle model of consumer spendng, older households were more lkely to report spendng most of ther tax cuts relatve to younger households. Interestngly, these relatonshps reverse (or at least dsappear) f t s assumed that the repayment of debt s subsequently converted to ncreased spendng. However, 10

12 t s dffcult to nterpret any of these unvarate relatonshps. Multvarate analyss s requred to dsentangle any potental ndependent relatonshp. Dfferences between checks and wthholdng As noted above, one notable dfference between the tax cut delvered through a one-tme advance chld credt rebate delvered n the form of a mal-out check and that delvered through gradual changes n tax wthholdng s that people were more lkely to notce the rebate checks than they were the wthholdng changes. In ths secton, we examne the dfferences n savng responses for those who report recevng both the rebate check and the wthholdng changes. In partcular, we ask whether people were more lkely to report spendng the rebate checks than they were the wthholdng changes. Note that the reducton n wthholdngs s a repeated tax cut that affects every paycheck between 2003 and 2006, whle the advance chld credt rebate check s a once-a-year payment that s only effectve n 2003 and If people understand these dfferences, they should save most of the advance chld credt rebate to smooth the boost to ncome over the year whle they should spend the wthholdng because t s already smooth. In comparng the responses among all households n Tables 4 and 5, t somewhat surprsng to see that the spendng response to the advance chld rebate s as large as the spendng response to the reducton n wthholdngs. The fact that there s lttle dfference between the spendng responses suggests that the hghly vsble form n whch the advance chld credt rebate was pad out had an effect. Indeed, the spendng response was slghtly larger out of the advance chld credt rebate. Table 6 reports the spendng responses to both the advance chld credt rebate and the reducton n wthholdngs among households that receved both tax cuts. In general, households seem to vew the rebate checks and wthholdng changes as very smlar wth roughly 65 percent of households plannng to use the rebate check and the ncrease n take-home pay n an dentcal manner (spend, save, or pay down debt). Even over the course of a year there s lttle dfference n the spendng plans for the advance chld credt rebate and the reducton n wthholdngs. In both cases, the share of households plannng on spendng ther tax cut ncreases by only about 5 to 10 percentage ponts when the tme horzon s expanded from a couple of 6 After 2006, the JGTRRA provson s dentcal to the EGTRRA provson on the reducton n margnal tax rates. 11

13 quarters to a year as ndcated n Table 4 and 5. Table 6 shows that the same s true for households that receved both tax cuts. The evdence suggests that households saw the advance chld credt rebate and changes n wthholdng as smlar. From a polcy perspectve, ths does suggest that rebates n general are lkely to have a larger mmedate effect on the economy, as households do not seem to smooth the spendng response to a one-tme tax cut. However, the evdence does not support the dea that, for a gven one-tme change n dsposable ncome, households wll spend a rebate check at a faster pace than they would changes n take-home pay. Estmatng a Margnal Propensty to Consume The unque advantage of the Mchgan survey data compared to household survey data on consumer expendtures s that t reveals the change n household spendng gven the tax cuts compared to what households beleved ther spendng would have been n the absence of the tax cuts. We estmate an emprcal model of the household spendng response S out of the tax cut τ that allows for both observed and unobserved heterogenety n the mpc: S = τ exp( α + γx + η ), (1) where exp( α + γx + η ) s the mpc out of the tax cut, X s a vector of household characterstcs and η represents random unobserved heterogenety that s uncorrelated wth τ and If the Mchgan survey measured each household s spendng response, the mpc would smply be gven by S X. / τ, and the aggregate mpc would mmedately follow. Unfortunately, the spendng response s unobserved. Rather, the survey nstrument measures only whether a household spent most of ther tax cut, saved most of ther cut, or pad down debt wth most of ther tax cut. If a household clamed that they dd not spend most of ther tax cut, they were then asked f they thought they would spend most of t wthn a year. As are result, S s treated as a latent varable and the parameters α and γ reman to be estmated. Let I = 1 f a household clams that they receved a tax cut and that they spent most of t where most s defned as beng greater than 50 percent, and I = 0 f a households clams to have receved a tax cut but they dd not spend most of t. The model follows from the defnton of the ndcator varable and (1): 12

14 ( I = X) = ( S > τ ) Pr 1 Pr 0.5 ( τ α γx η τ) ( η α γx ) = Pr exp( + + ) > 0.5 = Pr < ( ln 0.5) +. (2) Assumng η s normally dstrbuted and normalzng to a unt varance, (2) can be estmated as a standard Probt model. Table 7 reports the estmated mpc s for the advanced chld credt and the lower wthholdngs usng the self-reported spendng response ndcators. Both the ntal mpc and the mpc after one year are estmated. No demographcs are ncluded n the regresson, only a constant term. 7 The frst row drops all households that clam to have not receved the respectve tax cut. There s only a small dfference between the mpc s out of the two dfferent tax cuts. As ndcated n the frst row of Table 7, households boosted spendng by 25 percent of the advanced chld credt and by 22 percent of the reducton n wthholdngs. Stll, the fact that the mpc out of the advanced chld credt s a touch larger s consstent wth mental accountng hypothess whch clams that the smaller and more vsble tax cut should have provded a bgger boost to spendng. After one year, the mpled mpc out of the advanced chld credt ncreases to 30 percent and ncreases to 32 percent for the mpc out of lowered wthholdngs. Some households may have receved a tax cut that was not recorded n the survey, ether because the respondent was nattentve to the tax cut or because of a codng error. Usng the fnancal and demographc characterstcs reported n the survey, we mpute whether or not a household receved a tax cut. The spendng response of these households also needed to be mputed. Dependng on the decsons rules, the unobserved tax may have been ether saved or spent, or both. The bottom half of Table 7 reports the estmated mpc s assumng that all or none of these mputed tax cuts were spent. The mpc s jump to almost one-half when assumng the households wth an mputed tax cut spend most of t. As noted above, t s dffcult to nterpret the unvarate relatonshps reported n Table 4 and 5. Table 8 reports the results of the estmated probt model allowng for varaton n the mpc by household demographc and fnancal characterstcs. The frst column under the advance chld credt rebate and the frst column under the reduced wthholdng tax show the demographc effects on the mpc. The age effect for both tax cuts s sgnfcant wth older households have a 7 In ths case, the estmated mpc s smply exp( ˆ α ). Standard errors are corrected usng the delta method. 13

15 hgher mpc, whle the effect of marrage and race are nsgnfcant. The second columns under the two tax cuts keep the age effect and add household fnancal characterstcs. The age effects are unchanged. Homeowners have a smaller mpc out of the reducton n wthholdngs but ths s largely offset f they also own stocks, whch has a postve effect on the mpc. Asset ownershp has lttle effect on the mpc out of the advance chld rebate. The relatonshp between ncome and the mpc out of both tax cuts s nsgnfcant for ncomes below $75,000. However, among smlar aged households, those wth ncome between $75,000 and $100,000 have an mpc out of the reducton n wthholdngs that s 37 percent larger than households wth ncome less than $30,000 (and perhaps less than $75,000 snce ths s nsgnfcant). Households wth ncome greater than $100,000 have an mpc out the reducton n wthholdngs that s almost 50 percent larger. Ths postve relatonshp s at odds wth the conventonal wsdom that households wth hgher ncome tend to have a hgher propensty to save. However, the relatonshp hghlghted n Table 8 s between current ncome and spendng rather than between permanent or lfetme ncome and spendng. It s the latter relatonshp that has been shown to be negatve (Dynan, Sknner and Zeldes, 2004). Nevertheless, f hgher levels of current ncome are more assocated wth transtory or temporary ncome, t s dffcult to explan the postve relatonshp see n the data. The thrd columns under the two tax cuts report the effect of expected real ncome growth. The varable s equal to one f a household expects ther ncome to growth faster than the rate of nflaton and zero otherwse. Not surprsngly, households that expect ther real ncome to grow have an mpc that s about 26 percent larger than the mpc of households that expect ther real ncome to declne. The bottom two rows of Table 8 report the mean of the estmated mpc s as well as the standard devaton of the mpc s across households. As n Tables 4 and 5, there s lttle dfference n ether the mean or the standard devaton between the mpc out of the advance chld credt rebate and the reducton n wthholdngs. There s a notceable dfference n the range of mpc s. Whereas the mpc s out of the reducton n wthholdngs range between 0.13 and 0.72, the mpc s out of the chld tax credt only range from 0.17 to In general however, the spendng response out of the two tax cuts s strkngly smlar gven the dfferences n the way they were structured. We also examned varaton n the mpc out of the two tax cuts over the subsequent year. Although the mean of the mpc s were both boosted to about 0.33, the relatonshps 14

16 taxes. 8 The results are reported n columns 5 and 6 of Table 9. Applyng these mpc s to the between the mpc and household demographc and fnancal characterstcs were essentally unchanged. Aggregate Spendng Response The JGTRRA boosted dsposable personal ncome by roughly $100 bllon. As ndcated n the bottom row of Table 9, applyng the estmated constant mpc out of the reducton n wthholdngs from Table 7 to ths entre ncrease suggests total real spendng was rased by about $22 bllon n the thrd and possbly fourth quarter of Strkngly, ths s dentcal to the back-of-the-envelope estmate noted above based smply on a casual observaton of the pattern of total real personal consumpton expendture growth n Applyng the estmated constant mpc for spendng over the year suggests total real spendng was rased by about $31 bllon between 2003Q3 and 2004Q3. The results from Table 8 ndcated a sgnfcant and postve relatonshp between the mpc out of the tax cuts and ncome. Combned wth the fact that the magntude of the tax cut was larger at hgher levels of ncome n absolute terms, ths relatonshp suggests that the spendng response s larger than what s mpled by assumng a constant mpc. Table 9 reports the average dollar value of the tax cut by adjusted gross ncome (AGI) class along wth the number of households tax unts n each class. The mpc s by AGI class were estmated usng the same probt model from above for the spendng response to the reducton n wthholdng aggregate tax cut wthn each AGI class yelds a spendng response by AGI class, reported n the last two columns of the table. Not surprsngly, roughly 64 percent of the ntal spendng response can be attrbuted to households whose AGI s greater than $100,000. Ths share s unchanged for the spendng response over the twelve months followng the enacted tax cut. After accountng for an mpc that ncreases wth ncome, the ntal aggregate spendng response ncreases somewhat relatve to the assumpton of a constant mpc to $26.1 bllon. The mpled aggregate mpc defned as the rato of the aggregate spendng response to the aggregate tax cut s The aggregate spendng response over the year followng the tax cut s just under $38 bllon. 8 Usng the estmated mpc s out of the advance chld rebate ncreases the aggregate spendng response only slghtly. 15

17 Concluson The JGTRRA njected roughly $100 bllon nto the pocket books of U.S. households n the thrd quarter of There s lttle doubt that such a large nflow of resources boosted aggregate demand. However, the magntude of the ncrease s unclear and so the overall effectveness of the tax polcy remans an open queston. In ths paper, we examne the results from a household survey that explctly asked tax payers whether they spent most of ther tax cut or saved most t. Our analyss yelds three prncpal fndngs. Frst, we show that households were remarkably aware of recevng both the advance chld rebate that was maled out n checks to households n the late summer of 2003 and the boost to dsposable ncome that resulted from decreased wthholdng. Second, we fnd no dfference between the spendng response out of the maled-out advance chld-credt and the lowered wthholdng taxes a fndng that runs contrary to the dscusson at the tme on how to most effectvely stmulate spendng. Fnally, our results ndcate that roughly a quarter of the proceeds of the total tax cut were consumed wthn the frst two quarters of the enacted legslaton. The spendng response does not appear to vary sgnfcantly by household demographc characterstcs, although there s some evdence that households that own stocks and have hgher ncome spent a larger share of ther tax cut. However, the qualty of the avalable ncome data s suspect. These results are consstent wth the spendng response to the EGTRRA reported n Shapro and Slemrod (2003a, 2003b). We apply the estmated spendng response to aggregate data and fnd that household spendng was boosted by about $26 bllon n the thrd quarter of 2003 or roughly 1 percent of GDP at an annual rate. 16

18 References Agarwal, Sumt, Lu, Chunln, and Souleles, Ncholas S. The Response of Consumer Spendng and Debt to Tax Rebates -- Evdence from Consumer Credt Data. The Wharton School workng paper, Barro, Robert J. "Are Government Bonds Net Wealth." Journal of Poltcal Economy, 1974, 82 (6), Bertrand, Maranne, Mullanathan, Sendhl. "Do People Mean What They Say? Implcatons for Subjectve Survey Data." Amercan Economc Revew, 2001, 91 (2), Brownng, Martn, Lusard, Annamara. "Household Savng: Mcro Theores and Mcro Facts." Journal of Economc Lterature, 1996, Campbell, John Y., Mankw, N. Gregory. "Consumpton, Income, and Interest Rates: Renterpretng the Tme Seres Evdence," Blanchard, Olver J., Fscher, Stanely, NBER Macroeconomcs Annual Cambrdge, MA: MIT Press, 1989, Dynan, Karen E., Sknner, Jonathan, Zeldes, Stephen P. "Do the Rch Save More?" Journal of Poltcal Economy, 2004, 112 (2), Johnson, Davd S., Parker, Jonathan A., Souleles, Ncholas S. "The Response of Consumer Spendng to the Randomzed Income Tax Rebates of 2001." The Wharton School workng paper, February Mchel, Norbert J. and Rector, Ralph A. Was the 2001 Tax Rebate Effectve Stmulus Polcy? Usng the Consumer Expendture Survey to Test Whether Consumers Spent Ther Rebate Checks. Hertage Foundaton workng paper, June Mankw, N. Gregory. Press brefng, Washngton, D.C., Councl of Economc Advsers, February 9, Modglan, Franco, Stendel, Charles. "Is a Tax Rebate an Effectve Tool for Stablzaton Polcy?" Brookngs Papers on Economc Actvty, 1977, 1977 (1), Parker, Jonathan A. "The Reacton of Household Consumpton to Predctable Changes n Payroll Tax Rates." Amercan Economc Revew, 1999, 89 (4), Poterba, James M. "Are Consumers Forward Lookng? Evdence from Fscal Experments." Amercan Economc Revew, 1988, 78 (2), Shapro, Matthew D., Slemrod, Joel. "Consumer Response to Tax Rebates." Amercan Economc Revew, 2003, 93 (1), Shapro, Matthew D., Slemrod, Joel. "Consumer Response to the Tmng of Income: Evdence from a Change n Tax Wthholdng." Amercan Economc Revew, 1995, 85 (1), Shapro, Matthew D., Slemrod, Joel. "Dd the 2001 Tax Rebate Stmulate Spendng? Evdence from Taxpayer Surveys," Poterba, James, Tax Polcy and the Economy. Cambrdge: MIT Press, Souleles, Ncholas S. "Consumer Response to the Reagan Tax Cuts." Journal of Publc Economcs, 2002, Souleles, Ncholas S. "The Response of Household Consumpton to Income Tax Refunds." Amercan Economc Revew, 1999, 89 (4), Wlcox, Davd W. "The Constructon of U.S. Consumpton Data: Some Facts and Ther Implcatons for Emprcal Work." Amercan Economc Revew, 1992, 82 (4),

19 Table 1: Comparson of Mchgan Survey to March CPS (percent) Mchgan Survey March CPS Receved rebate check wth chldren Wthholdng changes age less than Table 2: Households wth Chldren Recevng Chld Credt Rebate Check (percent) Income ($1,000) Mchgan Survey March CPS Less than to to to to Greater than Table 3: Non-Elderly Households wth Reduced Wthholdng Tax (percent) Income ($1,000) Mchgan Survey March CPS Less than to to to to Greater than

20 Table 4: Responses to the Chld Credt Rebate N Percent Elgble Save Response (percent) Reduce Debt Spend Spend wthn a Year All Elgble Households Household Income Less than $30, $30,000 to $75, $75,000 to $100, Greater than $100, Corporate Equtes Don't own Own Educaton of Head No hgh school degree Hgh school degree Some College College degree Age of Head Less than to

21 Table 5: Responses to the Reduced Wthholdng Tax N Percent Elgble Save Response (percent) Reduce Debt Spend Spend wthn a Year All Elgble Households Household Income Less than $30, $30,000 to $75, $75,000 to $100, Greater than $100, Corporate Equtes Don't own Own Educaton of Head No hgh school degree Hgh school degree Some College College degree Age of Head Less than to Greater than

22 Table 6: Response Among Households Recevng Both Tax Cuts Response to: Reduced Wthholdng Tax Reduce Spend Total Debt Chld Credt Rebate Save Spend wthn a year Save (65) (14) (12) (91) (25) Pay down debt (31) (97) (16) (144) (36) Spend (15) (18) (38) (71) (42) Total (111) (129) (66) (306) Spend wthn a year (19) (26) (45) (57) Cell counts are n parentheses. Table 7: Estmated Margnal Propensty to Consume, Probt wth Constant Only Chld Credt Rebate Reduced Wthholdng Tax Spend Now Spend Year Spend Now Spend Year Self-Report Receve (0.02) (0.02) (0.01) (0.02) Impute Receve Save (0.01) (0.02) (0.01) (0.01) Spend (0.03) (0.03) (0.02) (0.03) Note: The table reports the estmated aggregate margnal propensty to consume across all households n the sample. The model assumes that the margnal propensty to consume vares across households due to a varaton n household preferences. The results are based on self-reported recept of the tax cut as well as the authors' mputaton of who should have receved the tax cut. For the mputaton, we ether assume that they unknowngly saved "most" of the tax cut or that they unknowngly spent "most" of the tax cut. Standard errors are shown n parentheses. 21

23 Table 8: Explanng Varaton n the Margnal Propensty to Consume (Spendng Now) Chld Credt Rebate Reduced Wthholdng Tax Constant (0.21) (0.23) (0.23) (0.13) (0.18) (0.18) Age: 44 or younger (0.18) (0.19) (0.19) (0.12) (0.12) (0.13) Age: 62 or older (0.21) (0.23) (0.23) Marred (0.17) (0.12) Black (0.31) (0.24) Own Home (0.20) (0.20) (0.15) (0.15) Own Stocks (0.18) (0.18) (0.14) (0.14) Income $30,000 to $75, (0.22) (0.22) (0.17) (0.17) Income $75,000 to $100, (0.26) (0.26) (0.21) (0.21) Inome greater than $100, (0.28) (0.29) (0.21) (0.21) Expected Real Income Growth (0.17) (0.13) Log-Lkelhood Margnal Propensty to Consume Mean Standard Devaton Standard errors n parentheses 22

24 Table 9: JGTRRA and the Aggregate Spendng Response Adjusted Gross Income Tax Cut MPC Spendng ($bll) Tax Unts (1,000) Per Unt Total Now Wthn Wthn Now ($) ($bll) Year Year Less than $30,000 74, $30,000 - $75,000 42, $75,000 - $100,000 9,518 1, Greater than $100,000 11,913 4, All 138, Memo: Constant MPC 138,