The effect of tax rebates on consumption expenditures: evidence from state tax rebates.

Overall, then, the estimated results suggest that the rebate checks may have increased spending in the quarter of receipt (though the coefficients are sometimes the wrong sign and often insignificant), and may have affected the composition of expenditures. For total expenditures, the magnitudes estimated are quite similar to those found elsewhere in the literature, with the estimate from the base specification implying that individuals spent about one-fourth of their rebate in the quarter of receipt, which is consistent with, thought slightly less than, the results found in Shapiro and Slemrod (2003) and Johnson et al. (2004).

To probe whether the effect of rebates may have been different among the different states that distributed them, in Table 8 rebate amounts and dummy variables for rebate receipt are entered separately by state. (27) In this table, most coefficients are of plausible magnitude, and many are significant. In the top panel, individuals in Oregon are estimated to have spent 1.4 percent of their rebate, Minnesotans are estimated to have spent 35 percent, and Wisconsinites are estimated to have spent 83 percent. However, only the coefficient on the Minnesota amount is significant. The coefficient on rebates in Connecticut, on the other hand, is negative and significant, with an implausible magnitude. In the dummy variable specifications, rebates in the same three states are estimated to have increased expenditures, though all coefficients are insignificant. Again, the coefficient for Connecticut is negative and significant. (28)

Considering the components of expenditure, in the top panel, in Oregon it appears that durables increased during the quarter of receipt, although this increase is insignificant. In Minnesota, nondurables significantly increased with the amount of the rebate and durables significantly decreased, but these results are not robust to changing the independent variable to a dummy for rebate receipt. In Wisconsin, the coefficients on durables and nondurables are positive in both specifications, and the coefficient on durables is significant at the 10 percent level.

Some of the results above suggest that individuals increased expenditures in response to the receipt of a rebate check, even though they had knowledge of this receipt at least two months prior to the checks being mailed, which is inconsistent with a simple permanent income/lifecycle hypothesis model. Others suggested an insignificant response to receipt of a rebate. However, either of these results could possibly be rationalized by appealing to credit constraints. For example, it may be that most individuals were able to smooth their consumption over quarters in which the rebate check was received, so for them the coefficient on rebate receipt is zero. However, some credit-constrained individuals desired to consume more than they could have absent receipt of the rebate check, so the receipt of the rebate check allowed them to spend an amount that would be closer to their unconstrained optimal amount. To probe this hypothesis, I follow Zeldes (1989) and Souleles (1999), and cut the sample to include only those who had low asset to income ratios. I ran this estimation both including those with asset to income ratios less than 0.15, and with only those with ratios less than 0.25. Since the results from the two specifications are very similar, I present only the results from the estimation including those with asset to income ratios less than 0.25 in the top panel of Table 9.

These results, however, provide little evidence either for or against a credit constraints story. The estimates in both the rebate amount and rebate dummy specifications are insignificant, but the standard errors are substantial. As a result, one cannot rule out that credit-constrained individuals responded with larger or smaller amounts of spending than the sample as a whole. (29)

However, in Table 9, I also cut the sample according to marital status and a stark result emerges. In the middle two panels, where only married respondents are included, the results are similar to the base specification, with an insignificant response of total expenditure and a significant response estimated only in the apparel regression. However, when single respondents are used in the bottom two panels, the marginal propensity to consumer out of the rebate is estimated to be in excess of one at 1.405, and split almost evenly between durables and nondurables, with coefficients of 0.769 and 0.635. Entertainment spending in particular is estimated to increase by 32 percent of the amount of the rebate in the quarter of receipt. These results suggest that, to the extent that being single is correlated with credit constraints, such constraints may be a cause of the response. On the other hand, it could also be that single individuals simply have shorter planning horizons than do married people, or have a higher marginal propensity to consume out of current income for some other reason. (30)

RESPONSE TO ANNOUNCEMENT OF REBATE

The results above are suggestive that, consistent with results found elsewhere in the literature, individuals may have increased spending in response to the receipt of a rebate check, with the results from the base specification implying an increase of about a fourth of the size of the rebate.

Given this result, one might wonder whether individuals responded to the announcement of the rebate as well. In the recent studies of tax rebates, such an effect could not be estimated, as there was no variation in the timing of announcement across individuals within the United States. However, as can be seen above in Table 1, the time of the signing of the enabling legislation for the tax rebates or the announcement of the amounts that the rebate checks would take varied widely across states, typically occurring two to three months prior to the receipt of the check.

When one adds in the information that households might have received during debates over this legislation, it is clear that households' information sets regarding their flow of future income changed before the actual receipt of the check. Thus, households may have reacted not only to the receipt of their rebate check, but also to the receipt of new information about their income stream. As such, households may have changed their expenditure path when they learned that they would receive a rebate of a certain size.

Unfortunately, it is difficult to pin down exactly when households' beliefs changed. For example, it could have been back when discussions about rebates initially started or when a certain political party increased their representation in the legislature. Thus, many changes in information are either unidentifiable or unobservable. However, one might expect the largest change in beliefs about how much of a rebate a household would receive would occur when enabling legislation is signed and/or rebate amounts are announced. Since these dates are observable, one can estimate whether households changed their expenditure path in response to such an announcement. (31)

To examine this empirically, I rerun the regressions above, but add to the specification a variable that represents the announced amount of the rebate that the household will receive in a subsequent month. The estimation equation, then, is of the form

[4] [[DELTA]C.sub.i,t] = [[alpha].sub.0] + [[alpha].sub.[alpha]][announcedrebate.sub.ist] +[[alpha].sub.r][rebate.sub.ist] + [summation over (t)][[alpha].sub.t][d.sub.t] +[summation over (s)][[alpha].sub.s][d.sub.s] + [a.sub.z][Z.sub.i] + [[epsilon].sub.it],

where announced [rebate.sub.ist] denotes the rebate amount anticipated to be received.

For some individuals, the respondent-quarter of announcement is the same as the respondent-quarter of receipt. For example, individuals interviewed in Connecticut in August, 1998, reported their expenditure over May, June and July of that year, which covers both the month of announcement and the quarter of receipt. However, for most individuals in the sample, the respondent-quarter of announcement and respondent-quarter of receipt will be two distinct quarters, and it is from these individuals' observations that separate announcement and receipt effects can be identified.

These results are presented in Table 10. Comparing this table to Tables 6 and 7, the received rebate results are essentially the same. Turning to the announced rebate coefficients, the coefficient in the rebate amount specification is very close to zero and is insignificant, suggesting that total expenditures did not increase in response to the announcement of the rebate. Looking at the subcomponents of expenditures yields an interesting pattern. The coefficient on the announced rebate amount in the durables equation is positive and significant, while the coefficient in the nondurables equation is negative. This suggests that households may have shifted some expenditures to purchase durable goods in anticipation of the rebate.

Looking at the dummy variable specification in the bottom panel, similar results are found. The coefficient on the announced rebate variable in the total expenditures specification is negative but small and insignificant. However, the coefficient on durable expenditures is positive, suggesting that individuals increased spending on durable goods by about $73 in the quarter that the rebate was announced, with a shift in expenditures away from nondurables. (32)

Overall then, there is little evidence that individuals increased total expenditures in response to the announcement of rebates, but some evidence that the composition of expenditures may have changed in response to announcements of the rebates.