The consumption effects associated with filing for personal bankruptcy.

1. Introduction

Personal bankruptcy in the United States is part of the social-insurance structure provided by the federal and state governments. Other programs in this network include, but are not limited to, Unemployment Insurance (UI), Temporary Assistance to Needy Families (TANF), and food stamps. While these latter three programs are targeted to specific groups, the recently unemployed for UI and low-income households for TANF and food stamps, personal bankruptcy is not. Much like these other programs, though, bankruptcy reduces income uncertainty and reduces wealth uncertainty.

Surprisingly little work has been done on the direct benefits to filing for bankruptcy. This is in contrast with the growing literature on the costs of bankruptcy. (1) Fay, Hurst, and White (2002) estimate the financial benefit to filing for bankruptcy, which equals the debt forgiven minus any lost assets. This measure is a good predictor of whether the household files for bankruptcy, but it is not clear that the financial benefit is the true measure of the welfare gain from bankruptcy. In fact, the financial benefit is most likely an overestimate of the household's gain in the year it files for bankruptcy. Unlike UI or TANF, bankruptcy provides no increase in money income for that period but rather increases net-of-debt income. This change is felt over the household's lifetime, and fliers will most likely smooth this effect over multiple years because households cannot file for bankruptcy again for six years. Further, using the financial benefit makes it difficult to compare results against the estimates of the consumption benefits from other programs that provide insurance (see Gruber 2000 for TANF, Browning and Crossley 2001 for UI, and Gunderson and Ziliak 2003 for food stamps).

For these reasons, an analysis of bankruptcy's effect on the household's consumption profile may be more relevant. Bankruptcy provides benefits that may affect consumption growth through two channels. First, forgiving unsecured debt provides the household with greater net-of-debt income. In the face of idiosyncratic shocks such as job loss or divorce, protecting net-of-debt income allows a household to stabilize consumption in lieu of debt servicing. In this way, households realize a consumption insurance benefit in the year they file for bankruptcy. (2) Second, bankruptcy may prevent the liquidation of certain assets, thereby insuring wealth and the flow of consumption services from these assets.

While other insurance programs also stabilize consumption, the unique aspects of bankruptcy are that it protects assets and that there are no eligibility requirements like there are for UI, TANF, and food stamps. As a result, bankruptcy could provide a consumption benefit that can be cashed in at a time of the household's choosing. In these ways, bankruptcy may be the most flexible insurance program, provided that it generates similar consumption benefits to other transfer programs.

In this article, we use the Panel Study of Income Dynamics (PSID; Survey Research Center 2002) to estimate the consumption insurance benefits of filing for personal bankruptcy. First, we estimate the responsiveness of food-consumption growth to the financial benefit associated with bankruptcy. This follows the literature that examines the mechanisms that provide partial insurance, such as the progressive income tax structure (Kniesner and Ziliak 2002), food stamps (Blundell and Pistaferri 2003; Gunderson and Ziliak 2003), unemployment insurance (Browning and Crossley 2001), and Medicaid (Gruber and Yelowitz 1999). The results in this article suggest that a 10% increase in the financial benefit increases food consumption growth by 0.9%, which is smaller than the effect of an increase in UI or TANF benefits. For UI and TANF, a 10% increase in the dollar benefits increases food consumption growth by approximately 2-3% (Gruber 1997, 2000).

Then we estimate how the choice of bankruptcy chapter that the household files under affects the growth in food consumption. The results indicate that (i) households that filed for bankruptcy under chapter 7 of the bankruptcy code experienced a 13% increase in food-consumption growth in the year they filed for bankruptcy and (ii) households that filed under chapter 13 experienced an insignificant change to their food consumption.

The remainder of the article is organized as follows. The next section discusses the relevant features of bankruptcy law crucial to the estimation, with an emphasis on the differences between chapter 7 and chapter 13. The third section clarifies the data, uses the previous literature on consumption insurance to motivate the empirical model, and addresses some of the concerns surrounding estimation. Section 4 presents the empirical results. Section 5 concludes with a discussion of our results.

2. Personal Bankruptcy Law and Process

An individual filing for bankruptcy chooses between chapter 7 and chapter 13. Of the 1.62 million personal bankruptcy filings in 2003, 1.15 million (71.2%) were filed under chapter 7. (3) Under chapter 7, the debtor forfeits all assets exceeding the bankruptcy exemption levels set by the state. A bankruptcy trustee sells the nonexempt assets and distributes the money to the creditors. In return, a chapter 7 filing discharges most unsecured debts. Secured debts are discharged only if the debtor forfeits the collateral.

The other option when filing for personal bankruptcy is chapter 13. Filers do not turn over assets to the bankruptcy court but instead propose a repayment plan for a portion of the outstanding debts. Essentially, chapter 13 fliers put themselves on a strict budget through this repayment plan. If the repayment plan is successfully completed, the filer receives a discharge of some of the unsecured debts. Because the discharge of debts does not occur until the repayment plan is completed, the full benefit of chapter 13 is not realized until then. ff the plan is not completed, the filer does not receive the discharge. The largest advantage to filing under chapter 13 is that fliers do not have to turn over any assets to the Bankruptcy Court, while chapter 7 fliers lose nonexempt assets.

Federal law gives bankruptcy judges the discretion to dismiss any bankruptcy filing that constitutes a substantial abuse of the process (Bankruptcy Reform Act of 1984). Dismissal of a bankruptcy filing request requires clear evidence of fraud on the flier's behalf, such as a transfer of property while the filer retains the property or transfers executed at much less than face value. In addition, the antifraud provision prevents accumulation of significant debt in the 12 months before filing. This prevents the household from using debt to fund current consumption in the months before filing. We argue that this provision insures that the financial benefit is exogenous to consumption growth, which becomes important in identifying the consumption insuring effect of bankruptcy.

3. Data and Empirical Model

Data

For all empirical work in the article, we use the PSID, which is a longitudinal household survey that began in 1968. The PSID follows 8000 U.S. households and collects economic, health, and social behavior data from each household and each member of the household. In 1996, the PSID included an extra set of questions on personal bankruptcy; households were asked whether they ever filed for bankruptcy. If the household filed for bankruptcy, the PSID collected additional information including the year and the bankruptcy chapter. Because the bankruptcy information was only asked in the 1996 wave, the household must have been interviewed in 1996 to be included in our sample. Because the 1996 survey took place during the spring of 1996 rather than at the end of the year, some households may have filed for bankruptcy in 1996 but after they were surveyed. Therefore, the last year used in our sample is 1995. If we included the 1996 wave in the panel specifications, we may incorrectly identify some bankruptcy fliers as households that did not file. (4)

The other important aspect of the PSID data is the expenditure questions. The PSID includes information regarding food eaten out, food eaten at home, and food-stamp usage. We sum these components to create one food-expenditure variable. Because the 1988 and 1989 waves of the PSID do not include the food questions, 1990 is the first year in our sample. Unfortunately, the PSID does not include a more comprehensive measure of consumption. (5) However, the PSID is the only data set that has the necessary bankruptcy information and longitudinal data on at least one measure of consumption. The focus on food consumption is not just a data constraint, however. From a policy perspective, social insurance programs should be designed to maintain reasonable standards of living in the face of idiosyncratic shocks to the household. While bankruptcy is not designed to target food consumption directly, it seems reasonable to evaluate this program's effect on one of the basic elements of the household's utility function. Further, if we find that bankruptcy affects an income inelastic good such as food, then we would expect that it would affect broader measures of consumption as well.

There are other restrictions placed on the sample. First, if the household had imputed values for food or income, it was excluded from our sample. Second, the household is required to have valid values for the independent variables in our specification. If there was a change in household composition, such as a divorce, during the sample period, the household is included in the sample. Previous studies that examine consumption changes in the PSID, such as Zeldes (1989), exclude these types of households. We include these households and control for changes in family size and marital status. If we excluded households that changed composition, we would lose potentially interesting households. Further, the number of bankruptcy filers would decrease because a number of bankruptcy filers divorced in the years around filing for bankruptcy. With these restrictions, the sample has 5589 unique households. Because we use an unbalanced panel between 1991 and 1995, the sample used has 21,397 observations.

Empirical Model

The theoretical grounding for the empirical model comes from the literature on consumption insurance (Cochrane 1991; Deaton 1992). In the standard model, a social planner allocates consumption under uncertainty by maximizing the weighted sum of individual household utilities subject to a resource constraint. With a utility function that displays constant relative risk aversion, the key implication of this literature is that the marginal utility of consumption is proportional across individuals (Deaton 1992). If there is full insurance, idiosyncratic shocks to income should not affect consumption growth (Cochrane 1991). Most empirical research on consumption insurance finds evidence inconsistent with full insurance (Cochrane 1991; Attanasio and Davis 1996; Hayashi, Altonji, and Kotlikoff 1996). Rather than testing for full insurance, this article cedes the point that there is not full insurance and, instead, we examine whether bankruptcy helps to provide partial consumption insurance through the dissolution of unsecured debt and the protection of assets.

The model to be estimated is a linearized consumption Euler equation of the form

(1) [DELTA]ln[C.sub.it] = [alpha]([benefit.sub.it]) + [eta][DELTA]ln([y.sub.it]) + [DELTA][X.sub.it][[beta].sub.1] + [Z.sub.i][[beta].sub.2] + [[tau].sub.t] + [DELTA][[epsilon].sub.it],

where [C.sub.it] represents the food consumption for household i in year t. [Benefit.sub.it] equals the financial benefit to filing for bankruptcy (described in detail below); [X.sub.it] represents a set of household characteristics and state-level characteristics that vary over time; [Z.sub.i] represents a set of household characteristics and state-level variables that are time invariant; [[tau].sub.t] represents time dummy variables that capture aggregate resources and the effects of changes in real interest rates; and, [[epsilon].sub.it] is a random error term. (6) By estimating the consumption growth equation rather than a model in levels, any unobserved household fixed effect is removed. Finally, because we have multiple observations for each household, the standard errors are calculated to allow for both within-household serial correlation and heteroskedasticity.

For the dependent variable, we use total food expenditures. Income, [y.sub.it], includes all sources of income including income from social programs such as UI. The variables in [X.sub.it] are family size, marital status, log family food needs, state unemployment rate, and state per capita income. The food needs variable comes from the PSID and is derived from the Census food-needs standards. Food needs varies by age and gender of the individuals in the household. The variables in [Z.sub.i] are age of the household head, race, gender, and education of the household head. (7)

The key regressor in Equation 1 is benefit, which is the financial benefit to filing. The financial benefit to filing is calculated as the debt forgiven in bankruptcy minus the assets lost (if any). Mathematically,

benefit = max{debt - max[(wealth - exemptions), 0], 0}.

Debt represents the unsecured debts of the bankruptcy filer in the year they filed. Wealth equals the household's nonhousing wealth plus its home equity. The exemptions are the relevant state bankruptcy exemptions, which vary not only by the state of residence and year but also by marital status. Following Fay, Hurst, and White (2002), the homestead exemption equals zero for households that rent. For homeowners that live in a state that allows an unlimited homestead exemption, this exemption is set equal to the market value of the home. Some states allow a filer to choose either the state or federal exemptions. In these states, the federal exemptions are used if they are higher. The debt and wealth variables come from the PSID. The exemption data come from Elias, Renauer, and Leonard (1995). (8)

The financial benefit provides a good estimate of the increase in the net-of-debt income that the household receives from filing. Having debt discharged frees up income that would have been used to repay loans. Conversely, losing assets may mean that the household has to spend additional income to make up for the lost flow of services. There may be some costs and benefits associated with bankruptcy that are not captured by benefit, such as lawyer fees and court filing fees, but those costs are small in relative terms and are excluded in the calculation. In an alternative specification of Equation 1, we replace benefit with a dummy variable equal to one in the year the household filed for bankruptcy and zero otherwise.

Estimation Issues

Several potential concerns with regard to estimation need to be addressed. First, there are two issues with the bankruptcy data in the PSID. One, the PSID bankruptcy rate is about half of the national filing rate. This means that there may be households in this sample that filed for bankruptcy but are not coded as such. If this underreporting is random, the coefficient on benefit would be biased toward zero (attenuation bias). Second, even if the bankruptcy rate is correct, there is still only a small percentage of households that filed for bankruptcy in the sample years (0.7%). We could add more years of data, but this solution has its shortcomings. One shortcoming is that the dependent variable, food expenditures, is not available in 1988 or 1989. Also, we are concerned that households are more likely to misreport the year they filed the earlier they filed, which produces measurement error in our main regressor. Finally, because the filing rate was much lower in the 1980s than in the 1990s, including earlier years adds less than 15 filers per year before 1988. Our best attempt to address this concern is to present the results of multiple specifications to help show that our findings are not likely to be spurious.

Another concern is that the regressor of interest (benefit in the first specification and the bankruptcy dummy variable in the second specification) may be endogenous to consumption growth, which is always a potential problem in studies of this type. While there is evidence to suggest that the timing of filing for bankruptcy is strategic and that households attempt to choose an optimal filing date (see Fay, Hurst, and White 2002), this does not necessarily imply that benefit is endogenous to consumption growth. When Fay, Hurst, and White (2002) discuss the strategic behavior of filers, it is to suggest that households are more likely to file when the financial benefit is highest, not that households run up debt or sell off nonexempt property in the year before filing. (9) Indeed, the antifraud provision in the bankruptcy law prohibits this type of activity in the 12 months before filing. Therefore, the benefit regressor is likely to be exogenous to decisions made in the current period. (10)

To be thorough, we address the potential endogeneity using a two-stage approach (see Wooldridge 2002, p. 118-22). A reduced form equation for benefit is first estimated. The residuals from this first stage estimation are appended to Equation 1. The test for exogeneity is then whether the coefficient on the residuals is statistically significant or not. If the coefficient is significant, then we reject the hypothesis that benefit is exogenous and must correct for this. In the first stage, we estimate

[benefit.sub.it] = [w.sub.it][delta] + [[alpha].sub.i] + [u.sub.it].

The vector of regressors ([w.sub.it]) includes the independent variables in Equation 1 and the instruments. The [[alpha].sub.i] is a household fixed effect.

We instrument for benefit using variables that are correlated with the decision to file for bankruptcy. Because benefit can be thought of as an interaction between the bankruptcy decision and the level of the benefit, we use instruments from the literature on the decision to file for bankruptcy (Elul and Subramanian 2002; Fay, Hurst, and White 2002). The instruments include the following: lagged income, whether the household head is currently unemployed, whether the household owns a business, the lag of the filing rate for the household's state of residence, the state unemployment rate, the standard deviation of state gross per capita income, and state fixed effects.

4. Results

Consumption Insurance Estimates

Table 1 presents statistics on food consumption growth for filers and nonfilers. For chapter 7 filers, column 2 shows that food consumption grew by 11.7% in the year they filed for bankruptcy. (11) Alternatively, food consumption declined by 0.5% for chapter 13 filers. Nonfilers in 1995 experienced negative growth in consumption (Table 1). Nonfilers experienced negative growth in food consumption in 1991 as well but had positive growth in 1992-1994.

Figure 1 shows the unconditioned response of food consumption growth for all bankruptcy filers around the time of filing. In the figure, year t represents the year the household filed for bankruptcy. The pattern that emerges illustrates a jump in consumption in the year the household files for bankruptcy. We take this as preliminary evidence that filing for bankruptcy provides contemporaneous consumption insurance. In the year after bankruptcy, though, there is a fall in food consumption growth on average. This fall in consumption may be a little misleading, though. In part, food consumption growth in the two years after bankruptcy may just be a reversion back to the mean. For nonfilers, average food consumption over all five years equals 0.7% per year. Filing for bankruptcy provides above-average food consumption growth, but the filers revert back to the average growth in the years after filing.

[FIGURE 1 OMITTED]

Next, we present the results of estimates of Equation 1 in Table 2. Before discussing the coefficient of interest--the benefit coefficient--the results of the test of endogeneity is discussed. To test for the endogeneity of the bankruptcy decision and benefit, we follow the two-stage procedure described above and include the residual from the first stage equation for benefit. (12) Importantly, the coefficient on the residual is statistically insignificant in Table 2, suggesting that we fail to reject the hypothesis that the benefit is exogenous to consumption growth. (13)

Given that benefit is exogenous, we can proceed with estimating Equation 1 using ordinary least squares (OLS). We identify the consumption insurance effect of bankruptcy by using the variation in the benefit across states of residence and differences in the household's debt and assets. The coefficient estimate reported in Table 2 suggests that a $1000 increase in the benefit to filing leads to a 0.88% increase in food consumption. Because the exemption levels are an important component in the variation of financial benefit, we can use the estimate in Table 2 to calculate the effects of various changes in the exemption levels on the consumption changes for the average filing household.

First, consider a 10% increase in all exemptions. This change increases the financial benefit by $68, on average, or a 1.3% increase. (14) This translates into a 0.08% increase in food consumption for the average filing household, which is relatively small. Second, we examine the effect of eliminating the unlimited homestead exemption and capping it at $100,000, which is one of several proposals currently under consideration in Congress. In our sample, the average financial benefit level falls $500 as a result. This decrease in the exemptions reduces the growth

in food consumption associated with bankruptcy by 0.5%.

To put these marginal effects in context of other social insurance programs, Gruber (2000) finds that a 10% increase in the UI replacement rate increases food consumption by 2.5%. For welfare, Gruber (1997) finds that a 10% increase in welfare benefits increases food consumption by 2.8%. In comparison to these cash transfer programs, the effect of a 10% increase in the bankruptcy exemptions is relatively small.

There may be a few reasons for this relatively small effect of bankruptcy. First, changes in the financial benefit affect net-of-debt income, not money income like UI or welfare. Second, the change in net-of-debt income is over the household's lifetime, not just one year. For example, suppose the household had $10,000 in credit-card-debt forgiven. In the year they actually filed, it is unlikely that they were planning on paying off all $10,000. More likely, the household planned on paying off a portion, which means that the current year change in net-of-debt income may be significantly less than $10,000. This means that the financial benefit is an overestimate of the increase in the current year net-of-debt income. Finally, if households perceive bankruptcy as a temporary shock, then the household would only alter consumption by a fraction of the increase in net-of-debt income.

We report the estimates of other coefficients in Table 2 to provide a sense of the appropriateness of the specification. These coefficients generally conform to expectations. The biggest surprise is the income coefficient, which is positive but insignificant. Because the income variable includes components that are known to provide consumption insurance (e.g., UI and TANF benefits), we would expect that this coefficient would be positive and significant. The coefficient on the change in family size is positive and significant, and the coefficient on the change in the census needs is positive but insignificant. However, these two variables are correlated ([rho] = 0.40), which may explain why the coefficient on the growth in census needs is statistically insignificant. (15)

There are several concerns regarding the validity of our estimate of the consumption response. First, because the benefit is a function of wealth and debt, there is concern that the results in Table 2 are due to wealth and not filing for bankruptcy. However, if these results are driven by some nonlinear wealth effect, then we would expect that the benefit to filing would be a significant predictor of consumption growth for nonfilers as well. In the second column of Table 2, we run the same regression but exclude fliers and set the benefit for nonfilers equal to their calculated benefit and not to zero as before. In this regression, the coefficient on the benefit is the wrong sign and insignificant.

Another way to address this issue is to separate the benefit into its two components: the debt forgiven and the nonexempt assets. We expect the coefficient on debt to be positive, indicating that increasing net-of-debt income increases the household's ability to insure consumption. Conversely, the coefficient on nonexempt assets is expected to be negative because it increases the number of assets lost when filing for bankruptcy and hinders the ability to insure consumption. In row A of Table 3, we present the results for the full sample. The coefficient on debt is positive and significant, while the coefficient on nonexempt assets is negative but statistically insignificant.

We conduct the same analysis using the sample of nonfilers, presented in row B of Table 3. As was the case for the benefit in Table 2, the coefficients on the two components of benefit are not statistically significant for this sample of nonfilers. Combined with the results from Table 2, the results in Table 3 suggest that the insurance effect of bankruptcy is for bankruptcy fliers only and is not driven by a wealth effect. The benefits of bankruptcy are only realized when the household actually files for bankruptcy.

A second issue is that the consumption benefit may depend on the bankruptcy chapter (7 or 13). As described above, chapter 7 is more generous with respect to current period decision because chapter 7 creates a contemporaneous increase in net-of-debt income. Alternatively, chapter 13 provides for a much smaller current period increase in net-of-debt income, but chapter 13 does a better job of protecting assets. We examine the possibility that filing for bankruptcy provides an insurance effect for consumption that differs by chapter filed in row C of Table 3. (16) As expected, the coefficient for chapter 7 fliers is positive and significant while the coefficient on chapter 13 is negative but insignificant. In Table 1, we show that consumption growth is negative for chapter 13 fliers in the year they filed, which matches the coefficient on chapter 13 fliers. However, the coefficient is statistically insignificant in Table 3, maybe because there are only 42 chapter 13 fliers in the sample.

A final concern regarding the validity of our estimates is that the control group in Table 2 includes households that would not financially benefit from filing for bankruptcy. This differs from previous work on the consumption effects of other government programs. There, researchers generally create a control group that consists of households that are eligible for the program but chose not to participate. As a result, a clear estimate of the average treatment effect of program participation is produced. However, given the absence of eligibility requirements for bankruptcy, we do not have the ability to construct a sample of those eligible for bankruptcy but that chose not to file.

Alternatively, we can utilize the information contained in the benefit variable as an indication of the household's ex ante likelihood of filing. Households with positive values for benefit at time t are more likely to file for bankruptcy. Row D in Table 3 presents an estimate of the impact of changes in benefit on the growth of food consumption for fliers, using a sample of nonfilers that have a positive value for the benefit variable as a control group. The insurance benefit (0.0084) in row D is quite similar to that provided in Table 2 (0.0088) and is significant at the 5% level. To address the concern of heterogeneity across chapter choice, we also interacted bankruptcy chapter dummy variables with the benefit. The results are not presented in a table, but they match the pattern from row C of Table 3, with only a chapter 7 filing eliciting a statistically significant response in food consumption.

The Participation Effect on Consumption Growth

Next, we examine whether bankruptcy provides an insurance benefit in a different manner by using a participation dummy variable rather than exploiting the variation in the benefit. We estimate Equation 1 using a dummy variable indicating whether the household filed for bankruptcy in year t in place of the benefit. We present the results in Table 4, using the same subsamples as in Tables 2 and 3. Across all samples, the results match those seen above. When ignoring chapter choice, the effect of filing for bankruptcy on consumption growth is positive and significant. When splitting fliers by chapter choice, the chapter 7 coefficient is positive and significant while the chapter 13 coefficient is statistically insignificant.

Using the same two-stage test as above, concerns about the endogeneity of the filing variables are addressed by the coefficients on the residuals. The coefficients used to obtain the residuals are presented in the second column of Appendix B. In Table 4, we test whether the bankruptcy decision is exogenous to consumption growth by examining whether the coefficients on the residuals are individually and jointly significant or not. Ignoring chapter choice, the coefficient on the residual is statistically insignificant. When separating by chapter choice, there is evidence that the chapter 13 variable is endogenous, with the chapter 13 residual significant at the 10% level. Using an F-test, however, we fail to reject the hypothesis that both the chapter 7 and chapter 13 residuals equal zero, suggesting that the filing decision is exogenous to consumption growth.

Overall, the results in Table 4 confirm what was found earlier. First, the results suggest that the bankruptcy decision is exogenous to food consumption growth. Second, filing for bankruptcy provides a consumption insuring benefit, and that this effect is seen for chapter 7 fliers and not chapter 13 fliers. The results in Table 4 suggest that filing for chapter 7 bankruptcy increases food consumption growth by 13% in the year of the filing.

Further Tests for Robustness of Results to Changes in the Dependent Variable

Finally, we present additional results to test the robustness of our results to different measures of consumption. To begin, we split the food consumption variable into the two parts provided by the PSID: food consumed at home, which includes food stamp expenditures; and food consumed away from home. This will tell us whether bankruptcy affects the distribution of food consumption between food at home and food away from home. Why might we be interested in this difference? As mentioned previously, social insurance programs are designed to maintain standards of living for households. If bankruptcy leads to fliers going to restaurants more often and/or spending more at restaurants, policymakers may likely question the impact of the program.

The results using food at home and food away separately are presented in Table 5. In the first row, the results suggest that a $1000 increase in the financial benefit increases the growth of food at home consumption by 0.1%. In the second row, the effect of the financial benefit on the growth of food away from home is positive but statistically insignificant. When we use the dummy variable rather than the financial benefit, the results are the same. These results suggest that filing for bankruptcy affects food consumed at home, not food consumed away from home. Thus, we may not need to be worried that bankruptcy fliers are using bankruptcy to spend large amounts of additional money at restaurants. We of course do not know whether the fliers are spending more money on food at home because they were consuming less than optimal before bankruptcy or whether they are spending more money on food at home by consuming products that are more expensive.

The next consumption issue is whether bankruptcy increases consumption other than or in addition to food consumption. It would not be surprising if filing for bankruptcy could have an effect on broader measures of consumption. By freeing up income that was used to service debt, the household could use this income for a variety of purchases, depending on the underlying preferences. Unfortunately, the PSID does not have a broader measure of consumption available in our sample years. Previous research addressed this problem with the PSID by imputing consumption. We follow this research by imputing total consumption using Consumer Expenditure Survey (CE) data, following the methodology first proposed in Skinner (1987). We use coefficients, however, from Fisher and Johnson (2003). The methodology and actual coefficients used to impute total consumption are presented in the Appendix.

The last row in Table 5 estimates the effect of the benefit on imputed total consumption. The estimated response is positive but smaller in magnitude than previous estimates and insignificant at conventional levels. When using the bankruptcy dummy variable rather than the financial benefit, the results are the same: The coefficient on the dummy variable is positive but statistically insignificant. The fact that the coefficients on benefit and the bankruptcy dummy variable are positive is encouraging. This suggests that there may actually be a positive and statistically significant effect, but we are unable to estimate it with adequate precision possibly because we are using an imputed and therefore noisy measure of total consumption. It is possible that we may be expecting too much from an imputed measure of consumption. On the other hand, perhaps bankruptcy does not affect a broader measure of consumption. This seems unlikely, though. Bankruptcy is not targeted at food consumption in the way food stamps are, and bankruptcy is not targeted at the poor (a group likely to need money for food) like TANF. Therefore, we should expect that bankruptcy would affect more than just food consumption. Future work should examine the effects of filing on broader measures of consumption.

5. Conclusion

Does bankruptcy provide a form of consumption insurance similar to other social insurance programs? Using information about bankruptcy contained in the PSID, we provide the first estimates of consumption insurance effects precipitated from filing for personal bankruptcy. To do this, we exploited two types of variation in the data. The first is the variation in the financial benefit to filing for the household, which is driven by the household's debt and assets at the time of the bankruptcy filing. We find a small impact on food consumption growth associated with changes in financial benefit to filing. The second source of variation is the bankruptcy filing decision itself. We find that households that filed for chapter 7 bankruptcy have average food consumption approximately 13% higher than nonfilers, while filing for chapter 13 does not affect consumption growth.

These results generate a variety of public-policy implications. First, we find that the consumption benefit to filing is not affected that much by changes in the bankruptcy exemptions because the financial benefit does not change much when the exemptions change. Consequently, policymakers could tighten bankruptcy legislation by lowering existing exemption levels, and this would have little detrimental impact on the consumption benefits to the household. As an example, if Congress capped the homestead exemption at $100,000 or $125,000, our results suggest that this decrease in the homestead exemption would reduce the growth in food consumption associated with bankruptcy by approximately 0.5%.

Second, the differing benefit by bankruptcy chapter is an important result to document. For the last few years, Congress has been debating legislation that would essentially force households that want to file for bankruptcy to use chapter 13 if they are able to repay at least 25% of their unsecured debts. While the full implications of this change are unknown, the results here indicate that chapter 13 does not provide a contemporaneous consumption benefit to fliers. Therefore, if one goal of bankruptcy is to provide consumption insurance, our results suggest that this potential change in the law would decrease the insuring benefits of bankruptcy. Consequently, faced with the option of filing under 13 or not filing, households that would have filed under chapter 7 under existing law may not file under chapter 13. This would obviously reduce filing rates, which may be the underlying goal of the new legislation.

This is not the first article to suggest that bankruptcy is part of the social insurance system. Posner (1995) argues that bankruptcy is social insurance for the nonpoor, and Fisher (2001) finds evidence that bankruptcy and UI are substitutes. However, if the goal of social insurance programs is to provide consumption insurance, the results presented here suggest that bankruptcy should not be seen as comparable. The consumption benefit associated with bankruptcy is small relative to estimates of benefits from other programs, which should not be surprising. While UI and TANF provide direct cash transfers, the benefits of bankruptcy do not translate into consumption benefits as cleanly. A marginal increase in the benefit to filing implies that the household either has additional dischargeable debt or is allowed to keep more property. The discharge of debt can free up income for current consumption, but keeping additional property does not necessarily imply that the household increases consumption as we measure it.

Finally, while UI and TANF may be better suited to provide consumption insurance, bankruptcy is unique in that it protects assets from creditors and does not have strict income and/or asset restrictions in order for a potential filer to be eligible. As a result, bankruptcy can provide some level of consumption insurance to households that might not be eligible for cash-transfer programs. In this regard, bankruptcy should be recognized as part of the social insurance system. And bankruptcy, like these other programs, has social costs associated with it. For bankruptcy, Gropp, Scholz, and White (1997) find that living in a state with higher bankruptcy exemptions reduced the availability of credit and increased the interest rate on automobile loans for low-asset households. The cash benefits of UI and TANF decrease labor supply (Meyer 1990; Montgomery and Navin 2000). (17) Because there are benefits and costs to all of these programs, a social planner or policymaker can alter the mix of these programs to provide an adequate level of social insurance while trying to minimize the collective social costs of each. Future research can use the results presented here to analyze the optimal mix of these social insurance programs to better guide policy.

Appendix A: Data Description

The household level data come from the Panel Study of Income Dynamics,
using the waves from 1990 to 1996. The following table presents the
variable names as they are identified in the article and the
corresponding variable from the PSID.

Food consumption              Sum of food at home, food away, and food
                                stamps
Bankruptcy dummy              From 1996 wave, equals one in year filed
                                for bankruptcy
Bankruptcy chapter            Chapter of bankruptcy household used (7
                                or 13)
Financial benefit             Uses the 1989, 1994, and 1999 wealth
                                supplements. The variables are debt and
                                wealth2, which includes home equity.
Exemptions                    Benefit also includes the state
                                bankruptcy exemptions, which come from
                                Elias, Renauer, and Leonard (1995) and
                                other editions in this series.
Income                        Total family income
Age of household head         Age of head
Black, white, and Other       Race of head
Married                       Marital status
Divorced-widowed-separated    Marital status
Family size                   Family size
Children                      Number of children in the household
Adults                        Family size minus children
Head is male                  Gender of household head
Education dummies             Education of head
Own home                      Own, rent, or other
State of residence (and
  region)                     FIPS state code
Census needs                  Census needs standard
State income growth           Per capita income from Statistical
                                Abstract
Unemployment rate             Statistical Abstract
Market value of home          House value
Whether unemployed            Current employment status
Lagged state filing rate      American Bankruptcy Institute:
                                abiworld.org

For variables at the state level, most of these values are found in the
Statistical Abstract of the United States (various years). The
variables not from the Statistical Abstract are the number of
bankruptcy filings in each state and the state bankruptcy exemptions.

Appendix B

Fixed Effect Model for the Financial Benefit and the Likelihood of
Filing for Bankruptcy

                                              Financial Benefit

Dependent Variable                      Coefficient    Standard Error

Instruments
  Lagged income                            -4.754         13.753
  Whether unemployed                       25.507         24.275
  Own business                            -16.226         36.315
  Lagged state filing rate                140.616         82.915 *
  State unemployment rate                   0.0406         0.119
  Standard deviation of state output        1.1007         0.239 **
  State fixed effects                       Yes
Variables from consumption function
  Change in log income                     -1.932          9.321
  Age                                      -1.247          7.738
  Black                                  -305.591        432.063
  White                                  -102.445        281.291
  Married                                  -7.111         42.560
  Gender                                  -31.492         78.491
  High school dropout                     -18.546        196.022
  High school graduate                    -15.299        169.895
  Some college                            -15.049        173.913
  Change in family size                    -2.938         10.449
  Change in census needs                  -29.372         55.173
  Change in unemployment                  -30.933         78.621
  State income growth                    -727.779        722.433
  Constant                               5634.434        818.641
Household fixed effects                     Yes
Year fixed effects                          Yes

                                            Likelihood of Filing

Dependent Variable                      Coefficient    Standard Error

Instruments
  Lagged income                            0.002          0.001 **
  Whether unemployed                       0.002          0.002
  Own business                            -0.004          0.003
  Lagged state filing rate                -0.408          0.601
  State unemployment rate                 -0.001          0.001
  Standard deviation of state output      -0.0001         0.0001
  State fixed effects                       Yes
Variables from consumption function
  Change in log income                     0.002          0.001 **
  Age                                      0.001          0.001 **
  Black                                   -0.169          0.031 **
  White                                   -0.045          0.020 **
  Married                                 -0.007          0.003
  Gender                                  -0.004          0.006
  High school dropout                      0.003          0.014
  High school graduate                    -0.004          0.012
  Some college                            -0.003          0.013
  Change in family size                   -0.001          0.002
  Change in census needs                  -0.001          0.004
  Change in unemployment                  -0.002          0.006
  State income growth                     -0.108          0.052 **
  Constant                                 0.071          0.045
Household fixed effects                     Yes
Year fixed effects                          Yes

The standard errors are robust to heteroskedasticity, using the Huber/
White procedure. This specification is used to calculate the residuals
used in Tables 2 and 4.

* Indicates that the coefficient is significant at the 10% level.

** Indicates that the coefficient is significant at the 5% level.

Appendix C: Imputing Consumption for the PSID Households

Because the PSID does not include a measure of total consumption, we impute consumption following the method in Fisher and Johnson (2003). This is an updated version of the imputation method first proposed in Skinner (1987). Due to data restrictions and certain aspects relevant to bankruptcy, we use a slightly different specification than the one Fisher and Johnson (2003) use. Specifically, the specification we use is

ln(total [consumption.sub.it]) = [beta]'[X.sub.it] + [gamma](food at [home.sub.it]) + [alpha](food eaten [out.sub.it]) + [e.sub.it].

Total consumption includes the amount that the consumer unit actually spends for current consumption plus the estimated service flows from homeownership and vehicles. This includes expenditures for food, housing, transportation, apparel, medical care, entertainment, and miscellaneous items for the consumer unit. Excluded are expenditures for pensions and social security, savings, life insurance, principal payments on mortgages, and gifts.

The vector of independent variables X includes the following: age, family size, region of residence, race, education, whether the household owns or rents, the market value of the home for homeowners, and yearly rent for renters. Table A.2 presents the coefficients.

The coefficients presented here differ from Fisher and Johnson (2003). First, while Fisher and Johnson present coefficients for four years, we only use data from 1994. We impute consumption from 1990 to 1995 using the 1994 coefficients. Second, our imputation uses fewer variables. Most important, the equation we use does not include number of vehicles or utility payments. In our sample, only the 1994 wave of the PSID includes both variables. Because bankruptcy can affect the number of vehicles a household owns, we do not want to use the 1994 values for number of vehicles or utilities in the other five years of data. Rather, we only use variables in the imputation that are available in all five years of the PSID.

Table C.1

Imputing Total Consumption Using the Consumer Expenditure (CE) Survey

In Total consumption    Coefficient    Standard error

Age 30-35                  0.0312          0.0144
Age 36-40                  0.0297          0.0155
Age 41-45                  0.0116          0.0159
Age 46-50                  0.0180          0.0169
Age 51-55                  0.0205          0.0183
Age 56-60                  0.0217          0.0191
Age 61-65                  0.0169          0.0198
Age 65 plus               -0.0458          0.0150
Northeast                 -0.0451          0.0120
Midwest                   -0.0244          0.0117
South                     -0.032           0.0111
Family size                0.0324          0.0034
White                      0.1231          0.0114
High school dropout       -0.2616          0.0113
High school graduate      -0.0906          0.0091
Market value of home     8.99E-07         4.28E-08
Own home                   0.6859          0.0147
Rent                     7.96E-05         2.20E-06
Food at home             8.65E-05         2.74E-06
Food eaten out           1.10E-04         2.93E-06
Constant                   8.8601          0.0202
Adjusted [R.sup.2]         0.732

Source: Coefficients provided by Fisher and Johnson (2003).

Table 1. Summary Statistics by Filing Status and Chapter Choice

                                   Nonfilers    Chapter 7   Chapter 13
                                     (1995)      Filers       Filers
                                      Mean        Mean         Mean

Income and consumption
  Income ($1000)                     49.941      35.579       33.982
  Change in log income                0.016       0.058       -0.028
  Change in log food consumption     -0.024       0.117       -0.005
Financial benefit
  Financial benefit ($1000)           2.509       5.490        6.207
Demographic characteristics
  Age                                41.722      35.845       34.279
  Black                               0.336       0.289        0.581
  White                               0.644       0.663        0.419
  Other                               0.020       0.048        0.000
  Single                              0.184       0.167        0.116
  Married                             0.597       0.512        0.628
  Divorced-widowed-separated          0.219       0.321        0.256
  Family size                         2.942       2.929        3.116
  Male                                0.753       0.774        0.651
  High school drop-out                0.209       0.200        0.071
  High school graduate                0.303       0.388        0.548
  Some college                        0.231       0.325        0.333
  College graduate                    0.258       0.088        0.024
  Own home                            0.567       0.369        0.465
Number of observations                4,998        89           42

Income, consumption, and the financial benefit are in 1996 dollars. The
change in log income equals ln([Y.sub.it]) -ln([Y.sub.it-1]), where
[Y.sub.it] equals the household's total income in year t. Similarly,
the change in log food consumption equals ln([C.sub.it-1])
- ln([C.sub.it-1]), where [C.sub.it] equals the household's food
consumption. For bankruptcy filers, we report statistics for the year
they filed for bankruptcy. For nonfilers, we only report the summary
statistics for one year, 1995. In the regressions, we use nonfilers
for 1991-1995.

Table 2. The Effect of Financial Benefit on Food Consumption Growth

                                   Filers and Nonfilers

                                Coefficient    Standard error

Financial benefit                 0.0088          0.0029 **
Change in log income              0.0011          0.0048
Age                              -0.0008          0.0002 **
Black                            -0.0153          0.0188
White                            -0.0189          0.0181
Married                           0.0001          0.0071
Gender                            0.0002          0.0082
High school dropout              -0.0004          0.0063
High school graduate             -0.0001          0.0049
Some college                      0.0009          0.0054
Change in family size             0.1217          0.0083 **
Change in census needs            0.0340          0.0385
Change in unemployment rate       0.0254          0.0321
State income growth               0.2485          0.2715
Year 1992                         0.0425          0.0156 **
Year 1993                         0.0249          0.0115 **
Year 1994                         0.0076          0.0130
Year 1995                        -0.0274          0.0142 *
Constant                          0.0476          0.0232
Financial benefit residual       1.33E-05        8.92E-06
Observations                      21,397

                                      Nonfilers Only

                                Coefficient    Standard error

Financial benefit                -2.10E-04        1.57E-03
Change in log income              0.0011          0.0048
Age                              -0.0008          0.0002 **
Black                            -0.0144          0.0181
White                            -0.0160          0.0177
Married                           0.0007          0.0072
Gender                           -0.0014          0.0082
High school dropout              -0.0021          0.0063
High school graduate             -0.0018          0.0050
Some college                     -0.0006          0.0054
Change in family size             0.1221          0.0083 **
Change in census needs            0.0345          0.0388
Change in unemployment rate       0.0294          0.0319
State income growth               0.2368          0.2720
Year 1992                         0.0440          0.0157 **
Year 1993                         0.0242          0.0115 **
Year 1994                         0.0102          0.0131
Year 1995                        -0.0256          0.0143 *
Constant                          0.0483          0.0219
Financial benefit residual          --              --
Observations                      21,266

The standard errors are robust to heteroskedasticity, using the
Huber/White procedure. The dependent variable is the change in log
food expenditures. In the first column of results, the financial
benefit to filing variable used in the specification equals zero
for all nonfilers. In the results restricted to nonfilers, the
financial benefit equals the observed financial benefit for all
households.

* Indicates that the coefficient is significant at the 10% level.

** Indicates that the coefficient is significant at the 5% level.

Table 3. Robustness Tests

                                      Financial            Nonexempt
                                    Benefit/1000          Assets/1000
Sample                            (Standard Error)     (Standard Error)

A
  Full sample (n = 21,397)               --            -0.0006 (0.0014)
B
  Nonfilers only (n = 21,266)            --             0.0003 (0.0005)
C
  Full sample (n = 21,389
  Chapter 7 filers               -0.0075 (0.0033) **          --
  Chapter 13 filers               0.0084 (0.0058)             --
D
  Financial benefit > 0                                       --
    (n = 6923)                   0.0084 (0.0029) **

                                 Debt Forgiven/1000
Sample                            (Standard Error)

A
  Full sample (n = 21,397)        0.0072 (0.0032) **
B
  Nonfilers only (n = 21,266)    -0.0009 (0.0006)
C
  Full sample (n = 21,389
  Chapter 7 filers                       --
  Chapter 13 filers                      --
D
  Financial benefit > 0
    (n = 6923)                           --

Robust standard errors are shown in parentheses. The specification
includes all variables shown in Table 2 as well. In row D, the
financial benefit for nonfilers is set to zero after we split the
sample. To split the sample though, only the nonfilers that
have a positive financial benefit are used.

* Indicates that the coefficient is significant at the 10% level.

** Indicates that the coefficient is significant at the 5% level.

Table 4. The Impact of Filing on the Growth of Food Consumption

Sample                        Full Sample           Full Sample

Filed dummy variable        0.0808 (0.0438) **           --
Filed residual              1.2128 (1.3767)              --
Chapter 7
Dummy variable                    --              0.1336 (0.0542) **
Residual                          --              0.1314 (0.2853)
Chapter 13
Dummy variable                    --             -0.0270 (0.0820)
Residual                          --              1.9469 (1.1216) *
Joint test of chapter
  residuals                       --                1.77 [0.17]

Sample                   Financial Benefit > 0

Filed dummy variable              --
Filed residual                    --
Chapter 7
Dummy variable              0.1312 (0.0542) **
Residual                   -0.2677 (0.5239)
Chapter 13
Dummy variable             -0.024 (0.0823)
Residual                   -0.0532 (2.1400)
Joint test of chapter
  residuals                 0.14 [0.87]

Robust standard errors are given in parentheses. The regressions above
also include the first difference of log income, family size, food
needs, state unemployment rate, and state output. In addition, the
model included the following information on the household head:
marital status, age, race, education, and gender. Year effects were
also included in all specifications. For the coefficients not shown,
the results are very similar to the results from Table 2. The last
row presents the results of a joint test of whether the chapter 7 and
chapter 13 residuals are statistically significant. The F-statistic
is given, with the p-value presented in brackets.

* Indicates that the coefficient is significant at the 10% level.

** Indicates that the coefficient is significant at the 5% level.

Table 5. Alternative Measures of Consumption

                                Financial Benefit Coefficient
Consumption Measure                    (Standard Error)

Food at home                           0.0010 (0.0003) **
Food away from home                    0.0004 (0.0003)
Total, imputed consumption             0.0003 (0.0002)

Robust standard errors are given in parentheses. The regressions above
also include the first difference of log income, family size, food
needs, state unemployment rate, and state output. In addition, the
model included the following information on the household head: marital
status, age, race, education, and gender. Year effects were also
included in all specifications. The imputed consumption measure comes
from the coefficients in Table C.1.

* Indicates that the coefficient is significant at the 10% level.

** Indicates that the coefficient is significant at the 5% level.

(1) There have been a number of studies examining the costs surrounding bankruptcy. For example, Gropp, Scholz, and White (1997) examine the effects of bankruptcy laws on the supply of credit and on the interest rates for automobile loans. Another article of note is Musto (2004), which estimates how filing for bankruptcy affects access to credit card debt in the years after the household files for bankruptcy.

(2) We define a consumption insurance effect following the definition in Browning and Crossley (2001). With incomplete markets, the household is able to insure a level of consumption against idiosyncratic shocks that are otherwise uninsured. The other type of consumption effect identified in the literature is referred to as consumption smoothing, whereby the household is able to consume at levels consistent with expectations of future labor income.

(3) Of the bankruptcy fliers in our sample, 68% filed under chapter 7 and 32% filed under chapter 13. Nationally, approximately 70% of all filings occurred under chapter 7. While it may be the case that bankruptcies are underreported in the PSID, it appears that the sample of bankruptcy fliers used in this study matches the distribution of fliers across bankruptcy chapters.

(4) In our sample, only five households filed for bankruptcy in 1996, while there were at least 23 filings in every other year from 1991 to 1995.

(5) In the empirical work below, we include a measure of imputed total consumption as one way to test the robustness of our results.

(6) This consumption equation has become relatively standard in the consumption insurance literature (see Cochrane 1991; Stephens 2003).

(7) The household-level variables come from the PSID. The state-level variables come from the Statistical Abstract.

(8) Because we use multiple editions of Elias, Renauer, and Leonard (1995), the bankruptcy exemptions vary across states and over time.

(9) In fact, Fay, Hurst, and White (2002) treat the financial benefit as exogenous to the bankruptcy decision.

(10) We believe our situation may be analogous to the situation in Souleles (1999), where he referred to a tax refund as predetermined in relation to current period consumption growth. While the household may be able to affect its financial benefit last period (i.e., 13 plus months before filing), legal restraints make benefit predetermined in relation to current period decisions. In fact, even a small amount of randomness in the decisions of bankruptcy judges should greatly reduce behavior that could make the decision to file endogenous to consumption growth.

(11) In the consumption literature that uses the PSID, there is concern about measurement error in consumption. As in Gruber (2000), we censor the largest 1% positive and negative deviations of consumption changes to the 99th and 1st percentile, respectively, to eliminate some of the considerable noise in this variable. This is an appropriate technique as long as the measurement error is not correlated with the regressors. Given the inherent randomness of the bankruptcy process and the possibility that large prefiling consumption (debt) changes by the household may be viewed as fraudulent and thus nondischargeable in bankruptcy, it is unlikely that the measurement error is correlated with the regressors.

(12) The coefficients for this first stage are contained in the first column of Table A.1.

(13) Another indicator that benefit is exogenous is that its coefficient does not change much when the residual is appended to the regression. In Table 2 with the residual appended, the coefficient on the benefit equals 0.0088. Without the residual, the coefficient equals 0.0085.

(14) Because the exemptions are not binding for all households, an increase in the exemptions does not increase the financial benefit for all households.

(15) Zeldes (1989) and Gruber (2000) only use the change in food needs in their specifications. We include both, which helps explain why the food needs coefficient is insignificant.

(16) Eight bankruptcy fliers in our sample did not report the bankruptcy chapter, and these fliers are excluded from all specifications that use the chapter variables.

(17) Meyer (1990) estimates that a 10-percentage point increase in the UI replacement rate increases the spell of unemployment by 1.5 weeks. Montgomery and Navin (2000) estimate that a 10% increase in Aid to Families with Dependent Children (AFDC) benefits decreases labor supply participation by 1%.

References

Attanasio, Orazio, and Steven Davis. 1996. Relative wage movements and the distribution of consumption. Journal of Political Economy 104:1227-62.

Bankruptcy Reform Act of 1984, Public Law 98-353, 97th Congress, (July 10, 1984) 312, 475.

Blundell, Richard, and Luigi Pistaferri. 2003. Income volatility and household consumption: The impact of food assistance programs. Journal of Human Resources 38(S): 1032-50.

Browning, Martin, and Thomas F. Crossley. 2001. Unemployment insurance benefit levels and consumption changes. Journal of Public Economics 80:1-23.

Cochrane, John H. 1991. A simple test of consumption insurance. Journal of Political Economy 99:957-76.

Deaton, Angus. 1992. Understanding consumption. Oxford: Oxford University Press.

Elias, Stephen, Albin Renauer, and Robin Leonard. 1995. How to file for bankruptcy. 5th edition. Berkeley, CA: Nolo Press.

Elul, Ronel, and Narayanan Subramanian. 2002. Forum-shopping and personal bankruptcy. Journal of Financial Services Research 21:233-55.

Fay, Scott, Erik Hurst, and Michelle J. White. 2002. The household bankruptcy decision. American Economic Review 92: 706-18.

Fisher, Jonathan D. 2001. The effect of transfer programs on personal bankruptcy. Bureau of Labor Statistics Working Paper No. 346.

Fisher, Jonathan D., and David S. Johnson. 2003. Consumption mobility in the U.S.: Evidence from two panel data sets. Unpublished paper, Bureau of Labor Statistics.

Gropp, Reint, J. Karl Scholz, and Michelle J. White. 1997. Personal bankruptcy and credit supply and demand. Quarterly Journal of Economics 112:217-51.

Gruber, Jonathan. 1997. The consumption smoothing benefits of unemployment insurance. American Economic Review 87: 192-205.

Gruber, Jonathan. 2000. Cash welfare as a consumption smoothing mechanism for divorced mothers. Journal of Public Economics 75:157-82.

Gruber, Jonathan, and Aaron Yelowitz. 1999. Public health insurance and private savings. Journal of Political Economy 107: 1249-74.

Gunderson, Craig, and James P. Ziliak. 2003. The role of food stamps in consumption stabilization. Journal of Human Resources 38(S):1051-79.

Hayashi, Fumio, Joseph Altonji, and Laurence Kotlikoff. 1996. Risk-sharing between and within families. Econometrica 64:261-94.

Kniesner, Thomas J., and James P. Ziliak. 2002. Tax reform and automatic stabilization. American Economic Review 92: 590-612.

Meyer, Bruce D. 1990. Unemployment insurance and unemployment spells. Econometrica 58:757-82.

Montgomery, Edward, and John C. Navin. 2000. Cross-state variation in Medicaid programs and female labor supply. Economic Inquiry 38:402-18.

Musto, David K. 2004. What happens when information leaves a market? Evidence from post-bankruptcy consumers. Journal of Business. 77:725--48.

Posner, Eric A. 1995. Contract law in the welfare state: A defense of the unconscionability doctrine, usury laws, and related limitations on the freedom to contract. Journal of Legal Studies 24:283-319.

Skinner, Jonathan. 1987. A superior measure of consumption from the panel study of income dynamics. Economics Letters 23:213-16.

Souleles, Nicholas S. 1999. The response of household consumption to income tax refunds. American Economic Review 89:947-58.

Stephens Jr., Melvin. 2003. '3rd of tha month': Do social security recipients smooth consumption between checks? American Economic Review 93:406-22.

Survey Research Center, Institute for Social Research, University of Michigan. 2002. Panel Study for Social Dynamics. Accessed July 2002. Available http://psidonhne.isr.umich.edu.

Wooldridge, Jeffrey M. 2002. Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.

Zeldes, Stephen P. 1989. Consumption and liquidity constraints: An empirical investigation. Journal of Political Economy 97:305-46.

Larry H. Filer II * and Jonathan D. Fisher ([dagger])

* Old Dominion University, Department of Economics, Norfolk, VA 23529, USA; E-mail [e-mail]; corresponding author.

([dagger]) Bureau of Labor Statistics, 2 Massachusetts Avenue, NE Room 3105, Washington, DC 20212-0001, USA; E-mail [e-mail].

The authors would like to thank the editor, two anonymous referees, Michelle White, Barry Adler, William Hoyt, Michel Robe, and participants of the American Law and Economics Association Conference. We would also like to acknowledge seminar participants at American University, The College of William and Mary, and the Bureau of Labor Statistics for helpful comments. All views expressed in this article are those of the authors and do not reflect the views or policies of the Bureau of Labor Statistics (BLS) or the views of other BLS staff members.

Received April 2004; accepted August 2004.