Further evidence of excess sensitivity of consumption? Nonseparability among goods and heterogeneity across households

Abstract

The standard theoretical framework for analysing households’ intertemporal decisions is the life-cycle/permanent income model. Among its implications, testing the model allows to analyse the response of consumption to fiscal policy. However, the empirical literature with microdata has yielded mixed results. This article examines the sensitivity of the results to the assumption of separability among goods and of homogeneity across households. For that purpose, we test a rational expectations permanent income model with household data drawn from the Spanish Family Expenditure Survey. This survey contains detailed information on total expenditure and the income presents large, exogenous quarterly changes due to an institutional feature. The article shows that assuming separability among commodities biases the test against the model. When separability is not imposed, we show that the rejection of the model depends on heterogeneity across households in terms of their members being unemployed or not. For those households permanently employed, the model cannot be rejected whatever their income status.

Acknowledgements

The authors would like to thank José M. Labeaga for helpful comments. Financial support of the Spanish Ministry of Science and Technology through the SEJ2007-65255 project is also acknowledged.

Notes

¹ See also the papers by Dow (1993), Shea (1994), and Lee and Kong (2000) with aggregate data.

² A similar proposal is that made by Levenson (1996), who analyses whether households in Taiwan increased their consumption after an announced reform to the Social Security that represented windfall retirements/severance benefits.

³ Poterba (1988) and Wilcox (1989) are examples of pioneering studies in this type of REPIH test, using aggregate data.

⁴ See Browning and Collado (2001) for a description of the annotation of the extra payments in the ECPF.

⁵ As commented below, the main differences between this study and that of Browning and Collado (2001) are that this article does not assume separability among goods, whilst heterogeneity across households is contemplated via the sample's segmentation. Finally, this study includes households with unemployed members and households where the spouse works.

⁶ See Browning and Crossley (1999) for theoretical and empirical results on consumption during an unemployment spell.

⁷ For those households without transitions into and out of employment, in 85% of all observations the sign of the quarterly income changes can be correctly predicted. In fact, despite a lack of official information on how widespread extra payments are, the analysis of the ECPF points to the fact that around 75% of all employees with no labour transitions receive extra payments.

⁸ Browning and Crossley (2001) calculate the welfare costs for Spanish households of automatically consuming all current income (measured as a percentage of annual spending) rather than following an optimally smoothed path, under the hypothesis that, during months with extra payments, double the normal income is paid. The authors conclude that the welfare costs stand at around 7%, very much higher than the figure for institutional features examined by Parker (1999) and Hsieh (2003), thus demonstrating the relevance of intrayearly planning in the Spanish case.

⁹ In order to have a direct measure of liquidity constraints, García (1999) substitutes the income variable for the change in households’ indebtedness, the latter obtained from National Accounts.

¹⁰ The reference period for each type of goods depends on the frequency of its purchase. Food expenditure corresponds to purchases made during the week of the interview, other nondurables to the previous month including the week of the interview, and durables to the previous three months including the week of the interview. The INE raises food spending and expenditure on other nondurables to a standard three-month period to homogenize the global expenditure period.

¹¹ The results were not affected when other minimum values for expenditure on durables were used.

¹² We estimated the Euler equations by GMM using the DPD programme written in GAUSS by Arellano and Bond (1998).

¹³ This type of heterogeneity could arise if each household had its own discount rate, which remained constant across time. In this case, the presence of persistent household-specific effects causes lagged consumption growth to have predictive power over current consumption growth. For this reason, to test their existence, Δln C _t−1, which would be correlated with the household-specific effect, was incorporated into the instrument set.

¹⁴ Attention was also paid to the possible correlation between age and unemployment. The analysis of the sample did not show a high degree of correlation between age and unemployment transitions.

¹⁵ See Hansen and Singleton (1982), Arellano and Bond (1991) and Bound et al. (1995) for the properties of the IV estimators when the instruments are weakly correlated with the endogenous variable.

¹⁶ The group aged between 61 and 64 was excluded to prevent transitions into retirement from distorting the results.

¹⁷ Zeldes (1989b) and most subsequent authors separate the sample on the basis of (liquid) wealth to income ratios. Unfortunately, wealth-related information is not available in the ECPF.

¹⁸ The null hypothesis of absence of second-order autocorrelation for the disturbance term (M2) could not be rejected. Neither could the null hypothesis of absence of persistent household-specific effects. These results were repeated in the remaining estimations. For the sake of brevity, tests of persistent household-specific effects are not reported, but are available upon request.

¹⁹ Note from Table 2 that the excess sensitivity is maintained when neither of the two groups of commodities is included as an explanatory variable.

²⁰ The results were not affected when retired households were excluded.

²¹ Brugiavini and Weber (1994) also obtain a negative correlation between nondurables and durables with cross-section data.