Credit card debt and consumption: evidence from household-level data

Abstract

This research investigates the relationship between credit card debt and consumption using household level data. This is a departure from the previous studies which have used aggregate measures of consumption and general debt such as the Debt Service Ratio or total revolving credit. We use a detailed monthly survey of credit card use to impute credit card debt to respondents from the Consumer Expenditure Survey sample. In contrast to some earlier studies using aggregate data, we find a negative relationship between debt and consumption growth. Our work shows that a $1000 increase in credit card debt results in a decrease in quarterly consumption growth of almost 2%. Investigations are also made into effects of debt within different age categories and into the impact of expected income growth on the debt–consumption relationship.

Notes

¹ The average credit card balance among revolvers was $5100 according to the 2004 SCF and $9205 according to Bankrate.com

² See Dunn et al. (2006) for sample characteristics. The SCF has been widely used in this literature, but the SCF appears only once in 3 years and is thus not suitable for this research.

³ The Federal Reserve has recently undertaken revisions to the DSR to make the process used in its calculation more in line with recent changes in financial markets and consumer behaviour, and a Financial Obligations Ratio has been developed (Dynan et al., 2003; Johnson, 2005).

⁴ A detailed comparison is available from the authors upon request.

⁵ The omitted month is April 2002.

⁶ It is conventional to use the variables in W in order to control for changes in household preferences. See Lusardi (1996) and Souleles (2004).

⁷ The results in this step conform to the previous literature. The results are available from the authors upon request.

⁸ The CEX has multiple observations on the same households, but for Step 2 we utilize only the first observation.

⁹ In preliminary investigations, the value 0.5 gave us the highest rate of correctly identifying the two groups.

¹⁰ Details of this adjustment process are available from the authors upon request.

¹¹ This may be related to the way in which we have designated goods as durable and nondurable since the CEX provides no conventions on this.

¹² We use the same University of Michigan consumer confidence question utilized by Souleles (2004) in a similar estimation method to impute income expectations to the CEX sample.

¹³ For more information on the description of the CEX survey, see CES Anthology (2003).

¹⁴ Sample selection was done in accordance with the conventions set in the previous literature that has used CEX data.