Does consumer debt cause economic recession? Evidence using directed acyclic graphs

Abstract

This study investigates the relationship between consumer debt and aggregate economic activity based on time series methods and directed acyclic graphs (DAG). Quarterly US data, measured over the period 1980 to 2003, on consumer debt, gross domestic product (GDP), interest rates, housing starts, and domestic auto sales, are analysed in an Error Correction Model (ECM). Contemporaneous innovations from this ECM are given a structural representation, using recent developments in DAG modelling. The ECM and DAG components are summarized using innovation accounting techniques (impulse response functions and forecast error variance decomposition). The DAG causal pattern reveals a causal flow from GDP to consumer debt; the subsequent innovation accounting results also show that consumer debt is not exogenous in contrast to GDP and other indicators. This result concurs with a previous study based on Granger causality, but contradicts other works that claim consumer debt is a root cause of aggregate economic performance.

Notes

¹ Based on the suggestion by (Spirtes et al. 2000), 10% and 20% significance level would be suitable for the sample size in the present study. It is also suggested that various levels of significance be considered in an attempt to achieve an unambiguous causal structure of the variables. The 30% level is used as it is the lowest significance level that produces an unambiguous causal ordering in this case.

² Beyond the causal flows from GDP and auto sales to consumer debt, causal flows revealed at 30% significance level by PC algorithm have also been used. The DAG pattern is used to determine the Cholesky ordering of the five variables. For the orders that are not revealed by DAG, i.e., interest rate vs. auto sales, housing starts vs. consumer debt, all four different orderings have been tested. The results indicate the change of ordering within each pair has little impact on impulse response functions and forecast error variance decomposition. One order is presented in this paper: auto sales, interest rate, GDP, housing starts, and consumer debt.