RDP 2022-06: Do Australian Households Borrow to Keep up with the Joneses? 6. Robustness Checks

In this section, I perform several robustness checks. First, the sample of non-moving households are not necessarily random. For example, people might move to a low inequality neighbourhood with the aim to reduce their debts. I follow Atalay et al (2020) in controlling for this selection through a Heckman selection model. The selection equation takes the form:

I model the household decision not to move as a function of the lagged indicators of their intention to move, their satisfaction with their neighbourhood, and their satisfaction with their local community, while controlling for a range of factors similarly specified in Equation (1). The satisfaction measures are arguably exogenous. Estimates using the Heckman selection model are similar to those using the baseline model (Table B2) with the association between investment debt and local income inequality standing out as particularly significant, both statistically and economically.

Second, I employ the top 10 per cent share of total income as an alternative measure of local income inequality. This measure is more sensitive to changes that affect the upper tail of the distribution than the Gini coefficient, and to the extent that households compare themselves against their most affluent neighbours, could be more suitable for my analysis. Table B4 shows that the choice of inequality indicator does not influence the qualitative results. A 1 percentage point increase in the top 10 per cent share of local income is associated with more than a 2 percentage point increase in investment debt and 0.3 percentage point increase in car debt, both as a proportion of income.

Next, I check the results across a number of other sub-samples for robustness. First I remove the self-employed. These households may be more responsive to macro shocks, and also tend to have more investment debt, and as such it is possible that they are driving some of the results. However, by restricting the sample to not self-employed households I find that investment debt accumulation following an increase in inequality remains the same (Table B6).

I also consider a sub-sample of households with no change in household structure to abstract from shocks that result in a change in household structure, which in turn affects debt-taking behaviour. Restricting the sample to households not going through structural changes between the wealth modules, roughly 80 per cent of the main sample, the results are, again, largely unchanged (Table B7).

Similarly, households could have changed occupations, as a result of local economic conditions that also affect local inequality, such as the industrial composition. This change could simultaneously affect their debt accumulation. However, restricting the sample to households whose heads did not change occupations between the wealth modules yields an even stronger link between local inequality and investment debt accumulation (Table B8).

Finally, I simplify the model specification by removing the interaction terms between SA3 macro shocks and local income inequality. Estimates are broadly unchanged, suggesting that the heterogeneous relationship between local income inequality and debt accumulation across household distributions is not confounded by other SA3 macro shocks (Table B10).