RDP 2009-07: Estimating Marginal Propensities to Consume in Australia Using Micro Data 7. Differences in Responses Across Households

As noted in the literature for the United States, there is likely to be heterogeneity in households' responses to an income shock. One explanation for this heterogeneity is the presence of liquidity constraints, usually proxied for by low income or low wealth. Liquidity-constrained households are likely to consume out of current, rather than permanent income, leading to a higher estimated MPC. In this paper we consider the level of income and other financial constraints as indicators of liquidity constraints as well as exploring another source of heterogeneity – household's perceptions of risk. If households are concerned about prospects for income, they may be more likely to save extra income rather than spend it.

Our first measure of liquidity constraints is based on income, which is standard in the literature. We classify a household as liquidity constrained if equivalised household income is less than the cut-off point for the bottom 30 per cent of equivalised income for the HILDA sample in each year,^[27] and it meets the low-income definition in all three periods.^[28] A dummy variable is created to identify low income households. This dummy variable is interacted with all of the explanatory variables allowing us to test for differences in the responses of the variables.^[29]

Table 8 shows the effect of liquidity constraints on the estimates of the MPC for both the tax cuts and the Baby Bonus. On non-durable expenditure, we find that households with low incomes spend more of their tax cuts and lump-sum payments than higher income households, but these are imprecisely estimated. The results are mixed for strictly non-durable expenditure. However, none of these differences are statistically significant.

Our second measure of constraints come directly from relevant questions in the HILDA Survey. Following La Cava and Simon (2003), households are classified as financially constrained or stressed if they answer yes to any of the options on the question: ‘Since January [year] did any of the following happen to you because of a shortage of money?’. The options are: could not pay electricity, gas or telephone bills on time; asked for financial help from friends or family; could not pay the mortgage or rent on time; pawned or sold something; was unable to heat home; went without meals; and/or asked for help from welfare/community organisations.

A dummy variable is created for households that are financially constrained in all three periods.^[30] Once again we interact the dummy variable with all (non-policy) variables and test for differences, however, we find none that are statistically significant. We thus only interact the financial constraint dummy with the policy variables to test for differences in the estimated MPCs for constrained and unconstrained households. The results are reported in Table 9. Surprisingly perhaps, we find that households that report constraints tend to spend less of their extra income than households that do not report constraints, although again these differences are not statistically significant.

There are two potential explanations for this result. First, only about 40 per cent of financially constrained households have low incomes (across the whole HILDA sample), suggesting that this variable is an imprecise measure of the existence of liquidity constraints. The inclusion of wealthier people, who are more likely to be under financial stress because they have large amounts of debt to repay, may explain why those households who report themselves to be financially constrained seem to spend less of the extra income than those that are not constrained. It seems likely that such households might use extra income to pay down debt, rather than increase expenditure. We test this hypothesis by splitting the sample into those with and without housing debt (since this is the only measure available in all three waves). However, we find that financially constrained households tend to spend less than unconstrained households regardless of whether or not they have housing debt. Second, since the survey question asks households if they have been in difficulty at any time since the beginning of the year, we cannot separate temporary from more persistent financial difficulty. While we have attempted to exclude temporary difficulty by only considering households that report difficulty in all three years, the measure is still quite imprecise.

Household perceptions of risk may also be an important driver of heterogeneity in household MPCs. If households are especially concerned about maintaining their income in the future, they may be more likely to save any extra income they receive, lowering their MPCs. In order to examine this precautionary savings motive, we use the self-reported unemployment risk variable for the household head. Respondents are asked to rank their feeling about the statement ‘I worry about the future of my job’ on a scale of one to seven (strongly disagree to strong agree). We classify households as worried if they answer ‘five’ or above in all three years.^[31] In contrast, for the Baby Bonus sample, we classify households as worried in any given year if they answer ‘five’ or above to the same question. We allow for this time variation since too few households were worried in all three years for reliable estimates.^[32]

To estimate the effect of perceived unemployment risk on the MPC, we interact the unemployment risk dummy variable with the tax variable in Equation (1) and the Baby Bonus variable in Equation (2). Our results, shown in Table 10, are consistent with our expectation that households facing more employment uncertainty will spend less out of additional income. The differences in the estimates of the MPC for ‘worried’ and ‘not worried’ households are not statistically significant for either the tax cuts or the Baby Bonus. However, they are noticeably different in terms of magnitude. Although the Baby Bonus should be interpreted with caution given the difficulties with sample size, our finding is consistent with Leigh (2009). Our results are not affected by restricting the sample to households where the reference person is employed.

Footnotes

Income is equivalised using the ABS method of dividing income by a weighted sum of the number of household members. Household members are assigned varying weights according to age and numbers of household members: 1 for the first person aged over 15 years and 0.5 thereafter, and 0.3 for each person aged under 15 years. The 30 per cent cut-off points are $21,831, $23,730 and $25,000 for Waves 5, 6 and 7 respectively. [27]

Only 14 per cent of households change their classification (in terms of this income threshold) over time, which creates a potential difficulty when using fixed effects or first differences. The issue is that the small number of observations that do change will have an unduly large influence on the coefficient estimates. To avoid this we apply a strict definition of low income, which removes all time variation in this variable. [28]

We find that the only variable (other than the Tax and Bonus variables) that has a statistically significant different effect between groups is the Wave 5 time dummy. As previously mentioned, there is an unusually high difference between the Wave 5 expenditure data and the other two waves. This suggests that whatever is causing this difference varies across income groups. However, further investigation shows that this relationship does not hold consistently across income deciles. Since there is a significant difference between the effect of the Wave 5 dummy for low-income earners and everyone else, we include separate Wave 5 dummy variables for these groups of households as well as allowing for separate MPCs. [29]

Time-series variation is again a potential issue in this specification, with only 26 per cent of households varying their response over time. [30]

Once again there is not very much time variation in this variable. Since the question is answered on a scale of 1–7, a substantial amount of the variation that does occur could be generated by arbitrary changes in a respondent's interpretation of the ranking from year-to-year. If time variation is included, we find that risk aversion has no effect on the estimate of the MPC. However, this is not surprising since there could be any number of unrelated reasons why households change their reported job security across time. [31]

There is also much more time variation in perceived job security in the Baby Bonus sample as compared with the tax cut sample. [32]