RDP 2003-07: Housing Wealth, Stock Market Wealth and Consumption: A Panel Analysis for Australia 1. Introduction

The dramatic changes in stock values and in house values over the last decade have renewed policy and academic interest in the effects of household wealth on consumption expenditure. It is sometimes argued that the effect of changes in housing wealth is larger than the effect of changes in stock market wealth[1]. The reasoning is two-fold: firstly, more people own houses than shares and secondly, financial innovation has made it easier to access capital gains from housing wealth. However, there are other factors that work in the opposite direction and so an assessment requires quantitative estimates of the effect of changes in each type of wealth on consumption.

Studies that estimate the effect on consumption of changes in housing wealth and stock market wealth often find that one of the coefficients is insignificant. For example, Tan and Voss (2003) use Australian data and find a strong long-run effect of stock market wealth but an insignificant housing wealth effect. In addition, Case, Quigley and Shiller (2001) report a number of US studies that use aggregate level or household level data and have had difficulty in finding a significant housing wealth effect. They suggest that this may be the result of multicollinearity of the two wealth variables, which might be overcome using state-level data. Using a panel of US states they find that the housing wealth effect is significant and larger than the stock market wealth effect.

In this paper we follow Case et al (2001) and use a state-level panel to estimate the effect of changes in wealth components on household consumption in Australia. However, we use a richer specification, which controls for changes in household debt. We also employ a wider range of econometric techniques in order to analyse the robustness of results, including instrumental variables to control for endogeneity, as well as a dynamic OLS estimator for panels, and mean group estimators based on a Seemingly Unrelated Regression (SUR) model for the state-specific wealth effects.

The next section discusses briefly the results from previous studies on consumption and wealth, followed by Section 3 which outlines some theoretical considerations for our chosen specification. Section 4 explains how the data set was constructed before we present the estimation results and a number of robustness tests. The robustness of the results through time is analysed in Section 5 followed by the conclusions in Section 6.


See, for instance, The Economist, Jan 11th 2003, ‘Living in never-never land’, p. 68. [1]