Consumption, wealth and expected stock returns in Australia: some further results

Abstract

This article re-examines the evidence that cay, the residual from a cointegrating regression of consumption on labour income and household wealth, is a useful predictor of excess stock returns in Australian data. In recursive samples beginning in 1976:q4 and ending from 1990:q1 to 2003:q1, cay is a strong predictor of excess returns. In samples that end thereafter, cay loses its predictive power for returns. This is due to a break-down in the cointegrating relation among consumption, labour income and household wealth following recent developments in the housing and stock markets.

Acknowledgements

I am grateful to Geoffrey Kingston, Paul Richardson, Glenn Otto and Graham Voss for helpful discussion and to the Australian Research Council for financial support (under Project RM02853).

Notes

¹ Fisher and Voss did find strong evidence for a cointegrating relation over the one sub-sample they did consider: 1976:q4 to 1995:q4. Ioannidis et al. (2006) found evidence for a cointegrating relation and that cay predicted stock returns over their full sample 1976:q4 to 1999:q3 of Australian data. Although their article was published after the Fisher and Voss article, they used an earlier vintage of data taken from Tan and Voss (2003) which ended in 1999:q3. Tang (2006) found evidence for a cointegrating relation among the levels (not log-levels) of per-capita consumption, labour income, housing wealth and financial wealth in Australian data for the period 1988:q2 to 2003:q1, although the coefficient estimates in the long-run relation were statistically significant only when the constant term was restricted to zero.

² Household net wealth is the sum of financial and nonfinancial wealth of households net of household debt. Nonfinancial wealth is the value of household dwelling assets and the value of household consumer durables. The largest component of household wealth in Australia is the value of household dwelling assets which comprise about two-thirds of the total.

³ Returns are defined in terms of logs and the quarterly excess stock return is expressed as a decimal. ASX is the abbreviation for the Australian stock exchange.

⁴ Regressing excess returns on a dummy variable with a value of one for 1987:q4 and a value of zero otherwise and using the residuals as the dependent variable in the predictive regression is exactly the same as setting the value of returns for 1987:q4 to zero. Alternatively, the dummy variable could be included in the predictive regression. I chose not to do this because I am interested in the explanatory power of cay in the predictive regression independent of the influence of the dummy variable on the regression R ². When the actual value of returns for 1987:q4 is used in the empirical analysis below, cay has little predictive power for returns in any sample. This occurs because the stock market crash observation makes it much more difficult to detect predictable variation in expected returns.

⁵ In Lettau and Ludvigson's framework, if the representative agent expects stock returns to increase in the future, the agent will temporarily increase consumption above the long-run trend it shares with labour income and net wealth. Thus, cay rises in anticipation of an increase in returns implying a positive coefficient on lagged cay in Equation 1.

⁶ I also estimated the long horizon regressions over the full sample but with cay constructed from the DOLS regression over the sub-sample. As for the full sample results reported in the table, cay had no predictive content for returns at any horizon.