On Measuring the Economic Significance of Asset Return Predictability

A number of recent studies have measured the quantitative effect of excess return predictability on the optimal consumption and portfolio choices of a rational investor, and they have used the utility costs of ignoring predictability as a natural measure of economic significance. We use a general equilibrium model as a laboratory for generating predictable excess returns and for assessing the properties of the estimated consumption/portfolio rules, under both the empirical and the true dynamics of excess returns. We find that conditional rules based on ordinary least squares estimates of excess returns are severely biased, and they have a large variance across multiple simulated histories of the model. In this experiment, we find the estimation issues to be so severe that the simple unconditional consumption and portfolio rules, from Merton (1969), actually outperform (in a utility cost sense) both simple and bias-corrected empirical estimates of conditionally optimal policies.