Conditional asset allocation using prediction intervals to produce allocation decisions

Traditional conditional asset allocation involves using key past economic and financial data to produce forecasts of expected returns for the various asset classes involved in the asset allocation decision. Ordinarily these are point forecasts and little or no use is made of the prediction interval in the decision-making process. The main reason is that forecast models are misspecified, hence error distributions are unknown and so model-based prediction intervals provide spurious measures of forecast uncertainty. Alternative approaches, such as Bayesian densities and bootstrapping, provide non-parametric forecast distributions, which are not model based and so reduce the uncertainty about the nature of the prediction interval. Here we investigate various different prediction intervals and attempt to quantify biases between model-based and non-model-based solutions. Asset allocation decisions based on prediction densities are investigated, and it is found that knowledge of a more accurate prediction interval is economically meaningful in an asset allocation context.