Title: On the usefulness of financial variables to predict the conditional distribution of the market portfolio
Authors: Azam Shamsi Zamenjani - University of New Brunswick (Canada) [presenting]
Abstract: The predictability of the conditional distribution of market returns using financial and macroeconomic variables is investigated. In contrast to the extant literature that mainly focuses on the mean forecasts, we study the predictability of the entire density. This provides us with useful information on the uncertainty around the point forecasts and tail events that is valuable in areas such as asset allocation and risk management. We consider a Bayesian nonparametric mixture model that allows the mixing distribution to change with time. The weights of the mixture are constructed as Probit transformations of a linear combination of the predictors. This allows us to study whether these predictors are useful in forecasting the unknown and time-varying density of the innovations. We compare statistical and economic measures of forecasting performance of the proposed model with a set of benchmark models. In spite of little or no improvement in point forecasts, certain variables display significant out-of-sample predictive ability for the stock return density, and increase economic value for investors when employed in portfolio decisions.