The Impact of Estimation Error on Latent Factor Model Forecasts of Portfolio Risk

Abstract

In this article, the authors measure the impact of estimation error on latent factor model forecasts of portfolio risk and factor exposures. In markets simulated with a Gaussian return generating process, the authors measure errors in forecasts for equally weighted and long-only minimum variance portfolios constructed from a universe of 500 securities. They find that an estimation period of 250 days may be adequate to accurately forecast risk and factor exposures for an equally weighted portfolio. In contrast, the risk of a long-only minimum variance portfolio is substantially under-forecast even with an estimation period of 1000 days. This underscores the importance of testing risk models on optimized portfolios.

Keywords: estimation error, latent factor model, simulation, minimum variance portfolio, equally weighted portfolio, underforecast

JEL Classification: G10