The Predictability of Real Estate Excess Returns: An Out-of-Sample Economic Value Analysis

Abstract

We study the recursive, out-of-sample realized predictive performance of a rich set of predictor choices and models, spanning linear and Markov switching frameworks when the forecast target is represented by excess NCREIF and equity NAREIT returns. We find considerable pockets of predictive power, especially at the short- and intermediate horizons and for private real estate returns, both in absolute term and in comparison to a simple, but powerful, historical sample mean benchmark. We then test whether such forecasting accuracy may translate to positive, risk-adjusted out-of-sample performance in a recursive mean-variance portfolio allocation exercise, selecting weights of stocks, bonds, cash, and real estate (private or public). Consistently, we find that especially in the case of private real estate, significant improvements in realized Sharpe ratios and mean-variance utility scores are achieved from a range of strategies, exploiting predictability at intermediate horizons, especially when supported by Markov switching models. These results are robust the inclusion of transaction costs and extend to public real estate.