This paper analyzes the role played by financial assets, direct real estate, and the Fama and French factors in explaining EREIT returns and examines the usefulness of these variables in forecasting returns. Four models are analyzed and their predictive potential is assessed by comparing three forecasting methods: time varying coefficient (TVC) regressions, vector autoregressive (VAR) systems, and neural networks models. Trading strategies on these forecasts are compared to a passive buy-and-hold strategy. The results show that EREIT returns are better explained by models including the Fama and French factors. The VAR forecasts are better than the TVC forecasts, but the best predictions are obtained with neural networks and especially when they are applied to the model using stock, bond, real estate, size, and book-to-market factors.
