Real estate investment trust return dynamics and value-at-risk under alternative classes of model specifications

Jung-Suk Yu

Abstract

ABSTRACT

This study compares the statistical and economic performances of two broad classes of model specifications for real estate investment trust (REIT) index returns. Empirical results suggest that the class of fat-tailed distributions, represented by the skewed t model with asymmetric generalized autoregressive conditional heteroscedasticity (GARCH), is more suitable for modeling REIT return dynamics than the new class of GARCH-jump models. We find that the superiority of the asymmetric fat-tailed (AFT) distributions stems from their ability to match the higher-order moments and high peaks around zero returns simultaneously. In addition, the value at-risk (VaR) analysis based on backtesting comparisons and out-of-sample forecasting indicates that the class of AFT distributions is a more effective risk management tool, because skewed models are capable of achieving the smallest number and magnitude of violations of the VaR compared to the GARCH-jump models. This finding was particularly important for the market turbulence during the 2008-9 crises.