Combining realized measures to forecast REIT volatility

Abstract

Purpose This study aims to show that the best-performing realized measures vary across markets when it comes to forecast real estate investment trust (REIT) volatility. This finding provides little guidance for practitioners on which one to use when facing a new market. The authors attempt to fill the hole by seeking a common estimator, which can study for different markets. Design/methodology/approach The authors do so by drawing upon the general forecasting literature, which finds that combinations of individual forecasts often outperform even the best individual forecast. The authors carry out the study by first introducing a number of commonly used realized measures and then considering several different combination strategies. The authors apply all of the individual measures and their different combinations to three major global REIT markets (Australia, UK and US). Findings The findings show that both unconstrained and constrained versions of the regression-based combinations consistently rank among the group of best forecasters across the three markets under study. None of their peers can do it including the three simple combinations and all of the individual measures. The conclusions are robust to the choice of evaluation metrics and of the out-of-sample evaluation periods. Originality/value The study provides practitioners with easy-to-follow insights on how to forecast REIT volatility, that is, use a regression-based combination of individual realized measures. The study has also extended the thin real estate literature on using high-frequency data to examine REIT volatility.