A Comparison of Forecasting Accuracy between Two Dynamic Conditional Correlation (DCC) Models

Abstract

This study compares two commonly used DCC-family models to predict the linkage between the US equity and REIT markets within a global minimum-variance portfolio. Equity and REIT portfolios are constructed using variance-covariance matrices, which represent forward-looking covariance information. These matrices are constructed out-of-sample with ex-ante forecasting. By assessing the predictive precision of each model, the study aims to determine which one produces the lowest forecasting errors and performs better economically. According to a statistical comparison, ex-ante correlation forecasts based on the Asymmetric DCC model were more accurate than those based on the standard DCC model. An empirical comparison of the economic performance of these two models in a dynamic portfolio allocation framework reveals that, despite its complexity, the Asymmetric DCC model exhibits similar economic performance characteristics to the standard DCC model. Despite the lack of emphasis on the economic overperformance of the Asymmetric DCC model, investors who recalibrate their portfolios weekly will benefit from reduced forecast errors and the ability to create more efficient portfolios by using an asymmetric model instead of a standard model.