Forecasting Structural Changes in Short and Long-Run Correlations : A Mixed Regime DCC-MIDAS Model

We propose a novel approach to modelling structural changes in asset returns correlations. Our framework allows for breaks of different type in the conditional and unconditional correlation components by capturing abrupt regime switches in the short-run correlations and smooth transitions between regimes in the long-run correlations. We associate regime switches in the long-run correlations with low-frequency macroeconomic variables. This approach increases the in-sample fit and the forecasting accuracy of the correlation component models by explicitly accounting for parameter instability over time. Results from an extensive forecasting exercise reveal its benefits in terms of statistical and economic criteria at medium and long-horizon forecasting and over periods of pronounced instability