(Re)correlation : A Markov Switching Multifractal Model with Time Varying Correlations

The paper develops a Markov switching multifractal model with dynamic conditional correlations. The objective is to give more flexibility to the initial bivariate Markov switching multifractal model [MSM] (Calvet et al. (2006)) by introducing some time dependency in the comovement structure. The new defined model is applied to stock index data (CAC, DAX, FTSE, NYSE) between 1996 and 2008 and compared to both the standard MSM and the DCC model of Engle and Sheppard (2002). The MSMDCC models present, in sample, better fit than the MSM and DCC models. Moreover, by combining these two setups, MSMDCC improves forecast performances for longer horizons, and provides a better understanding of market comovements during crisis episodes