Forecasting volatility under fractality, regime-switching, long memory and student-t innovations

摘要

We examine the performance of volatility models that incorporate features such as long (short) memory, regime-switching and multifractality along with two competing distributional assumptions of the error component, i.e. Normal vs Student-t. Our precise contribution is twofold. First, we introduce a new model to the family of Markov-Switching Multifractal models of asset returns (MSM), namely, the Markov-Switching Multifractal model of asset returns with Student-t innovations (MSM-t). Second, we perform a comprehensive panel forecasting analysis of the MSM models as well as other competing volatility models of the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) legacy. Our cross-sections consist of all-share equity indices, bond indices and real estate security indices at the country level. Furthermore, we investigate complementarities between models via combined forecasts. We find that: (i) Maximum Likelihood (ML) and Generalized Method of Moments (GMM) estimation are both suitable for MSM-t models, (ii) empirical panel forecasts of MSM-t models show an improvement over the alternative volatility models in terms of mean absolute forecast errors and that (iii) forecast combinations obtained from the different MSM and (FI)GARCH models considered appear to provide some improvement upon forecasts from single models.