Predicting tail risks by a Markov switching MGARCH model with varying copula regimes

To improve the dynamic assessment of risks of speculative assets, we apply a Markov switching MGARCH approach to portfolio risk forecasting. More specifically, we take advantage of the flexible Markov switching copula multivariate GARCH (MS-C-MGARCH) model of Fulle and Herwartz (2022). As an empirical illustration, we take the perspective of a risk-averse agent and employ the suggested model for assessments of future risks of portfolios composed of a high-yield equity index (S&P 500) and two safe-haven investment instruments (i.e., Gold and US Treasury Bond Futures). We follow recent suggestions to employ the expected shortfall as a prime assessment of tail risks. To accurately evaluate the merits of the new model, we back-test the risk forecasting for daily returns over 10 years for heterogeneous market environments including, for example, the COVID-19 pandemic. We find that the MS-C-MGARCH model outperforms benchmark volatility models (MGARCH, C-MGARCH) in predicting both value-at-risk and expected shortfall. The superiority of the MS-C-MGARCH model becomes stronger, when the share of comparably risky assets in the portfolio is relatively large.