Forecasting VaR and ES through Markov-switching GARCH models: does the specication matter?


We compare some of the most common specifications of Markov Switching GARCH (MS-GARCH) models in terms of their risk forecasting ability for exchange rates (EUR/USD, JPY/USD, CAD/USD and DKK/USD). Specifically, we compare out-of-sample forecasts for the value at risk and the expected shortfall. Additionally, we present a brief introduction to the implemented MSGARCH models as well as a discussion of the finite sample properties of parameter estimates and risk forecast based on Monte Carlo experiments. The results based on Monte Carlo experiments and empirical data suggest that the models implemented are robust to Markov switching volatility misspecification for forecasting both risk measures. For both, Monte Carlo simulations and empirical data, the forecasting performance of all of them improves as the sample size.

Área do Conhecimento