Forecasting Intraday VaR and ES using a Markov-Switching MC-GARCH Model

Abstract

This research adopts a comparative study in modelling high frequency asset returns and forecasting Value-at-Risk (VaR) and Expected Shortfall (ES) for three datasets namely Gold, FTSE 100 and Lloyds Banking Group PLC (LLOY). This study develops the hybrid Markov-Switching MC-GARCH (MSMC-GARCH) model. The regime switching methodology aims to capture the presence of structural breaks in financial prices in order to reduce the estimation bias of the MC-GARCH model. The application of the MSMC-GARCH model to the one-minute return series of each asset is investigated. An empirical comparison is performed with the benchmark MC-GARCH model under five error specifications. The intraday VaR and ES are forecasted at the 95%, 99% and 99.5% levels for an out-of-sample set. The forecasting accuracy of each model is assessed using various backtesting procedures. On the basis of VaR forecasting performance, the backtesting results revealed that the MSMC-GARCH model under non-gaussian assumptions best forecasted VaR for all assets, with only one exception being at the 99.5% VaR. The results of the ES forecasting accuracy further confirmed the superiority of the MSMC-GARCH model over the MC-GARCH model for all assets.