Abstract: Much literature in Econophysics reveals that the volatility in financial markets presents multifractal features. Thus, measuring and forecasting the market volatility accurately is very important for financial risk management. Based on the earlier research of multifractal volatility and its model, an out-of-sample dynamic VaR forecasting method is proposed in this paper. The empirical results on two backtesting techniques show that, on high-risk levels, VaR model based on multifractal volatility produces much better out-of-sample VaR forecasts than eight popular linear and nonlinear GARCH models.

Key words: multifractal, volatility, out-of-sample dynamic VaR, forecasting, backtesting