Abstract: Value-at-Risk (VaR), is a widely used measure of financial market risk, which is conceptually simple and intuitive. Accurate measurement of VaR is important for financial risk managers and regulators. Traditionally, the measurement of market risk (VaR) doesn't use the information content of option and high-frequency data fully, and mainly based on the single-factor volatility models such as the GARCH and stochastic volatility models, which leads to the loss of information and the lack of flexibility of the model.
With the rapid development of derivative (option) markets, the option data become readily available. The option prices reflect market sentiment and/or investors' expectations about future stock market volatility. And a growing body of research has found that option data contain important (forward-looking) information for volatility forecasting and risk measurement. At the same time, with the advanced computers and communications technology, high-frequency financial data are now widely available, which usually summarized in terms of realized volatility measure, providing much more detailed information about the current level of volatility.
Motivated by the above interpretation, by extracting the forward-looking and current information of option and high-frequency data based on a flexible two-factor stochastic volatility model, a market risk measurement volatility model is proposed for measuring VaR. To estimate the model parameters, the continuous particle filters-based maximum likelihood estimation method is developed. Using iVX index and realized volatility measure as the proxies of the Shanghai 50ETF option and high-frequency data, an empirical study for the proposed market risk measurement volatility model is presented. The results show that the two-factor stochastic volatility model incorporated with option and high-frequency information can provide more accurate volatility estimates even in the rapidly changing market environment, which leads to more accurate VaR estimates than other volatility models, including the GARCH model incorporated with only historical information, the realized GARCH model incorporated with high-frequency information and the single-factor stochastic volatility model incorporated with option and high-frequency information. In particular, the proposed model can improve the accuracy of VaR estimates more significantly over others under the extreme risk condition.
The empirical results highlight the values of the information content of option and high-frequency data and the two-factor volatility in market risk management. And our model provides an efficient and promising tool to measuring VaR.

Key words: VaR, option pricing, high-frequency data, two-factor stochastic volatility, particle filters