Can High Frequency Information Help Predict VaR In Oil Market?

Crude oil price volatility is closely related to derivative pricing,risk management and portfolio selection.Therefore,it is very important for investors,practitioners and regulators to accurately measure the market risk in the crude oil market.VaR(Value at Risk),as the most mainstream index to measure market risk,is widely used by financial institutions at home and abroad.At present,the crude oil market VaR measurement methods mainly include historical simulation method and GARCH model,which estimate based on traditional low-frequency data.Historical simulation method is widely used in practice,while scholars prefer to use parameterized GARCH method.However,the industry and academia have not reached a consensus on which method has the best prediction accuracy.In recent years,with the emergence of high frequency data,scholars have proposed various Realized measures based on high frequency data(RV、RK,etc.),which provides rich tools for scholars to predict VaR in crude oil market with highfrequency information.Compared with low-frequency volatility,realized volatility contains richer intraday information,which provides potential for more accurate prediction of VaR in the crude oil market.However,there are still many disputes about the implementation methods in academic circles,which makes it difficult to apply VaR prediction in crude oil market.In general,there is still no consistent conclusion on how to choose a more accurate model to predict VaR in the crude oil market.Therefore,based on a unified backtest framework,this paper carries out rigorous posterior analysis on the accuracy of different types of prediction models,thus providing systematic and scientific guidance for risk managers in the crude oil market,which has important theoretical and practical significance.The main research work and innovation of this paper are as follows:First,this paper provides the first study to apply a large number of realized measures to VaR prediction in the crude oil market.In this paper,a large number of VaR prediction models are formalized compared under a unified back measurement framework,including GARCH models(RV,RK,RQ,etc.)that adopt various highfrequency realized measures.And historical simulation method(THS,BRW)and traditional GARCH model(GARCH,EGARCH,GJR-GARCH,etc.)based only on daily low frequency data.Secondly,because different markets have different microstructure effects,this paper further studies which realized measures is most suitable for predicting the value at risk of crude oil market,providing guidance for practitioners.Specifically,this paper incorporates 30 realized measures into the realized GARCH model,which spans 10 different classes of estimators and uses hf data at 3 frequencies(1 min,5min,and 10 min).The main conclusions of this paper are as follows:First,the empirical results of this paper show that high frequency information is helpful to predict VaR in crude oil futures market.The realized GARCH models incorporating RQ volatility showed relatively good risk measurement accuracy,especially the models with 1-min high frequency information almost all passed the test.This paper argues that RQ measures is more accurate than historical simulation method and traditional GARCH model in predicting the value at risk of crude oil market.Second,not all implemented methods are helpful for predicting VaR in the oil market.In fact,most of the implemented GARCH models that include implemented metrics(e.g.,RV,RK,RRV,RS,etc.)have worse predictive performance than traditional GARCH models based on daily data.Thirdly,the choice of sample frequency will significantly affect the prediction accuracy of the implemented method.The results of this paper show that the implemented method based on sampling data at 1-minute frequency shows better predictive performance than the implemented method at 5-minute and 10-minute frequencies.Finally,we find that the historical simulation methods used most by financial institutions show the worst prediction accuracy.Therefore,in order to avoid potential losses,this paper does not recommend practitioners to use this model to predict the var of crude oil market.