Forecasting And Modeling For Crude Oil Market Volatility

The vigorous development of the global economy has promoted the status of energy in the strategic and economic situations of various countries nowadays.As one of the most important raw materials for industrial production in the world,also as a non-renewable resource with financial attributes,the fluctuation of crude oil prices affected by market supply and demand,the stock and foreign exchange market and the trend of global policies.In addition,the impact of oil price shock brought by crude oil itself on the economy and financial market of various countries can not be underestimated.The volatility of crude oil prices has deepened with its own degree of financialization,and its relationship with financial transactions such as asset pricing,investment portfolio selection,hedging strategies and risk management in the financial market has become more and more closely connected.Therefore,exploring and understanding the trend of crude oil prices,accurately estimating and predicting the fluctuations of crude oil prices are essential for academic researchers,market participants,and policy makers.In recent years,under the influence of the background of frequent macro and micro economic events（such as financial crisis,global epidemic outbreak,imbalance between domestic and foreign supply and demand,etc）,international crude oil prices have also become more and more volatile.With such a complicated environmental background,the main issues studied in this article are as follows.First of all,examing whether the information of international stock markets have predictive effect on the volatility of the crude oil futures market,selecting the stock market index that has the greatest impact on the volatility of the crude oil market under the structure of the heterogeneous autoregressive joint model.Secondly,on the premise that the US stock market has an impact on the crude oil market volatility,considering the cross-market effect of extracting the crude oil-US market common information contained in the residuals,and then exploring its incremental contribution to the forecast of volatility of the crude oil market.Furthermore,based on the the multivariate heterogeneous autoregressive model and the Markov regime switching,we examine the differences in predicting the volatility of crude oil futures markets by different non-parametric measurement variables.This research will focus on the above issues one by one and discuss the analysis results.Chapter 3 of this thesis based on the Heterogeneous autoregressive realized volatility（HAR-RV）model,considering that the crude oil market is affected by stock market fluctuations,we construct the benchmark model with the information of the crude oil market itself and select 16 realized volatility indexes from 16 different stock markets as exogenous variables to construct 16 competitive models,respectively.Comparing the empirical results,we find the US stock market has the greatest effect on the prediction of crude oil market price volatility than other competitive models.We explain these conclusions separately from theoretical and statistical significance.In order to obtain a robustness result,we test it from different prediction steps,different rolling windows,different realized measurement methods,or during the international epidemic.Compared with other markets,the US stock market has always maintained a steady impact on the price fluctuations of the crude oil futures market.In particular,the prediction ability of the linear joint model jointly constructed by crude oil and the United States has obvious advantages.Based on the conclusions of Chapter 3,Chapter 4 of this thesis further proposes the effect of nonlinear cross-market volatility between crude oil futures and the US stock market on crude oil prices volatility based on two kinds linear heterogeneous autoregressive models（HAR-RV and HAR-RVU）.At the same time,Chapter 4 examines the predictive effect of common information of multiple structures on the volatility of crude oil prices,from the results,we can find that the common information can make up for the shortcomings of traditional measurement and forecasting models,and also it can help to understand and characterize the dynamic interaction mechanism of the financial markets.This chapter explores for the first time that the effectiveness of a multivariate heterogeneous autoregressive model constructed by combining multivariate GARCH and HAR-RV models to predict the future volatility of crude oil prices.The results of the study find that compared to linear models,multivariate models can utilize the DCC-GARCH structure to better extract common information between markets and avoid the over-fitting defects of linear models that are easily caused by too many parameters,so as to achieve the purpose of optimizing the out-of-sample prediction accuracy of crude oil futures market price volatility.In order to compare the effectiveness and universality of the multivariate model,the author also enumerates the comparison of the crude oil futures market’s out-of-sample forecasts on the US stock market,also,the empirical results show that the multivariate nonlinear model has irreplaceable ability than the unit linear models.On the basis of the multivariate heterogeneous autoregressive（MHAR-RV-DCC）model,Chapter 5 of this thesis fully considers the structural mutation characteristics of financial market information fluctuations,and embeds Markov regime switching（MS）to investigate the prediction performance of the crude oil market volatility prediction model under the transition of high and low states,and explores whether the Markov mechanism conversion multivariate model constructed by non-parametric measurement variables will help the government,scholars and practitioners to judge the future market.Based on the previous two chapters,this chapter introduces the negative rate of return leverage factor and signed jump variation（SJV）to construct a linear model of 5 types of different variables,and constructs a total of 20 models about the corresponding linear crude oil-US joint model,nonlinear multivariate MHAR-RV-DCC models,nonlinear MS-MHAR-RV-DCC models.Furthermore,we test the in-sample fitting and out-of-sample forecasting,and a series of robustness tests(including MCS test,DoC direction test,out-of-sample R_oos²,DM test).In comparison,our newly proposed multivariate MHAR-RV-DCC model and MS-MHAR-RV-DCC model have stronger predictive capabilities than other high-frequency volatility prediction models,especially when the SJV is introduced into the MS-MHAR-RV-DCC model,it shows that MS-MHAR-RV-SJV-DCC has the best predictive ability.The above empirical results not only affirm the advantages of the newly constructed multivariate and Markov regime swicthing models in the application of crude oil futures market forecasting,but also broaden the research ideas and specific methods for characterization and prediction of crude oil market volatility in the future.