Forecast The Realized Volatility Of Crude Oil Futures From The Cojumps Perspective

Crude oil is a kind of global commodity which can affect economic activity and financial markets.Hedging is one of the most important functions of futures markets.Therefore,people who need to hedge oil price risk are not only limited to oil producers and refiners,financial market participants and policy makers,but also ordinary business managers who tend to use financial derivatives to improve business performance.Therefore,it is very important to study the volatility of crude oil futures prices for the global economy and financial markets.Financial assets' high frequency return fluctuates substantially in an approximately continuous period during one day,this kind of discontinuous change is known as jump,which is usually associated with specific macroeconomic news.When two or more assets or indices jump at the same time,that is the cojump,will bring a harder shock on the capital market.Modelling and forecasting financial assets' volatility are important branch by making use of jump and cojump.Use jump and cojump to model and forecast the volatility of crude oil futures prices,is a key issue.With the rapid development of financial econometrics,the use of intraday highfrequency data for volatility testing and modeling to predict oil price volatility has attracted a deep concern of domestic and foreign scholars.HAR-RV model(Heterogeneous Auto Regressive model of Realized Volatility),which is a representative and popular model based on high-frequency data,has developed rapidly in recent years.Based on HAR-RV model,to detect whether the jumps occur significantly,we employ the two daily tests to detect the daily jumps and the four intraday tests to detect the intraday jumps,these jump tests are the most commonly used in academics.With different test methods identifying jumps to construct new volatility models,and analyzing the forecasting ability of the old and new volatility models.On the basis of using “the co-exceedance rule” to identify cojump between the crude oil futures market and the American stock market.Then,through a series of technical paths,on the one hand,discuss the prediction ability of new crude oil volatility models by adding signed jumps and signed cojumps(distinguished positive and negative component).On the other hand,the jumps and cojumps are decomposed into large and small jumps(or cojumps)respectively,and analyze the influence of large and small jumps(or cojumps)on model estimation and prediction by constructing new crude oil volatility models.Further,the author treat jumps(and cojumps)as point processes and model their occurrence intensity(or probability)using the Hawkes process model.By empirical analysis,it compares the results of different models' forecasting ability which contain different jump components,include jump(or cojumps)intensity,the signed jumps(or cojumps)as well as large and small jumps(or cojumps).Especially,the author compare the superior volatility models mentioned in the previous chapters,to verify the role of jumps and jumps' decomposition components and their intensity variables in terms of volatility forecasting.Through these series of studies,the main empirical results are as follows:(1)LM test has a superior performance among the six tests to detect whether the jumps occur significantly in the crude oil futures market.To quantitatively compare the significant difference in the predictive ability of each model,the author use two loss functions.And use an advanced statistical test,the Model Confidence Set(MCS)test,to choose a subset of models cont,aining all possible superior models from the initial model set.The author take a series of robust tests to reexamine whether or not the empirical results are robust,such as the alternative of volatility-realized kernel(RK),different forecasting window,recursive window,different forecasting horizons and so on.In general,the out-of-sample performance of HAR-CRV-LM model not only can defeat HAR-RV model but also can defeat other volatility models whose jumps detected by other jump tests.The in-sample estimated parameters of HAR-CRV-LM model are positive,shows a positive influence on future volatility,the existence of jump component will aggravate the future volatility.These results are consistent with previous research.(2)Signed jump and signed cojump are introduced.The in-sample results suggest that the behavior of long-term investors has a greater impact on one-ahead-step predictions.Our empirical results reveal several noteworthy findings.First,the effects of signed jumps and cojumps based on the daily and intraday jump tests on future volatility are asymmetric,and the negative components are much more powerful in forecasting volatility.Moreover,our proposed models,including the signed jump and cojump components,are able to generate higher forecasting accuracy,and we find that disentangling the effects of positive and negative jumps and cojumps can significantly improve forecasts of future volatility.Lastly,the findings are reliable for various robustness checks and our study provides some new insights into forecasting oil price realized volatility.(3)Large and small jumps(or cojumps)are introduced.The authors' results provide strong evidence that the model combines large and small jumps to obtain significantly superior predictive performance.More specifically,the large and small components of the jump have a positive effect on future volatility over the medium and long term.Large jumps averages in weekly and monthly lead to larger fluctuations throughout the day,and large jumps have a stronger effect than small jumps.HAR-CJ*LS-CIJLS model which contains the large and small jump information and large and small cojump information(the cojumps are detected on the basis of using “the co-exceedance rule” by the four intraday tests),has an outstanding performance in predicting volatility.This means that the large and small cojumps between the oil futures market and the stock market contain more useful predictive information than cojumps themselves.Through a series of robustness tests,the conclusion is proved to be robust and reliable.(4)Hawkes process model is used to analyze the probability of oil price jump(i.e.jump intensity).The author add jump intensity and cojump intensity into the volatility models constructed in the previous chapters to build new volatility models,and emphatically compare forecasting performance between the superior volatility models mentioned in the previous chapters.Firstly,modeling jump intensity shows that the jump intensity is negatively correlated with future volatility,indicating the higher the probability of future jump,the smaller the volatility of future oil price.Secondly,instead of treating the oil market as standing alone,the author consider its link with the fundamental economy,and explore whether the extreme co-movements(i.e.cojumps)between the oil market and fundamental economy have explanatory powers for future oil volatility.The author estimate the cojump intensity from the Hawkes process model,and study their usages for future oil volatility,respectively.The results show that the cojump intensity has a positive effect on future volatility,suggesting that the greater probability of cojump between the crude oil market and the stock market,and the more likely to increase of the RV in the future.In other words,the more likely it is that oil-related economic fundamentals will drive a jump in oil prices,the more volatile oil prices will be in the future.It is find that modeling jumps in terms of their cojump intensity largely improve the accuracy of volatility forecasts in oil market.Thirdly,the author then also decompose the cojumps into their positive and negative components,large and small components.The results suggest that extending the HAR model towards the three jump-related dimensions,including jump(or cojumps)intensity,the signed jumps(or cojumps)as well as large and small jumps(or cojumps),can significantly improve the oil volatility forecasting accuracy compared with the benchmark model(HAR-RV-CJ model).These results are widely consistent across a variety of robustness tests.To our best knowledge,this is the first study to exploit the impacts of co-movements in three jump-related dimensions between the oil market and another market on future oil volatility forecasting.Collectively,this study sheds new light on how to harness jump components for oil volatility forecasting.Last but not the least,the author applied the research results into the real stock exchange market,tried to transform the research results into real economic behavior,and achieved some achievements.Further more,this paper analyzes the impact of oil volatility forecasting on the management innovation of Chinese petroleum related enterprises,and puts forward some suggestions.