Research On Energy Volatility Forecasting Based On Uncertainty Indices

As an important support for the economic development of each country,price volatility in the energy markets have an important impact on the economic development process,social stability and national security of each country.In recent years,influenced by various uncertainties such as trade protectionism,unilateralism and geopolitical conflicts,the volatility of the international energy market has intensified,increasing the instability of China’s foreign exchange expenditure and international trade,bringing great uncertainties to China’s energy security,economic development transformation and energy consumption structure upgrading.Against the backdrop of China’s reform and opening-up entering a period of hard work and deep water,modelling and forecasting the volatility of the international energy market by combining various international uncertainty indices is of great practical importance to Chinese government departments in preventing and resolving risks in the international energy market,enhancing national energy security and ensuring the healthy and stable development of the real economy.Meanwhile,in the environment of high uncertainty in global financial,economic and political situation,modelling and forecasting energy market volatility using multiple uncertainties,and discussing which methods can more effectively utilize key information from multiple uncertainty indices to improve the prediction accuracy of energy volatility,has important theoretical implications for further improving and developing existing research related to energy volatility forecasting,asset pricing and risk management.Based on the discussion of the research background and research implications,this thesis focuses on the following questions,using international crude oil and natural gas markets as representatives: What are the differences in the predictive power of investor sentiment and economic policy uncertainty indices in forecasting long-term(monthly)volatility in energy market,and can the MIDAS-RV model show significant advantages when used to forecast monthly volatility in energy markets? Can the MIDAS-RV model show a clear advantage when used to forecast monthly energy market volatility? Is it possible to find an index with the most predictive power for energy market volatility among the many uncertainty indices?Can the MIDAS-RV models incorporating the dimensionality reduction approach produce relatively robust forecasts for the monthly volatility of energy market by extracting the combined information of multiple uncertainty indices? Can the introduction of Least Absolute Shrinkage and Selection Operator(LASSO)technique and Markov regime switching technique based on the MIDAS-RV model fully utilize the advantage of variable selection of LASSO method and the flexibility of Markov regime switching method to produce robust and excellent out-of-sample forecasting effect on the monthly volatility of energy markets?Chapter 2 of this thesis provides some introductions to 5 energy markets and 15 uncertainty indices.This chapter briefly discusses the reasons for our focus on the volatility of WTI crude oil spot,WTI crude oil futures,Brent crude oil spot as well as natural gas spot and futures.Besides,this chapter describes the methodology for constructing the uncertainty indices and discusses the potential reasons for their impact on energy volatility.Chapter 3 of this thesis compares the ability of Global Economic Policy Uncertainty Index(GEPU),the U.S.Economic Policy Uncertainty Index(USEPU),Volatility Index(VIX)and Investor Sentiment Index in improving the forecasting accuracy of the monthly volatility of energy markets.It is found that the Investor Sentiment Index and VIX only improve the prediction accuracy of the crude oil market,significantly,while they have no significant effect on the volatility of natural gas futures and spot.While both GEPU and USEPU are not robust in predicting the volatility of both crude oil and natural gas markets,only USEPU among all indicators can generate economic value beyond the benchmark model in both crude oil and natural gas markets.The results of the supplementary analysis further reveal that the MIDASRV model can better captures the multi-order lagged effects of economic policy uncertainty indices on energy volatility,producing out-of-sample forecasting accuracy for energy volatility that exceeds that of AR models.Chapter 4 of the thesis introduces 15 uncertainty indices based on the MIDAS-RV model in an attempt to find the uncertainty indices that have the most predictive power for the volatility of international enery markets.It is found that the Petroleum Market Equity Market Volatility Tracking Index(PMEMV)and the Financial Stress Index(FSI)have relatively robust forecasting ability for crude oil market volatility.And,PMEMV’s forecasting ability is mainly reflected in low volatility periods,while FSI’s forecasting ability is mainly reflected in high volatility periods.In terms of multi-step forecasting,PMEMV still exhibits significant out-of-sample forecasting power.However,in the WTI crude oil market,the Geopolitical Risk Index(GPR)outperforms the PMEMV in terms of forecasting power,while in the Brent crude oil spot market,the PMEMV has better forecasting performance.However,the predictive power of all uncertainty indices for natural gas asset volatility is dynamically changing.On the one hand,uncertainty indices only have strong long-term predictive effect on natural gas market volatility,and on the other hand,no uncertainty indicator can produce accurate volatility forecast values for natural gas assets at both high and low volatility levels.Chapter 5 of this thesis introduces principal component analysis(PCA),scaled PCA as well as partial least squares(PLS)to the MIDAS-RV model,focusing on whether the MIDASRV-PCA,MIDAS-RV-PLS,and MIDAS-RV-SPCA models can produce robust out-ofsample forecasts of volatility in the five energy markets by capturing the comprehensive information of the 15 uncertainty indices.The results show that only the comprehensive information captured by the PLS and SPCA methods have significant positive impact on the volatility of the five energy markets simultaneously.The out-of-sample forecasting assessment results indicate that the MIDAS-RV-PLS model produces more robust forecasts for all five energy market volatilities than the combination forecasting approaches.Also,the out-of-sample forecasting accuracy of the MIDAS-RV-PLS model outperforms the combination forecasting methods even for longer-term forecasts.However,both the combination forecasting approaches and the dimensional reduction models have significant forecasting power for energy volatility only at high volatility levels.Chapter 6 of this thesis further introduces the LASSO method and Markov regime switching technique on the basis of the MIDAS-RV model to construct the MIDAS-LASSO model and the MS-MIDAS-LASSO model,investigating whether they can produce robust and superior out-of-sample forecasts for energy market volatility by dynamically screening the most predictive uncertainties in the forecasting process.It is found that both the combination forecasting method and the dimensionality reduction models struggle to outperform the MIDAS-LASSO and MS-MIDAS-LASSO modes in predicting monthly volatility of energy markets.Meanwhile,the MS-MIDAS-LASSO model’s out-of-sample forecasting ability for energy market volatility outperforms the MIDAS-LASSO model in most scenarios.Even for longer-term forecasting,the MIDAS-LASSO and MS-MIDAS-LASSO models still have excellent forecasting performance.Also,the MIDAS-LASSO and MS-MIDAS-LASSO models maintain their excellent forecasting power even at different volatility levels.Finally,mean-variance investors focusing on constant Sharpe ratios with different risk appetites and different investment horizons can achieve relatively desirable economics benefits using the forecasting results of the MS-MIDAS-LASSO as well as the MIDAS-LASSO models.