A Study Of Brent Crude Oil Spot Price Forecasting Based On Identification Of Sufficient Necessary Conditions

Since the reform and opening up,China’s economy has entered a stage of rapid development,and the demand for crude oil from industries such as industrial and agricultural production,national defense and transportation has been increasing day by day.However,China’s basic energy situation of "rich in coal,poor in oil and little in gas" makes the crude oil production unable to make up for the huge crude oil consumption gap.The stable operation of China’s economy and society will depend on a large number of crude oil imports for a long time has become an unchangeable fact in the process of China’s current social development.Therefore,it is crucial to analyze the factors influencing the formation of the daily spot price of Brent and forecast its future trend and fluctuation to ensure the safety of China’s crude oil imports and ensure the high-speed and stable development of China’s economy and society.The formation process of daily Brent crude oil spot price is extremely complex.Thus,a comprehensive and adequate analysis of the impact factors will not only contribute to understanding the price formation mechanism but also conduce to predict the future trend of daily Brent crude oil spot price more accurately.To address this reality,this paper centers around the following three aspects,namely,the construction of the set of impact factors,the analysis of the factors,and the forecasting of Brent crude oil spot price.Firstly,a set of 30 influencing factors is constructed for a comprehensive analysis of daily Brent crude oil spot prices,taking into account five types of factors:crude oil market fundamentals,crude oil futures market,macro-financial market,OPEC policy and uncertainty.At the same time,some proxies are proposed to indirectly reflect the influence of supply,demand and policy factors,given the limitations of some indicators,such as low reporting frequency and difficult data availability.Second,based on the constructed set of influencing factors,the local correlation and leading lag relationship between each influencing factor and the daily Brent crude oil spot price are analyzed using wavelet coherence and phase difference in wavelet analysis to find the factors that precede the daily Brent crude oil spot price change in the set of influencing factors,which are the sufficient condition.After that,necessary conditions analysis is introduced to further identify the necessity of the sufficient conditions for the daily Brent crude oil spot price,so as to find out the sufficient necessary conditions affecting the daily Brent crude oil spot price change and eliminate the redundant factors in the set of factors.Wavelet coherence analysis and necessary conditions analysis are not limited to the assumptions of data stationarity and linearity,and are more suitable for analyzing complex data sets.Finally,bi-directional long short term memory neural network with both "long memory" and "short memory" is used to build the forecasting sub-models to analyze the effectiveness of each sufficient necessary condition on the daily Brent crude oil spot price forecasting study.At the same time,in order to combine the effects of each sufficient necessary condition and break the limitation of the single condition in performance improvement,the multi-objective bonobos optimization algorithm is used to optimize the weights of each sub-model,and the weighted average of each sub-model is the final combined prediction model with both high prediction accuracy and high prediction direction accuracy.The results of the study confirm that firstly,among the set of 30 influencing factors constructed,only 16 influencing factors precede the change of daily Brent crude oil spot price and are sufficient conditions for the analysis of daily Brent crude oil spot price,while the remaining 14 factors lag the change of daily Brent crude oil spot price and do not influence its price formation.Among them,the spot price of OPEC’s crude oil basket is a strong sufficient necessary condition;the Baltic Crude Oil Freight Index,Hang Seng Index,USD/CNH exchange rate and Google Search Trends Index are moderately sufficient necessary conditions;the SSE Index,Geopolitical Risk Index,Geopolitical Act Risk Index,VIX,Gold ETF Volatility Index,Dow Jones Volatility Index and Crude Oil ETF Volatility Index are weakly sufficient necessary conditions;while Dollar Index,Brent crude oil futures-spot spreads,spot spreads of Brent crude and OPEC basket crude,and Systematic Risk Index are sufficient but not necessary conditions,and their necessity results are not statistically significant.Second,each of these single sufficient and necessary conditions positively affects the forecast performance of daily Brent crude oil spot price,but the direction and the degree of action are not consistent.For example,the strong sufficient necessary condition has the largest improvement in directional accuracy and the weakest improvement in accuracy,while USD/CNH exchange rate has the largest improvement in accuracy and the second largest improvement in directional accuracy after the spot price of OPEC crude oil basket.Finally,the combined forecasting strategy can combine the advantages of all sufficient necessary conditions to improve forecast accuracy without reducing or even improving the directional accuracy.In the one-step forecast,for example,the combined forecasting model achieves the performance improvement of 6%-7%in forecast accuracy metrics and over 37%in forecast directional accuracy metrics.