Data Analysis & Mining Of Shanghai Crude Oil Futures Prices And Their Corresponding Spot Prices

Bulk commodities are material commodities in large quantities used for industrial or agricultural production / consumption.At the same time,in the financial investment market,bulk commodities can also be traded as futures,options and other financial instruments,that is,they have both commodity and financial attributes.Data mining of the commodity futures prices and their spot prices can not only help us understand the intrinsic characteristics of them,explore the correlationship between the futures prices and the spot prices,but also provide an effective prediction of the spot prices.This paper aims to select the representative Shanghai crude oil futures as the research object,conduct data analysis / mining on its futures prices and the corresponding spot prices,explore their relationships,especially pay attention to the impact of the crude oil futures prices on the crude oil spot prices.The main contents of this thesis are as follows.Firstly,the basic statistical analysis of the price series is carried out,which includes the descriptive statistical analysis and the multifractal analysis.The former mainly includes the price chart,the change of logarithmic rate of the returns,the descriptive statistical summaries and the cross-correlation function graphs.By calculating the Hurst index,the latter describes the complex fluctuation phenomenon of the financial time series such as the futures prices and their spot prices.Secondly,this paper tackles the correlationship between the futures prices and their spot prices.The specific research steps are as follows.(1)Three test forms(i.e.,with intercept term,without intercept term and with trend term)are utilized to conduct the ADF stationarity test on the prices time series data.(2)A VAR model is established and its optimal lag order is determined according to a few guidelines.(3)The Johansen cointegration test is performed on the time series data of the prices.(4)On the basis of establishing the VECM,the Granger causality test is conducted on the prices time series data to determine whether there is a guiding relationship between the futures prices and the spot prices;in addition,the impulse response function and the variance decomposition are resorted to explore the degree of influence between the futures prices and the spot prices.Finally,based on the codynamic relationship between the futures prices and their spot prices,four basic machine learning models,namely CNN,GRU,LSTM and SVR,are used to predict the spot prices,and the prediction effect is evaluated according to RMSE,MAPE and other indicators.In order to improve the prediction effect,this paper combines the basic machine learning models in pairs,and adds some relating input variables such as the settlement prices,the opening prices and the high / low prices that can affect the spot prices.According to the above analysis,the following conclusions can be drawn:(1)the returns of Shanghai crude oil futures and Shengli crude oil spot have the characteristic of state persistence;(2)the prices of Shanghai crude oil futures and Shengli crude oil spot are mutually guided;(3)among the four basic machine learning models,the GRU model has the best prediction effect;among the combined models,the prediction effect of the LSTM-SVR model is better than that of the single LSTM model and the single SVR model;for the CNN,GRU,LSTM,SVR and LSTM-SVR models,adding input variables could significantly improve the prediction effect.