Research On Nowcasting Of A-Share Coal Industry Revenue Growth Based On Random Forest Model

Coal is the biggest consumer energy in China.People’s livelihood construction,economic development,and scientific and technological progress can not leave the development of coal industry.As the micro-main body of the coal industry,the development of coal enterprises is affected by marketization.The year-on-year growth of the operating revenue of the industry reflects the changes of the industry demand in the market,and is one of the important reference indicators of the industry prosperity.Taking the industry revenue growth rate as the point of view,this paper construct a nowcasting model of revenue growth of the coal industry according to the economic framework of the coal industry.Firstly,the existing economic framework of coal is summarized and expanded.Supply and demand are the basis of determining price,transportation is the bridge connecting supply and demand,and is a part of the formation of terminal price.Inventory can be regarded as the reservoir of price,which determines the price fluctuation,and finally sold in the industry.The above constitute the economic framework of"supply-transportation-inventory-demand-price-enterprise revenue".Then the revenue growth rate formula is disassembled,and the expression of the relationship between revenue growth rate and price growth rate,volume growth rate,volume and price growth rate is obtained.This formula is used as the basis for the nowcasting model construction in this paper,and the economic data related to the industry is collected and the characteristic project is constructed with this formula:It includes the year-over-year term and the interactive derivative term of volume and price of the original data,and conducts the time difference correlation analysis and data alignment,which not only retains the principle of time series prediction,but also meets the data requirements of subsequent modeling.Based on the characteristic data,a fusion model of rolling forecast framework of revenue growth and random forest algorithm is constructed,and the relevant model parameters are set.Then the nowcasting models of three forecast targets of industry revenue growth are constructed with industry economic data from January 2011 to September 2021.The results show that the rolling model is better than the traditional multi-step prediction model under each evaluation index,and the rolling window model is better than the rolling starting point model,with R~2 exceeding 0.8,average error fluctuation less than 10%,accuracy over 90%,and inflection point victory rate closing to 65%.Thus,it is verified that the relationship between variables and forecast targets changes with the passage of time in economic forecasting,and the experience that more training sets,better result is not always correct in a changing economic model.The rolling window model has achieved better results in the application of goodness of fit,error,accuracy and inflection point victory rate,and is more suitable for the coal cyclical economic industry.Finally,based on the disassembly of the feature engineering principle and the rolling model neutron model,this paper analyzes the variables that have important influence on the three prediction objectives.By analyzing the duration,characteristics and lead time of the variables,the validity of the feature engineering in this paper and the heterogeneity of the rolling model sub-models are verified.According to the temporal characteristics of variables and their comprehensive influence on the three prediction objectives,the importance analysis of social network variables based on temporal weighting is constructed,and the scoring ranking of variables is given.It is found that the variables objectively are screened by the model overlaps with previous studies,which verifies the rationality of the model.Meanwhile,new potential effective variables are obtained for the reference of subsequent research.