Revenue Forecast Of Listed Company Based On Combination Model

The stock market plays a decisive role in the financial market.Both investors and operators of listed companies hope they could have a good understanding of the compay’s business conditions and the development situation in order to make investment choices.The revenue is an important indicator of whether a company is robust or not.Therefore,the ability of revenue-forecasting in investment has a big influence on the result of decision-making.This article establishes four forecasting models to predict the cumulative quarterly revenue of 300 A-share listed companies on June 30,2017.The related theories of XGBoost model,RF model,and SVR model and their respective applications are also described.These three types of models are widely used in classification and regression problems.Based on the characteristics of the original financial data,this paper innovatively combine the characteristics of market data,and establishes XGBoost and RF regression prediction models.The prediction of these two models shows that the main financial characteristics that affect revenue are operating expenses,cash paid for goods and services,business taxes and surcharges,so do market value and turnover value of listed companies.The results shows that there is a direct correction between the revenue of listed companies and the stock market.The MAPEs of both models are about 4%,which means a good prediction effect.Based on artificially constructed features,this paper establishes a regression model for forecasting quarterly revenue based on SVR,which predicts the quarterly growth rate of revenue.Experiments show that the prediction results of this model are better than the previous two models,with the MAPE is reduced to 3.1%.In order to further improve the prediction accuracy of the model and make it interpretability,this paper creates a combined model based on the fusion of the aforementioned three models,and integrate the results of the three models mentioned above as the final prediction result,which reduced the MAPE to nearly 1%.