Prediction of return on equity of the energy industry based on equity characteristics

Abstract

We take the return on equity of energy enterprises as the research object to predict it. Our research adopts a new framework to solve multivariable time series problems. Compared to a single regression model, this model focuses more on the results of the regression equation rather than the coefficients of each indicator. Compared to the single machine learning regression method, this model can use the two-way encoder representation of the Transformers model to embed text data into the data, and then use the XGBoost model for regression model processing after PCA dimensionality reduction processing, thereby improving the accuracy of model prediction. Comparative experiments have verified that the method we use has advantages in terms of prediction accuracy.