Reliability of analyst recommendations based on the DS-LightGBM model

Abstract

Securities analysts play a crucial role in the stock market, and their stock recommendations have an important impact on investors' investment decisions. The key to fully leveraging the value of analyst recommendations as an information resource lies in determining the reliability of these recommendations. This study proposes a method for predicting the reliability of analyst recommendations based on the Dempster–Shafer evidence theory and the LightGBM model (DS–LightGBM). The DS–LightGBM model is constructed by incorporating the LightGBM algorithm into evidence theory, which consists of three dimensions: analyst characteristics, rating characteristics, and company characteristics. In the process of reliability prediction, the initial step involves assessing the reliability of the evidence, followed by employing the DS synthesis rule to fuse the information, along with the explainability provided by the SHAP method. The effectiveness of the proposed method is validated through experiments using analysts and A–share market data in China. When compared to the prediction outcomes of random forest, AdaBoost, and similar models, it becomes evident that the DS–LightGBM model exhibits superior prediction accuracy. Additionally, this model effectively measures the contribution and relevance of features, thereby improving the model's explainability and dependability. Consequently, it offers investors, brokers, and other information users a more precise foundation for decision–making purposes.