Forecasting Model of High Transfer Stock —Based on Integrated Learning

Abstract

High transfer is often considered by many investors and researchers as the embodiment of the company&s good operation ability and growth ability, which is of great research value to individual investors, enterprises and institutions. The purpose of this paper is to study the prediction of high transfer based on integrated learning, select the important factors that affect the company’s dividend and share expansion scheme through the design of multi-dimensional statistical indicators, and build a classification model of predicting high transfer by means of the fusion of various gradient promotion algorithms. First of all, according to the annual data, daily data and basic data of stocks, the factors that have great influence on the implementation of high transfer scheme of listed companies are screened out. Then, taking important factors as explanatory variables and whether the next year is high transfer or not as dependent variables, four integrated learning algorithms, XGBoost, LightGBM, CatBoost and Random Forest, are used to construct classification prediction models. With the above four classifiers as primary learners, multi-response ridge regression as secondary learners, and Stacking fusion model was adopted, the test shows that the evaluation indexes of the fusion model are slightly better than those of a single primary learner, and the accuracy and precision are about 88 % and 65% respectively, in addition, the AUC can up to 0.8746. A novel machine learning algorithm is used to analyze the high transfer stock of listed companies in the eighth year. The model has good prediction ability and generalization ability, and has been improved under the integrated learning mode, which has certain reference significance.