Analysis of Executive Transaction Characteristics Based on Machine Learning Cross-Validation

Abstract

This paper adopts the machine learning cross-validation method, selects the transaction data of the increase and reduction of the stock holdings of executives of A-share listed companies from 2008 to 2018, and conduct 10 cross-checks on the data set to analyze the characteristics of executive transactions. On this basis, the accuracy of the model was further tested with 2019 data. The research found that: (1) among the four models of machine learning, the decision tree model is more prominent than regression analysis, random forest model and support vector machine model in terms of recall, precision, accuracy and F1 measurement. And the data forecasting ability is the best, which can accurately predict the trading behavior of executives; (2) the stock market environment of listed companies has been constantly changing, and the cumulative excess returns of executives’ share holding transactions have shown an inverted V shape, that is, the short-term return is negative and the long-term return is positive; (3) the short-term stock return is always negative before and after the executive share reduction transaction.