Stock Price Manipulation Detection Based on Machine Learning Technology: Evidence in China

Stock price manipulation has become a big concern in stock markets, especially in emerging markets like China. This paper aims to employ machine learning methods to detect the stock price manipulation in China to increase the market fairness and transparency. Based on the information given by China Securities Regulatory Commission, we took the difference of stocks between manipulated time and normal time based on their daily return, trading volume, stock price volatility and market value. We used them as explanatory variables. Then we employed single model, Support Vector Machine (SVM), and ensemble model, Random Forest (RF) for detection. Test performance of classification accuracy, sensitivity and specificity statistics for SVM were compared with the results of RF. As a result, we found that both of them have a meaningful accuracy while RF outperforms SVM. We also found that daily return and market value have a bigger effect on detection than other explanatory variables do.