Study On The Methods Of Identifying Stock Price Manipulation Through Neural Network Models

Since the world's first stock exchange was established in Amsterdam in 1680, the development of stock market has been constantly plagued by stock price manipulation. Price manipulation has seriously distorted the stock market's mechanism of price discovery and impeded its fairness and effectiveness. Price manipulation has also damaged the stock market's resource allocation function and greatly harmed the interests of the majority of the ordinary investors. Since established 20 years ago, China's stock market has also suffered greatly from price manipulation. Price manipulation and other irregularities post a big threat on the healthy development of an emerging market and these negative impacts should not be overlooked. Therefore, the study of stock price manipulation is of great practical significance both to the global market and to China.This thesis combines the results of theoretical studies by foreign scholars on price manipulation with analysis of the Chinese market and focuses on the characteristics of stock price manipulation in China. This paper has collected all the penalty announcements of price manipulation published by China Securities Regulatory Commission since 2001. Based on these cases, this paper studies the classification of price manipulation methods in China and analyzes the manipulated stocks regarding their industry distribution and capital and financial situation. It is found that the main type of price manipulation in China is transaction-based manipulation and transaction-based manipulation in China can be further divided into two different types - short-term manipulation and long-term manipulation. These two types of manipulation differ significantly in terms of the structure of manipulators, the number of accounts used, manipulation duration and the mechanism of how price is influenced. In addition, empirical research on the manipulation objects discovers that the manipulated stocks are widely distributed among all industries, though general stocks and financial stocks are more vulnerable to manipulation. The majority of the manipulated stocks have a small-scale equity, assets and liabilities. Their profit-making abilities are comparatively poor and they have a low asset-liability rate, which might be took advantage by the manipulators.Furthermore, this paper discusses the changes of rate of return, turnover rate andβcoefficient before manipulation, during the initial stage of manipulation, the whole manipulation period and after manipulation. It also examines the significance of the difference before and after the manipulation. Results show that before manipulation, the manipulated stocks have a relatively low rate of return and turnover rate; during the initial stage of manipulation, the rate of return and turnover rate rise significantly; the averaged rate of return and turnover rate of the whole period of manipulation are at normal standards; when the manipulation is over, the turnover rate increases rapidly and the rate of return slumps to below average market return. Besides, before manipulation, the manipulated stock has a positive correlation with the market index and during the manipulation, the correlation between the manipulated stock and the market index weakens or becomes a negative one.Finally, this paper uses Enterprise Miner of SAS to establish a model which can use a stock's price, turnover, volatility, abnormal rate of return and technical analysis data of six successive days to determine if the stock has been manipulated during the six days. This model assesses the effectiveness of three different data mining models - tree model, regression model and neural network model. It turns out that the neural network model has performed best. ?...