Using Random Forest And QGA-SVR For The Purpose Of Stock Selection

At present in the China’s stock market there are more than two thousand shares. Every day a huge amount of financial data is produced. Stock investors are eager to get effective investment information, and select shares with profit potential to build the effective investment portfolio. In order to select shares with profit potential from a large number of shares, quantitative stock selection has become a hot issue in the field of investment decision-making research. Because the stock market is a high-dimension, nonlinear and complicated system with much noise, the traditional linear model cannot solve this kind of problems well. To solve these problems, in the field of quantitative stock selection, this paper builds the corresponding value investment oriented financial index system, introduces the advanced data mining method, and completes the following research work.With the research object of stock selection, from the perspective of value investment, according to the study of the early literatures, this paper first selects 16 financial index in six categories of listed companies as input characteristic variables, including rationality of price per share, profitability, financial leverage level, liquidity of assets, capital usage efficiency and the growth ability of the company. Second, since these financial characteristic variables have redundant information, in order to allow investors to actually see the influence of each financial index on stock returns, this paper selects the random forest algorithm(RF) in the combination algorithm, evaluates the financial indexes, and selects the important financial indexes. Third, combined with the high-dimensional complexity of the stock market, this paper uses support vector regression machine(SVR) with superior performance in the treatment of high-dimensional and nonlinear system as the basic method framework, uses the superior quantum genetic algorithm(QGA) in the global search algorithm for the dynamic parameter optimization of the support vector regression machine, and deeply optimizes the penalty factor c, nuclear parameter g and slack variable p of support vector regression machine respectively. In order to verify and study the optimization of parameters by quantum genetic algorithm(QGA), it is compared with genetic algorithm, which ensures the accurate effect of support vector regression machine(SVR). Fourth, the support vector regression machine(SVR), random forest(RF) and quantum genetic algorithm(QGA) is combined to build the comprehensive stock selection model process RF-QGA-SVR. Fifth, this paper uses the data of top 200 shares with the highest market value in the A-share market from 2003 to 2014 as the empirical object, uses the RF-QGA-SVR model established to forecast the stock returns, ranks according to the forecasted results, constructs the investment portfolio, respectively screens out the top 10, 20 and 30 stock portfolios, and compares with the market benchmark portfolio returns, so as to obtain the effectiveness of the integrated model RF-QGA-SVR constructed. In the process of stock selection, this paper summarizes financial indexes of listed companies, measures the importance of financial indexes, and hopes that the financial indexes, consideration sequence and RF-QGA-SVR integrated stock selection model can provide scientific stock selection idea to value investors in the actual investment environment.