Application Of SVM In Security Investment Analysis

SVM (Support vector machine) is a new tool which is used to solve machine learning problems with optimal methods, invented by Vapnik in 1990s. Recently, it gained great developments both in theoretic and arithmetic. SVM is a useful method against "Curse of Dimensionality" and "over fit". In the paper, firstly we convert the analysis of security investment to learning problems. Then, Let SVM learn the behavior of security market, gained a classified model (or discrimination function) which is similar to rules in technical trading system. The main research we had done as follows.First, introduce several technical indicators which popular in China security market, and according to rules of indicators we research SHSE index and SZSE index. The results indicate that two of three rules is useful.Second, we write program code for the SVM arithmetic with matlab.Third, we established SVM classified model for predicting stock price trend and research three stocks and two indexes. The results indicate that predicting accuracy is around 60%, the stability for predicting indexes is better than predicting stocks.Finally, we established SVM classified model for selecting good stocks. We research all stocks in Shanghai Exchange from 1999 to 2005. The selected stocks have higher EPS(Earning Per Share) than the average of all.The innovation of this paper is that we established the two SVM classified model: price trend predicting model and stock selecting model. The former, assume that technical indicators is useful for security investment analysis, and whether future price up or down reflects in technical indicators. So regard technical indicators as input vector, we gained a discrimination function, which put out price up or down. The second, assume that financial indicators of a company will affect its profit next year. So regard financial indicators an input vector, we regard a discrimination function which put out the company's profit high or low.