Stock Selection Strategy Using Trajectory Clustering Based On SOM Algorithm

In recent decades, quantitative investment in developed capital markets has become a mainstream investment. With the rising of complexity of domestic capital market, the traditional investment that is mainly based on subjective judgment has been challenged. Quantitative investment has received more and more investors and domestic scholar’s attention due to its unique advantages. Based on this background, this paper chooses the current domestic A-share market as the research object, attempting to build a quantitative stock selection strategy.The paper describes the research background, expounds the relevant quantitative stock selection research achievements, defines the research problems and put forward a Self-Organizing Map (SOM) related clustering model for picking profitable stock combination. Due to the fact that the stock selection strategy built in this paper is as popular as those mainstream stock selection method, the second chapter of this paper introduces the SOM algorithm in detail with its applications on two groups of artificial data sets, and derived a so-called trajectory clustering based on SOM algorithm. The proposed algorithm has been utilized to choose certain stock combination which is able to outperform the market.The main research work of this paper is divided into two stages:the construction stage of stock selection strategy by clustering and the stage of simulated trading test on this strategy. In this paper, the third chapter firstly illustrates the basic idea of stock selection strategy by clustering, then presents the completed building process of this strategy. The fourth chapter firstly builds the simulated trading algorithm of the strategy, and gives the explanation of this algorithm, then simulated trading test is done using the stock selection strategy. In testing experiments, this paper selected six different test periods:15/06/10-15/07/2214/01/03-15/08/10 14/01/03-15/01/0513/01/04-14/01/1011/12/30-13/01/04 and 11/01/04-12/01/04. First a single transaction test is done in 15/06/10-15/07/22 test period to show specific transaction information. Then for each test period a 100 transaction test is done to examine the effect of the stock selection strategy. The 100 test results of the above six test period is summarizes in the end of the fourth chapter, the results show that the testing period for 15/06/10-15/07/22 14/01/03-15/08/10 and 13/01/04-14/01/10 the test winning percentage is significantly higher than 50%, obtaining 72% 75% and 91% respectively, while the testing period for 11/12/30-13/01/04 14/01/03-15/01/05 and 11/01/0412/01/04 the winning percentage is only 46%,27% and 26%. Despite the poor performance of index in some test period (15/06/10-15/07/22 and 13/01/04-14/01/10), strategy has showed certain advantages, for test period (11/01/04-12/01/04) in which the performance of index is really disappointing, the test result is very unsatisfactory. Overall, the stock selection strategy built in this paper has not yet gained good and stable performance, but the thought and method incorporated in the strategy has certain reference significance for stock investment, further detailed research is worth being done by investors.The last chapter of this paper summarizes and concludes the research work, and suggests some possible future works.