Research On Effectiveness Of Technical Analysis Based On Kernel Regression Pattern Recognition

Technical analysis is one of the earliest used security investing analysis method by investors.It is an analysis approach that uses market behavior in past times and now to find classical rules predicting future trends.The effectiveness of technical analysis is that whether we can get higher yield if we use technical analysis rather than not.The very topic is always being researched by scholars,and it is not ensured till now.This paper uses kernel regression & pattern recognition to discuss if technical analysis is effective.Kernel regression is an excellent data-smoothing algorithm,whose basic idea is weighted mean.The weights are determined by kernel functions,which is dynamic.We smooth the stock price data by kernel regression to look for nonlinear relations in the price.Pattern recognition is a process of training machine to recognize things,which is an important area in artificial intelligence.This paper tries to use auto-detecting method in Lo.'s essay and neutral networks method to recognize locations of specified technical patterns in smoothing stock price data respectively,so as to calculate yields after the patterns,reviewing whether technical patterns can bring excess profits.This paper uses KS detection method and bootstrap-based detection method to detect conditional & unconditional yields to explore the effectiveness of technical analysis.This paper chooses 84 representative stocks in A-share stock market of 2011 to 2015,and uses their close price to do empirical analysis.The result shows that the stock market of our country is far from weak-effectiveness,and some of the technical patterns can produce excess profits.Besides,in the process of pattern recognition,compared to auto-detecting method,the neutral networks can find more locations of technical patterns,and can overcome the problem of oversensitivity.