Design And Research Of Quantitative Trading System Basing On Data Mining Methods

By the end of 2013,the size of quantitative investment and hedge funds has reached 2.7 trillion dollars.Capital size of various funds and other institutions has approximately accounted for 30% of the total global investment.In many large stock exchanges around the world,near 50% of volume is from different style of quantitative investments.In order to build quantitative trading strategies,statistical analysis of the information in the securities and futures market is firstly needed,and then quantitative models are back-tested using historical data.Models which have good effect and are stable will be applied in practical operations.Combining with practical applications for quantitative trading,this article designs a quantitative trading system basing on data mining methods.Main development tool is the numerical calculation software—MATLAB.Four core modules are designed to support simple trading decisions,including quantitative stock-selecting,strategy back-testing,timing analysis and portfolio management.The main contents and results are as follows.(1)Investment portfolio is built using a multi-factors stock-selecting model,and portfolio can outperform performance benchmark over a long period.A stock's rise and fall classifier is built basing on the SVM algorithm to predict all stocks' future ups and downs.More importantly,a two-stage clustering model is built by combining hierarchical clustering with K-means clustering.Finally the most profitable stock category is selected.(2)The classical double moving average trend strategy is back-tested according to CSI 300 stock index futures' 5-minute closing data.A higher annual Sharpe ratio is got in the sample's test set using parameters scanning to optimize parameters.(3)Stock price trend is predicted using a ARIMA model and a grey model.Market index trend is predicted using a Markov model and a SVM regression model.The results show that the predictions for stocks have large errors and predictions for market index have higher accuracy.(4)20 stocks which have highest return rates in the past period of time are allocated basing on the mean-variance model.By setting allocation ratios of stocks and industries,every stock's percent can be obtained when the portfolio risk is set to minimum.3 most representative mutual funds are analyzed and compared using performance evaluation ratios,and overall performance is assessed.Portfolio's VaR is calculated using 3 VaR model's computing methods.