Chinese Stock Index Futures Regression Prediction Based On SVM

In china， stock index futures has been introduced in April16，2010. Since thelaunch of Chinese stock index futures, regulators and investors have an urgent need todraw on some theories and methods about prediction. While previous studies aremostly limited to the development and implementation of trading rules and the way,or take other financial index as the subject of study which lack of deep research onparameters choice, resulting in accuracy rate is not very high.Starting from the selection of indicators, the paper selects eight indicators toforecast the next day’s opening price, such as the opening price, the highest price, thelowest price, the closing price, the growth rate, the vibration amplitude, the totalhands and the turnover, which can reflect the transactions of the day. And wedownload all daily trading data of April16,2010to June15,2012from a stock indexfutures trading system named Tonghuashun. We use genetic algorithm (GA) andparticle swarm optimization algorithm (PSO) to optimize the support vector machine(SVM) with four different kernel functions to form eight programs, and we havecompared the accuracy and the timeliness of all the programs by empirical study.Experiments show that: the linear kernel function SVM optimized by PSO is the bestmethod in regression prediction of Chinese stock index futures. Then the paper doesregression prediction on The Shanghai Composite Index and The PetroChina dailyK-line index, and gets good results. It indicates that the model can also be applied toother financial index forecast, and has broad applicability.The study has achieved the desired effect, not only constructed linear kernelsupport vector machine optimized by particle swarm regression prediction model ofChina’s stock index futures, enriched the content of Chinese stock index futures, butalso explored the parameters optimization and kernel function selection of SVM, which will contribute to SVM parameter optimization and kernel function selectiontheory.