Study On A New Method Of Index Replication Strategy

The key of indexation investment management was to choose the appropriate methods and techniques to construct the optimal portfolio,to minimize the tracking error between the tracking portfolio and the target index.Based on the research of scholars at home and abroad,this paper put forward a new index replication method of picking stock by using clustering,and then using the time weighted support vector regression to obtain the optimal weight of the portfolio.This paper used the Shanghai and Shenzhen 300 index as a benchmark index,built a tracking portfolio which consisted of 30 constituent stocks.In the stock selection phase,firstly weused the risk characteristics of the system that is the beta coefficient of the constituent stocks.According to the different characteristics of the trend of risk system to hierarchically cluster the stock,the stock portfolio built in this way can effectively spread the risk and solve the lack of industry classification.The efficiency of the cluster was analyzed,by comparing with the industry classification of the CSI of Shanghai and Shenzhen 300 constituent stocks,it was proved that the cluster was effective.Then,after the calculation of how many stocks should be selected per class,according to the average market value of stock per class and the volume index to synthetically pick stock,finally obtained a portfolio of 30 stocks.In the stage of stocks weight optimization in portfolio,considering that the financial data had time series,that is,the closer the data to the present,the stronger the impact on the future,the remoter the data were,the smaller the impact on the future,so it was very necessary to consider the time factor in the index tracking method.Based on above,this paper constructed a support vector regression(SVR)index replication model based on time weight on the basis of the above 30 stock portfolios,hoping to solve the time series problems of financial data when constructing the tracking portfolio.Calculated the optimal weight of the selected stocks by time weighted support vector regression(SVR)model,and ultimately obtained the optimal tracking portfolio of index replication tracking.In the empirical analysis,this paper compared the established time weighted SVR index replication model to the traditional quadratic programming weight optimization model and the SVR model without time weight.The results show that there was the optimum tracking effect and the smallest tracking error inner and out of sample in tracking portfolio obtained from the SVR index replication model.which were revealed not only on the tracking error inner of the sample,but also the tracking error was minimum on test set out of the sample.Moreover,thetraditional quadratic programming weight optimization model had the worst performance of tracking effect,although the tracking effect on test set was similar to the common SVR index replication model,but the performance on the test set was obviously not as good as the SVR model,because the SVR model was based on structural risk minimization principle which can improve the generalization abilities of the model well.