Study Of Indexing Portfolio Construction Based On Random Forest

Indexing investment is an investment model, which means that individuals purchase some or all component stocks to trace the index in a relatively mature securities markets, and based on risk diversity fully and passive management to minimize the risks, to gain market average revenue. The index fund has been developing rapidly since the first opening index fund appears in 2002 in China. The Shanghai 50 ETF option as the first stock options in China is traded in Shanghai Stock Exchange, which will promote the development of index fund greatly.The stratified sampling method is based on analyzing the structure of the securities market, and then chooses the most representative index to fitting the target index. This paper extends the application of random forests, which is used to selection the component stocks index investing, and it will enrich the construction of stratified sampling methods. Specifically, the paper selects the Shanghai 50 index as the benchmark index in the study of the applicability of random forest in the stock selection which is the first step index portfolio construction. The importance of each index component stocks is calculated by random forest model. 10 relatively important constituent stocks are selected,then the further analysis of characteristics of this 10 stocks in the index constituent stocks can be used to explain the applicability of random forest stock; secondly, in the study the second step process of index portfolio construction which is equity distribution, the weight allocation is given by a capital mode on Tracking Error Minimization and a cointegration model on price sequence after finishing the selection of index tracking component stocks. The article analyzes the performance of the two models, and compares the performance of random forest and other stratified sampling method in different capital allocation model, and then get conclusion of the applicability of random forest in the whole process of portfolio construction.After the Simulation verification, this paper finds that the stocks chosen by random forests have a better performance in fitting index trend compared to the greatest weight sampling method, industry and cluster stratified sampling method. Furthermore, MAD model based on random forest is the best among all the stock portfolio weight ratio models. This article also indicates the Tracking Error Minimization model has a littler error and a better fitting effect, but the cointegration model has a better performance. In rebalancing tests, the low tracking error makes the latter rebalancing of low cost and volume.