FOF Investment Strategy Research Based On Ensemble Learning

Along with the domestic fund market unceasing and the international market conformity,the fund that domestic design also become more and more abundant.The first batch of publicly listed FOF funds was approved in September,2017,which marked that the development of public funds entered a new era.The emergence of FOF funds not only enriches China's financial product system,but also helps investors select funds through professional management,so that risks can be dispersed twice.In addition,FOF funds can also help investors to allocate assets globally and prevent the risk of unilateral market decline.As public offering funds approved by the public,domestic scholars' fund research more and more,the fund's performance mainly depends on the asset allocation ability of fund managers and sons' fund selection ability,at present domestic scholar's study are mostly limited to the asset allocation,especially the research on risk parity,the mean variance model of asset allocation,few people study how to select funds,while a portfolio choice for fund performance is critical.Aiming at this situation,this paper adopts Xgboost,Lightgbm,Catboost and other methods in integrated learning to select investment targets,and combines the minimum variance,risk parity,and equal weight model to construct FOF fund,in an attempt to provide FOF fund managers and FOF fund researchers with a new way of selecting investment targets and allocating weights.The theoretical part of this paper mainly expounds the principles of the five models in integrated learning(including Xgboost,Lightgbm,Catboost,GBDT,RF),and then introduces the commonly used asset allocation models,including mean variance model,risk evaluation model,and other weight models,and respectively illustrates the advantages and disadvantages of these methods.In the empirical analysis part,this paper refers to a large number of literatures,selects 26 indicators that can most affect fund performance,and then constructs 759 new features based on these 26 indicators through feature engineering.Finally,785 features(including the original 26 features and the newly constructed 759 features)are used for modeling.Then,based on the data of the last 4 years,integrated learning method(including Xgboost?Lightgbm?Catboost?GBDT?Random Forest)was used to predict the rise and fall(classification)and cumulative return rate(regression)of the sample fund in the next period,and AUC,Recall,Precision and RecallPrecision graph were used to evaluate the effect of the model.In view of the final score of the model,select the public offering fund with the top 10 rating as the investment target of the FOF fund,and then conduct the combination according to the Equal Weight method,Risk Parity method and Minimum Variance method,and conduct fitting comparison with the out-of-sample data from October 2017 to September 2018,and then in this paper the results of the two 'investment strategy and the CSI 500,the first issue of the six public offering' funds to use the annual rate of return,volatility,maximum withdrawal and other indicators are evaluated to prove the effectiveness of this strategy.It is hoped that the results of empirical analysis can provide FOF fund managers?researchers and investors with some reference for the FOF investment.