| In an environment where China’s securities investment market is on a volatile trend and the international situation is complex and volatile,the performance of the fund market also shows obvious sector rotation along with the stock market,and diversification for more stable returns gradually becomes the mainstream demand of most investors.Based on this,the development of FOF(Fund of Funds)is also gaining more and more attention.In 2017,China’s public FOF was officially launched,which is different from the traditional fund operation mode,and through flexible allocation of various equity asset products,FOF avoids exposing risk to a fixed equity asset,and has smaller retracement when the market has sector rotation,thus reducing losses.With the official approval of China’s public FOF,the number of domestic scholars studying FOF has gradually increased,and the two core steps of building a FOF are fund screening and asset allocation.The prerequisites for constructing a top-performing FOF portfolio still need to be improved,and there is still relatively little research on how to screen high-quality funds in China.Therefore,this paper aims to build a system of using integrated learning models from public fund screening to asset allocation model allocation weights on the basis of existing research,using all open-end public funds as alternative funds,and finally empirically analyzing that the combination of two models can build a reasonable and robust FOF portfolio.With the perspective of constructing a FOF portfolio strategy,this paper systematically investigates the principles of three integrated algorithms,XGBoost,LightGBM and CatBoost,and three asset allocation models,the minimum variance model,the risk parity model and the equal weight model,in the theoretical part.The three integrated learning algorithms are also the latest models in the Boosting algorithm,which are highly efficient in solving forecasting problems,have a wide range of application scenarios,handle large-scale data and have high accuracy rates.In this paper,we use three models to train indicator data,perform data pre-processing,feature factor screening,parameter tuning,and quarterly rolling test the error value of the three models in the position period as the result of model construction,analyze and compare the learning performance of the three algorithms,and comprehensively select the one with the smallest error value as the integrated model for screening funds,and then use the model’s predicted return of the next period for the positioned funds as the basis for determining the fund Then,we use the model’s prediction of the next period’s return of the fund positions as the basis for determining the underlying funds,and then obtain the underlying funds for each position period.The SLSQP algorithm in Python and Wind’s asset allocation module are then used to calculate the weights of the above three asset allocation models for each position period and make a visual return curve.The optimal asset allocation model is then selected.From the results of the study,the innovative application of the integrated learning approach to FOF fund screening is feasible,and the CatBoost algorithm performs the best regression prediction for 19 funds in the holding period.Subsequently,three asset allocation models were used for backtesting analysis,and the risk parity model was able to achieve substantial returns with stable overall risk control and performed the best.The empirical evidence shows that the theoretical system proposed in this paper for constructing the FOF framework is feasible,and also provides a reference for constructing FOF portfolios in the future. |
