Research On Fund Quantitative Investment Strategy Based On Ensemble Learning

In recent years,the domestic fund market has experienced rapid growth,and the number of different funds has exploded.Given the large number of funds,it has become a big problem for investors to select suitable funds.The rise of quantitative investment and machine learning offers a new way to implement.Based on the above,this paper takes the public offering of funds as the research object and uses ensemble learning algorithms combined with portfolio optimization models to try to develop a quantitative investment strategy for funds with low risk and stable returns to solve the fund selection problem and provide investors with a feasible investment scheme.The investment strategy of this paper is divided into two parts:fund screening and asset allocation.In fund screening,we select four types of funds,including equity,bond,hybrid and money market funds,as the fund pool and use the performance data of the last five years to construct 12 characteristic factors.Then,the XGBoost,LightGBM and CatBoost algorithms are selected through ensemble learning to build the fund prediction model,and time series cross-validation is used to adjust the parameters.Different from previous studies,this paper integrates three algorithms by Stacking,explores models with better performance,and then selects the models with best performance to screen funds and build investment portfolios.The empirical result shows that the LightGBM model has the best performance in predicting equity and money market funds,while the Stacking model can better predict the returns of hybrid and bond funds.In asset allocation,based on traditional portfolio optimization models such as the equal weight model,the minimum variance model,and the risk parity model,we also introduce a relaxed risk parity model that incorporates investors’ subjective views into asset allocation to better meet the needs of investors with different risk preferences.The backtest result shows that the equal weight model cannot withstand the risk of a market decline,that returns are not stable,and that overall performance is poor.Both the minimum variance model and the risk parity model have good risk aversion ability,with the minimum variance model having lower risk,while the risk parity model has higher returns.The relaxed risk parity model has achieved the highest annual returns among all portfolios,and accordingly its risk is higher than that of the risk parity model,which is at an intermediate level among all portfolios.