Research On The Fund Performance Prediction Method And Application Based On External Attention Mechanism

Securities investment fund plays an important role in our capital market.The development of securities investment fund also guarantees the perfection and stability of the market and provides more investment options for investors.Today,with the rapid development of fund industry,with the continuous expansion of fund scale,fund as an important commodity in the capital market audience is also increasing,and in the process of selecting funds,fund performance has naturally become the focus of attention of investors at all levels.For investors,the focus is on the fund can bring their own income,the use of fund related indicators to predict the fund is the focus of many investors.Therefore,this paper incorporates an external attention mechanism and incorporates different levels of data related to fund performance into the model to construct macro-industry-micro fund level data.The study is also conducted from two aspects,namely,the fusion of hierarchical data and the fund performance prediction method based on fused data.Based on fund net worth data,fund return data and macro industry related data,the fund performance is predicted by machine learning methods and the impact of different levels of feature indicators on fund performance prediction is analyzed.The specific conclusions obtained from the empirical study based on the external attention mechanism and fund performance forecasting methods are as follows:First,the attention module upstream of the model is constructed based on the external attention mechanism,which provides fused data based on different levels of indicators for the downstream performance classification task.It is found that the loss indicators RMSE and MAE of the model are minimized by fusing fund-related macro data,industry data and the fund’s own data through the external attention mechanism.Secondly,the prediction evaluation results based on the model show that the accuracy rate of the three types of fund performance prediction reaches98.86%/99.67%/99.23%,and the F1 score also reaches 96.29%/97.56%/98.34%.And on the basis of the integration of external attention mechanism,the contribution of different characteristics of the model was compared and analyzed,and the impact of different characteristics on fund performance was found from Sharpely value.Finally,considering the marginal contribution of different characteristic indexes to fund performance forecast,the investment suggestions of related funds are given.That is,different investment strategies should be adopted for different types of funds in different market environments.When investors choose funds,it provides more in-depth evaluation methods and different characteristics of the analysis method.