Assessment method for relevance between fund manager management capability and fund performance

Abstract

The invention relates to a method for evaluating the relevance between fund manager management capability and fund performance, and aims to predict the management capability of a fund manager and the contribution of the management capability to the fund performance by establishing a scientific model so as to further optimize fund management and investment decision. The method mainly comprises the following steps: project conversion: converting a research problem into two main exploration directions: one is whether the management capability of a fund manager can be predicted through historical performance, and the other is the prediction contribution degree of the management capability of the fund manager to the fund yield; and data selection: selecting weighted return rate data of 88 fund managers from 2014 to 2021, 66 fund related indexes and 2 fund manager indexes, constructing a feature system, and ensuring a data coverage period.

Description

Technical Field

[0001] This application relates to the technical field of funds, and particularly relates to a method for evaluating the correlation between a fund manager's management ability and fund performance. Background Art

[0002] Nowadays, the phenomenon of multiple management is becoming more and more common, which has accelerated the transformation of fund research to the research of fund manager ability to a certain extent. Individual investors are often confused about whether to focus on the fund or follow the fund manager. To get the answer to this question, it is necessary to first clarify the relationship between the fund manager's management ability and fund performance, and whether there is a two-way influence between them.

[0003] The LSTM model has the advantages of being good at processing long sequence data and having strong generalization ability, so it is suitable for processing financial time series data with characteristics such as heteroscedasticity, sequence correlation, and nonlinearity; the random forest has strong interpretability and supports the evaluation of variable importance. Summary of the Invention

[0004] To solve the above technical problems or at least partially solve the above technical problems, this application provides a method for evaluating the correlation between a fund manager's management ability and fund performance, which can perform more accurate evaluations.

[0005] The purpose of the present invention is to study the two-way influence relationship between a fund manager's management ability and fund performance. In view of the increasingly common phenomenon of multiple management nowadays, the daily weighted return rate indicators of a total of 88 fund managers from 2014 to 2021 are selected as the data set. Considering that financial time series data has characteristics such as heteroscedasticity, sequence correlation, and nonlinearity, the LSTM model is selected for time series data prediction, and the conclusion that a fund manager's management ability can be reflected by fund performance is given; for reverse prediction, a data set composed of 66 fund-related indicators and 2 fund manager-related indicators is constructed, and the monthly data of 124 funds are selected for research. The random forest is used to model and predict fund performance, and the conclusion that fund management ability has a certain impact on fund performance is obtained.

[0006] The present invention provides the following technical solutions: It is concluded that there is a two-way influence between a fund manager's management ability and fund performance, including the following steps: 1. The performance of the funds managed historically can reflect the management ability of the fund manager: Step 1, first transform the research topic into "exploring whether the management ability indicators of a fund manager can be predicted by their historical performance"; Step 2, data selection, that is, select the weighted return rate indicator data of 88 fund managers from 2014 to 2021; Step 3, data preprocessing, including missing value processing, smoothing processing, and normalization processing; Step 4, model prediction. Select the LSTM model to predict time series data; Step 5, compare and evaluate the results. Compare the model prediction results with the true values, and draw conclusions based on the differences between the two.

[0007] Furthermore, the original research topic of Step 1 is "Whether the management ability of fund managers is reflected by the performance of the funds they have managed historically". Since the management ability index of fund managers is calculated by weighting the return rates of the funds they have managed, it is necessary to abstract the original research topic, that is, "Explore whether the management ability index of fund managers can be predicted by their historical performance", in order to make more direct prediction research.

[0008] Furthermore, in Step 2 of data selection, considering that the macro economy has a certain impact on fund performance, the fund performance in the same period may have commonalities. To exclude the influence of external economic factors, it is necessary to divide the management ability index of fund managers according to the same period. Under the comprehensive consideration of theoretical feasibility and model accuracy, the weighted return rate index data of 88 fund managers from 2014 to 2021 are finally selected.

[0009] Furthermore, in Step 3, the missing value processing is divided into two aspects. At the time series level, for the cases where the net value of closed-end funds has too many missing values and there is data for on-site funds on non-working days, deletion processing is performed; at the individual level, for new fund managers, since the time series data is short and it is difficult to reflect their management ability, this part of the data is selected to be discarded. Data smoothing processing is because when selecting daily data for modeling, the management ability index of fund managers has characteristics such as large volatility and rapid oscillation, and it is too difficult to directly use this data for prediction, so it is necessary to perform data smoothing operations on it first. Normalization processing is to eliminate the influence of dimensions, so that the preprocessed data is limited to [-1, 1], which is suitable for comprehensive comparison and evaluation, thus eliminating the adverse effects caused by singular sample data.

[0010] Furthermore, in Step 4 of model prediction, the LSTM model is adopted. Since financial time series data has characteristics such as heteroscedasticity, serial correlation, and non-linearity, it is difficult for traditional econometrics to achieve good prediction results. Therefore, the long short-term memory model is selected to model the management ability index of fund managers. It mainly includes model establishment and compilation, model fitting, and result visualization.

[0011] Furthermore, in Step 5, the predicted values are compared with the true values to summarize the evaluation results. Since the model data volume is too large to be effectively compared, in order to obtain more accurate conclusions, the problem is transformed into comparing the deciles of the two. The main reason is to better grasp the overall situation of the data, obtain more data information, and retain the impact of macroeconomic factors on individuals.

[0012] 2. The management ability of fund managers will also have a certain impact on the performance of the funds they manage: Step 1: First, transform the research topic into "the contribution degree of fund manager management ability indicators to the prediction of fund returns"; Step 2: Data selection, that is, use 66 fund-related indicators and 2 fund manager indicators to construct a feature system, and select 124 funds managed by 88 fund managers who meet the fund tenure from 2014 to 2021 and have been proven to have management capabilities before. Step 3: Modeling prediction, select the random forest model; Step 4: Comparison of modeling results, that is, comparison of model fitting effects, errors, model residuals, and evaluation of variable importance; Step 5: Robustness analysis. The importance of fund manager indicators cannot always be ranked among the top, and thus conclusions are drawn.

[0013] Furthermore, the original research topic in Step 1 was "Research on the impact degree of fund manager management ability on fund performance". Since the random forest is used to model and predict fund returns, it is transformed into "the contribution degree of fund manager management ability indicators to the prediction of fund returns".

[0014] Furthermore, in Step 2, a total of 66 feature vectors are selected to construct an index system for fund prediction, and the weighted return rate index of the fund manager and the net value index of the fund manager of this return rate are selected as relevant indicators reflecting the management ability of the fund manager.

[0015] Furthermore, in Step 3, the random forest model is selected for modeling because the evaluation of variable importance it supports makes the research more convenient and intuitive.

[0016] Furthermore, in Step 4, by comparing the fitting effects of the random forest models with and without fund manager management ability indicators in the index system, it is verified whether it can improve the prediction accuracy of fund returns. Four indicators, namely MSE, RMSE, R2, and MAD, are selected to compare and analyze the model fitting effects and errors; the random forest model evaluates variable importance through two indicators, namely "%IncMSE" and "IncNodePurity".

[0017] Further, in step 5, 124 funds were selected for random forest prediction respectively. Since there are many characteristic variables used in the model, in fact, the management ability index of the fund manager is relatively stable at the position of contributing degree ranking in the top 1 / 3. The fund manager index does not always maintain its importance at the forefront, which may be determined by the artificial selection of the fund manager's management ability index, and other characteristics may also convey the information of the fund manager's management ability.

[0018] The beneficial effects of the present invention compared with the prior art are as follows: Using the LSTM model to predict time series data can correctly fit the prediction effect of the time trend change direction, and the MAPE value is 1.7; adopting the random forest modeling to predict the fund yield supports the evaluation of variable importance.

[0019] The beneficial effects of the additional technical features of the present invention will be described in the specific implementation part of this specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the flow schematic diagram of the present application; Figure 2 is the structure diagram of the LSTM model; Figure 3 is the structure diagram of the random forest model; Figure 4 is the distribution diagram of the average value of the fund manager's management ability index; Figure 5 is the distribution diagram of all values of the fund manager's management ability index; Figure 6 is the management ability index diagram of the fund manager Bi Tianyu; Figure 7 is the management ability index diagram of the fund manager Bi Tianyu after moving average; Figure 8 is the model iteration process diagram; Figure 9 is the prediction effect diagram of the LSTM model; Figure 10 is the comparison diagram of the fitting effect of the random forest model; Figure 11 is the reverse cumulative distribution function diagram of the absolute value of the model residual without adding the fund manager index; Figure 12 is the reverse cumulative distribution function diagram of the absolute value of the model residual with the addition of the fund manager index; Figure 13 is the variable importance evaluation diagram of the model without adding the fund manager index; Figure 14 is the variable importance evaluation diagram of the model with the addition of the fund manager index; Figure 15It is the evaluation diagram of the importance of the model variables of Fund Manager No. 1; Figure 16 It is the evaluation diagram of the importance of the model variables of Fund Manager No. 2; Figure 17 It is the evaluation diagram of the importance of the model variables of Fund Manager No. 3; Figure 18 It is the evaluation diagram of the importance of the model variables of Fund Manager No. 4; Figure 19 It is the evaluation diagram of the importance of the model variables of Fund Manager No. 5; Figure 20 It is the evaluation diagram of the importance of the model variables of Fund Manager No. 6. Specific implementation manners

[0021] Next, the technical solutions in the present application will be described in conjunction with the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present application, rather than all of the embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0023] Combined with Figures 1 - 2 As shown, the full name of the LSTM model is the long short-term memory model. It is a variant of the traditional recurrent neural network and a deep learning modeling method; it can process whether to delete the transmission of information by opening and closing the gates.

[0024] Combined with Figure 3 As shown, a more intuitive explanation of the construction structure of the random forest model is given. Select the data of n different sample data sets and input them into the decision tree learner to construct n different decision tree models. When constructing the decision tree, bootstrap sampling is performed, and the results are integrated according to the mean or voting of the decision tree models.

[0025] Combined with Figure 4 and Figure 5 As shown, descriptive statistics are performed on the dataset of the fund manager's management ability. The data distribution is relatively uniform and shows the characteristics of being dense in the middle and sparse at both ends. Therefore, it is considered that this dataset can better reflect the distribution of the fund manager's management ability indicators.

[0026] Combined with Figure 6 and Figure 7 As shown, since it is too difficult to directly use the data for prediction, it is necessary to perform a smoothing operation on it first. The moving average method is selected. Taking the fund manager Bi Tianyu as an example, the images of the management ability indicators before and after the moving average are given. After smoothing, the data greatly alleviates the oscillation situation and the extreme values are also improved.

[0027] Combined with Figure 8 As shown, the model iteration process is given. The sample size of each Batch is defined as 128, the storage space of the random sampling buffer is 100, and 20 iterations are carried out. Finally, the obtained model accuracy is the training set MAPE value of 4.93 and the test set MAPE value of 1.77.

[0028] Combined with Figure 9 As shown, the model prediction results and the true values are plotted as curves for comparison. The true values are represented by the blue line and the predicted values are represented by the orange line. For the management ability index of this fund manager, the predicted values of the LSTM model effectively fit the trend of the time series and successfully predicted the rise and fall of the true values. However, for the prediction of extreme values, the model still cannot be accurate enough and further improvement is still needed.

[0029] Combined with Figure 10 As shown, four indicators of MSE, RMSE, R^2, and MAD are selected to compare the fitting effect and error of the model. MSE, RMSE, and MAD represent the model error situation. The smaller the data, the better the model fitting. Therefore, the MSE and RMSE indicators indicate that the model after adding the fund manager index is better, and the MAD indicator indicates that the model without adding the fund manager index is better; the closer the R^2 value is to 1, the stronger the model fitting ability.

[0030] Combined with Figure 11 and Figure 12 As shown, by comparing the two figures, it is found that the descent speed of the image after adding the fund manager index is slightly slower than that without adding the fund manager index, and the convergence speed of the model residuals is slower. This may be due to the difference in the number of model features.

[0031] Combined with Figure 13 and Figure 14As shown, the random forest model evaluates variable importance through two metrics: "%IncMSE" and "IncNodePurity". "%IncMSE" is measured by randomly assigning values to each predictive variable. If the predictive variable is more important, the error of the model prediction will increase after its value is randomly replaced. The larger this value, the more important the variable. "IncNodePurity" is measured by the sum of squared residuals and represents the impact of each variable on the heterogeneity of the observations at each node of the classification tree, thereby comparing the importance of variables. The larger this value, the more important the variable. By comparing the variable importance evaluation results of the two models, it is found that the predictive contribution of Rmrf (the excess return of the market index) is always relatively high, indicating that market trends have a greater impact on the performance of this fund. The management ability index of the fund manager also ranks among the top, suggesting that the management ability index of the fund manager can improve the model prediction effect, that is, the management ability of the fund manager has an impact on the performance of the funds he manages.

[0032] Combined with Figures 15 - 20 , they are randomly selected partial model effect displays, from which it can be concluded that not all management ability indicators of fund managers make extremely important contributions to the prediction of fund performance. For some funds, the relevant indicators of the fund manager have relatively high weights and rank among the top; for some funds, the management ability indicators of the fund manager are ranked lower in the chart, indicating that by introducing a large amount of data for feature selection, the impact of the management ability indicators of the fund manager on fund performance will be diluted. This may be because the management ability indicators of the fund manager are determined manually.

[0033] It should be noted that in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Additionally, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising said element. Moreover, in the description of the embodiments of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B; "and / or" in this text is merely a description of the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Also, in the description of the embodiments of the present application, "a plurality of" means two or more than two.

[0034] The above are only specific embodiments of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments described herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. An evaluation method for the correlation between the management ability of a fund manager and the fund performance, characterized in that It includes the following steps: Obtain the weighted return rate index data of 88 fund managers during the period from 2014 to 2021; Preprocess the weighted return rate index data to obtain the preprocessed index data; Use the LSTM model to perform time series prediction on the preprocessed index data to obtain the prediction results; Compare the prediction results with the actual index data to evaluate the prediction effect of the historical performance of fund managers on their management ability; Construct a feature system, which includes fund-related indicators and fund manager indicators; Select the data from 2014 to 2021 of 124 funds managed by fund managers with management talents from the 88 fund managers; Use the random forest model to perform modeling prediction on the data of the 124 funds; Evaluate the prediction results of the random forest model, and the evaluation includes model fitting effect evaluation, error evaluation, model residual evaluation and variable importance evaluation; Conduct robustness analysis based on the evaluation results to obtain the prediction contribution degree of the fund manager management ability index to the fund return rate.

2. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The step of obtaining the weighted return rate index data includes: dividing the fund manager management ability index according to the same period to exclude the influence of macroeconomic factors on the fund performance.

3. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The preprocessing includes: deleting the data with too many missing net value in closed-end funds; deleting the situation where there is data in the on-site funds on non-working days; deleting the data of new fund managers whose time series data length is not enough to reflect the management ability; smoothing the daily data to reduce data volatility; normalizing the data to the interval [-1, 1].

4. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The time series prediction using the LSTM model includes: Establish and compile the LSTM model; Perform model fitting on the preprocessed index data; Visualize the prediction results.

5. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The comparison of the prediction results with the actual index data includes: Calculate the deciles of the predicted value and the actual value; Compare the decile distribution of the predicted value and the actual value.

6. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The fund-related indicators include 66 feature vectors; The fund manager indicators include the fund manager weighted return rate index and the fund manager net value index.

7. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The evaluation of the prediction results of the random forest model includes: Analyze the model fitting effect and error using the MSE, RMSE, R2 and MAD indicators; Evaluate the variable importance using the %IncMSE and IncNodePurity indicators.

8. The method for evaluating the correlation between the management ability of a fund manager and the fund performance according to claim 1, wherein The robustness analysis includes: Perform random forest prediction on each of the 124 funds respectively; Analyze the stability of the fund manager management ability index in the variable importance ranking; Evaluate the information transmission effect of other features on the fund manager management ability.

Images