A real-time fund valuation method based on LSTM and market factors

Abstract

The present invention discloses a real-time fund valuation method based on LSTM and market factors. By deeply integrating the prediction advantages of LSTM with the impact analysis of market factors, a more accurate and comprehensive real-time fund valuation model is constructed to meet the market's urgent demand for efficient and accurate valuation methods.

Description

Technical Field

[0001] The present invention relates to the field of fund investment, and in particular to a real-time fund valuation method based on LSTM and market factors. Background Art

[0002] With the increasing complexity of financial markets and the growing demand for real-time information from investors, traditional fund valuation methods have been unable to meet the needs of efficient market operations due to their lag and lack of precision. Especially in the face of high-frequency trading and rapid market changes, accurate and timely fund valuation has become the key to improving investment decision-making efficiency and risk control capabilities. As a special recurrent neural network in deep learning, the long short-term memory network (LSTM) has shown great potential in the field of financial forecasting with its powerful learning and prediction capabilities for time series data. By capturing the nonlinear relationships and long-term dependency characteristics in historical data, LSTM can more effectively predict the changing trend of fund net value.

[0003] Disadvantages of existing technology:

[0004] There are three main types of current technical solutions:

[0005] Estimating net worth based on the position penetration method;

[0006] Use traditional statistical models or machine learning algorithms to estimate net worth;

[0007] Use deep learning methods such as LSTM alone to do it. Summary of the invention

[0008] In view of the above problems, the present invention is proposed to provide a real-time fund valuation method based on LSTM and market factors that overcomes the above problems or at least partially solves the above problems.

[0009] According to one aspect of the present invention, a real-time fund valuation method based on LSTM and market factors is provided, and the real-time valuation method comprises:

[0010] Step S1: Collect historical adjusted net value data of funds and daily index data;

[0011] Step S2: convert the adjusted net value and the adjusted closing price of the index into yields and align them by date;

[0012] Step S3: Calculate the most relevant indexes corresponding to multiple funds;

[0013] Step S4: Estimate the net asset value rate of return for the latest day to construct the latest data set, and divide it into a training set and a validation set;

[0014] Step S5: Merge the fund data with the same most relevant market factors to build and train LSTM models respectively:

[0015] Step S6: Calculate the model-dependent input parameters in advance and record the parameters;

[0016] Step S7: During intraday trading, the 14-day yield of the comprehensive fund and the yield of the corresponding market index during the trading session are requested from the LSTM model to obtain the real-time estimated yield r`;

[0017] Step S8: Estimate the real-time net value of the fund to be evaluated on day t+1 according to the net value of the fund to be evaluated on day t and the real-time rate of return of the fund to be evaluated on day t+1 estimated in step S7;

[0018] Step S9: Backtest the deviation between the actual value and the estimated value of the fund.

[0019] Optionally, the step S1: collecting the fund's historical adjusted net asset value data and index daily line data specifically includes: collecting the fund's historical adjusted unit net asset value data and index daily line data with intraday trading data.

[0020] Optionally, the step S2: converting the adjusted net value and the adjusted closing price of the index into a rate of return and aligning them by date specifically includes: for historical data, aligning the adjusted net value of the fund and the real-time index closing price with intraday transactions according to the transaction date dimension, and converting them into a rate of return.

[0021] Optionally, the step S3: calculating the most relevant indexes corresponding to the multiple funds specifically includes: calculating the indexes with the highest correlation coefficients for the multiple funds respectively according to the correlation coefficient matrix.

[0022] Optionally, the step S4: estimating the latest day's net asset value yield for the latest constructed data set, and splitting the training set and the validation set specifically includes: estimating the latest day's net asset value yield for the latest constructed data set using the net asset value yield of the past 14 days + the market factor yield of the day, and splitting the training set and the validation set in a ratio of 8:2.

[0023] Optionally, the input parameters that the model depends on calculated in advance specifically include: the fund's yield in the past 14 days and the dependent market index.

[0024] Optionally, estimating the real-time net value of the fund to be evaluated on day t+1 according to the net value of the fund to be evaluated on day t and the real-time yield of the fund to be evaluated on day t+1 estimated in step S7 specifically includes:

[0025] The real-time net worth calculation formula is as follows:

[0026]

[0027] Among them, y tis the net value of the fund to be valued on day t, It is the estimated real-time rate of return of the fund to be valued on day t+1.

[0028] Optionally, the step S9: backtesting the deviation between the actual fund value and the estimated value specifically includes: backtesting the deviation between the estimated fund net value and the actual net value to evaluate whether the online conditions are met.

[0029] The present invention provides a real-time valuation method for funds based on LSTM and market factors, and the real-time valuation method comprises: step S1: collecting historical adjusted net value data and index daily line data of funds; step S2: converting the adjusted net value and the adjusted closing price of the index into yields and aligning them by date; step S3: calculating the most relevant indexes corresponding to multiple funds; step S4: estimating the net value yield of the latest day to construct the latest data set, and dividing the training set and the verification set; step S5: merging the fund data with the same most relevant market factors to construct training LSTM models respectively: step S6: calculating the input parameters that the model depends on in advance and recording the parameters; step S7: during intraday trading, requesting the LSTM model to obtain the real-time estimated yield r' by combining the 14-day yield of the comprehensive fund and the corresponding market index yield during the intraday trading; step S8: estimating the real-time net value of the fund to be estimated on the t+1 day according to the net value of the fund to be estimated on the t day and the real-time yield of the fund to be estimated on the t+1 day estimated in step S7; and step S9: backtesting the deviation between the actual value of the fund and the estimated value. Build a more accurate and comprehensive real-time fund valuation model to meet the market's urgent demand for efficient and accurate valuation methods.

[0030] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0032] Figure 1 A flowchart of a real-time fund valuation method based on LSTM and market factors provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0034] The terms "comprises" and "having" and any variations thereof in the description embodiments, claims and drawings of the present invention are intended to cover non-exclusive inclusions, for example, including a series of steps or units.

[0035] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0036] like Figure 1 As shown in the figure, a real-time fund valuation method based on LSTM and market factors includes:

[0037] Step S1: Collect historical adjusted net value data of funds and daily index data;

[0038] Step S2: convert the adjusted net value and the adjusted closing price of the index into yields and align them by date;

[0039] Step S3: Calculate the most relevant index corresponding to different funds;

[0040] Step S4: Use the net asset value rate of return in the past 14 days + the market factor rate of return on the day to estimate the net asset value rate of return on the latest day to construct the latest data set and divide the training set and the validation set into a ratio of 8:2;

[0041] Step S5: Merge the fund data with the same most relevant market factors to build and train LSTM models respectively:

[0042] Step S6: Calculate the model-dependent input parameters in advance, such as the fund's yield in the past 14 days and the dependent market index, and record the parameters;

[0043] Step S7: During intraday trading, the 14-day yield of the comprehensive fund and the yield of the corresponding market index during the trading session are requested from the LSTM model to obtain the real-time estimated yield r`;

[0044] Step S8: Estimate the real-time net value of the fund to be evaluated on day t+1 based on the net value of the fund to be evaluated on day t and the real-time rate of return of the fund to be evaluated on day t+1 estimated in step S7. The real-time net value calculation formula is as follows:

[0045]

[0046] Among them, y t is the net value of the fund to be valued on day t, is the estimated real-time rate of return of the fund to be valued on day t+1;

[0047] Step S9: Backtest the deviation between the actual value and the estimated value of the fund.

[0048] In step S1, the fund's historical adjusted unit net value data and index daily line data with intraday trading data are collected;

[0049] The sample table of fund historical adjusted unit net value data is as follows:

[0050]

[0051] The sample table of industry index data with intraday trading is as follows:

[0052] Serial number Transaction Date CSI 300 Shanghai Composite Index CSI 500 Dividend Index 1 20240701 3478.21 ... ... ... 2 20240702 3450.34 ... ... ... 3 20240703 3442.34 ... ... ... ... ... ... ... ... ...

[0053] In step S2, for historical data, the adjusted net value of the fund and the closing price of the real-time index with intraday transactions need to be aligned according to the transaction date dimension and converted into a rate of return.

[0054]

[0055] In step S3, the index with the highest correlation coefficient is calculated for different funds through the correlation coefficient matrix;

[0056] Through correlation coefficient calculation, it was found that the "China Merchants Industry Select Stock Fund" has the highest correlation coefficient with the Dividend Index among the CSI 300, Shanghai Composite Index, CSI 500 and Dividend Index, which is 92%.

[0057] In step S4, the latest data set is constructed by using the net asset value rate of return of the past 14 days + the market factor rate of return of the day to estimate the net asset value rate of return of the latest day:

[0058] The following is the net asset value data of China Merchants Industry Select Stock Fund on some trading dates.

[0059] Serial number Transaction Date Net return 1 20240701 0.002101 2 20240702 0.001818 3 20240703 -0.007670 4 20240704 0.003213 5 20240705 0.004272 6 20240708 0.002552 7 20240709 0.002932 8 20240710 0.002211 ... ... ...

[0060] The data samples converted into the required ones are as follows: xn represents the yield sequence itself lagged by n days, x_r represents the daily yield of the corresponding index corresponding to the trading day, and each row of data including empty data is removed. The remaining row of data represents a sample. Use x1, x2, x3...x_r to predict y as the task to train the LSTM model.

[0061]

[0062]

[0063] Based on the table obtained in S4, the fund data with the same most relevant market factors are merged to build and train LSTM models respectively;

[0064] Since this data construction method will cause a large number of data samples to be discarded, in order to increase the data samples, consider merging data samples of the same index for training together. On the one hand, it can reduce overfitting, and on the other hand, it can expand the data samples and make the LSTM model training more sufficient.

[0065] In addition, funds of the same index refer to that in step S3, there may be multiple funds that are highly correlated with the same index, for example, XX Home Appliances Mixed Stock A and XX Home Appliances Index Enhanced Fund are highly correlated with the Home Appliances Index.

[0066] Based on step S6, the model-dependent input parameters are calculated in advance, such as the fund's yield in the past 14 days and the dependent market index, and the parameters are recorded.

[0067] According to step S7, during intraday trading, the 14-day yield of the comprehensive fund and the yield of the corresponding market index during the trading session are requested to the LSTM model to obtain the real-time estimated yield r`.

[0068] For example, for the China Merchants Industry Select Stock Fund, when estimating its net asset value yield at 2024-07-11 10:00:00, it is necessary to obtain the yield of the dividend index at the corresponding time point from the exchange, such as 0.03213, and combine it with the model parameters pchglist recorded in advance into [0.002211, 0.002932, 0.002552,..., 0.03213] to request the LSTM model algorithm service and obtain the estimated yield r`.

[0069] The real-time estimated net value is calculated based on the net value of the fund to be evaluated on day t and the rate of return estimated in step S8.

[0070] For example, if the net value of the previous day is 1.2503 and the estimated rate of return is 1.52%, then the real-time estimated net value is: 1.25*(1+1.52%)=1.2693.

[0071] In step S9, the deviation between the estimated fund net value and the actual net value is backtested to evaluate whether the upper limit conditions are met.

[0072] Beneficial effects: The present invention proposes a real-time fund valuation method based on LSTM and market factors, aiming to build a more accurate and comprehensive real-time fund valuation model by deeply integrating the prediction advantages of LSTM with the impact analysis of market factors, so as to meet the market's urgent demand for efficient and accurate valuation methods.

[0073] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time fund valuation method based on LSTM and market factors, characterized in that: The real-time valuation method comprises: Step S1: Collect historical adjusted net value data of funds and daily index data; Step S2: convert the adjusted net value and the adjusted closing price of the index into yields and align them by date; Step S3: Calculate the most relevant indexes corresponding to multiple funds; Step S4: Estimate the net asset value rate of return for the latest day to construct the latest data set, and divide it into a training set and a validation set; Step S5: Merge the fund data with the same most relevant market factors to build and train LSTM models respectively: Step S6: Calculate the model-dependent input parameters in advance and record the parameters; Step S7: During intraday trading, the 14-day yield of the comprehensive fund and the yield of the corresponding market index during the trading session are requested from the LSTM model to obtain the real-time estimated yield r`; Step S8: Estimate the real-time net value of the fund to be evaluated on day t+1 according to the net value of the fund to be evaluated on day t and the real-time rate of return of the fund to be evaluated on day t+1 estimated in step S7; Step S9: Backtest the deviation between the actual value and the estimated value of the fund.

2. A real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step S1: collecting the fund's historical adjusted net asset value data and index daily line data specifically includes: collecting the fund's historical adjusted unit net asset value data and index daily line data with intraday trading data.

3. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step S2: converting the adjusted net value and the adjusted closing price of the index into a rate of return and aligning them by date specifically includes: for historical data, aligning the adjusted net value of the fund and the real-time index closing price with intraday transactions according to the transaction date dimension, and converting them into a rate of return.

4. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step S3: calculating the most relevant indexes corresponding to the multiple funds specifically includes: calculating the indexes with the highest correlation coefficients for the multiple funds respectively according to the correlation coefficient matrix.

5. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step S4: estimating the latest day's net asset value yield rate to construct the latest data set, and splitting the training set and the validation set specifically includes: estimating the latest day's net asset value yield rate to construct the latest data set according to the net asset value yield rate of the past 14 days + the market factor yield rate of the day, and splitting the training set and the validation set according to the ratio of 8:

2.

6. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The input parameters that the model depends on calculated in advance specifically include: the fund's return rate in the past 14 days and the dependent market index.

7. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step of estimating the real-time net value of the fund to be evaluated on day t+1 according to the net value of the fund to be evaluated on day t and the real-time yield of the fund to be evaluated on day t+1 estimated in step S7 specifically includes: The real-time net worth calculation formula is as follows: Among them, y t is the net value of the fund to be valued on day t, It is the estimated real-time rate of return of the fund to be valued on day t+1.

8. The real-time fund valuation method based on LSTM and market factors according to claim 1, characterized in that: The step S9: backtesting the deviation between the actual fund value and the estimated value specifically includes: backtesting the deviation between the estimated fund net value and the actual net value to evaluate whether the online conditions are met.

Images