Weight confirmation method and device, equipment and medium
The timing characteristics in fund information and the mutual influence characteristics between funds are extracted through the timing chart neural network model, and the daily holding weight is predicted, which solves the investment decision-making problems caused by financial reporting lag, and improves the prediction accuracy and return on investment.
Patent Information
- Application Number
- CN202411650942.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-14
AI Technical Summary
In the prior art, there is a serious lag in guiding trading decisions based on half-year or once a year, which causes investors to miss the best trading opportunity and increase investment risks.
The timing graph neural network model is used to extract the timing feature vectors in fund information through the timing convolution network, and the adjacent feature vectors of the target fund and other funds are extracted through the graph attention network, and the daily position weight is predicted in combination with the prediction network.
It improves the accuracy of daily holding weight prediction, avoids investment risks caused by decision-making lag, and improves the return on investment.
Smart Images
Figure CN119624652A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of machine learning technology, and in particular to a weight confirmation method, device, equipment and medium. Background Art
[0002] In the field of technology finance, especially in a complex environment that requires close follow-up of market dynamics and high-frequency trading, the more accurate and timely investors are in understanding the fund's holdings, the more advantageous they will be in the ever-changing market. The fund's holdings not only show the specific proportion of various financial products in the fund, but also reveal the flow of funds between different asset classes.
[0003] Accurate position information can help investors better understand the investment logic and market judgment of financial products. In related technologies, investors can rely on regularly released financial reports to obtain position information. These financial reports are usually published every six months or every year and can provide detailed position data, including the composition of the investment portfolio, detailed classification of assets and liabilities, etc.
[0004] However, there is a serious lag in guiding trading decisions based on semi-annual or annual financial reports. This lag may not only cause investors to miss the best trading opportunities, but also increase investment risks due to the inability to respond to market fluctuations in a timely manner. Summary of the invention
[0005] Based on the above problems, the present application provides a weight confirmation method, device, equipment and medium, which can improve the accuracy of daily position weight prediction, reduce investment risks and improve investment return rate.
[0006] The embodiments of the present application disclose the following technical solutions:
[0007] In a first aspect, the present application discloses a weight confirmation method, the method comprising:
[0008] Obtain fund information of the target fund;
[0009] The fund information is input into a time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, the adjacency feature vector between the target fund and other funds is extracted through the graph attention network, and the daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacency feature vector through the prediction network.
[0010] Optionally, the method further comprises:
[0011] The model parameters of the time series graph neural network model are updated according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
[0012] Optionally, the method further comprises:
[0013] Obtaining actual holding information of the target fund;
[0014] If the actual position information does not match the predicted position information corresponding to the daily position weight, the model parameters of the time series graph neural network model are updated according to the actual position information and the predicted position information.
[0015] Optionally, the obtaining fund information of the target fund includes:
[0016] Obtain original information of target funds;
[0017] The fund information is obtained by performing at least one of removing invalid position information and / or erroneous position information in the original information, unifying the format of the original information, and completing the original information.
[0018] Optionally, the time series graph neural network model is constructed by the following steps:
[0019] Get the fund information of the sample fund set;
[0020] According to the fund information of the sample fund set, a graph network of the sample fund set is determined, wherein the sample funds in the sample fund set are primary nodes of the graph network, the financial products held by the sample funds are secondary nodes of the graph network, and the nodes corresponding to the sample funds and the secondary nodes corresponding to the financial products held by the sample funds are connected by a first-class edge, and the weight of the first-class edge is positively correlated with the holding ratio of the sample fund to the financial products, and the sample funds include a first sample fund and a second sample fund. If the first sample fund is related to the second sample fund, the primary node corresponding to the first sample fund and the primary node corresponding to the second sample fund are connected by a second-class edge, and the weight of the second-class edge is positively correlated with the correlation ratio of the first sample fund to the second sample fund;
[0021] Inputting the graph network of the sample fund set into a time-series graph neural network model, the time-series graph neural network model includes a time-series convolutional network, a graph attention network and a prediction network, extracting a time-series feature vector from the graph network of the sample fund set through the time-series convolutional network, extracting an adjacency feature vector between the sample fund and other funds in the sample fund set through the graph attention network, and confirming the daily holding weight of the sample fund according to the time-series feature vector and the adjacency feature vector through the prediction network;
[0022] A time-series graph neural network model is trained based on the graph network of the sample fund set and the daily holding weights of the sample funds.
[0023] In a second aspect, the present application discloses a weight confirmation device, the device comprising: an acquisition module and a confirmation module;
[0024] The acquisition module is used to acquire the fund information of the target fund;
[0025] The confirmation module is used to input the fund information into a time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, the adjacent feature vector between the target fund and other funds is extracted through the graph attention network, and the daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacent feature vector through the prediction network.
[0026] Optionally, the device further includes: a first updating module;
[0027] The first updating module is used to update the model parameters of the time series graph neural network model according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
[0028] Optionally, the device further includes: a second updating module;
[0029] The second updating module is used to obtain the actual holding information of the target fund; if the actual holding information does not match the predicted holding information corresponding to the daily holding weight, the model parameters of the time series graph neural network model are updated according to the actual holding information and the predicted holding information.
[0030] Optionally, the acquisition module is specifically used to: obtain the original information of the target fund; obtain the fund information by removing invalid holding information and / or erroneous holding information in the original information, unifying the format of the original information, and completing the original information.
[0031] Optionally, the time series graph neural network model is constructed by the following units:
[0032] A first acquisition unit, used to acquire fund information of a sample fund set;
[0033] A second determining unit is configured to determine a graph network of the sample fund set according to the fund information of the sample fund set, wherein the sample funds in the sample fund set are primary nodes of the graph network, the financial products held by the sample funds are secondary nodes of the graph network, and the nodes corresponding to the sample funds and the secondary nodes corresponding to the financial products held by the sample funds are connected by a first-class edge, and the weight of the first-class edge is positively correlated with the holding ratio of the sample fund to the financial products, and the sample funds include a first sample fund and a second sample fund. If the first sample fund is related to the second sample fund, the primary node corresponding to the first sample fund and the primary node corresponding to the second sample fund are connected by a second-class edge, and the weight of the second-class edge is positively correlated with the correlation ratio of the first sample fund to the second sample fund;
[0034] A third input unit is used to input the graph network of the sample fund set into a time-series graph neural network model, wherein the time-series graph neural network model includes a time-series convolutional network, a graph attention network, and a prediction network, extract a time-series feature vector from the graph network of the sample fund set through the time-series convolutional network, extract an adjacency feature vector between the sample fund and other funds in the sample fund set through the graph attention network, and confirm the daily holding weight of the sample fund according to the time-series feature vector and the adjacency feature vector through the prediction network;
[0035] The fourth training unit is used to train a time-series graph neural network model based on the graph network of the sample fund set and the daily holding weights of the sample funds.
[0036] In a third aspect, the present application discloses a weight confirmation device, the device comprising: a memory and a processor;
[0037] The memory is used to store programs;
[0038] The processor is used to execute the program to implement the various steps of the weight confirmation method as described in the first aspect.
[0039] In a fourth aspect, the present application discloses a computer-readable medium having a computer program stored thereon, and when the computer program is executed by a processor, the various steps of the weight confirmation method as described in the first aspect are implemented.
[0040] Compared with the prior art, this application has the following beneficial effects:
[0041] The embodiment of the present application provides a weight confirmation method, device, equipment and medium. The weight confirmation method inputs the acquired fund information of the target fund into a time-series graph neural network model, extracts the time-series feature vectors in the fund information through the time-series convolution network of the time-series graph neural network model, and can improve the sensitivity to changes in fund holdings and the accuracy of prediction. The adjacent feature vectors of the target fund and other funds are extracted through the graph attention network of the time-series graph neural network model, and the mutual influence between different funds can be captured. The prediction network of the time-series graph neural network model combines the time-series feature vectors and the adjacent feature vectors to predict the daily holding weight, and can improve the accuracy of the daily holding weight prediction. Furthermore, the predicted daily holding weight can prevent investors from missing the best trading opportunities due to decision-making delays, thereby reducing investment risks and improving investment returns. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0043] Figure 1 A flowchart of a weight confirmation method provided in an embodiment of the present application;
[0044] Figure 2A A flowchart of another weight confirmation method provided in an embodiment of the present application;
[0045] Figure 2B A schematic diagram of a timing graph neural network model provided in an embodiment of the present application;
[0046] Figure 3 A schematic diagram of a weight confirmation device provided in an embodiment of the present application;
[0047] Figure 4 A schematic diagram of a computer-readable medium provided for an embodiment of the present application. DETAILED DESCRIPTION
[0048] First, the technical terms involved in this application are explained:
[0049] The holding weight refers to the proportion or ratio of a certain financial product (such as stocks, bonds, etc.) in the financial product portfolio. Simply put, it is the proportion of a certain asset held by the user in the total investment amount of the user. For example, suppose user A has an investment portfolio with a total value of 100,000 yuan. There are three financial products in this portfolio: product A is worth 30,000 yuan, product B is worth 50,000 yuan, and product C is worth 20,000 yuan. Then, the holding weight of product A is 30% (30,000 / 100,000), the holding weight of product B is 50% (50,000 / 100,000), and the holding weight of product C is 20% (20,000 / 100,000). By adjusting the holding weights of different assets, investors can optimize asset allocation to achieve the purpose of diversifying risks and increasing returns.
[0050] As described above, investors can rely on regularly released financial reports to obtain information about their holdings. These financial reports are usually published every six months or annually and can provide detailed holdings data, including the composition of the investment portfolio, detailed classification of assets and liabilities, etc.
[0051] However, there is a serious lag in guiding trading decisions based on semi-annual or annual financial reports. This lag may not only cause investors to miss the best trading opportunities, but also increase investment risks due to the inability to respond to market fluctuations in a timely manner.
[0052] After research, the inventor proposed a weight confirmation method, device, equipment and medium. The weight confirmation method inputs the acquired fund information of the target fund into a time-series graph neural network model, extracts the time-series feature vector in the fund information through the time-series convolution network of the time-series graph neural network model, can improve the sensitivity to fund position changes and prediction accuracy, extracts the adjacent feature vectors of the target fund and other funds through the graph attention network of the time-series graph neural network model, can capture the mutual influence between different funds, and predicts the daily position weight through the prediction network of the time-series graph neural network model combined with the time-series feature vector and the adjacent feature vector, can improve the accuracy of the daily position weight prediction. Furthermore, the predicted daily position weight can prevent investors from missing the best trading opportunity due to decision lag, thereby reducing investment risks and improving investment returns.
[0053] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0054] See also Figure 1, which is a flow chart of a weight confirmation method provided in an embodiment of the present application. The method includes:
[0055] S101: Obtain fund information of a target fund.
[0056] The fund information of the target fund includes the historical net value data, transaction record data, position details data and market performance data of the target fund.
[0057] Among them, the historical net value data of the target fund can reflect the changing trend of the value of the target fund over time. The historical net value data can usually be obtained from the official website of the fund company, stock exchanges, third-party financial service platforms and other channels.
[0058] The target fund's transaction record data includes all the target fund's buying and selling activities, including transaction date, transaction subject, transaction quantity and transaction price, etc. Transaction record data can usually be obtained from the fund company's regular reports, transaction databases or professional financial data service platforms.
[0059] The detailed data of the target fund's holdings refers to all the financial products (including stocks, bonds, cash, etc.) currently held by the target fund and their proportions, which can reveal the investment preferences and diversification of the target fund. The detailed data of holdings can usually be found in the fund's quarterly report, semi-annual report and annual report, or can be obtained through the fund company's official website or third-party financial service platform.
[0060] The market performance data of the target fund refers to the performance of the target fund relative to the market index or other similar funds, which can help investors evaluate the relative value and competitiveness of the target fund. Market performance data can usually be obtained from fund rating agencies, stock exchanges, financial news websites and other channels.
[0061] S102: Input the fund information into the time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, and the adjacency feature vector between the target fund and other funds is extracted through the graph attention network. The daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacency feature vector through the prediction network.
[0062] The Temporal Graph Neural Network (TGNN) model includes a Temporal Convolutional Neural Network (TCN), a Graph Attention Network (GAT), and a Feed-Forward Neural Network (FFN). By integrating the TCN network, the GAT network, and the FFN network, an accurate prediction of the daily holding weight of the target fund is achieved. In the TGNN model, the outputs of the GAT network and the TCN network are connected to the input of the FFN network to form an end-to-end learning framework.
[0063] Specifically, the TCN network is used to process time series data and can capture the trends and patterns of data changes over time. In the weight confirmation method provided in the embodiment of the present application, the TCN network is responsible for extracting the time series feature vector from the fund information of the target fund (such as historical net value data, transaction record data), that is, reflecting the behavior pattern of the target fund changing over time.
[0064] The GAT network is used to learn the representation of nodes by focusing on the relationship between nodes. In the weight confirmation method provided in the embodiment of the present application, the GAT network is used to analyze the relationship between the target fund and other funds, thereby extracting the adjacent feature vector. The adjacent feature vector can reveal the degree of mutual influence between funds, such as whether some funds tend to conduct similar trading activities in the same period.
[0065] The FFN network is a feedforward neural network used to make predictions based on the input feature vector. In the weight confirmation method provided in the embodiment of the present application, the FFN network is used to predict the future daily holding weight of the target fund based on the time series feature vector output by the TCN network and the adjacent feature vector output by the GAN. The prediction results of the daily holding weight can help investors plan investment strategies in advance and optimize asset allocation.
[0066] In summary, the embodiment of the present application discloses a weight confirmation method, which inputs the acquired fund information of the target fund into a time-series graph neural network model, extracts the time-series feature vectors in the fund information through the time-series convolutional network of the time-series graph neural network model, and can improve the sensitivity to changes in fund holdings and the accuracy of prediction, extracts the adjacent feature vectors of the target fund and other funds through the graph attention network of the time-series graph neural network model, and can capture the mutual influence between different funds, and predicts the daily holding weight by combining the time-series feature vector and the adjacent feature vector through the prediction network of the time-series graph neural network model, and can improve the accuracy of the daily holding weight prediction. Furthermore, the predicted daily holding weight can prevent investors from missing the best trading opportunities due to decision-making delays, thereby reducing investment risks and improving investment returns.
[0067] See also Figure 2A , which is a flow chart of another weight confirmation method provided by an embodiment of the present application. The method includes:
[0068] S201: Obtain fund information of a sample fund set, wherein the fund information includes position information.
[0069] Collect detailed information on each sample fund in the sample fund set, especially its holdings information.
[0070] Specifically, position information can be obtained in the following ways:
[0071] In some specific implementations, the holdings information of the sample funds can be obtained through the quarterly reports of the sample funds (e.g., within 15 working days after the end of each quarter). For example, the holdings information may include the top ten financial products held by the sample funds and their proportion in the total assets of the sample funds, the holdings information may also include the allocation ratio of the sample funds in various industries, and the holdings information may also include the holding ratio of the sample funds in different asset categories.
[0072] In some other specific implementations, the holdings information of the sample funds can be obtained through the semi-annual reports of the sample funds (e.g., within 60 working days after the end of the first half of the year). For example, the holdings information may include detailed information on all financial products held by the sample funds, including name, quantity, market value, and proportion. Similarly, the holdings information may also include the allocation ratio of the sample funds in various industries, and the holdings information may also include the holding ratio of the sample funds in different asset categories.
[0073] In other specific implementations, the holdings information of the sample fund can be obtained through the annual report of the sample fund (for example, within 90 working days after the end of each year). The holdings information may include the financial statements, audit reports and management reports of the sample fund. Similarly, the holdings information may include detailed information on all financial products held by the sample fund, including name, quantity, market value and proportion. The holdings information may also include the allocation ratio of the sample fund in various industries. The holdings information may also include the holding ratio of the sample fund in different asset categories.
[0074] S202: Pre-process the position information to obtain processed position information.
[0075] The acquired holdings information of the sample funds may contain some invalid or erroneous holdings information, which may be caused by data entry errors, system anomalies or other reasons. In order to ensure the accuracy and completeness of the holdings information, it is necessary to clean up these invalid or erroneous holdings information. In the first example, since the number of positions cannot be negative, the negative number of positions can be set to 0 or deleted. In the second example, since the holding ratio is a value between 0 and 100%, the holding ratio that is less than 0 or greater than 100% can be set to 0 or 100% or deleted.
[0076] After obtaining the holding information of the sample funds, the information format of the holding information can also be unified, which helps to reduce the complexity and error rate of the handling of the holding information. In the first example, the format of all date data can be ensured to be consistent, such as unified into the YYYY-MM-DD format. In the second example, the format of all numerical data (such as the number of positions, market value, etc.) can be ensured to be consistent, such as unified into two decimal places. In the third example, the format of the codes and names of all financial products can be ensured to be consistent to avoid data matching errors caused by inconsistent formats, such as unifying the codes of financial products into a fixed number of digits (such as 6 digits) and the names of financial products into full names with capitalized first letters.
[0077] After obtaining the holding information of the sample fund, if there are missing values in the holding information, the holding information of the sample fund can be supplemented. In the first example, for missing time series data (such as the number of positions or market value), linear interpolation can be used to fill in the missing values. For example, if the number of positions of a financial product on a certain day is missing, the number of positions held on the previous day and the next day can be used for linear interpolation. In the second example, for certain non-time series data (such as the average holding ratio), the mean or median of the field can be used to fill in. In the third example, other reliable data sources, such as market indices, industry reports, etc., can be used to complete the data. It should be noted that the specific information completion method should be selected according to the actual situation and data characteristics, and this application does not limit the specific information completion method.
[0078] S203: Construct a graph network of the sample fund set based on the processed holding information.
[0079] First, the position information at each time point is extracted from the processed position information, and the position information is organized into a format suitable for time series analysis, such as sorting by date, so that the changes in the position information over time can be clearly seen.
[0080] Secondly, the sample funds in the sample fund set are taken as the first-level nodes of the graph network, and the financial products held by the sample funds are taken as the second-level nodes of the graph network. That is to say, the second-level nodes represent the specific objects of fund investment, such as stocks, bonds, etc. If the sample fund has a holding relationship with a certain financial product, then the first node corresponding to the sample fund needs to be connected with the second-level node corresponding to the financial product through a type of edge, and the length (or weight) of the type of edge is positively correlated with the sample fund's holding ratio of the financial product. In other words, the higher the holding ratio, the "heavier" or "longer" the edge.
[0081] Subsequently, in addition to the direct relationship between the sample funds and the financial products, the association between the financial products can be further constructed. This association can be established based on information such as market indexes and industry classifications. For example, if two financial products belong to the same industry, an edge can be established between the secondary nodes corresponding to the two financial products to represent this association, which helps to reveal the mutual influence and potential risks between financial products. Similarly, the association between the sample funds can also be constructed. This association can be established based on factors such as the correlation between the funds, investment strategies, and management teams. For example, if the sample funds include a first sample fund and a second sample fund, and the first sample fund and the second sample fund are related (for example, the investment strategies of the two funds are similar or the management teams overlap), the first-level nodes corresponding to the first sample fund and the first-level nodes corresponding to the second sample fund are connected by a second-class edge, wherein the weight of the second-class edge is positively correlated with the correlation ratio of the first sample fund and the second sample fund.
[0082] Finally, use visualization tools (such as Matplotlib, Gephi, etc.) to display the constructed graph network. Visualization helps to more intuitively understand the holding structure of sample funds and the relationship between financial products. Through visualization, you can clearly see which funds hold which financial products, which financial products are related, and which funds are related.
[0083] S204: Input the graph network of the sample fund set into the time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector is extracted from the graph network of the sample fund set through the time series convolutional network, and the adjacency feature vector between the sample fund and other funds in the sample fund set is extracted through the graph attention network. The daily holding weight of the sample fund is confirmed according to the time series feature vector and the adjacency feature vector through the prediction network.
[0084] See also Figure 2B , which is a schematic diagram of a time-series graph neural network model provided in an embodiment of the present application. Figure 2B It can be seen that the TGNN model includes a TCN network, a GAT network and a FFN network, and the outputs of the GAT network and the TCN network are connected to the input of the FFN network.
[0085] Specifically, by inputting the graph network into the GAT network, the GAT network can analyze and capture the complex relationships between funds, between funds and financial products, and between financial products based on the graph network. Furthermore, in the GAT network, each node (whether as a primary node of a fund or as a secondary node of a financial product) carries a feature vector. Using the attention mechanism, the GAT network can dynamically evaluate the importance of different nodes, and then learn the relationship weights between these nodes. This process enables the GAT network to highlight those nodes that are more critical to a given task (in the embodiment of the present application, the accurate prediction of daily position weights), and integrate the node information into the updated node feature vector (i.e., the adjacent feature vector), which not only retains the characteristics of the original node, but also incorporates the influence of its neighboring nodes.
[0086] By inputting the graph network into the TCN network, the TCN network can output node feature vectors containing time series characteristics (i.e., time series feature vectors) by using causal convolution and dilated convolution. These time series feature vectors reflect the node status changing over time.
[0087] As the last layer of the TGNN model, the FFN network receives the outputs from the GAT network and the TCN network, namely the adjacency feature vector and the time series feature vector. By combining these two types of feature vectors, FFN is able to generate the final prediction result, namely the daily holding weight of the target fund. The design of FFN enables the TGNN model to consider the relationship between funds (extracted through the GAT network) while also considering the impact of time factors on fund holdings (extracted through the TCN network), thereby improving the accuracy of the prediction.
[0088] S205: Train a time series graph neural network model based on the graph network of the sample fund set and the daily holding weights of the sample funds.
[0089] S206: Obtain fund information of the target fund.
[0090] S207: Input the fund information into the time series graph neural network model to confirm the daily holding weight of the target fund.
[0091] It should be noted that steps S206-S207 are similar to steps S101-S102 and will not be described in detail here.
[0092] S208: Update the model parameters of the time series graph neural network model according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
[0093] The target financial product refers to the specific financial product (such as stocks, bonds, etc.) whose holding weight the TGNN model attempts to predict. The actual return of the target financial product refers to the actual rate of return or return of the target financial product in a specific time period (such as daily, weekly or monthly). The total actual return of the target fund refers to the overall rate of return or return obtained by the target fund through its investment portfolio (which may include multiple financial products, including the target financial product) in the same time period.
[0094] Specifically, first calculate the actual return of the target financial product in a specific time period and compare it with the total actual return of the target fund in the same time period to obtain the return ratio. This return ratio is a relative indicator used to evaluate the performance of the target financial product relative to the overall performance of the target fund. The higher the return ratio, the greater the contribution of the target financial product to the overall performance of the target fund. At the same time, it is necessary to obtain the daily holding weight of the target fund on the target financial product. The holding weight reflects the allocation strategy of the financial product at different time points. Finally, according to the calculated return ratio and daily holding weight, update the parameters of the time series graph neural network model.
[0095] Through step S208, the TGNN model can continuously learn and adapt to changes in the financial market, improve the accuracy of daily position weight predictions, optimize portfolio allocation, and improve investment returns.
[0096] S209: Obtain the actual holding information of the target fund.
[0097] Actual holdings information refers to the actual holdings data of the target fund in a specific time period (such as daily, weekly or monthly), including the holding ratio of each financial product. Actual holdings information usually comes from the fund's regular reports (such as quarterly, semi-annual and annual reports) or public data from the exchange.
[0098] S210: If the actual position information does not match the predicted position information corresponding to the daily position weight, the model parameters of the time series graph neural network model are updated according to the actual position information and the predicted position information.
[0099] Specifically, the actual position information and the predicted position information are first obtained, and the difference between the actual position information and the predicted position information is compared by using methods such as mean square error (MSE) and root mean square error (RMSE). If the difference between the actual position information and the predicted position information is significant (for example, the difference between the actual position information and the predicted position information is greater than the difference threshold), the model parameters of the TGNN model are updated using the back propagation algorithm or other optimization methods based on the actual position information and the predicted position information. Among them, the back propagation algorithm is a commonly used neural network training algorithm, which updates the model parameters by calculating the gradient of the loss function, thereby minimizing the prediction error, making the prediction results of the TGNN model closer to the actual daily position weight, thereby improving the accuracy and robustness of the TGNN model.
[0100] It should be noted that the above process is not completed in one go, but is repeated every day to form a forecasting iteration process. Each iteration will update the model parameters based on the latest market data and fund performance, so that the forecast results can continue to approach the actual situation.
[0101] In summary, the embodiment of the present application discloses a weight confirmation method, which inputs the acquired fund information of the target fund into a time-series graph neural network model, extracts the time-series feature vectors in the fund information through the time-series convolutional network of the time-series graph neural network model, and can improve the sensitivity to changes in fund holdings and the accuracy of prediction, extracts the adjacent feature vectors of the target fund and other funds through the graph attention network of the time-series graph neural network model, and can capture the mutual influence between different funds, and predicts the daily holding weight by combining the time-series feature vector and the adjacent feature vector through the prediction network of the time-series graph neural network model, and can improve the accuracy of the daily holding weight prediction. Furthermore, the predicted daily holding weight can prevent investors from missing the best trading opportunities due to decision-making delays, thereby reducing investment risks and improving investment returns.
[0102] See also Figure 3 , which is a schematic diagram of a weight confirmation device provided in an embodiment of the present application. The weight confirmation device 300 includes: an acquisition module 301 and a confirmation module 302.
[0103] An acquisition module 301 is used to acquire fund information of a target fund;
[0104] Confirmation module 302 is used to input fund information into a time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, and the adjacent feature vectors of the target fund and other funds are extracted through the graph attention network. The daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacent feature vector through the prediction network.
[0105] Optionally, the weight confirmation device 300 further includes: a first updating module;
[0106] The first updating module is used to update the model parameters of the time series graph neural network model according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
[0107] Optionally, the weight confirmation device 300 further includes: a second updating module;
[0108] The second updating module is used to obtain the actual holding information of the target fund; if the actual holding information does not match the predicted holding information corresponding to the daily holding weight, the model parameters of the time series graph neural network model are updated according to the actual holding information and the predicted holding information.
[0109] Optionally, the acquisition module 301 is specifically used to: obtain the original information of the target fund; obtain the fund information by removing invalid holding information and / or erroneous holding information in the original information, unifying the format of the original information, and completing the original information.
[0110] Optionally, the time series graph neural network model is constructed by the following units:
[0111] A first acquisition unit, used to acquire fund information of a sample fund set;
[0112] A second determination unit is used to determine a graph network of the sample fund set according to the fund information of the sample fund set, wherein the sample funds in the sample fund set are primary nodes of the graph network, the financial products held by the sample funds are secondary nodes of the graph network, the secondary nodes corresponding to the sample funds and the nodes and the financial products held by the sample funds are connected by a first-class edge, the weight of the first-class edge is positively correlated with the proportion of the sample funds' holdings of the financial products, the sample funds include a first sample fund and a second sample fund, if the first sample fund and the second sample fund are related, then the primary nodes corresponding to the first sample fund and the primary nodes corresponding to the second sample fund are connected by a second-class edge, the weight of the second-class edge is positively correlated with the correlation ratio of the first sample fund and the second sample fund;
[0113] A third input unit is used to input the graph network of the sample fund set into a time-series graph neural network model, the time-series graph neural network model includes a time-series convolutional network, a graph attention network and a prediction network, extract a time-series feature vector from the graph network of the sample fund set through the time-series convolutional network, extract an adjacency feature vector between the sample fund and other funds in the sample fund set through the graph attention network, and confirm the daily holding weight of the sample fund according to the time-series feature vector and the adjacency feature vector through the prediction network;
[0114] The fourth training unit is used to train the time series graph neural network model based on the graph network of the sample fund set and the daily holding weights of the sample funds.
[0115] In summary, the embodiment of the present application discloses a weight confirmation device, which inputs the acquired fund information of the target fund into a time-series graph neural network model, extracts the time-series feature vectors in the fund information through the time-series convolutional network of the time-series graph neural network model, and can improve the sensitivity to changes in fund holdings and the accuracy of prediction, extracts the adjacent feature vectors of the target fund and other funds through the graph attention network of the time-series graph neural network model, and can capture the mutual influence between different funds, and predicts the daily holding weight by combining the time-series feature vector and the adjacent feature vector through the prediction network of the time-series graph neural network model, and can improve the accuracy of the daily holding weight prediction. Furthermore, the predicted daily holding weight can prevent investors from missing the best trading opportunity due to decision-making delays, thereby reducing investment risks and improving investment returns.
[0116] The embodiment of the present application also provides a corresponding weight confirmation device and a computer-readable medium for implementing the weight confirmation method provided in the embodiment of the present application.
[0117] Among them, the weight confirmation device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute instructions or codes so that the device executes a weight confirmation method of any embodiment of the present application.
[0118] See also Figure 4 , which is a schematic diagram of a computer-readable medium provided in an embodiment of the present application. The computer-readable medium 400 stores a computer program 411, which implements the above-mentioned Figure 1 The steps of the weight confirmation method.
[0119] It should be noted that in the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0120] It should be noted that the machine-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
[0121] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device.
[0122] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.
[0123] Although several specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present application. Certain features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination.
[0124] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.
Claims
1. A weight confirmation method, characterized in that: The method comprises: Obtain fund information of the target fund; The fund information is input into a time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, the adjacency feature vector between the target fund and other funds is extracted through the graph attention network, and the daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacency feature vector through the prediction network.
2. The method according to claim 1, characterized in that The method further comprises: The model parameters of the time series graph neural network model are updated according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
3. The method according to claim 1, characterized in that The method further comprises: Obtaining actual holding information of the target fund; If the actual position information does not match the predicted position information corresponding to the daily position weight, the model parameters of the time series graph neural network model are updated according to the actual position information and the predicted position information.
4. The method according to claim 1, characterized in that: The obtaining of fund information of the target fund includes: Obtain original information of target funds; The fund information is obtained by performing at least one of removing invalid position information and / or erroneous position information in the original information, unifying the format of the original information, and completing the original information.
5. The method according to claim 1, characterized in that The timing graph neural network model is constructed in the following steps: Get the fund information of the sample fund set; According to the fund information of the sample fund set, a graph network of the sample fund set is determined, wherein the sample funds in the sample fund set are primary nodes of the graph network, the financial products held by the sample funds are secondary nodes of the graph network, and the nodes corresponding to the sample funds and the secondary nodes corresponding to the financial products held by the sample funds are connected by a first-class edge, and the weight of the first-class edge is positively correlated with the holding ratio of the sample fund to the financial products, and the sample funds include a first sample fund and a second sample fund. If the first sample fund is related to the second sample fund, the primary node corresponding to the first sample fund and the primary node corresponding to the second sample fund are connected by a second-class edge, and the weight of the second-class edge is positively correlated with the correlation ratio of the first sample fund to the second sample fund; Inputting the graph network of the sample fund set into a time-series graph neural network model, the time-series graph neural network model includes a time-series convolutional network, a graph attention network and a prediction network, extracting a time-series feature vector from the graph network of the sample fund set through the time-series convolutional network, extracting an adjacency feature vector between the sample fund and other funds in the sample fund set through the graph attention network, and confirming the daily holding weight of the sample fund according to the time-series feature vector and the adjacency feature vector through the prediction network; A time-series graph neural network model is trained based on the graph network of the sample fund set and the daily holding weights of the sample funds.
6. A weight confirmation device, characterized in that: The device comprises: an acquisition module and a confirmation module; The acquisition module is used to acquire the fund information of the target fund; The confirmation module is used to input the fund information into a time series graph neural network model, which includes a time series convolutional network, a graph attention network and a prediction network. The time series feature vector in the fund information is extracted through the time series convolutional network, the adjacent feature vector between the target fund and other funds is extracted through the graph attention network, and the daily holding weight of the target fund is confirmed according to the time series feature vector and the adjacent feature vector through the prediction network.
7. The device according to claim 6, characterized in that The device further includes: a first updating module; The first updating module is used to update the model parameters of the time series graph neural network model according to the ratio of the actual return of the target financial product to the total actual return of the target fund, and the daily holding weight of the target fund.
8. The device according to claim 6, characterized in that The device further comprises: a second updating module; The second updating module is used to obtain the actual holding information of the target fund; if the actual holding information does not match the predicted holding information corresponding to the daily holding weight, the model parameters of the time series graph neural network model are updated according to the actual holding information and the predicted holding information.
9. A weight confirmation device, characterized in that: The device comprises: a memory and a processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the weight confirmation method as described in any one of claims 1 to 5.
10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the weight confirmation method according to any one of claims 1 to 5 is implemented.