Value fluctuation prediction model generating method, prediction method, apparatus therefor, and program therefor

Abstract

To provide a technique for predicting a value of financial products which can easily be used by a person lacking dedicated knowledge on the financial products.SOLUTION: The present invention is directed to a prediction model generating method for predicting information relating to a value of a financial product. The method has a preparation step (S1) of preparing a learning data including a time-serial value fluctuation data 23 relating to fluctuation of a value of a financial product, a time-serial financial index data 21 related to fluctuation of the financial index, and a time-serial economical index data 22 relating to fluctuation of the economic index, and a learning step (S3) of carrying out machine learning using the learning data to generate a learned prediction model 24 for use in predicting information relating to the value of the financial product at a desired point of time. A period between an end time of a data period of the financial index data 21 and the economical index data 22 and the value fluctuation data 23 is a predetermined period of time or longer.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method, an apparatus and a program for predicting fluctuations in the value of a financial product. [Background technology]

[0002] There are various types of financial products, such as investment trusts (funds), foreign exchange, stocks, bonds, commodity futures, and gold bullion, and the trading prices of these financial products fluctuate over time. Accurately predicting fluctuations in the trading prices of financial products is difficult, and various techniques have been proposed. For example, Patent Document 1 discloses a technique for predicting the prices of financial products. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-175435 Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, with the widespread use of internet-enabled smartphones and tablet devices, it has become possible to view fluctuations in the prices of financial products anytime, anywhere. Interest and motivation in asset formation has also been increasing year by year, with individuals such as ordinary office workers, housewives, and students showing interest in trading financial products. However, the reality is that many of these individuals lack specialized knowledge about financial products.

[0005] The technology disclosed in Patent Document 1 allows a user to select and set various conditions used in predicting the price of a financial product, thereby appropriately reflecting the user's investment stance in the price prediction. However, the technology disclosed in Patent Document 1 requires users to have specialized knowledge about financial products. Requiring individuals, such as ordinary office workers, housewives, and students, to acquire such specialized knowledge is an obstacle for financial institutions, such as securities companies, banks, and life insurance companies, that are trying to acquire new customers. There is a need for a technology for predicting the value of financial products that can be easily used even by people who lack specialized knowledge about financial products.

[0006] The present invention aims to provide a technology for predicting the value of financial products that can be easily used even by people who lack specialized knowledge about financial products. [Means for solving the problem]

[0007] The present invention for solving the above problems includes, for example, the following aspects. (Section 1) a preparation step of preparing learning data including time-series value fluctuation data relating to fluctuations in the value of financial instruments, time-series financial index data relating to fluctuations in financial indexes, and time-series economic index data relating to fluctuations in economic indexes; a learning step of performing machine learning using the learning data to generate a trained prediction model that predicts information about the value of the financial product at a desired time; Including, A prediction model generation method, wherein the period between the end of a data period for the financial index data and the economic index data and the time of the value fluctuation data is a predetermined time or longer. (Section 2) Item 2. The prediction model generation method according to Item 1, wherein the learning step sets each of the indicators included in the financial indicator data and the economic indicator data as explanatory variables, and sets the value fluctuation data as a target variable to learn the prediction model. (Section 3) further comprising an index fluctuation suppression step of performing data processing to suppress irregular fluctuations in the indexes included in the financial index data and the economic index data; 3. The prediction model generation method according to item 1 or 2, wherein the learning step performs machine learning using the learning data including the financial indicator data and the economic indicator data in which irregular fluctuations in indicators have been suppressed. (Section 4) A method for predicting information about the value of a financial instrument, comprising: a prediction time acquisition step of acquiring a desired prediction time; a prediction step of predicting information about the value at the acquired time point by inputting financial indicator data and economic indicator data whose end point of a data period is at least the predetermined time before the acquired time point into a prediction model generated by the prediction model generation method according to any one of claims 1 to 3; A prediction method, including: (Section 5) Item 5. The prediction method according to Item 4, further comprising a contribution calculation step of calculating the contribution of each of the indicators included in the financial indicator data and the economic indicator data, which are set as explanatory variables in the trained prediction model, when the trained prediction model predicts information about the value. (Section 6) a fluctuation value acquisition step of acquiring a fluctuation value of an index that is included in the financial index data or the economic index data and whose fluctuation is desired; an index value correction step of correcting the value of the index set as an explanatory variable in the trained prediction model using the acquired fluctuation value; further comprising Item 6. The prediction method according to item 4 or 5, wherein the prediction step predicts information about the value using the trained prediction model in which the value of the index has been corrected. (Section 7) 7. The prediction method according to any one of items 4 to 6, wherein the prediction step predicts information about the value for each of a plurality of different financial instruments, and ranks and displays each of the plurality of different financial instruments based on the predicted information about each of the values. (Section 8) The prediction time point acquisition step acquires a plurality of desired prediction time points, 8. The prediction method according to any one of items 4 to 7, wherein the prediction step predicts information about the value of the target financial product at each of the desired multiple points in time by inputting each of the acquired multiple points in time into the trained prediction model. (Section 9) a preparation means for preparing learning data including time-series value fluctuation data relating to fluctuations in the value of financial instruments, time-series financial index data relating to fluctuations in financial indexes, and time-series economic index data relating to fluctuations in economic indexes; a learning means for performing machine learning using the learning data to generate a trained prediction model that predicts information about the value of the financial product at a desired time; Equipped with A prediction model generation device, wherein the period between the end of a data period for the financial index data and the economic index data and the time of the value fluctuation data is equal to or longer than a predetermined time. (Section 10) A prediction device for predicting information about the value of a financial product, a prediction time acquisition means for acquiring a desired prediction time; A prediction means for predicting information about the value at the acquired time point by inputting financial index data and economic index data whose end point of a data period is earlier than the acquired time point by the predetermined time or more into a prediction model generated by the prediction model generation device according to item 9; A prediction device comprising: (Section 11) On the computer, A prediction model generation program for executing each step of the prediction model generation method according to any one of items 1 to 3. (Section 12) On the computer, A prediction program for executing each step of the prediction method according to any one of items 4 to 8. [Effects of the Invention]

[0008] According to the present invention, it is possible to provide a technology for predicting the value of financial products that can be easily used even by people who lack specialized knowledge about financial products. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating a schematic configuration of a prediction system according to an embodiment. [Figure 2] FIG. 10 is a diagram illustrating an image of data used when making a prediction. [Figure 3] FIG. 10 is a diagram illustrating an image of data used when making a prediction. [Figure 4] FIG. 10 is a diagram illustrating an image of data used when making a prediction. [Figure 5] FIG. 2 is a block diagram illustrating functions of a prediction device according to an embodiment. [Figure 6] 10A and 10B are diagrams for explaining data processing that an index fluctuation suppression unit applies to original series data. [Figure 7] 1 is a flowchart illustrating a procedure in which a prediction device according to an embodiment performs machine learning on a prediction model. [Figure 8] 1 is a flowchart illustrating an example of a procedure in which a prediction device according to an embodiment predicts information about the value of a single financial product. [Figure 9] 10 is an example of a screen display of a terminal device used by a user. [Figure 10] 10 is an example of a screen display of a terminal device used by a user. [Figure 11]1 is a flowchart illustrating an example of a procedure in which a prediction device according to an embodiment predicts information about the value of a plurality of financial products. [Figure 12] 10 is an example of a screen display of a terminal device used by a user. [Figure 13] 13 is an example of a screen display when the variation value of an index that the user desires to vary is changed from the state shown in FIG. 12. [Figure 14] 10 is another example of a screen display of a terminal device used by a user. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and drawings, the same reference numerals will denote the same or similar components, and therefore, redundant descriptions of the same or similar components will be omitted. [Prediction System] <System Overview>

[0011] FIG. 1 is a diagram illustrating a schematic configuration of a prediction system according to an embodiment.

[0012] A prediction system 100 according to one embodiment is a system that acquires a prediction time point desired by a user 9 and predicts information related to the value of a financial product at the desired time point. The prediction system 100 includes a prediction device 1 and a terminal device 2. The prediction device 1 and the terminal device 2 are connected directly or indirectly via a wired or wireless connection, for example, so that data can be transmitted and received via a network 3. The network 3 can be a public network such as the Internet, or a private network such as an in-house intranet or VPN. In another embodiment, the prediction device 1 and the terminal device 2 can be integrated into one device in a standalone format that does not use the network 3.

[0013] The prediction device 1 predicts information related to value by acquiring the desired time point for prediction, selecting appropriate financial indicator data 21 and economic indicator data 22 according to the acquired time point, and inputting them into the trained prediction model 24. The user 9 inputs the desired time point for prediction to the terminal device 2, for example, via the input unit 32B. The prediction device 1 acquires the user 9's desired time point for prediction from the terminal device 2, for example, via the network 3.

[0014] The prediction model 24 is trained in advance by the prediction device 1. In this embodiment, the prediction device 1 also generates the trained prediction model 24. When the prediction device 1 generates the trained prediction model 24, the prediction device 1 can also be expressed as a prediction model generation device 1.

[0015] The prediction device 1 prepares learning data including time-series financial index data 21 relating to fluctuations in financial indexes, time-series economic index data 22 relating to fluctuations in economic indexes, and time-series value fluctuation data 23 relating to fluctuations in the value of financial products. The prediction device 1 performs machine learning on a prediction model 24 using the prepared learning data, and generates a trained prediction model 24 for predicting information about the value of financial products at a desired time point. As will be described later, the learning data used by the prediction device 1 for machine learning has a predetermined time (preferably two months) or more between the end of the data period of the financial index data 21 and the economic index data 22 and the time of the value fluctuation data 23.

[0016] Financial products that are the subject of prediction include, for example, investment trusts (funds), foreign exchange, stocks, bonds, commodity futures, and gold bullion. Information related to values that are the subject of prediction includes information that directly represents value, such as the asset value and trading price of an investment trust, as well as information that represents value fluctuations, such as the rate of change and predicted results of change (increase or decrease) of the investment trust. In this embodiment, as an example of a financial product, an investment trust managed by an insurance company in a separate account for variable insurance is targeted, and an example of predicting the asset value (hereinafter, asset value will be simply referred to as value) of the investment trust managed in the separate account is described. The rate of change can be calculated from the proportion of the asset value (trading price) that has changed between a base point and the desired prediction point. The base point can be, for example, the end of the month preceding the date on which the prediction is made.

[0017] In this embodiment, the prediction device 1 is operated by an insurance company that manages variable insurance, and the terminal device 2 is used by a customer (user 9) who has signed a variable insurance contract to predict the future value of an investment trust that is the main investment target of the separate account for the variable insurance that the customer has signed a contract for. The user 9 is not limited to the customer who has signed a variable insurance contract, but may also be, for example, a sales representative for the variable insurance or a sales representative at an agency that sells variable insurance.

[0018] This section provides additional information about variable insurance managed by insurance companies. Insurance companies sell and manage variable insurance products that incorporate separate accounts (in this embodiment, investment trusts are the primary investment assets) to customers. A customer's asset allocation (also called a portfolio) is constructed using one or more separate accounts corresponding to the type of investment assets (in this embodiment, investment trusts). As a result, variable insurance products managed by insurance companies include one or more separate accounts corresponding to the type of investment assets. In addition to a form in which the customer selects one or more separate accounts to manage, this also includes a form that incorporates a fund wrap service (a service in which an investment advisor advises the customer on the combination of separate accounts to manage based on the customer's desired return and risk levels). Insurance companies offer customers the benefits of separate accounts established by insurance companies that were not available to customers with traditional investment trusts or fund wrap products. The benefits of a special account include, for example, in terms of taxation when managing assets, the deferral of tax on investment gains during rebalancing and the tax exemption of investment gains during switching, and, for example, in terms of taxation when inheriting an estate, a tax exemption limit of 5 million yen applies per legal heir.

[0019] 2 to 4 are diagrams for explaining the image of data used when making a prediction. For simplicity, in the explanation referring to FIGS. 2 to 4, the time point at which the user 9 desires a prediction is fixed to the last day of the month. The relationship between the data period of the learning data and the time point at which a prediction is desired will be explained with reference to FIGS. 2 to 4.

[0020] The learning data used for machine learning by the prediction device 1 according to one embodiment has a predetermined time (preferably two months) or more between the end of the data period of the financial index data 21 and the economic index data 22 and the time of the value fluctuation data 23.

[0021] As shown in FIGS. 2 to 4, in a prediction model generation method and a prediction method according to an embodiment, multiple pieces of training data are prepared in advance. A predetermined time (e.g., two months) is always provided between the prepared training data used as explanatory variables and the data used as a dependent variable. That is, when preparing the training data, a predetermined time (e.g., two months) is provided between the data period of financial index data 21 and economic index data 22 set as explanatory variables and the data period of value fluctuation data 23 set as a dependent variable. For example, as shown in FIG. 3, when a customer (user 9) wants to know the value of a financial product (investment trust) at the end of the month one month from the present (e.g., September 16, 2023) (October 31, 2023), the prediction device 1 selects financial index data 21 and economic index data 22 (data period: up to August 31, 2023) with an appropriate end date in the data period according to the time at which the user 9 desires to make a prediction (the end of the month one month from the present), and inputs the selected index data 21 and 22 into a trained prediction model 24. The trained prediction model 24 outputs the predicted value of the financial product as of the end of the month one month from now. This makes it possible to prevent the prediction model from being excessively influenced by the most recent financial index data, improves the prediction accuracy of the information on the value of the financial product predicted by the prediction device 1, and prevents the prediction result from changing excessively even when the user 9 views the prediction result.

[0022] 2 illustrates an example in which a user 9 uses the prediction device 1 on, for example, September 1st to predict the value of a financial product at the end of October, the end of November, and the end of December. In all three cases shown in FIG. 2, the index data 21 and 22 up to the day before the prediction execution date are input into the prediction model 24 to obtain a prediction result, but a predetermined time of two months is ensured between August 31st, the last day of the data period for the index data 21 and 22, and the time at which the user 9 desires to make a prediction (the end of October, the end of November, or the end of December).

[0023] 3 illustrates an example in which user 9 uses prediction device 1 on, for example, September 16th to predict the value of a financial product at the end of October, the end of November, and the end of December. In the case of FIG. 3, when predicting the value at the end of October, which is one month from now, unlike the case of FIG. 2 in which the value at the end of October is predicted, the last day of the data period for index data 21 and 22 input into prediction model 24 is August 31st, not September 15th, which is the day before the prediction execution date. This is because the time between September 16th, which is the prediction execution date, and October 31st, which is the time when user 9 desires to make a prediction, is less than the predetermined time of two months.

[0024] 4 illustrates an example in which a user 9 uses the prediction device 1 on October 1st to predict the value of a financial product at the end of November, the end of December, and the end of January of the following year. All three cases shown in FIG. 4 are similar to the three cases shown in FIG. 2.

[0025] A trained prediction model 24 is created for each desired prediction point in time for each target financial product (investment trust). For example, if there are seven types of target investment trusts and the desired prediction points in time are the end of the month one month from now, the end of the month two months from now, and the end of the month three months from now, then a total of 21 trained prediction models 24 (7 types x 3) are created.

[0026] As explained above with reference to Figures 2 to 4, the learning data used by the prediction device 1 for machine learning has a predetermined time (preferably two months) or more between the end of the data period of the financial index data 21 and the economic index data 22 and the time of the value fluctuation data 23. The prediction device 1 selects the financial index data 21 and the economic index data 22 whose end date of the data period is appropriate according to the time when the user 9 desires to make a prediction, inputs the selected index data 21, 22 into the trained prediction model 24, and outputs the predicted value of the financial product.

[0027] This ensures a certain amount of time between the end of the data period of the learning data and the time when the user 9 desires a prediction, making it possible to prevent the prediction model from being excessively influenced by the most recent financial index data and improving the prediction accuracy of information regarding the value of financial products predicted by the prediction device 1. When prediction accuracy is improved, excessive changes in the prediction results are prevented, even in cases where the user 9 looks at the prediction results every day, and confusion among users 9 who lack specialized knowledge about financial products is reduced. <Hardware configuration>

[0028] The prediction device 1 and the terminal device 2 can be configured using, for example, a general-purpose computer, a tablet terminal, a smartphone, etc. The prediction device 1 and the terminal device 2 can be configured entirely by a general-purpose computer, or partly by a tablet terminal or a smartphone.

[0029] The prediction device 1 includes, as its hardware configuration, a processor (not shown) such as a CPU for processing data, a main memory (not shown) used by the processor as a work area for data processing, an auxiliary memory 20 used for temporarily storing data, a communication interface unit (communication I / F unit) 31, and an input unit 32A. The auxiliary memory 20 stores data 21, 22, and 23 as learning data, a prediction model 24, a prediction model generation program P, and the like. A ,Prediction Program P B etc. are stored as appropriate. The communication I / F unit 31 can be various wired or wireless connections such as Ethernet (registered trademark), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc. The input unit 32A can be configured with, for example, a mouse, a keyboard, etc.

[0030] The terminal device 2 includes, as hardware components, an input unit 32B and a display unit 33B. The input unit 32B is a device that accepts input from the user 9. The display unit 33B is a device that displays information to the user 9. The input unit 32B can be configured, for example, by a mouse, a pointing device, a keyboard, etc., and the display unit 33B can be configured, for example, by a liquid crystal display, an organic EL display, etc. The input unit 32B and the display unit 33B can also be configured as an integrated touch panel. [Device configuration]

[0031] FIG. 5 is a block diagram illustrating the function of a prediction device according to an embodiment.

[0032] The prediction device 1 includes a processing unit 10. The processing unit 10 includes, as functional blocks, a preparation unit 11, an index fluctuation suppression unit 12, a learning unit 13, a financial product name acquisition unit 14 (investment trust name acquisition unit 14), a prediction time acquisition unit 15, a fluctuation value acquisition unit 16, an index value correction unit 17, a contribution calculation unit 18, and a prediction unit 19. Of these multiple functional blocks 11 to 19, the functional blocks 11 to 13 perform machine learning on a prediction model 24 to generate a trained prediction model 24, and the functional blocks 14 to 19 acquire a prediction time point desired by a user 9 and predict information regarding the value of a financial product (e.g., the asset value of an investment trust) at the desired time point using the trained prediction model.

[0033] These functional blocks 11 to 19 included in the processing unit 10 are implemented by the processor of the prediction device 1 in accordance with the prediction model generation program P A and prediction program P B The prediction model generation program P can be implemented as software by loading the program into the main storage device of the prediction device 1 and executing it. Alternatively, these function blocks 11 to 19 can be implemented as hardware using an integrated circuit or the like. A is a computer program for executing the function blocks 11 to 13. Bis a computer program for executing the function blocks 14 to 19. The device 1 executes the prediction model generation program P A When executed, it functions as a prediction model generation device 1, and the prediction program P B When this is executed, it functions as a prediction device 1. <Learning data>

[0034] The learning data includes time-series financial index data 21 relating to fluctuations in financial indexes, time-series economic index data 22 relating to fluctuations in economic indexes, and time-series value fluctuation data 23 relating to fluctuations in the value of financial products. These learning data are data used for training the prediction model 24, and are a set of data 21, 22 set as explanatory variables in the prediction model 24 and data 23 set as a target variable in the prediction model 24. A plurality of learning data (e.g., one learning data per week) are prepared according to the last day of the data period. In this embodiment, one learning data is prepared per week. In this embodiment, the learning data is pre-stored in the auxiliary storage device 20 of the prediction device 1.

[0035] The financial index data 21 is time-series data relating to fluctuations in financial indexes. The economic index data 22 is time-series data relating to fluctuations in economic indexes. The value fluctuation data 23 is time-series data relating to fluctuations in the value of financial products. In this embodiment, the value fluctuation data 23 is data indicating fluctuations (price movements) in the asset value of financial products (investment trusts).

[0036] Examples of financial indicators and economic indicators used in this embodiment are shown in Tables 1 to 3. Of these indicators shown in Tables 1 to 3, the indicators shown in Table 1 are financial indicators, and the indicators shown in Tables 2 and 3 are economic indicators. Preferably, the financial indicators are financial market indicators such as those of the foreign exchange market, stock market, and bond market, as exemplified in Table 1, and the economic indicators are macroeconomic indicators such as those exemplified in Tables 2 and 3. Note that it is also possible to use other financial market indicators such as those of the commodity market and real estate market as financial indicators.

[0037] [Table 1]

[0038] [Table 2]

[0039] [Table 3] <Prediction model>

[0040] Various known machine learning models classified as supervised learning can be used for the prediction model 24. In this embodiment, a gradient boosting decision tree is used for the prediction model 24. In other embodiments, instead of a gradient boosting decision tree, a multi-factor model, which is a type of multiple regression model, or a multi-layer artificial neural network for deep learning can be used for the prediction model 24. <Function block for generating predictive models>

[0041] The preparation unit 11 prepares learning data including time-series value fluctuation data 23 relating to fluctuations in the value of financial products, time-series financial index data 21 relating to fluctuations in financial indexes, and time-series economic index data 22 relating to fluctuations in economic indexes. The preparation unit 11 prepares multiple pieces of learning data according to the last date of the data period of the learning data. The preparation unit 11 prepares one piece of learning data per week, for example. As a result, multiple pieces of learning data with different last dates of the data period are prepared and stored in the auxiliary storage device 20 of the prediction device 1.

[0042] FIG. 6 is a diagram illustrating the data processing performed by the index fluctuation suppression unit on the original series of data. The index fluctuation suppression unit 12 performs data processing to suppress irregular fluctuations in the indexes included in the financial index data 21 and the economic index data 22. The index fluctuation suppression unit 12 performs the data processing illustrated in FIG. 6 to suppress irregular fluctuations in the indexes on the original series of data. This allows the trends and volatility of the financial and economic indexes shown in Tables 1 to 3 to be more appropriately captured and reflected in the machine learning of the prediction model 24. Table 4 shows variations in the data processing methods performed by the index fluctuation suppression unit 12.

[0043] [Table 4]

[0044] For example, according to the processing method "3MSteep / 3MMean," a value is calculated in which the steepness (steepness) of the target index value over the past three months is used as the numerator and the mean (mean) of the past three months is used as the denominator. The calculated value is used as a new index for the explanatory variable of the prediction model 24. In this embodiment, a total of 83 types of financial and economic indexes are used, as shown in Tables 1 to 3, and as shown in Table 4, there are eight variations in the processing method applied to the original series data of each index. Therefore, in this embodiment, a total of 664 types of indexes are used as explanatory variables of the prediction model 24. However, it is also possible not to use indexes that have a low correlation with the value of the financial product to be predicted as explanatory variables. The index fluctuation suppression unit 12 can be configured as desired.

[0045] The learning unit 13 performs machine learning using the learning data to generate a trained prediction model 24. The prediction model 24 is a machine learning model that predicts information about the value of financial products. The learning unit 13 sets each index included in the financial index data 21 and the economic index data 22 as an explanatory variable, and sets the value fluctuation data 23 as a target variable, to train the prediction model 24. <Function block that performs predictions using the generated prediction model>

[0046] The financial product name acquisition unit 14 (investment trust name acquisition unit 14) acquires the name of the financial product (investment trust) for which the user 9 wishes to make a prediction. The prediction time acquisition unit 15 acquires the time for prediction desired by the user 9. In this embodiment, the name 25 of the investment trust for which a prediction is desired and the desired time 26 for prediction are acquired via the input unit 32B of the terminal device 2 and the network 3. Note that, for example, if the name of the financial product (investment trust) to be predicted is fixed, the financial product name acquisition unit 14 can be omitted.

[0047] The fluctuation value acquisition unit 16 acquires the fluctuation value of an index that is included in the financial index data 21 or the economic index data 22 and is desired to fluctuate. The type of index whose value is to fluctuate is preferably an index related to interest rates, such as the U.S. FF rate shown in Table 3 or an index related to the Bank of Japan's policy interest rate shown in Table 2. The index value correction unit 17 corrects the value of the index that is set as an explanatory variable in the trained prediction model 24 using the fluctuation value acquired by the fluctuation value acquisition unit 16. In this embodiment, the fluctuation value 27 of the index that is desired to fluctuate is acquired via the input unit 32B of the terminal device 2 and the network 3. The fluctuation value acquisition unit 16 and the index value correction unit 17 can have any configuration. The process of acquiring the fluctuation value of an index and correcting the value of the index will be described in detail later.

[0048] The contribution calculation unit 18 is each of the indicators included in the financial indicator data 21 and the economic indicator data 22, which are set as explanatory variables in the trained prediction model 24, and calculates the contribution of each of the indicators when information about value is predicted by the trained prediction model 24. The contribution calculation unit 18 can have any configuration. The process of calculating the contribution will be described in detail later.

[0049] The prediction unit 19 selects appropriate financial index data 21 and economic index data 22 acquired by the prediction time acquisition unit 15 according to the time point of prediction desired by the user 9, and inputs the selected index data 21, 22 into the trained prediction model 24. As described with reference to Figures 2 to 4, the prediction unit 19 inputs financial index data 21 and economic index data 22 whose data period ends a predetermined time (for example, two months) or more before the time point of prediction desired by the user 9 into the trained prediction model 24, thereby predicting information regarding the value of the financial product at the time point of prediction desired by the user 9. [Machine learning procedure]

[0050] FIG. 7 is a flowchart illustrating a procedure in which a prediction device according to an embodiment performs machine learning of a prediction model.

[0051] In step S1 (preparation step), the preparation unit 11 prepares learning data including time-series value fluctuation data 23 relating to fluctuations in the value of financial products (investment trusts), time-series financial index data 21 relating to fluctuations in financial indexes, and time-series economic index data 22 relating to fluctuations in economic indexes. The preparation unit 11 prepares multiple pieces of learning data according to the last date of the data period of the learning data. As described with reference to FIG. 2, the last dates of the data periods of the multiple pieces of learning data are different. The learning data is stored in the auxiliary storage device 20 of the prediction device 1 via the network 3 or an input unit 32A such as a keyboard and a mouse.

[0052] In step S2 (index fluctuation suppression step), the index fluctuation suppression unit 12 performs data processing on each of the indexes included in the financial index data 21 and the economic index data 22 to suppress irregular fluctuations in the indexes.

[0053] In step S3 (learning step), the learning unit 13 performs machine learning using the learning data to generate a trained prediction model 24 that predicts information about the value of a financial product at the desired time point. The learning unit 13 sets each of the indicators included in the financial indicator data 21 and the economic indicator data 22 as explanatory variables, sets the value fluctuation data 23 as a target variable, and trains the prediction model 24. In this embodiment, since data processing is performed in step S2 to suppress irregular fluctuations in the indicators, a total of 664 types of indicators are used as explanatory variables in step S3. The generated trained prediction model 24 is stored in the auxiliary storage device 20 of the prediction device 1. [Prediction procedure]

[0054] Fig. 8 is a flowchart illustrating an example of a procedure in which a prediction device according to an embodiment predicts information about the value of a single financial product. Figs. 9 and 10 are examples of screen displays on a terminal device used by a user. In the procedure shown in steps S4 to S7 below, information about the value of a single financial product (investment trust) for which the user 9 wishes to make a prediction is predicted at the time of the prediction desired by the user 9.

[0055] In step S4 (issue acquisition step), the financial product issue acquisition unit 14 acquires the issue of the financial product (investment trust) for which the user 9 wishes to make a prediction. As illustrated in FIG. 9, in this embodiment, the portion of the screen indicated by reference numeral 41 is a pull-down menu, and the user 9 selects the issue of the financial product for which the user 9 wishes to make a prediction from the pull-down menu. In the example shown, "Growth Balanced" is selected as the issue of the financial product for which the user 9 wishes to make a prediction. The issue 25 of the financial product selected by the user 9 is stored in the auxiliary storage device 20 of the prediction device 1. Note that, as indicated by reference numeral 42, the asset allocation (portfolio) of the issue 25 of the financial product selected by the user 9 is displayed on the display unit 33B of the terminal device 2.

[0056] In step S5 (prediction time acquisition step), the prediction time acquisition unit 15 acquires the prediction time desired by the user 9. As illustrated in FIG. 9, in this embodiment, the portions of the screen indicated by the reference numerals 43a to 43c are alternative menus, and the user 9 selects by pressing one of the buttons 43a to 43c corresponding to the desired prediction time. In the illustrated example, "end of October" has been selected as the desired prediction time, and the corresponding button 43a is displayed in reverse video. The desired prediction time 26 selected by the user 9 is stored in the auxiliary storage device 20 of the prediction device 1.

[0057] In step S6 (prediction step), the prediction unit 19 inputs financial indicator data 21 and economic indicator data 22 whose data period ends a predetermined time (e.g., two months) or more before the time of prediction desired by the user 9 into the trained prediction model 24, thereby predicting information regarding the value of the financial product at the time of prediction desired by the user 9. Note that when performing machine learning of the prediction model 24, data processing to suppress irregular fluctuations in the indicators is performed on the indicator data 21 and 22 in step S2, and if the number of indicators used as explanatory variables of the prediction model 24 is increasing, the types of indicator data 21 and 22 input into the prediction model 24 in this step can also be adjusted accordingly.

[0058] As shown by reference numeral 44a in FIG. 9 , in this embodiment, the prediction unit 19 displays the rate of change of the financial product (investment trust) selected by the user 9 on the display unit 33B of the terminal device 2 as a prediction result by the prediction unit 19. The rate of change of the financial product can be calculated from the proportion of the asset value (trading price) that has changed between a base point in time and the desired prediction point in time. The base point in time can be, for example, the end of the month prior to the date on which the prediction is made. The rate of change is calculated by, for example, the prediction unit 19. Furthermore, as shown by reference numeral 44b, if the value of the financial product selected by the user 9 has increased, the prediction unit 19 can also display an image 44b indicating this on the display unit 33B of the terminal device 2.

[0059] In step S7 (contribution degree calculation step), the contribution degree calculation unit 18 calculates the contribution degree of each of the indicators included in the financial indicator data 21 and the economic indicator data 22, which are set as explanatory variables in the trained prediction model 24, when information related to value is predicted by the trained prediction model 24. The contribution degree calculation unit 18 displays the calculated contribution degree of each indicator on the display unit 33B of the terminal device 2. The contribution degree calculation unit 18 can use, for example, a calculation method using the SHAP (Shapley additive explanations) method to calculate the contribution degree of each indicator.

[0060] As illustrated in FIG. 10 , in this embodiment, the contribution 45 of each indicator 46 is displayed in numerical value and bar graph order in order of magnitude (absolute value) of the contribution 45. In the illustrated example, the indicator (explanatory variable) that contributes most to the prediction result by the prediction unit 19 is "China Retail Sales (Year-on-Year)_6Msteep," which is the economic indicator "China Retail Sales (Year-on-Year)" shown in Table 3, with the index fluctuation suppression processing of "6Msteep." The magnitude of the contribution is "-0.29." In the illustrated example, to make it easier for the user 9 to understand the prediction result by the prediction unit 19, the contribution calculation unit 18 displays this indicator that contributes most to the prediction result as a focus indicator on the display unit 33B of the terminal device 2, as indicated by reference numeral 47 in FIGS. 9 and 10 . Displaying the indicators with high contributions and their characteristics on the screen improves the user 9's interpretability of the prediction result by the prediction unit 19.

[0061] Fig. 11 is a flowchart illustrating an example of a procedure in which a prediction device according to an embodiment predicts information relating to the value of a plurality of financial products. Fig. 12 is an example of a screen display on a terminal device used by a user. In the procedure shown in steps S8 to S11 below, information relating to the value of a plurality of financial products (investment trusts) with different asset allocation ratios that have been prepared in advance is predicted at the time of prediction desired by the user 9. In this embodiment, the investment trusts with different asset allocation ratios are of seven types: stable balanced, growth balanced, stable growth balanced, Japanese stock, US stock, global stock, and US bond.

[0062] In step S8 (prediction time acquisition step), similarly to step S5, the prediction time acquisition unit 15 acquires the prediction time points desired by the user 9. As illustrated in FIG. 12, in this embodiment, the prediction time points desired by the user 9 are fixed to three types, as indicated by reference numerals 43a to 43c on the screen, and the prediction device 1 processes the procedures shown in the following steps S9 to S11 in parallel for each of the prediction time points 43a to 43c desired by the user 9. The prediction time acquisition unit 15 acquires multiple prediction time points desired by the user 9.

[0063] In step S9 (fluctuation value acquisition step), the fluctuation value acquisition unit 16 acquires the fluctuation value of an index that is included in the financial index data 21 or the economic index data 22 and that is desired to fluctuate. In this embodiment, the index that is fluctuated when the prediction device 1 makes a prediction is the U.S. FF rate shown in Table 3. In this embodiment, the portion indicated by reference numeral 48 in Figures 12 and 13 is a slide bar, and the fluctuation value acquisition unit 16 acquires the fluctuation value of the index when the user 9 moves the slide bar 48. In the example shown in Figure 12, the fluctuation value of the index is "±0.0%" (current value).

[0064] In step S10 (index value correction step), the index value correction unit 17 corrects the value of the index set as an explanatory variable in the trained prediction model 24, using the fluctuation value acquired by the fluctuation value acquisition unit 16. In the example shown in Fig. 12, the fluctuation value of the index is "±0.0%" (current value), and the index value correction unit 17 has not yet corrected the index value.

[0065] In step S11 (prediction step), similar to step S6, the prediction unit 19 inputs financial indicator data 21 and economic indicator data 22, the end point of which is a data period that is more than a predetermined time (e.g., two months) before the time of prediction desired by the user 9, into the trained prediction model 24, thereby predicting information related to the value of the financial product at the time of prediction desired by the user 9. Unlike step S6, in step S11, the prediction unit 19 predicts information related to the value of the financial product using the trained prediction model 24 in which the index value has been corrected by the index value correction unit 17. In the example shown in FIG. 12, the fluctuation value of the index is "±0.0%" (current value), and the prediction unit 19 predicts information related to the value of the financial product using the trained prediction model 24 in which the index value has not yet been corrected.

[0066] In this embodiment, the prediction unit 19 predicts information about the value of each of a plurality of different financial products, and ranks and displays each of the plurality of different financial products based on the information about their respective predicted values. As shown in Fig. 12, the display unit 33B of the terminal device 2 displays seven investment trusts with different asset allocation ratios in a ranked format in descending order of the rate of change.

[0067] Thereafter, the prediction device 1 performs the procedure shown in steps S9 to S11 for each of the prediction times 43b and 43c desired by the user 9 in the same manner as for the prediction time 43a, and displays the prediction results shown in Fig. 12 on the display unit 33B of the terminal device 2. The prediction unit 19 inputs each of the multiple time points desired by the user 9 into the trained prediction model 24, thereby predicting information about the value of the target financial product at each of the multiple desired time points.

[0068] FIG. 13 is an example of a screen display on a terminal device used by a user, and is an example of a screen display when the variation value of an index that the user desires to vary is changed from the state shown in FIG.

[0069] In the example shown in FIG. 13 , when the user 9 moves the slide bar 48, the fluctuation value acquisition unit 16 acquires the index fluctuation value of "-0.25%." Accordingly, the index value correction unit 17 uses the acquired fluctuation value of "-0.25%" to correct the value of the index set as an explanatory variable in the trained prediction model 24. The prediction unit 19 predicts information about the value of a financial product using the trained prediction model 24 whose index value has been corrected by the index value correction unit 17. The prediction device 1 performs processing for each of the prediction times 43b and 43c desired by the user 9 in the same manner as for the prediction time 43a, and displays the prediction results illustrated in FIG. 13 on the display unit 33B of the terminal device 2. In FIG. 13 , seven mutual funds are ranked in order of their rate of change at the end of October, November, and December, which are the end of one month, two months, and three months from now, assuming that the U.S. federal funds rate has decreased by 0.25% from the current value.

[0070] In this way, the prediction device 1 according to one embodiment can easily calculate a prediction result regarding the value of a financial product when the user 9 assumes that a certain financial or economic indicator has fluctuated. The display unit 33B of the terminal device 2 used by the user 9 displays the prediction results corresponding to the multiple prediction time points desired by the user 9 together, for example, in a ranking format. This allows the user 9 to easily understand the prediction result regarding the value of the financial product.

[0071] As described above, the prediction device 1 according to one embodiment can provide a technology for predicting the value of financial products that can be easily used even by people who lack specialized knowledge about financial products.

[0072] As will be explained with reference to Figures 2 to 4, the learning data used by the prediction device 1 for machine learning has a predetermined time (preferably two months) or more between the end of the data period of the financial index data 21 and economic index data 22 and the time of the value fluctuation data 23. This ensures a certain amount of time or more between the end of the data period of the learning data and the time at which the user 9 desires a prediction, making it possible to prevent the prediction model from being excessively influenced by the most recent financial index data and improving the prediction accuracy of information regarding the value of financial products predicted by the prediction device 1. With improved prediction accuracy, even if the user 9 looks at the prediction results every day, for example, daily changes in the prediction results are prevented, reducing confusion for users 9 who lack specialized knowledge about financial products.

[0073] The prediction unit 19 displays, for example, in a ranking format, the prediction results corresponding to the multiple prediction time points desired by the user 9. This allows the user 9 to easily understand the prediction results regarding the value of the financial product.

[0074] The index fluctuation suppression unit 12 performs data processing on the original series data to suppress irregular fluctuations in the index, as illustrated in Fig. 6. This allows the trends and volatility of the financial and economic indexes shown in Tables 1 to 3 to be more appropriately captured and reflected in the machine learning of the prediction model 24, thereby improving the prediction accuracy of the prediction model 24.

[0075] The contribution calculation unit 18 displays the index that contributes most to the prediction result as a noteworthy index. By displaying the index with the highest contribution and its characteristics, the user 9 can improve their interpretability of the prediction result.

[0076] The fluctuation value acquisition unit 16 and the index value correction unit 17 enable the user 9 to easily calculate a prediction result regarding the value of a financial product when the user 9 assumes that a certain financial index or economic index has fluctuated. [Other forms]

[0077] Although the present invention has been described above with reference to specific embodiments, the present invention is not limited to the above-described embodiments.

[0078] FIG. 14 shows another example of a screen display on a terminal device used by a user. In the above-described embodiment, as shown by reference numerals 44a and 44b in FIG. 9 , the prediction unit 19 displays the rate of change of the financial product (investment trust) selected by the user 9 and related information on the display unit 33B of the terminal device 2 as the prediction result by the prediction unit 19. However, the information displayed on the display unit 33B of the terminal device 2 for the financial product selected by the user 9 is not limited to this. For example, as shown in FIG. 14 , the prediction unit 19 may display information 49 indicating a correspondence with past predictions on the display unit 33B of the terminal device 2 on a screen displaying the actual results (past data) of past predictions by the prediction unit 19. The information 49 indicating a correspondence with past predictions is information resulting from a comparison between past prediction results (priority predictions) by the prediction unit 19 and actual price changes for a certain financial product. The information 49 can also be referred to as the accuracy or incorrectness of the price prediction by the prediction unit 19. As shown by reference numeral 49, the accuracy or incorrectness can be displayed using, for example, a graphic. Such a comparison can be performed by the prediction unit 19, for example, by the prediction device 1 updating and accumulating learning data daily. This allows the user 9 to easily understand how a certain financial product has actually moved in price in the past and what kind of prediction results the prediction unit 19 has output in response to that price movement.

[0079] In the above embodiment, the prediction device 1 is realized as an integrated device, but the prediction device 1 does not need to be an integrated device, and the CPU, memory, auxiliary storage device 20, etc. may be located in separate locations and connected to each other via a network.

[0080] In the above embodiment, the functional blocks 11 to 19 constituting the processing unit 10 are realized by software, but some or all of these functional blocks 11 to 19 may be realized as hardware. The processing of the functional blocks 11 to 19 constituting the processing unit 10 does not need to be performed by a single processor, but may be distributed and processed by multiple processors. Some or all of the functions of the processing unit 10 and the data items in the auxiliary storage device 20 may be cloud-based in another server device (not shown) connected via the communication I / F unit 31.

[0081] Examples of the present invention will be described below to clarify the features of the present invention. [Example]

[0082] In Example 1, the accuracy of prediction results using a trained prediction model was verified. As training data, time-series data of financial indicators and economic indicators covering a 15-year period from 2003 to 2017, as well as time-series data of investment trust price fluctuation prediction results (binary classification of "up" or "down" in this example), were prepared. The 83 types of financial indicators and economic indicators shown in Tables 1 to 3 were used. Eight types of data processing shown in Table 4 were applied to the original series data of each of these 83 indicators, resulting in time-series data for a total of 664 indicators. The seven types of investment trusts targeted for training and prediction were those shown in Table 5. Using the training data prepared in this way, machine learning of a prediction model using gradient boosting decision trees was performed.

[0083] Next, at the end of each month from January 2018 to April 2023, the trained prediction model was used to predict the rise or fall (increase or fall) of each of the seven mutual funds for the end of the month one month later, the end of the month two months later, and the end of the month three months later from the end of that month. The predicted rise or fall results of the trained prediction model were then compared with the actual rise or fall results of the mutual funds, and the percentage of correct predictions was calculated. The accuracy rate was calculated as a correct answer when the predicted rise or fall results matched the actual results, and an incorrect answer when they did not. The results are shown in Table 5. As shown in Table 5, a rise / fall prediction accuracy of over 50% was achieved for all seven types of mutual funds used in the predictions. [Table 5] [Explanation of symbols]

[0084] 1. Device (prediction device, prediction model generation device) 2. Terminal Device 3 Network 9 User 10 Processing section 11 Preparation Department 12 Index fluctuation suppression unit 13 Learning Department 14 Financial Products Identification Department (Investment Trust Identification Department) 15. Prediction time acquisition section 16 Fluctuation value acquisition unit 17 Index value correction section 18 Contribution calculation section 19 Prediction Department 20 Auxiliary storage 21 Financial Indicator Data 22 Economic indicator data 23 Value Fluctuation Data 24 Predictive Models 31 Communication interface section (communication I / F section) 32B input section 33B Display section 100 Prediction System P A Prediction Model Generator PB Prediction Program

Claims

1. a preparation step of preparing learning data including time-series value fluctuation data relating to fluctuations in the value of financial instruments, time-series financial index data relating to fluctuations in financial indexes, and time-series economic index data relating to fluctuations in economic indexes; a learning step of performing machine learning using the learning data to generate a trained prediction model that predicts information about the value of the financial product at a desired time; Including, A prediction model generation method, wherein the period between the end of a data period for the financial index data and the economic index data and the time of the value fluctuation data is a predetermined time or longer.

2. 2. The prediction model generation method according to claim 1, wherein the learning step sets each of the indicators included in the financial indicator data and the economic indicator data as explanatory variables and sets the value fluctuation data as a target variable to learn the prediction model.

3. further comprising an index fluctuation suppression step of performing data processing to suppress irregular fluctuations in the indexes included in the financial index data and the economic index data; The prediction model generating method according to claim 1 , wherein the learning step performs machine learning using the learning data including the financial index data and the economic index data in which irregular fluctuations in the indexes have been suppressed.

4. A method for predicting information about the value of a financial instrument, comprising: a prediction time acquisition step of acquiring a desired prediction time; a prediction step of predicting information about the value at the acquired time point by inputting financial index data and economic index data whose end point of a data period is at least the predetermined time before the acquired time point into a prediction model generated by the prediction model generation method according to claim 1; A prediction method, including:

5. 5. The prediction method according to claim 4, further comprising a contribution calculation step of calculating the contribution of each indicator included in the financial indicator data and the economic indicator data, which are set as explanatory variables in the trained prediction model, when the trained prediction model predicts information about the value.

6. a fluctuation value acquisition step of acquiring a fluctuation value of an index that is included in the financial index data or the economic index data and whose fluctuation is desired; an index value correction step of correcting the value of the index set as an explanatory variable in the trained prediction model using the acquired fluctuation value; further comprising The prediction method according to claim 4 , wherein the prediction step predicts the information about the value using the trained prediction model in which the value of the index has been corrected.

7. 5. The prediction method according to claim 4, wherein the prediction step predicts the information regarding the value for each of a plurality of different financial instruments, and ranks and displays each of the plurality of different financial instruments based on the predicted information regarding each of the values.

8. The prediction time point acquisition step acquires a plurality of desired prediction time points, 5. The prediction method according to claim 4, wherein the prediction step predicts information regarding the value of the target financial product at each of the desired multiple points in time by inputting each of the acquired multiple points in time into the trained prediction model.

9. a preparation means for preparing learning data including time-series value fluctuation data relating to fluctuations in the value of financial instruments, time-series financial index data relating to fluctuations in financial indexes, and time-series economic index data relating to fluctuations in economic indexes; a learning means for performing machine learning using the learning data to generate a trained prediction model that predicts information about the value of the financial product at a desired time; Equipped with A prediction model generation device, wherein the period between the end of a data period for the financial index data and the economic index data and the time of the value fluctuation data is equal to or longer than a predetermined time.

10. A prediction device for predicting information about the value of a financial product, a prediction time acquisition means for acquiring a desired prediction time; a prediction means for predicting information about the value at the acquired time point by inputting financial index data and economic index data whose end point of a data period is earlier than the acquired time point by the predetermined time or more into a prediction model generated by the prediction model generation device according to claim 9; A prediction device comprising:

11. On the computer, A prediction model generation program for executing each step of the prediction model generation method according to any one of claims 1 to 3.

12. On the computer, A prediction program for executing each step of the prediction method according to any one of claims 4 to 8.

Images