WO2019017032A1

WO2019017032A1 - Computer system, method and program for accumulating asset having value which fluctuates over time

Info

Publication number: WO2019017032A1
Application number: PCT/JP2018/016279
Authority: WO
Inventors: 一弥榊原
Original assignee: 株式会社じぶん銀行
Priority date: 2017-07-20
Filing date: 2018-04-20
Publication date: 2019-01-24
Also published as: JP6381844B1; JP2019021320A; SG11202000540UA; JP2019021298A

Abstract

Provided are a computer system, a method and a program which make it possible to predict a particular value, such as the highest value or lowest value (i.e. maximum value or minimum value) during a prescribed period, of the value of an asset for which the value fluctuates irregularly over time. A computer system 200 for accumulating foreign currency that has a value which fluctuates over time is configured to: predict a time, within a prescribed period, when the value of the foreign currency will be the lowest (Step S501); and purchase the foreign currency at the predicted time when the value of the foreign currency is to be the lowest.

Description

Computer system, method and program for accumulating assets whose value changes as time passes

The present invention relates to computer systems, methods, and programs for accumulating assets that change in value over time.

Periodically accumulated foreign exchange security trading system that periodically purchases various foreign currency denominated foreign currencies (hereinafter referred to as "foreign currency"), accumulates the purchased foreign currency, and transmits the profit and loss evaluation information of the accumulated foreign currency to the terminal Are known (see, for example, Patent Document 1).

JP, 2004-192587, A

However, in the conventional system, when foreign currency is regularly purchased, it is not analyzed whether the value of the foreign currency to be purchased is the optimal price at the time of purchase or whether or not the purchase time is optimal. There is a problem that the purchaser of the is not optimized for purchasing.

The present invention has been made in view of such problems, and the highest or lowest value (that is, the maximum value or the minimum value) of the value of an asset whose value fluctuates with time over a predetermined period To provide a computer system, method, and program that make it possible to predict when to be a specific value, such as a specific ranking value (eg, second lowest value, tenth highest value etc) To aim.

The present invention provides a computer system for accumulating foreign currency whose value changes as time passes, and the computer system predicts when the value of the foreign currency will be the lowest price in a specific rank within a predetermined period of time And purchasing the foreign currency at the predicted time.

In one embodiment, the predicting includes dividing the predetermined period into a plurality of periods, and when the value of the foreign currency becomes a low price of a specific rank among the plurality of periods after the division. And determining a period of time to which it belongs.

In one embodiment, the predicting further comprises: (1) further dividing into a plurality of periods the period in which it is determined that the time when the value of the foreign currency becomes a low in a specific rank belongs to; Among the multiple periods after division, determining the period to which the time when the value of the foreign currency becomes a specific low price belongs, and when the value when the value of the foreign currency becomes a low low in a specific order And (d) repeating the determined period until the value of the foreign currency is as long as the value of the specific low price.

In one embodiment, the determining includes determining a period in which the average value of the value of the foreign currency is low among the plurality of divided periods.

In one embodiment, the determining determines a period to which a time when the value of the foreign currency becomes a low price of a specific rank among a plurality of divided periods using each of a plurality of prediction models. And, of the plurality of periods after the division, a period determined to be a period to which a time when the value of the foreign currency becomes a low in a specific rank belongs to the foreign currency Determining that it is the period to which the value falls in a particular low position.

In one embodiment, the determining determines a period to which a time when the value of the foreign currency becomes a low price of a specific rank among a plurality of divided periods using each of a plurality of prediction models. And for each of the first half of the plurality of divided periods, the number of prediction models determined to be the period to which the time when the value of the foreign currency is a specific low price belongs. And determining that a period in which the number of the prediction models is equal to or more than a predetermined number is a period in which the time when the value of the foreign currency is a specific low price belongs.

In one embodiment, the determining further includes changing the predetermined number according to a relative position of the divided period with respect to the entire predetermined period.

In one embodiment, the plurality of time periods is two time periods.

In one embodiment, the predicting includes suppressing predicting a period within a predetermined suppression period from a start point of the predetermined period as a period when the value of the foreign currency becomes a specific rank low price. .

In one embodiment, the predicting and the purchasing are repeated.

In one embodiment, the purchasing is performed automatically without a user's input operation for purchasing the foreign currency at a time when the predicted value of the foreign currency becomes a low in a specific rank. .

In one embodiment, the specific low price is the lowest price.

In one embodiment, the low price of the specific rank is the nth lowest value (n is an integer of 2 or more).

The present invention provides a method for accumulating foreign currency whose value changes as time passes, the method is executed in a computer system, and the method determines the value of the foreign currency within a predetermined time period. The method may include predicting when the ranking will be low and purchasing the foreign currency at the predicted time.

The present invention provides a program for accumulating foreign currency whose value changes as time passes, and when the program is executed in a computer system, the value of the foreign currency is ranked in a specific order within a predetermined period. Allowing the computer system to perform processing including predicting when the price will be low and purchasing the foreign currency at the predicted time.

The present invention is configured to predict when the value of the foreign currency will be a specific rank value within a predetermined period of time based on past value fluctuations of the foreign currency whose value changes as time passes. Provide a computer system.

In one embodiment, the computer system of the present invention is further configured to issue the foreign currency transaction request at the predicted time.

The present invention provides a program to be executed in a computer system, said program, when executed, said foreign country within a predetermined period of time based on past value fluctuations of foreign currency whose value fluctuates as time passes. The computer system is caused to perform processing including predicting when the value of the currency will be a particular rank value.

In one embodiment, the processing further comprises issuing the foreign currency transaction request at the predicted time.

In accordance with the present invention, the value of an asset whose value fluctuates with time over time is the highest or lowest (ie maximum or minimum) or value of a particular rank (eg 2) Computer systems, methods, and programs can be provided that allow one to predict particular values, such as the second lowest value, the tenth highest value, etc.).

Furthermore, according to the present invention, it is possible to increase the evaluation value of the asset by periodically or continuously purchasing or accumulating the asset based on the prediction of the time when the value becomes a specific value. Methods and programs can be provided.

The figure which shows an example of the data flow between the user, a bank, and a foreign exchange market in AI foreign currency reserve deposit. A figure showing an example of composition of computer system 200 for realizing a new financial product. FIG. 8 is a view showing an example of the data configuration of information on accumulation of value fluctuation assets stored in the database unit 220. The figure which shows an example of the multilayer perceptron 450 which specific value prediction AI utilizes. 5 is a flowchart illustrating an example of a process 500 in the computer system 200 for realizing a new financial product. The flowchart which shows one example of the processing which predicts the time when value of the value fluctuation asset becomes the lowest in the specified value prediction AI within the specified period with step S501. 5 is a flowchart illustrating an example of processing in a case where the processing of step S602 is performed using a plurality of prediction models. Processing in the case of performing the process of step S602 using a plurality of prediction models, and limiting the prediction of the first half of a predetermined period as the time when the value of the value fluctuation asset becomes the lowest value in the predetermined period The flowchart which shows an example. The graph which shows an example of the time change of a US dollar price (yen). The graph which shows an example of the time change of a US dollar price (yen). The graph which shows an example of the time change of a US dollar price (yen). The graph which shows an example of the time change of a US dollar price (yen). The graph which shows an example of the time change of an Australian dollar price (yen). The graph which shows an example of the time change of an Australian dollar price (yen). The graph which shows an example of the time change of an Australian dollar price (yen). The graph which shows an example of the time change of an Australian dollar price (yen). FIG. 7 shows an example of a computer system 200 'that is an alternative configuration of computer system 200 for implementing a new financial product.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

1. AI Foreign Currency Deposits The inventor of the present invention has developed a new financial instrument called "AI foreign currency deposits". AI foreign currency reserve deposits use artificial intelligence (AI) to predict the day when the target foreign currency will be the lowest in a given period, and the day when the target foreign currency is predicted to be the lowest. It is an accumulated deposit that accumulates the target foreign currency every predetermined period by purchasing. According to AI foreign currency reserve, it is possible to purchase foreign currency at a favorable exchange rate than conventional foreign currency reserve. As a result, it is possible to further reduce the risk due to exchange rate fluctuations. Here, AI is an abbreviation of "Artificial Intelligence", a technology for artificially implementing human intelligence (or intelligence that surpasses human intelligence) on a computer system, or such technology Refers to a computer system implemented.

FIG. 1 shows an example of data flow between a user, a bank and a foreign exchange market in an AI foreign currency funded deposit. In the example shown in FIG. 1, it is assumed that the target foreign currency is accumulated monthly from April. Each step shown in FIG. 1 will be described below.

Step S101: The user 10 applies to the bank 20 for an AI foreign currency reserve deposit. This starts the data flow of AI foreign currency reserve deposits. The application of the AI foreign currency savings deposit is made, for example, by the user 10 accessing the bank system of the bank 20 using a user device (for example, a smart phone) and operating the user device according to a predetermined application procedure. Alternatively, such an application may be made as the user 10 submits an application for an AI foreign currency reserve deposit to the window of the bank 20. In the application, the user 10 sets, for example, a reserve currency, a reserve amount, a target amount, and the like. For example, in the application, the user 10 may set to accumulate the AUD for 10,000 yen and 1,000,000 yen. At this time, the user 10 can not set the timing of funding (for example, on what day every month the AUD is to be purchased).

The bank 20 receives an application for an AI foreign currency-funded deposit and processes for the application. When the procedure by the bank 20 is completed, the bank system of the bank 20 executes processing for AI foreign currency reserve deposits.

The banking system of the bank 20 predicts, for example, the first day of April when the target foreign currency will be lowest in April. Such prediction is performed using AI (Artificial Intelligence). For example, suppose that April 14 is predicted to be the day when the target foreign currency becomes the lowest price.

Step S102: On April 14, the bank 20 transmits a purchase request to the foreign exchange market 30 to purchase the target foreign currency using the Japanese yen of the amount set by the user 10.

Step S103: In response to the purchase request from the bank 20, the foreign exchange market 30 sells the target foreign currency to the bank 20 at the exchange rate of April 14. The bank 20 accumulates the target foreign currency purchased for April from the foreign exchange market 30 in the account of the user 10.

Step S104: The bank 20 reports to the user 10 that the foreign currency to be purchased for April is accumulated in the account of the user 10.

The banking system of the bank 20 predicts, for example, the first day of May when the target foreign currency will be lowest in May. Such prediction is performed using AI (Artificial Intelligence). For example, suppose that it is predicted that May 2nd will be the day when the target foreign currency becomes the lowest price.

Step S105: The bank 20 transmits a purchase request to the foreign exchange market 30 on the 2nd of May using the Japanese yen of the amount set by the user 10 to purchase the target foreign currency.

Step S106: In response to the purchase request from the bank 20, the foreign exchange market 30 sells the target foreign currency to the bank 20 at the exchange rate of May 2. The bank 20 accumulates the target foreign currency purchased for May from the foreign exchange market 30 in the account of the user 10.

Step S107: The bank 20 reports to the user 10 that the target foreign currency purchased for May has been accumulated in the account of the user 10.

The banking system of the bank 20 predicts, for example, the first day of June when the target foreign currency becomes the lowest in June. Such prediction is performed using AI (Artificial Intelligence). For example, it is assumed that June 28 is predicted to be the day when the target foreign currency becomes the lowest price.

Step S108: The bank 20 transmits a purchase request to the foreign exchange market 30 on the 28th of June using the Japanese yen of the amount set by the user 10 to purchase the target foreign currency.

Step S109: In response to the purchase request from the bank 20, the foreign exchange market 30 sells the target foreign currency to the bank 20 at the exchange rate of June 28. The bank 20 accumulates the target foreign currency purchased for June from the foreign exchange market 30 in the account of the user 10.

Step S110: The bank 20 reports to the user 10 that the target foreign currency purchased for June has been accumulated in the user 10 account.

Thus, in AI foreign currency reserve deposits, the day when the target foreign currency becomes the lowest in a predetermined period is predicted using AI (Artificial Intelligence), and the day when the target foreign currency is predicted to be the lowest. Target foreign currency is purchased. In this way, foreign currency is accumulated every predetermined period. Thus, according to the AI foreign currency reserve, it is possible to purchase and accumulate foreign currency at a favorable exchange rate than a conventional foreign currency reserve which purchases foreign currency on a fixed date every month. As a result, it is possible to further reduce the risk due to exchange rate fluctuations. This allows mark-to-market valuation of foreign currency assets to be increased at the end of any foreign currency purchase cycle.

Note that the forecast of the day when the target foreign currency will be the lowest price by AI is just a prediction, and it is not a guarantee that the lowest price will be reached on that day. However, by improving the prediction accuracy of the AI, it is possible to execute the accumulation at a more favorable exchange rate more reliably.

Although the embodiment described above with reference to FIG. 1 illustrates an example in which the prediction of the day when the target foreign currency is the lowest according to the AI is performed on the first day of the transaction month, the timing of the prediction is not limited thereto. . For example, the prediction may be made before reaching the trading period (e.g., the day before the trading month).

In the embodiment described above with reference to FIG. 1, AI is used to predict the day when the target foreign currency becomes the lowest price in a predetermined period, but the present invention is not limited thereto. Using AI, it becomes the fifth lowest value on a day other than the day when the target foreign currency becomes the lowest price within the predetermined period, for example, the day when the target foreign currency becomes the second lowest value within the predetermined period The day having the 10th lowest value, the day having the 14th lowest value, etc. may be predicted. This is a more effective method when accumulating foreign currency that has high randomness of value change and is difficult to predict. In the case of foreign currency that has high randomness of value change and is difficult to predict, by not aiming at the lowest price, it is possible to avoid the highest price without taking the lowest price, so the final expected value improves.

In the embodiment described above with reference to FIG. 1, the bank 20 reports to the user 10 every time the target foreign currency is accumulated, but the necessity and timing of the report to the user 10 are arbitrary. For example, the bank 20 may not actively report to the user 10, or may report to the user after a certain number of accumulations have been completed.

Although the embodiment described above with reference to FIG. 1 describes AI foreign currency-funded deposits as an example of the new financial product, the new financial product is not limited thereto. For example, trading is not limited to purchasing or funding. The transaction may be a sale. At this time, using AI (Artificial Intelligence), the day when the target foreign currency becomes the highest within a predetermined period is predicted, and the target foreign currency is sold on the day when the target foreign currency is predicted to become the highest. For example, the object of the transaction is not limited to foreign currency. The object of the transaction may be any asset whose value changes as time passes (hereinafter, also referred to as “value fluctuation asset”). Examples of variable value assets include, but are not limited to, foreign currency, mutual funds, gold, commodities (crude oil, corn etc) etc. In other words, new financial instruments use artificial intelligence (AI) to target value-changing assets within specified time periods (maximum value, minimum value, 14th lowest value, or 5th highest value, etc.) Forecasting the time (for example, seconds, minutes, hours, days, or weeks) to be a specific ranking value, and within the time when the target value fluctuation asset is predicted to be the lowest price or the specific ranking low A portfolio of value change assets by purchasing the target value change asset or selling the target value change asset within the time the target value change asset is expected to be at a high or a specific high. It is a commodity that makes it possible to carry out the formation of With this new financial instrument, it is possible to buy or sell value-changing assets at a price that is more advantageous than conventional funded financial instruments. As a result, it is possible to further reduce the risk of value change and increase the value of the evaluation.

2. Configuration of Computer System for Realizing a New Financial Product FIG. 2 shows an example of the configuration of a computer system 200 for realizing the above-mentioned new financial product.

The computer system 200 is, for example, a computer system used in the bank 20.

Computer system 200 is connected to asset trading computer system 300 via network 400. Here, the type of the network 400 does not matter. The network 400 may be, for example, the Internet or a LAN.

The asset trading computer system 300 is a computer system that includes a server device and the like that can execute processing for trading of value fluctuation assets. The asset trading computer system 300 is, for example, a computer system used in the foreign exchange market 30. The asset trading computer system 300 may be, for example, a computer system managed by a trading brokerage company or trading party such as a short-listed company participating in the foreign exchange market 30. The asset trading computer system 300 is configured to perform processing to sell value changing assets in response to a request to purchase value changing assets. The asset trading computer system 300 is also configured to perform processing to purchase value changing assets in response to a request to sell the value changing assets. The computer system 200 is able to trade value-varying assets by being connected to the asset trading computer system 300.

The computer system 200 includes a server device 210 and a database unit 220 connected to the server device 210. The server device 210 includes an interface unit 211, a processor unit 212, and a memory unit 213.

The interface unit 211 is configured to control communication via the network 400. The interface unit 211 can transmit information to the outside of the server device 210, and can receive information from the outside of the user device 210. For example, the interface unit 211 transmits a request for purchase of a value change asset to the asset transaction computer system 300 via the network 400. For example, the interface unit 211 transmits, to the asset trading computer system 300, a request for sale of value fluctuation assets via the network 400. For example, the interface unit 211 transmits information to the database unit 220 connected to the server device 210, and receives information from the database unit 220.

The processor unit 212 controls the overall operation of the server device 210. The processor unit 212 reads a program stored in the memory unit 213 and executes the program. This enables the server device 210 to function as a device that executes a desired step. For example, the processor unit 212 can perform part or all of the process of accumulating value fluctuation assets. The processor unit 212 may be implemented by a single processor or may be implemented by a plurality of processors.

The memory unit 213 stores a program for executing a process in the server apparatus 210, data required to execute the program, and the like. In the memory unit 213, for example, a program for trading value-varying assets or a program for accumulating value-varying assets (for example, a program for realizing the processing shown in FIGS. 5 and 6A to 6C described later) Part or all is stored. Here, it does not matter how the program is stored in the memory unit 213. For example, the program may be pre-installed in the memory unit 213. Alternatively, the program may be installed in the memory unit 213 by being downloaded via the network 400. The memory unit 213 can be implemented by any storage means.

The database unit 220 stores, for each user, information on the accumulation of value fluctuation assets. The information on the accumulation of the value change asset includes the information on the accumulation destination of the value change asset. For example, information on the value fluctuation asset accumulated in the value fluctuation asset accumulation destination (for example, purchase time of the value fluctuation asset, purchase price of the value fluctuation asset, purchase amount of the value fluctuation asset) , Accumulated amount of value fluctuation assets, etc.). Information on where to deposit value change assets is information on value change assets sold from where value change assets are stored (for example, sale time of value change assets, sale price of value change assets, sale amount of value change assets, etc.) May also be included. If the value change asset is a foreign currency, the information on the deposit destination of the value change asset is the information on the foreign currency reserve account, and information on the foreign currency set up in the foreign currency reserve account (for example, purchase date and time of foreign currency, purchase price of foreign currency This may include foreign currency purchases, foreign currency reserve totals, etc.) and information on foreign currencies sold from foreign currency funding accounts (eg, foreign currency sales date, foreign currency sales prices, foreign currency sales volumes, etc.).

The database unit 220 may store, for example, past value fluctuation data. The past value change data can be used for learning of a specific value prediction AI described later to predict future value change. The historical value change data used for learning is the value change data of the target asset with respect to the currency for purchase (for example, Japanese yen) (for example, when purchasing and accumulating the US dollar in Japanese yen, the value change of the US dollar relative to Japanese yen Data) can be included. The historical value change data used for learning includes the value change data of the target asset relative to the currency for purchase as well as the change in value of the target asset relative to other currencies (eg, US dollar, Australian dollar, euro, NZ dollar, etc.) Data or data such as S & P 500 index, Nikkei Stock Average, German stock index, or WTI crude oil price may be included. The database unit 220 can also store, for example, learned value fluctuation tendencies. When a specific value prediction AI to be described later uses a multilayer perceptron, the database unit 220 may store a weight vector of each node established by learning.

FIG. 3 shows an example of a data configuration of information on accumulation of value fluctuation assets stored in the database unit 220. As shown in FIG.

In the example shown in FIG. 3, when the value change asset is a foreign currency, an example of the data configuration of the information stored in the database unit 220 is shown.

The database unit 220 stores, for each user, information on a foreign currency funding account. The database unit 220 includes a database 221 for storing information on a first user's foreign currency accumulation account and a database 222 for storing information on a second user's foreign currency accumulation account. The database 221 for storing the information of the first user's foreign currency funding account is, for example, the foreign currency funding account number of the first user who started the dollar-denominated financial deposit from April 2017, the foreign currency purchase date, Includes foreign currency purchase price and foreign currency reserve capital. For example, the database 221 stores information that the first user purchased $ 113 on April 14, 2017, and the accumulated total amount became $ 113, and the first user is stored on May 2, 2017. Stores $ 119 and stores information that the accumulated amount has reached $ 232, and stores information that the first user purchased $ 125 on June 28, 2017 and the accumulated amount has reached $ 357 Then, the first user may purchase $ 110 on July 10, 2017, and store information that the accumulated total amount is $ 467. The database 222 for storing the information of the second user's foreign currency funding account is, for example, the foreign currency funding account number of the second user who started the euro-denominated financial deposit from July 2016, the date and time of purchase of the foreign currency, Includes foreign currency purchase price and foreign currency reserve capital. For example, the database 221 may store information that the second user purchased 520 euros on July 12, 2016, and the accumulated total amount is 520 euros.

Within a given time period (eg, seconds) when the value of the value-changing asset is at a specific value (for example, a specific ranking value such as the lowest price, the highest price, the 14th lowest value, or the 5th highest value) An AI (hereinafter, also referred to as “specific value prediction AI”) capable of predicting hour, minute, hour, day, or week) is realized by the computer system 200 having the configuration described above. The specific value prediction AI uses the multilayer perceptron 450 as a prediction model.

FIG. 4 shows an example of a multilayer perceptron 450 used by the specific value prediction AI.

The multilayer perceptron 450 has an input layer, a hidden layer, and an output layer. In the example shown in FIG. 4, the multilayer perceptron 450 is shown to have two hidden layers, but the number of hidden layers is not limited thereto. The multilayer perceptron 450 can comprise one or more hidden layers. Also, each layer of the input layer, hidden layer, and output layer of multilayer perceptron 450 can include any number of nodes. For example, each layer may include nodes, and each hidden layer may include the same number of nodes so that the nodes decrease as going from the input layer to the output layer.

The specific value prediction AI can predict future changes in the value of value change assets by learning past value change data using the multi-layer perceptron 450. The past value change data to be learned includes, for example, value change data of the asset to be predicted relative to the currency for purchase (for example, Japanese yen). Additionally, historical value change data to be learned may include value change data of the asset being forecasted for a currency other than the currency for purchase (e.g., USD, AUD, EUR, NZD, etc.), or , Asset price data other than the asset to be forecasted, such as the S & P 500 index, Nikkei Stock Average, the German stock index, or WTI crude oil prices. The specific value prediction AI can establish the weight vector of each node of the multilayer perceptron 450 by learning past value variation data as training data. For example, the specific value prediction AI performs learning using data of a fixed period of at least one kind of value fluctuation data as teacher input data, and value fluctuation data of an asset to be predicted after the predetermined period as teacher output data. be able to.

The teacher output data may be, for example, an average value of a predetermined period of the asset to be predicted after a predetermined period. The multilayer perceptron 450 having a weight vector of each node established by learning with such teacher output data outputs prediction of an average value of a future predetermined period, when value fluctuation data of the immediately preceding period is input. Will be able to

The output data for teachers is, for example, whether the asset to be predicted has the lowest price in the first half of a certain period after a certain period (for example, January, 10 business days, 6 business days, 2 business days, etc.) It may be a binary value (0 or 1) indicating whether it is the lowest price. The multilayer perceptron 450 having a weight vector of each node established by learning based on such teacher output data may receive a certain period of time (for example, January, 10) when value fluctuation data of the immediately preceding period is input. It is possible to output a forecast as to whether the price will be the lowest in the first half of the business day, 6 business days, 2 business days, etc.).

The output data for teacher is, for example, a period in which an asset to be predicted is divided into a plurality of periods after a certain period (for example, January, 10 business days, 6 business days, 2 business days, etc.) It may be a value indicating which period of the period the lowest price will be. The multilayer perceptron 450 having a weight vector of each node established by learning based on such teacher output data may receive a certain period of time (for example, January, 10) when value fluctuation data of the immediately preceding period is input. It is possible to output a forecast of which period of the business day, 8 business days, 4 business days, etc.) is divided into a plurality of periods.

Specific value prediction AI, using multi-layer perceptron 450, from the past value fluctuation data, for example, "The month of the big uptrend is likely to be the lowest in the first half of the month, the second half of the month is likely to be the highest "The big downtrend month is likely to have the lowest price in the second half of the month, and the first half of the month is likely to have the highest price." , You can learn the value fluctuation trend based on the trend such as "If the trend is up, it is easier to get the lowest price on the weekend than at the end of the week". The specific value prediction AI is based on the amount of fluctuation such as “always falls after a day rising above a certain threshold” or “always rises after a day falling above a certain threshold” from past value fluctuation data, for example. Value change tendency can be learned. The learning result is stored in the database unit 220 together with past fluctuation data.

The specific value prediction AI utilizes the multi-layer perceptron 450, and based on the learning result, the value of the value fluctuation asset has a specific value (for example, the lowest price, the highest value, the 14th lowest value, or Predict a time (for example, seconds, minutes, hours, days, or weeks) to be a specific rank value such as the fifth highest value. The input to the multilayer perceptron 450 for processing of the prediction may be value change data of a period immediately prior to the predetermined period to be predicted. The input to the multi-layer perceptron 450 for processing the forecast may be value change data of the forecasted asset relative to the buying currency (e.g. Japanese yen), but a currency other than the buying currency (e.g. The asset price data of the target asset against the Euro, NZ Dollar, etc.) or asset price data other than the asset to be forecasted such as the S & P 500 Index, Nikkei Stock Average, German Stock Price Index, or WTI crude oil price It may be input in the process of prediction. The input to the multi-layer perceptron 450 may be, for example, value change data for 16 days immediately before a predetermined period to be predicted, value change data for immediately preceding 32 days, or value change data for immediately preceding 64 days. It should be noted that the more data that is input to the multilayer perceptron 450, the better the prediction accuracy, but the more computational complexity increases as the data increases. In addition, it should be noted that the prediction accuracy improves as the amount of past value fluctuation data learned increases, but also the calculation amount for learning increases as the data increases. Furthermore, as the number of types of historical value fluctuation data learned increases, the prediction accuracy improves more, but when unrelated data is learned for prediction, the specific value prediction AI becomes confused and the prediction accuracy decreases. It should be noted that there is a risk of

The multilayer perceptron 450 is an example of a prediction model, and the specific value prediction AI can predict when the value of the value fluctuation asset becomes a specific value using any other prediction model. For example, the specific value prediction AI can use CNN (Convolutional Neural Network), recurrent neural network (RNN) or the like as a prediction model to predict when the value of the value fluctuation asset becomes a specific value. In general, CNN has better prediction accuracy than multilayer perceptron 450.

The specific value prediction AI can use a plurality of prediction models to predict when the value of the value fluctuation asset becomes a specific rank value. The plurality of prediction models may be, for example, a plurality of prediction models that differ in the type or amount of past value fluctuation data used for learning. The plurality of prediction models may be, for example, a plurality of prediction models that differ in type or amount of value fluctuation data of an immediately preceding period input for prediction. The plurality of prediction models may be, for example, a plurality of prediction models having different types of neural networks to be used, such as a prediction model using the multilayer perceptron 450, a prediction model using a CNN, and a prediction model using an RNN. The specific value prediction AI improves prediction accuracy by using each of a plurality of prediction models, taking majority of the prediction results, and predicting when the value of the value fluctuation asset becomes a specific rank value. It is possible.

In the example shown in FIG. 2, each component of the server device 210 is provided in the server device 210, but the present invention is not limited to this. It is also possible that any one of the components of the server device 210 is provided outside the server device 210. For example, when each of the processor unit 212 and the memory unit 213 is configured by separate hardware components, each hardware component may be connected via an arbitrary network. At this time, the type of network does not matter. Each hardware component may be connected via, for example, a LAN, may be wirelessly connected, or may be wired.

In the example shown in FIG. 2, the database unit 220 is provided outside the server device 210, but the present invention is not limited to this. It is also possible to provide the database unit 220 inside the server apparatus 210. At this time, the database unit 220 may be implemented by the same storage unit as the storage unit that implements the memory unit 213, or may be implemented by a storage unit different from the storage unit that implements the memory unit 213. In any case, the database unit 220 is configured as a storage unit for the server device 210. The configuration of the database unit 220 is not limited to a specific hardware configuration. For example, the database unit 220 may be configured by a single hardware component or may be configured by a plurality of hardware components. For example, the database unit 220 may be configured as an external hard disk drive of the server device 210 or may be configured as storage on a cloud connected via the network 400.

3. Process in a Computer System for Realizing a New Financial Product FIG. 5 shows an example of a process 500 in a computer system 200 for realizing a new financial product. Here, a process 500 for accumulating assets whose value changes as time passes will be described as an example.

In step S501, the specific value prediction AI implemented by the computer system 200 predicts when the value of the value fluctuation asset will be the lowest value in a predetermined period. The specific value prediction AI predicts, for example, when the value of the value fluctuation asset will be the lowest value in a predetermined period, by the process described later with reference to FIG. Here, the predetermined period may be an interval for accumulating value fluctuation assets. For example, when accumulating value changing assets monthly, the predetermined period may be January (or 23 business days). For example, when accumulating value change assets weekly, the predetermined period may be one week (or five business days). For example, when accumulating value change assets daily, the predetermined period may be one day (or nine hours). The time when the value of the value fluctuation asset becomes the lowest value in a predetermined period may be, for example, any time unit such as seconds, minutes, hours, days, or weeks.

At the time when the value of the value fluctuation asset predicted in step S501 reaches the lowest price, in step S502, the processor unit 212 of the server apparatus 210 executes a process of purchasing the value fluctuation asset. Specifically, the processor unit 212 transmits a purchase request for value fluctuation asset to the asset transaction computer system 300 via the interface unit 211, and executes processing for purchasing a desired value fluctuation asset from the asset transaction computer system. . The processor unit 212 may store information of the purchased value fluctuation asset in the database unit 220.

By repeating steps S501 and S502, value fluctuation assets are accumulated for each predetermined period. The information on the accumulated value fluctuation asset is stored in the database unit 220.

It should be noted that after step S501, prior to step S502, the user inputs in order to confirm with the user about purchasing the value change asset when the value of the value change asset predicted in step S501 is the lowest price. An operation may be performed. However, step S502 is preferably performed automatically without input operation by the user. The need for the user to confirm each time a value change asset is purchased is annoying for the user and impairs convenience. Regarding performing step S 502 automatically, for example, the user's consent may be obtained at the time of application for value change asset accumulation by the user.

In the above-described example, in step S501, a time when the value of the value fluctuation asset becomes the lowest value in a predetermined period is predicted, but in step S501 'instead of step S501, the value of the value fluctuation asset in the predetermined period There is a specific rank low (for example, the second lowest value, the fifth lowest value, the 10th lowest value, the 14th lowest value, ie, the nth lowest value (n is an integer of 2 or more)) It may be possible to predict when it will be. This is a more effective approach when the value fluctuations of value fluctuation assets are highly random and difficult to predict. By not aiming for the lowest price, you can avoid the highest price without taking the lowest price, which will improve the final expected value. This makes it possible to achieve stable asset purchases.

The process in computer system 200 may be a process for selling an asset whose value changes as time passes. In this case, in step S501, the specific value prediction AI predicts when the value of the value fluctuation asset becomes the highest value within a predetermined period, and reaches the time when the value of the predicted value fluctuation asset becomes the highest value. In step S502, the processor unit 212 of the server apparatus 210 may execute a process of selling the value change asset. Even in this case, as in the case of step S501 'described above, the value of the value fluctuation asset within a predetermined period is high at a specific rank (for example, the second highest value, the fifth highest value, the tenth highest) The value, the 14th highest value, that is, the n-th highest value (n is an integer of 2 or more) may be predicted, which has high randomness of the value change of the value change asset, This is a more effective method when it is difficult to predict, and by not aiming for the highest price, the lowest price can be avoided even if the highest price is not taken, thus improving the final expected value. Can achieve stable asset sales.

That is, in the processing in the computer system 200, in step S501, the specific value prediction AI determines that the value of the value fluctuation asset has a specific value (highest value, lowest value, nth lowest value, or nth value) within a predetermined period. The server device 210 is predicted in step S 502 when it is predicted that the time to become a high value (n is an integer of 2 or more) and a time when the value of the predicted value fluctuation asset becomes a specific value. The processor unit 212 of may be a process of executing a process of trading (buying or selling) value change assets.

It may be possible to switch, for example, at predetermined intervals, to predict the time when the value of the specific ranking is reached instead of predicting the time when the lowest price / highest price is reached. For example, when it is predicted that the trend will be strong (value change is linear) in a certain month, it may be possible to predict when the lowest price / highest price will be obtained, or in another month. When there is no trend (the value fluctuation curves up and down), instead of predicting when it will be the lowest price / maximum value, it will be forecast time when it will be a specific ranking value. May be This prediction may be performed, for example, using any trend and / or volatility prediction techniques.

Instead of predicting when the lowest price / highest price will be reached, predicting when it will be a value of a specific rank may be switched for each target value change asset. For example, if you analyze historical value fluctuation data in a certain value fluctuation asset and it is judged that there is no trend in more than 9/12 months in a year, when targeting that value fluctuation asset Instead of predicting when the lowest price / highest price will be reached, it is also possible to predict when the value of a specific ranking will be reached.

When predicting the time to become a specific ranking value, the specific ranking may be set by the specific value prediction AI or may be set manually, in consideration of price fluctuations of value fluctuation assets. In addition, the specific order may be set individually for each of the target value change assets. For example, a back test may be used to verify the optimum order in which order the specific order should be.

FIG. 6A shows an example of a process of predicting in step S501 when the value of the value fluctuation asset is the lowest value in the predetermined value prediction AI within a predetermined period. The specific value prediction AI predicts when the value of the value fluctuation asset will be the lowest value in a predetermined period by binary search, as described below. In binary search, since an appropriate prediction model can be used according to the width of the search, prediction can be performed with high accuracy. For example, if the search width is long-term, a prediction model for predicting long-term fluctuation can be used, and as the search width becomes short, a prediction model for predicting short-term fluctuation can be used. In general, a model that predicts short-term fluctuations has better prediction accuracy than a prediction model that predicts long-term fluctuations.

In step S601, the specific value prediction AI divides a predetermined period into two periods. For example, if the predetermined period is February (46 business days), the specific value prediction AI may have 46 business days, the first period including the first business day to the 23rd business day, and the 24th business day to It may be divided into a second period including the 46th business day. For example, in the case where the predetermined period is January (23 business days), the specific value prediction AI includes 23 business days, the first period including the first business day to the twelfth business day, and the thirteenth business day to It may be divided into a second period including the 23rd business day. For example, if the predetermined period is one week (5 business days), the first period including the first business day to the third business day, and the second period including the fourth business day to the fifth business day Can be divided into Thus, the two periods after division may be the same length or different lengths. The two time periods after division may each include a time period of any length.

In step S602, the specific value prediction AI determines, of the two periods after the division in step S601, the period to which the time when the value of the value fluctuation asset becomes the lowest value belongs. The period in which the time when the value of the value fluctuation asset becomes the lowest value belongs can be determined, for example, using a prediction model that predicts the fluctuation of the prediction target period. In order to determine the period in which the time when the value of the value fluctuation asset becomes the lowest falls, the specific value prediction AI determines, for example, a period in which the average value of the value of the value fluctuation asset is low among two periods after division. Do. This is statistically based on the fact that the time when the value of the value change asset is the lowest is likely to be in the period when the average value of the value change asset is lower. At this time, the specific value prediction AI learns past value fluctuation data as teacher data so that prediction of an average value in a predetermined future period can be output.

In step S602, the specific value prediction AI determines a period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, out of the two periods after the division in step S601, using a plurality of prediction models. You may At this time, for example, the process shown in FIG. 6B may be performed.

FIG. 6B shows an example of processing in the case where the processing of step S602 is performed using a plurality of prediction models.

In step S602-1, the specific value prediction AI uses, for each of the plurality of prediction models, a period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, out of the two periods after the division in step S601. judge. For example, when the specific value prediction AI uses five prediction models, the first prediction model is used to determine when the value of the value fluctuation asset is the lowest value in the two periods after the division in step S601. The period to which it belongs is determined, and the period to which the time when the value of the value fluctuation asset becomes the lowest belongs is determined using the second prediction model, of the two periods after the division in step S601, ... Of the two periods after the division in step S601, the prediction model of No. 5 is used to determine the period to which the time when the value of the value fluctuation asset becomes the lowest value belongs. In the determination of the period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, for example, the past value fluctuation data is learned as teacher data so that the prediction of the average value of the predetermined period in the future can be output. Depending on the specific value prediction AI, it may be performed by determining a period in which the average value of the value of the value fluctuation asset is low among the two periods after the division. The determination of the period to which the time when the value of the value fluctuation asset is the lowest falls belongs, for example, so that it can output a prediction as to whether it will be the lowest in the first half of the future or the second half in the second half It may be performed by a specific value prediction AI that is learning past value fluctuation data as teacher data.

Next, in step S602-2, the specific value prediction AI is a period to which the time when the value of the value fluctuation asset becomes the lowest price by more prediction models belongs to the two periods after the division in step S601. The period determined to be determined as the period to which the value of the value fluctuation asset becomes the lowest value belongs. The prediction accuracy can be improved by using a plurality of prediction models to determine the period in which the time when the value of the value fluctuation asset is the lowest falls. For example, according to the first prediction model, it is determined that the first half of the two periods after the division in step S601 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. The prediction model determines that the first half of the two periods after the division in step S601 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, and the third prediction model It is determined that the second half of the two periods after the division in S601 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, and the fourth prediction model determines that the division in step S601 The first half of the two periods is determined to be the period to which the value of the value fluctuation asset becomes the lowest, and the fifth forecasting model If it is determined that the second half of the two periods after the division at 601 belongs to a period to which the value of the value fluctuation asset becomes the lowest value, the two periods after the division at step S601. The number of forecast models determined that the first half period is the period to which the time when the value of the value-changing asset is the lowest belongs is the second half period of the two periods after the division in step S601. Since there are more than the number of prediction models determined to be the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, the specific value prediction AI is a majority vote and among the two periods after the division in step S601. Determine the first half period as the period to which the value of the value fluctuation asset becomes the lowest.

For example, the number of forecast models judged that the period in which the value of the value fluctuation asset becomes the lowest belongs to the first half period, and the period in which the period when the value of the value fluctuation asset becomes the lowest belongs in the second period. If the number of prediction models determined to be present is the same, the predetermined period may be determined as the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. It is preferable to determine the period of as the period in which the value of the value fluctuation asset has the lowest value. As described later, this is because the potential risk is greater in the first half period.

Referring back to FIG. 6A again, in step S603, the specific value prediction AI determines that the value of the value fluctuation asset has the lowest value during the period when it is determined that the time when the value of the value fluctuation asset has the lowest value belongs in step S602. It is determined whether it is the same length as the time. If it is determined that the period in which it is determined in step S602 that the time when the value of the value fluctuation asset becomes the lowest is not the same length as the time when the value of the value fluctuation asset becomes the lowest (in the case of No) The process proceeds to step S604. When it is determined that the period in which it is determined in step S602 that the day when the value of the value fluctuation asset is the lowest is included is the same length as the time when the value of the value fluctuation asset is the lowest (in the case of Yes) , S606 and the process ends. This is because it is possible to determine that the period in which it is determined in step S602 that the time when the value of the value fluctuation asset becomes the lowest is the time when the value of the value fluctuation asset becomes the lowest.

In step S604, the specific value prediction AI further divides the period in which it is determined that the time when the value of the value fluctuation asset is the lowest falls into two more periods. For example, when it is determined that the period when the value of the value fluctuation asset becomes the lowest value belongs to 11 business days, the specific value prediction AI includes 11 business days and the first business day to the sixth business day. It may be divided into a first period and a second period including the seventh business day to the eleventh business day. For example, when it is determined that the period when the value of the value fluctuation asset becomes the lowest value belongs to 6 business days, the specific value prediction AI includes 6 business days, and includes the first business day to the third business day. It may be divided into a first period and a second period including a fourth business day to a sixth business day. Thus, the two periods after division may be the same length or different lengths. The two time periods after division may each include a time period of any length.

In step S605, the specific value prediction AI determines, of the two periods after the division in step S604, the period to which the time when the value of the value fluctuation asset becomes the lowest value belongs. The period in which the time when the value of the value fluctuation asset becomes the lowest value belongs can be determined, for example, using a prediction model that predicts the fluctuation of the prediction target period. In order to determine the period in which the time when the value of the value fluctuation asset becomes the lowest falls, the specific value prediction AI determines, for example, a period in which the average value of the value of the value fluctuation asset is low among two periods after division. Do. This is statistically based on the fact that the time when the value of the value change asset is the lowest is likely to be in the period when the average value of the value change asset is lower. At this time, the specific value prediction AI learns past value fluctuation data as teacher data so that prediction of an average value in a predetermined future period can be output.

In step S605, the specific value prediction AI determines a period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, out of the two periods after the division in step S604, using a plurality of prediction models. You may At this time, for example, processing similar to the processing shown in FIG. 6B may be performed.

In step S605-1, the specific value prediction AI uses, for each of a plurality of prediction models, a period to which the time when the value of the value fluctuation asset becomes the lowest value belongs, out of the two periods after the division in step S604. judge. For example, when the specific value prediction AI uses five prediction models, the first prediction model is used to determine when the value of the value fluctuation asset is the lowest value in the two periods after the division in step S604. The period to which it belongs is determined, and the period to which the time when the value of the value fluctuation asset becomes the lowest belongs is determined using the second prediction model, of the two periods after the division in step S604,. By using the prediction model of 5, it is determined, of the two periods after the division in step S604, the period in which the time when the value of the value fluctuation asset is the lowest value belongs. In the determination of the period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, for example, the past value fluctuation data is learned as teacher data so that the prediction of the average value of the predetermined period in the future can be output. Depending on the specific value prediction AI, it may be performed by determining a period in which the average value of the value of the value fluctuation asset is low among the two periods after the division. The determination of the period to which the time when the value of the value fluctuation asset is the lowest falls belongs, for example, so that it can output a prediction as to whether it will be the lowest in the first half of the future or the second half in the second half It may be performed by a specific value prediction AI that is learning past value fluctuation data as teacher data.

Next, in step S605-2, the specific value prediction AI is a period to which, among the two periods after the division in step S604, the time at which the value of the value fluctuation asset becomes the lowest according to more prediction models belongs. The period determined to be determined as the period to which the value of the value fluctuation asset becomes the lowest value belongs. For example, according to the first prediction model, it is determined that the second half of the two periods after the division in step S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. The prediction model determines that the first half of the two periods after the division in step S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, and the third prediction model It is determined that the second half of the two periods after the division in S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, and the fourth prediction model determines that the division in step S604 The first half of the two periods is determined to be the period to which the value of the value fluctuation asset becomes the lowest, and the fifth forecasting model If it is determined that the second half of the two periods after the division at 604 belongs to the period in which the value of the value-changing asset becomes the lowest value, the two periods after the division at step S 604 The number of prediction models determined that the latter half of the period is the period to which the time when the value of the value-changing asset is the lowest belongs is the first half of the two periods after the division in step S604. Since the value fluctuation asset value is greater than the number of prediction models determined to be the period to which the low price period belongs, the specific value prediction AI is a majority vote among two periods after the division in step S604. The second half period is determined as the period to which the value of the value fluctuation asset becomes the lowest.

After step S605, the process proceeds to step S603 again. In step S603, the specific value prediction AI has the same length as the period in which the value fluctuation asset has the lowest value when it is determined that the time when the value fluctuation asset has the lowest value belongs in step S605. It is determined whether or not. When it is determined that the period in which it is determined in step S605 that the time when the value of the value fluctuation asset becomes the lowest is not the same length as the time when the value of the value fluctuation asset becomes the lowest (in the case of No) Proceed to step S604 again. Then, until it is determined that the period in which it is determined in step S605 that the time when the value of the value fluctuation asset becomes the lowest belongs is the same length as the time when the value of the value fluctuation asset becomes the lowest. Steps S605 are repeated. When it is determined that the period in which it is determined in step S605 that the time when the value of the value fluctuation asset is the lowest belongs is the same length as the time when the value of the value fluctuation asset is the lowest (in the case of Yes) , S606 and the process ends. This is because it is possible to determine that the period in which it is determined that the time when the value of the value fluctuation asset becomes the lowest is the time when the value of the value fluctuation asset becomes the lowest.

The time predicted by the specific value prediction AI in step S501 (steps S601 to S605) is the first half of the predetermined period, in particular, the beginning of the predetermined period (for example, the beginning of the month), and the prediction is made in step S502. If the process of purchasing value change assets is executed at the same time, the risk of value change during the remaining period of a predetermined period will be incurred. That is, if you purchase value change assets in the first half of a given period, especially at the beginning of the given period (for example, the beginning of the month), the uncertainty of the value change in the remaining period of the given period is large, There is a possibility that the return is not appropriate. This is particularly noticeable for assets that are highly random in value change. For example, if it is predicted that the price will be the lowest at the beginning of the month and you purchase value change assets at the beginning of the month, if the trend of value change assets changes in the middle of the month, the value of value change assets purchased with the lowest price forecasted However, the result may be the highest price. In this case, the prediction result by the specific value prediction AI becomes unstable, and eventually, the accumulation result of the value fluctuation asset also becomes unstable.

In order to avoid the case where such a risk and return are not appropriate, the specific value prediction AI in step S501 is in the first half of the predetermined period, in particular, at the beginning of the predetermined period, within the predetermined period. When forecasting when the value of a value fluctuation asset will be the lowest price, it is preferable to limit it to cases where the confidence of the forecasting model is high. If the degree of confidence in the forecasting model is high, it can be said that the risk that the trend of the value fluctuation asset changes in the middle of the month and the maximum value is eventually grasped is extremely low. As a result, even if the asset has a high degree of randomness in value change, it is possible to avoid a case where the risk and the return do not match, and to obtain a stable forecast result.

In the case where the degree of confidence of the prediction model is high, for example, the predetermined number or more of the plurality of prediction models may perform the same prediction. This is because when more prediction models make the same prediction, the accuracy is high. For example, in step S602, when the first half of the predetermined period is predicted as the time when the value of the value-changing asset becomes the lowest value within the predetermined period, the case is limited to a case where the confidence by the prediction model is high. The process shown in FIG.

FIG. 6C performs the process of step S602 using a plurality of prediction models, and limits the time when the first half of the predetermined period is predicted as the time when the value of the value fluctuation asset becomes the lowest value in the predetermined period. An example of processing in the case is shown.

Step S602-1 is the same process as step S602-1 shown in FIG. 6B. The specific value prediction AI determines, using each of the plurality of prediction models, a period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, out of the two periods after the division in step S601.

In step S602-2 ′, it is determined that the specific value prediction AI is a period in which the first half of the two periods after the division in step S601 is a period to which the time when the value of the value fluctuation asset becomes the lowest price belongs. Determine the number of models. For example, when each of the five prediction models is used, the first prediction model allows the first half of the two periods after the division in step S601 to have a time when the value of the value fluctuation asset becomes the lowest value. It is determined that the period is the period, and it is determined by the second prediction model that the first half of the two periods after the division in step S601 is the period to which the time when the value of the value fluctuation asset becomes the lowest price belongs. It is determined by the third prediction model that the second half of the two periods after the division in step S601 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. According to the prediction model, it is determined that the first half of the two periods after the division in step S601 is the period to which the time when the value of the value fluctuation asset is the lowest belongs. If it is determined by the fifth prediction model that the second half of the two periods after the division in step S601 belongs to the period in which the value of the value fluctuation asset has the lowest value, The number of prediction models determined to be the period to which the first half period of the two periods after the division in S601 belongs is the period to which the value of the value fluctuation asset becomes the lowest price is determined to be three.

Next, in step S602-3, the specific value prediction AI determines whether the number of prediction models determined in step S602-2 'is equal to or greater than a predetermined number. If the number of prediction models is equal to or greater than the predetermined number, the process proceeds to step S602-4. If the number of prediction models is less than the predetermined number, the process proceeds to step S602-5. Here, the predetermined number may be any number larger than half of the total number of prediction models used. The predetermined number may be, for example, the total number of prediction models used, or 80% of the total number.

In step S602-4, the specific value prediction AI determines that the first half of the two periods after the division in step 601 is a period to which the time when the value of the value-changing asset is the lowest value belongs. This is because the degree of confidence of multiple prediction models is high and the risk of selecting the first half period is low.

In step S602-5, the specific value prediction AI determines that the second half of the two periods after the division in step 601 is a period to which the time when the value of the value-changing asset is the lowest value belongs. This is because the degree of confidence of multiple prediction models is low, and it can be said that the risk of selecting the first half period is high.

In step S605, the first half of the predetermined period is also predicted as the time when the value of the value-changing asset becomes the lowest value in the predetermined period, as shown in FIG. The same process as the process to be performed may be performed.

Step S605-1 is the same process as step S605-1 described above. The specific value prediction AI uses each of the plurality of prediction models to determine the period to which the time when the value of the value fluctuation asset becomes the lowest price belongs, out of the two periods after the division in step S604.

In step S605-2 ′, it is determined that the specific value prediction AI is a period in which the first half of the two periods after the division in step S604 is a period to which the time when the value of the value-changing asset is the lowest value belongs. Determine the number of models. For example, when each of the five prediction models is used, the first prediction model allows the first half of the two periods after the division in step S604 to have a time when the value of the value fluctuation asset becomes the lowest value. It is determined that the period is the period, and it is determined by the second prediction model that the first half of the two periods after the division in step S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest price belongs. It is determined by the third prediction model that the second half of the two periods after the division in step S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. According to the prediction model, it is determined that the first half of the two periods after the division in step S604 is the period to which the time when the value of the value fluctuation asset becomes the lowest belongs. If it is determined by the fifth prediction model that the first half of the two periods after the division in step S604 is a period to which the time when the value of the value fluctuation asset is the lowest belongs. It is determined that the number of prediction models determined to be the period to which the first half period of the two periods after the division in S604 belongs is the period in which the value of the value-changing asset is the lowest value belongs.

Next, in step S605-3, the specific value prediction AI determines whether the number of prediction models determined in step S605-2 'is equal to or greater than a predetermined number. If the number of prediction models is equal to or greater than the predetermined number, the process proceeds to step S605-4. If the number of prediction models is less than the predetermined number, the process proceeds to step S605-5. Here, the predetermined number may be any number larger than half of the total number of prediction models used. The predetermined number may be, for example, the total number of prediction models used, or 80% of the total number.

The predetermined number may be changed, for example, according to the position of the divided period with respect to the entire predetermined period. For example, two periods after the division in step S604 may be any one of the first half of the predetermined period (for example, the predetermined period is 23 working days, and the two periods after the division in step S604 may be In the case of the first period including one business day to the sixth business day and the second period including the seventh business day to the eleventh business day), the predetermined number may be high. In particular, as the period after division is closer to the start point of the predetermined period, it is preferable to increase the predetermined number. This is because the closer the divided period is to the start of the predetermined period, the greater the risk to be incurred. For example, if the predetermined period is 23 business days, and the two periods after the division in step S604 are the first period including the first business day and the second period including the second business day, Preferably, the predetermined number is a maximum. For example, when the period after division is somewhat away from the start point of the predetermined period (for example, the predetermined period is 23 business days, and the two periods after the division in step S604 are the third business day ... Even in the first period including the fourth business day and the second period including the fifth business day to the sixth business day), the predetermined number may be maximized in consideration of the risk. For example, two periods after the division in step S604 are any one of the second half of the predetermined period (for example, the predetermined period is 23 business days, and the two periods after the division in step S604 are the first In the case of the first period including 13 business days to 15 business days and the second period including 16th business day to 18th business days, the predetermined number may be low. If the predetermined number is the minimum value, the processing is equivalent to that of step S602-2.

The predetermined number may be set by the specific value prediction AI, or may be set manually, in consideration of price fluctuations of the value fluctuation asset. Also, the predetermined number may be set individually for each of the target value fluctuation assets.

In step S605-4, the specific value prediction AI determines that the first half of the two periods after the division in step 604 is a period to which the time when the value of the value-changing asset is the lowest value belongs. This is because the degree of confidence of multiple prediction models is high and the risk of selecting the first half period is low.

In step S605-5, the specific value prediction AI determines that the second half of the two periods after the division in step 604 is a period to which the value of the value fluctuation asset has the lowest value. This is because the degree of confidence of multiple prediction models is low, and it can be said that the risk of selecting the first half period is high.

In order to avoid the case in which the risk and the return do not match, in step S501, additionally or alternatively, the specific value prediction AI is used for the first half of the predetermined period, in particular, the beginning of the predetermined period. The prediction of the time when the value of the value fluctuation asset becomes the lowest price within a predetermined period may be suppressed. This is achieved, for example, by providing a predetermined suppression period and suppressing prediction of the time within the predetermined suppression period as the time when the value of the value fluctuation asset is the lowest price within the predetermined period.

The predetermined suppression period is a period starting from the start of the predetermined period, and the length of the period may be any length. The length of the predetermined suppression period may be changed, for example, according to the target asset. For example, in the case of an asset in which the trend is easy to read (for example, when purchasing the euro in Japanese yen), the length of the predetermined suppression period can be shortened. For example, in the case of an asset where the trend is difficult to read (e.g., when purchasing the NZD for Japanese Yen), the length of the predetermined suppression period can be increased.

The following table is an example of the length of the predetermined suppression period when the predetermined period is January.

The length of the predetermined suppression period may be set by the specific value prediction AI or may be set manually, in consideration of the price fluctuation of the value fluctuation asset.

Here, as an example, the process of step S501 in the computer system 200 in the case of accumulating the US dollar every month will be described with reference to FIGS. 7A to 7D. It is assumed that the predetermined period is January (23 business days). The period when the value of the value fluctuation asset is the lowest is on a daily basis. 7A to 7D are graphs showing an example of the time variation of the US dollar price (yen). The solid line indicates the actual fluctuation value, and the broken line indicates the fluctuation value predicted using a prediction model that predicts the fluctuation of the prediction target period.

In step 601, the specific value prediction AI divides a predetermined period (23 business days) into two periods on the first day of the month. As shown in FIG. 7A, the specific value prediction AI includes a predetermined period (23 business days), a first period 701 consisting of the first business day to the 12th business day, and a 13th business day to the 23rd business day. It is divided into a second period 702 consisting of days.

In step S602, the specific value prediction AI determines, of the first period 701 and the second period 702, the period to which the day when the US dollar price is the lowest falls. The specific value prediction AI determines a period in which the average value of the US dollar price is low in the first period 701 and the second period 702 using a prediction model that predicts fluctuations in 23 days, and calculates the average of the US dollar prices A period with a low value is determined to be a period to which a day when the US dollar price becomes the lowest value belongs. In this example, the specific value prediction AI uses an average value 710 of the first period 701 predicted using a prediction model that predicts fluctuations in 23 days, and a prediction model that predicts fluctuations in 23 days The average value 720 of the predicted second period 702 is compared, and it is determined that the second period 702 is a period to which a day when the US dollar price becomes the lowest value belongs.

As an example, it is a multilayer perceptron comprising an input layer, three hidden layers, and an output layer, wherein the number of nodes is 31, 16, 8, 4, 1 and decreases from the input layer toward the output layer In the first period 701 and the second period 702, a period in which the average value of the US dollar value is low is determined by the specific value prediction AI that uses the multilayer perceptron as a prediction model that predicts fluctuations in 23 days. When processing was performed to determine that the period in which the average price value is low is the period to which the day when the US dollar price is the lowest falls, the following maximum prediction accuracy was obtained. The value change data used for learning the specific value prediction AI is only the value change data of the US dollar relative to the Japanese yen.

Here, the term "random" is a model that predicts the day when the lowest price is randomly obtained, and was used as a comparison target. Also from this result, it can be seen that the prediction accuracy is further improved as more data are input to the multilayer perceptron.

Further, as one example, a specific value prediction AI using CNN as a prediction model determines a period in which the average value of the US dollar price is low among the first period 701 and the second period 702, and the average value of the US dollar price is low. When it is determined that the term belongs to the term to which the day when the US dollar price is the lowest falls, the following maximum prediction accuracy can be obtained by changing the value fluctuation data used for learning.

Thus, the prediction model used and the value fluctuation data to be learned are closely related to the prediction accuracy. By selecting an appropriate forecasting model, value variation data to learn, forecasting accuracy of over 70% can be achieved.

In step S603, the specific value prediction AI determines whether the second period 702 is one day. Since the second period 702 is a period of 11 days consisting of the thirteenth business day to the twenty-third business day, the process proceeds to step S604.

In step S604, the specific value prediction AI divides the second period 702 into two periods on the first day of the second period 702. As shown in FIG. 7B, the specific value prediction AI comprises the second period 702, the third period 703 consisting of the thirteenth business day to the seventeenth business day, and the eighteenth business day to the twenty-third business day. It is divided into a fourth period 704.

Although the example illustrated in FIG. 7B indicates that step S604 is performed on the first day of the second period 702, the timing at which each step of the process of step S501 is performed is arbitrary. For example, on the first day of the month, the process of step S501 (the process of predicting the day when the value of the value fluctuation asset is the lowest value in a predetermined period) may be completely completed, or the value of the value fluctuation asset is As part of the processing in step S501 (processing to predict the date when the value of the value fluctuation asset becomes the lowest value in a predetermined period) is performed each time the date when the lowest price day is determined to belong. It is also good.

In step S605, the specific value prediction AI determines, of the third period 703 and the fourth period 704, the period to which the day when the US dollar price is the lowest falls. The specific value prediction AI determines a period in which the average value of the US dollar price is low in the third period 703 and the fourth period 704 using a prediction model that predicts 13-day fluctuation, and calculates the average of the US dollar price A period with a low value is determined to be a period to which a day when the US dollar price becomes the lowest value belongs. In this example, the specific value prediction AI uses an average value 730 of the third period 703 predicted using a prediction model that predicts 13-day fluctuation, and a prediction model that predicts 13-day fluctuation. The average value 740 of the fourth predicted period 704 is compared, and it is determined that the third period 703 is a period to which a day when the US dollar price is the lowest is included.

Returning to step S603, the specific value prediction AI determines whether the third period 703 is one day. Since the third period 703 is a five-day period consisting of the thirteenth business day to the seventeenth business day, the process proceeds to step S604 again.

In step S604, the specific value prediction AI divides the third period 703 into two periods. As shown in FIG. 7C, the specific value prediction AI comprises the third period 703, the fifth period 705 consisting of the thirteenth business day to the fourteenth business day, and the fifteenth business day to the seventeenth business day. It is divided into a sixth period 706.

In step S605, the specific value prediction AI determines, of the fifth period 705 and the sixth period 706, the period to which the day when the US dollar price is the lowest falls. The specific value prediction AI determines a period in which the average value of the US dollar price is low in the fifth period 705 and the sixth period 706 using a prediction model that predicts 5-day fluctuation, and calculates the average of the US dollar prices A period with a low value is determined to be a period to which a day when the US dollar price becomes the lowest value belongs. In this example, the specific value prediction AI uses an average value 750 of the fifth period 705 predicted using a prediction model that predicts fluctuations in five days, and a prediction model that predicts fluctuations in five days. The average value 760 of the predicted sixth period 706 is compared, and it is determined that the fifth period 705 is a period to which a day when the US dollar price becomes the lowest value belongs.

Returning to step S603, the specific value prediction AI determines whether or not the fifth period 705 is one day. Since the fifth period 705 is a two-day period consisting of the thirteenth business day to the fourteenth business day, the process proceeds again to step S604.

In step S604, the specific value prediction AI divides the fifth period 705 into two periods. As shown in FIG. 7D, the specific value prediction AI divides the fifth period 705 into a seventh period 707 consisting of a thirteenth business day and an eighth period 708 consisting of a fourteenth business day.

In step S605, the specific value prediction AI determines, of the seventh period 707 and the eighth period 708, the period to which the day when the US dollar price is the lowest falls. The specific value prediction AI determines a period in which the average value of the US dollar price is low in the seventh period 707 and the eighth period 708 using a prediction model that predicts fluctuations of two days, and the average of the US dollar prices A period with a low value is determined to be a period to which a day when the US dollar price becomes the lowest value belongs. In this example, the specific value prediction AI uses an average value 770 of the seventh period 707 predicted using a prediction model that predicts fluctuations for two days, and a prediction model that predicts fluctuations for two days The average value 780 of the predicted eighth period 708 is compared, and it is determined that the eighth period 708 is a period to which a day when the US dollar price becomes the lowest value belongs.

Returning to step S603, the specific value prediction AI determines whether the eighth period 708 is one day. Since the eighth period 708 is a one-day period consisting of the fourteenth business day, the process proceeds to step S606 and ends.

As a result of the above processing, it is predicted that the day the US dollar will be the lowest in this month is the 14th business day. At the 14th business day, in step S502, the processor unit 212 purchases the US dollar and accumulates it.

Here, as a further example, the processing of step S501 in the computer system 200 in the case of accumulating the AUD every month will be described with reference to FIGS. 8A to 8D.

8A to 8D are graphs showing an example of the time change of the Australian dollar price (yen). The solid line indicates the actual fluctuation value, and the broken line indicates the fluctuation value predicted using a prediction model that predicts the fluctuation of the prediction target period. It is assumed that the predetermined period is January (23 business days). The period when the value of the value fluctuation asset is the lowest is on a daily basis. The length of the predetermined suppression period is 4 days. The specific value prediction AI uses five prediction models to perform prediction. The prescribed number shall be varied as follows.

In step 601, the specific value prediction AI divides a predetermined period (23 business days) into two periods on the first day of the month. As shown in FIG. 8A, the specific value prediction AI includes a predetermined period (23 business days), a first period 801 consisting of the first business day to the 12th business day, and a 13th business day to the 23rd business day. And a second period 802 consisting of days.

In step S602-1, the specific value prediction AI determines, of each of the first period 801 and the second period 802, the period to which the day when the AUD price is the lowest falls, using each of the five prediction models. Do. The specific value prediction AI uses, for each of five prediction models that predict fluctuations in 23 days, a period in which the day when the AUD price becomes the lowest among the first period 801 and the second period 802 belongs. Determine The prediction model may be, for example, a plurality of prediction models that differ in the type or amount of past value fluctuation data used for learning, for example, the type or amount of past value fluctuation data input for prediction is A plurality of different prediction models may be used. For example, a plurality of prediction models having different types of neural networks, such as a prediction model using multilayer perceptron 450, a prediction model using CNN, and a prediction model using RNN It may be In this example, it is assumed that each prediction model predicts as follows.

In step S602-2 ', the specific value prediction AI is a period to which the first low price of the first period 801, which is the first half of the first period 801 and the second period 802, belongs. Determine the number of prediction models determined to be From the results in Table 5, the number of prediction models is determined to be three.

In step S602-3, the specific value prediction AI determines whether the number of prediction models determined in step S602-2 'is equal to or greater than a predetermined number. Since the predetermined number of the first business day to the twelfth business day is 3 and the number of prediction models determined in step S602-2 'is 3 or more, the process proceeds to step S602-4.

In step S602-4, the specific value prediction AI determines that the first period 801 is a period to which a day when the AUD price becomes the lowest value belongs.

In step S603, the specific value prediction AI determines whether the first period 801 is one day. Since the first period 801 is a 12-day period consisting of the first business day to the twelfth business day, the process proceeds to step S604.

In step S604, the specific value prediction AI divides the first period 801 into two periods. As shown in FIG. 8B, the specific value prediction AI comprises a first period 801, a third period 803 consisting of a first business day to a sixth business day, and a seventh business day to a twelfth business day. And the fourth period 804.

Although the example illustrated in FIG. 8B indicates that step S604 is performed on the first day of the third period 803, the timing at which each step of the process of step S501 is performed is arbitrary. For example, on the first day of the month, the process of step S501 (the process of predicting the day when the value of the value fluctuation asset is the lowest value in a predetermined period) may be completely completed, or the value of the value fluctuation asset is As part of the processing in step S501 (processing to predict the date when the value of the value fluctuation asset becomes the lowest value in a predetermined period) is performed each time the date when the lowest price day is determined to belong. It is also good.

In step S605, the specific value prediction AI determines a period to which a day in which the AUD price is the lowest falls in the third period 803 and the fourth period 804. The specific value prediction AI uses five prediction models that predict 13-day fluctuation to determine the period to which the Australian dollar price becomes lowest in the third period 803 and the fourth period 804. Do. In this example, it is assumed that each prediction model predicts as follows.

In step S605-2 ′, the specific value prediction AI is a period in which the third day of the third period 803, which is the first half period of the third period 803 and the fourth period 804, belongs to the day when the AUD price becomes the lowest. Determine the number of prediction models determined to be From the results in Table 6, the number of prediction models is determined to be four.

In step S605-3, the specific value prediction AI determines whether the number of prediction models determined in step S605-2 'is equal to or greater than a predetermined number. Since the predetermined number of the first business day to the sixth business day is five and the number of prediction models determined in step S605-2 'is not five or more, the process proceeds to step S605-5.

In step S605-5, the specific value prediction AI determines that the fourth period 804, which is the second half period, is a period to which a day when the AUD price is the lowest falls.

Returning to step S603, the specific value prediction AI determines whether or not the fourth period 804 is one day. Since the fourth period 804 is a six-day period consisting of the seventh business day to the twelfth business day, the process proceeds again to step S604.

In step S604, the specific value prediction AI divides the fourth period 804 into two periods. As shown in FIG. 8C, the specific value prediction AI comprises a fourth period 804, a fifth period 805 consisting of the seventh business day to the ninth business day, and a tenth business day to the twelfth business day. And the sixth period 806.

In step S605, the specific value prediction AI determines the period to which the day in which the AUD price is the lowest falls in the fifth period 805 and the sixth period 806. The specific value prediction AI uses five prediction models that predict 5-day fluctuation to determine the period to which the Australian dollar price becomes lowest among the fifth period 805 and the sixth period 806. Do. In this example, it is assumed that each prediction model predicts as follows.

In step S605-2 ', the specific value prediction AI is a period in which the fifth day of the fifth period 805 which is the first half period of the fifth period 805 and the sixth period 806 belongs to the day when the AUD price becomes the lowest. Determine the number of prediction models determined to be From the results in Table 7, the number of prediction models is determined to be five.

In step S605-3, the specific value prediction AI determines whether the number of prediction models determined in step S605-2 'is equal to or greater than a predetermined number. Since the predetermined number of the seventh business day to the ninth business day is 4 and the number of prediction models determined in step S605-2 'is 4 or more, the process proceeds to step S605-4.

In step S605-4, the specific value prediction AI determines that the fifth period 805 is a period to which a day when the AUD price becomes the lowest value belongs.

Returning to step S603, the specific value prediction AI determines whether the fifth period 805 is one day. Since the fifth period 805 is a three-day period consisting of the seventh business day to the ninth business day, the process proceeds to step S604 again.

In step S604, the specific value prediction AI divides the fifth period 805 into two periods. As shown in FIG. 8D, the specific value prediction AI includes a fifth period 805, a seventh period 807 consisting of the seventh business day, and an eighth period 808 consisting of the eighth business day to the ninth business day. Divide into and.

In step S605, the specific value prediction AI determines the period to which the day in which the AUD price is the lowest falls in the seventh period 807 and the eighth period 808. The specific value prediction AI uses five prediction models that predict two-day fluctuation to determine the period to which the low Australian price day belongs in the seventh period 807 and the eighth period 808. Do. In this example, it is assumed that each prediction model predicts as follows.

In step S605-2 ', the specific value prediction AI is a period in which the seventh day of the seventh period 807, which is the first half period of the seventh period 807 and the eighth period 808, belongs to the day when the AUD price becomes the lowest. Determine the number of prediction models determined to be From the results in Table 8, the number of prediction models is determined to be four.

In step S605-3, the specific value prediction AI determines whether the number of prediction models determined in step S605-2 'is equal to or greater than a predetermined number. Since the predetermined number of the seventh business day is 4 and the number of prediction models determined in step S605-2 'is 4 or more, the process proceeds to step S605-4.

In step S605-4, the specific value prediction AI determines that the seventh period 807 is a period to which a day when the AUD price becomes the lowest value belongs.

Returning to step S603, the specific value prediction AI determines whether the seventh period 807 is one day. Since the seventh period 807 is a one-day period consisting of the seventh business day, the process proceeds to step S606 and ends.

By the above processing, it is predicted that the day the Australian dollar will be the lowest in this month is the seventh business day. At the seventh business day, in step S502, the processor unit 212 purchases AUD and accumulates it.

When predicting the time of becoming a value of a specific rank instead of predicting the time of becoming the lowest price, in step S602 and step S605, a period of time when the value of an asset becomes a specific rank is determined It is also good. For example, a case will be described in which a period of time in which the value of an asset becomes the 14th lowest value belongs to the 23 working days will be described.

After 23 working days are divided into two periods in step S601, in step S602, one of the two divided periods is selected which has the lowest price. That is because 14> 23/2.

Next, after step S603, 11 or 12 business days are divided into two periods in step S604, and then in step S605, the period in which the low price (minimum value) exists among the two periods after division select. This is because (14-23 / 2) <23/4.

Next, after step S603, 5 or 6 business days are divided into two periods in step S604, and in step S605, the period in which the low price (local minimum) exists in the two periods after division is select. This is because (14-23 / 2) <23/8.

Next, after step S603, 2 or 3 business days are divided into two periods in step S604, and in step S605, one of the two periods after division does not have a low price (minimum value). select. (14-23 / 2)> 23/16.

This is repeated until it is determined as Yes in step S603.

Thus, in the same manner as the binary search described above, it is possible to determine the period to which the time when the asset value becomes a specific rank belongs.

In the example described above, it has been described that the binary search which divides the target period into two periods predicts the time when the value of the value fluctuation asset becomes the value of the specific rank within the predetermined period, but the present invention It is not limited to binary search. It is also within the scope of the present invention to divide the period of interest into more than two periods to predict when the value of the value-changing asset will be the value of a particular rank within a given period of time.

For example, in step S601, the specific value prediction AI divides the predetermined period into m periods instead of dividing it into two periods (m is an integer larger than 2), and in step S602, the specific value prediction AI Is the period to which the value of the value fluctuation asset belongs to a specific rank value (for example, lowest price, highest value, n lowest value, n highest value) of m periods after division May be determined. When determining the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, this judgment may, for example, teach past value fluctuation data so that the prediction of the average value of the predetermined period in the future can be output. The specific value prediction AI learned as data may be performed by determining a period in which the average value of the value of the value fluctuation asset has the lowest value among the m periods after division. In the determination of the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, for example, output the prediction of which of the periods obtained by dividing a future period into a plurality of periods. It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that it can be done. When determining the period to which the time when the value of the value fluctuation asset reaches the highest value belongs, this judgment is, for example, teaching past value fluctuation data so that the prediction of the average value of the predetermined period in the future can be output. The specific value prediction AI learned as data may be performed by determining, out of m divided periods, a period in which the average value of the value of the value fluctuation asset is the highest. The determination of the period to which the time when the value of the value fluctuation asset becomes the highest belongs, for example, is to output a prediction of which of the periods obtained by dividing a future period into a plurality of periods. It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that it can be done. When determining the period in which the time when the value of the value fluctuation asset becomes the value of the specific rank belongs, this judgment may, for example, be the past value fluctuation so that the prediction of the average value of the future predetermined period can be output. It may be performed by determining the period in which the average value of the value of the value fluctuation asset is in a specific order among the m periods after division by the specific value prediction AI learning data as teacher data . For example, in the period in which a certain future period is divided into a plurality of time periods, the determination of the period to which the time when the value of the value fluctuation asset becomes the value of the specific order falls within the specific order value It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that the prediction of can be output.

For example, in step S604, the specific value prediction AI further divides the period in which it is determined that the time when the value of the value fluctuation asset has a specific rank value belongs to m periods, and in step S605, the specific value prediction AI The AI belongs to the m periods after the division when the value of the value fluctuation asset has a specific rank value (for example, the lowest price, the highest value, the nth lowest value, the nth highest value). The period may be determined. When determining the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, this judgment may, for example, teach past value fluctuation data so that the prediction of the average value of the predetermined period in the future can be output. The specific value prediction AI learned as data may be performed by determining a period in which the average value of the value of the value fluctuation asset has the lowest value among the m periods after division. In the determination of the period to which the time when the value of the value fluctuation asset becomes the lowest belongs, for example, output the prediction of which of the periods obtained by dividing a future period into a plurality of periods. It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that it can be done. When determining the period to which the time when the value of the value fluctuation asset reaches the highest value belongs, this judgment is, for example, teaching past value fluctuation data so that the prediction of the average value of the predetermined period in the future can be output. The specific value prediction AI learned as data may be performed by determining, out of m divided periods, a period in which the average value of the value of the value fluctuation asset is the highest. The determination of the period to which the time when the value of the value fluctuation asset becomes the highest belongs, for example, is to output a prediction of which of the periods obtained by dividing a future period into a plurality of periods. It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that it can be done. When determining the period in which the time when the value of the value fluctuation asset becomes the value of the specific rank belongs, this judgment may, for example, be the past value fluctuation so that the prediction of the average value of the future predetermined period can be output. It may be performed by determining the period in which the average value of the value of the value fluctuation asset is in a specific order among the m periods after division by the specific value prediction AI learning data as teacher data . For example, in the period in which a certain future period is divided into a plurality of time periods, the determination of the period to which the time when the value of the value fluctuation asset becomes the value of the specific order falls within the specific order value It may be performed by specific value prediction AI which is learning past value fluctuation data as teacher data so that the prediction of can be output.

At this time, for example, in step S602-2 or step S605-2, the specific value prediction AI determines that the value of the value fluctuation asset has a specific rank value according to more prediction models in m periods after division. The period determined to be the period to which the time of day belongs is determined to be the period of time to which the value of the asset becomes the value of the specific rank.

For example, in step S602-2 'or step S605-2', for each of the first half of the m periods after division, the specific value prediction AI ranks such that the value of the value fluctuation asset has a specific value in each of the first half periods. The number of prediction models determined to be the period to which the time having the value of 1 belongs is determined, and then in step S602-3 or step S605-3, the specific value prediction AI is determined for each of the first half period. It is determined whether the number of predicted models is equal to or greater than a predetermined number. If it is more than a predetermined number, it is determined that the period is a period to which the value of the value fluctuation asset has a value of a specific rank.

For example, in the case of m = 6, in step S602-2 ′ or step S605-2 ′, the specific value prediction AI has a value fluctuation asset value period for each of the first three periods of the six divided periods. Determine the number of forecasting models that have been determined to be the period to which the time when value becomes a particular rank value belongs. Next, in step S602-3 or step S605-3, the specific value prediction AI determines whether or not the number of the determined prediction models is a predetermined number for each of the first three periods, and the predetermined number or more is determined. If so, the period is determined to be a period to which the value of the value fluctuation asset has a value of a specific rank. For example, each of the number of prediction models determined for the first one of the first three periods and the number of prediction models determined for the third one of the first three periods is less than a predetermined number, If the number of forecasting models determined for the second one of the first three periods is greater than or equal to the predetermined number, the value of the value fluctuation asset is specified for the second one of the first three periods. It is determined that it is a period to which the timing of the ranking value belongs.

At this time, when there is no period in which the number of prediction models is equal to or more than a predetermined number, for example, the entire second half period of the plurality of periods is a period to which the time when the value of the value change asset becomes a specific rank value It may be determined that the process proceeds to step S603, or the entire period excluding the first period among the plurality of periods is a period to which the time when the value of the value change asset becomes the value of the specific rank belongs You may decide and proceed to step S603. This makes it possible to avoid selecting the beginning of the predetermined period when the confidence degree by the prediction model is low, and as a result, it is possible to avoid the case where risk and return do not match. .

4. Alternative Configuration of Computer System for Implementing a New Financial Product FIG. 9 shows an example of a computer system 200 'that is an alternative configuration of computer system 200 for implementing the above-described new financial product. In FIG. 9, the same elements as the elements shown in FIG. 2 are denoted by the same reference numerals, and the description thereof is omitted here. The computer system 200 'causes the user device 100 to predict when the value of the value change asset becomes a specific value within a predetermined period, and the value change occurs when the value of the predicted value change asset becomes a specific value. The server apparatus 210 performs the transaction of the asset.

The computer system 200 ′ includes a server device 210, a database unit 220 connected to the server device 210, and the user device 100. The server device 210, the user device 100, and the asset transaction computer system 300 are mutually connected via the network 400. Here, the type of the network 400 does not matter. The network 400 may be, for example, the Internet or a LAN.

The user device 100 is a terminal used by the user 10 who uses the service of accumulating value change assets, and may be any terminal device such as a smartphone, a tablet, a personal computer and the like. The user device 100 includes an interface unit 111, a processor unit 112, and a storage unit 113.

The interface unit 111 is configured to control communication via the network 400. The interface unit 111 can transmit information to the outside of the user device 100, and can receive information from the outside of the user device 100. For example, the interface unit 111 transmits, via the network 400, to the server apparatus 210, a prediction of a date when the value of the value fluctuation asset is the lowest price in a predetermined period.

The processor unit 112 controls the overall operation of the user device 100. The processor unit 112 reads a program stored in the storage unit 113 and executes the program. This enables the user device 100 to function as a device that performs a desired step. For example, the processor unit 112 can perform part of the process of accumulating value change assets. The processor unit 112 may be implemented by a single processor or may be implemented by a plurality of processors.

The storage unit 113 stores a program for executing a process in the user device 100, data required to execute the program, and the like. In the storage unit 113, for example, a program for trading value-varying assets, or a program for accumulating value-varying assets (for example, a program for realizing the processing shown in FIGS. 5 and 6A to 6C described above) Part of) is stored. Here, it does not matter how the program is stored in the storage unit 113. For example, the program may be preinstalled in the storage unit 113. Alternatively, the program may be installed in the storage unit 113 by being downloaded via the network 400. The storage unit 113 may be implemented by any storage means.

The storage unit 113 may also store, for example, past value fluctuation data. Historical value change data can be used to predict future value changes. Historical value change data used for learning includes value change data for the asset in question against the currency for purchase (for example, Japanese yen), as well as against other currencies (for example, US dollar, Australian dollar, euro, NZ dollar, etc.) It may include value change data of the target asset, or data such as S & P 500 index, Nikkei Stock Average, German stock index or WTI crude oil price. The database unit 220 can also store, for example, learned value fluctuation tendencies. The database 220 may store the weight vector of each node of the multilayer perceptron 450 established by learning.

The specific value prediction AI may be realized by the user device 100 having the configuration described above. The specific value prediction AI can predict when the value of the value fluctuation asset will be the lowest value using any prediction model, as described above.

In the example shown in FIG. 9, each component of the user apparatus 100 is provided in one user apparatus 100, but the present invention is not limited to this. It is also possible that any of the components of the user device 100 is provided outside the user device 100. For example, when each of the processor unit 112 and the storage unit 113 is configured by different hardware components, each hardware component may be connected via an arbitrary network. At this time, the type of network does not matter. Each hardware component may be connected via, for example, a LAN, may be wirelessly connected, or may be wired.

Processing in computer system 200 ′ is similar to processing 500 in computer system 200. However, the processing of step S501 is different from the processing 500 of the computer system 200 in that the specific value prediction AI realized by the user device 100 performs the processing of step S501 and the processor unit 212 of the server device 210 performs the processing of step S502. .

Although the example described above describes predicting the day when the value of the value fluctuation asset is the lowest, the present invention predicts a unit of time (e.g., hour, minute, second, etc.) shorter than "day". It is also within the scope of the present invention to predict time units (eg, weeks, months etc.) longer than "days". For example, when accumulating value change assets daily, the

computer system

200 or 200 ′ performs processing similar to that described above with reference to FIG. 5 and FIGS. It is possible to predict when is the lowest price and to purchase value-changing assets when the value of the predicted value-changing asset becomes the lowest. Here, when the value of the value fluctuation asset is the lowest, it may be an arbitrary unit of time. When the value of the value fluctuation asset is the lowest price, it may be, for example, in seconds, minutes, or hours.

The present invention is not limited to the embodiments described above. It is understood that the scope of the present invention should be interpreted only by the claims. It is understood that those skilled in the art can implement the equivalent scope based on the description of the present invention and common technical knowledge, from the description of the specific preferred embodiments of the present invention.

The present invention determines the value of an asset whose value fluctuates with time over time, such as the highest or lowest price (ie, maximum or minimum) or a particular rank value (eg, second) The present invention is useful as providing a computer system, method, and program that make it possible to predict when to be a specific value such as a cheap value, a tenth highest value, etc.). Furthermore, the present invention makes it possible to increase the evaluation value of an asset by periodically or continuously purchasing or accumulating the asset based on the prediction of when it becomes a specific value. And it is useful as what provides a program.

Reference Signs List 10 user 20 bank 30 forex market 100 user device 200, 200 'computer system 210 server device 220 database unit 300 asset trading computer system 400 network

Claims

Predicting when the value of the foreign currency will be a low in a particular rank within a predetermined time period;
A computer system configured to: purchase the foreign currency at the predicted time.
Said predicting is
Dividing the predetermined period into a plurality of periods;
The computer system according to claim 1, further comprising: determining a period to which a time when the value of the foreign currency is a specific low price belongs, out of the plurality of divided periods.
Said predicting is
(1) further dividing the period in which it is determined that the time when the value of the foreign currency is the low price of a specific rank belongs to a plurality of periods,
(2) Of the plurality of periods after the division, determining the period to which the time when the value of the foreign currency is a specific low price belongs, and the value of the foreign currency is a low price of a specific order The computer system according to claim 2, further comprising: repeating a period in which it is determined that the period belongs to the same length as a period in which the value of the foreign currency becomes a specific rank low.
The computer system according to claim 2 or 3, wherein the determination includes determining a period in which the average value of the value of the foreign currency is low among the plurality of divided periods.
The above determination is
Determining, using each of a plurality of prediction models, a period to which a time when the value of the foreign currency becomes a low price of a specific rank among the plurality of periods after the division;
Among the multiple periods after the division, the value of the foreign currency is determined by a period determined by more prediction models to be the period to which the time when the value of the foreign currency becomes a low in a specific rank belongs. The computer system according to any one of claims 2 to 4, comprising determining that the time when the ranking is low belongs to a period to which it belongs.
The above determination is
Determining, using each of a plurality of prediction models, a period to which a time when the value of the foreign currency becomes a low price of a specific rank among the plurality of periods after the division;
Determining, for each of the first half of the plurality of divided periods, the number of prediction models that are determined to be periods to which the time when the value of the foreign currency is a specific low price belongs; When,
The method according to any one of claims 2 to 4, comprising: determining a period in which the number of the prediction models is equal to or more than a predetermined number as a period to which a time when the value of the foreign currency becomes a specific low price belongs. Computer system as described in.
The computer system according to claim 6, wherein the determination further includes changing the predetermined number in accordance with a relative position of the divided period with respect to the entire predetermined period.
The computer system according to any one of claims 2 to 7, wherein the plurality of time periods is two time periods.
4. The method according to claim 1, wherein the predicting includes suppressing prediction of a period within a predetermined suppression period from a start point of the predetermined period as a period when the value of the foreign currency is a low in a specific rank. 8. The computer system according to any one of 8.
The computer system according to any one of claims 1 to 9, wherein the predicting and the purchasing are performed repeatedly.
The method according to claim 1, wherein the purchasing is automatically performed without a user's input operation for purchasing the foreign currency at a time when the predicted value of the foreign currency becomes a low in a specific rank. 10. A computer system according to any one of 10.
The computer system according to any one of claims 1 to 11, wherein the low price in the specific rank is the lowest price.
The computer system according to any one of claims 1 to 11, wherein the low price of the specific rank is the n-th lowest value (n is an integer of 2 or more).
A method implemented in a computer system, said method comprising
Predicting when the value of the foreign currency will be a low in a particular rank within a predetermined time period;
And C. purchasing the foreign currency at the predicted time.
A program executed on a computer system, said program being executed on a computer system
Predicting when the value of the foreign currency will be a low in a particular rank within a predetermined time period;
A program for causing the computer system to perform processing including: purchasing the foreign currency at the predicted time.
A computer system configured to predict when the value of the foreign currency will be a particular rank value within a predetermined period of time based on past value fluctuations of the foreign currency whose value changes as time passes.
The computer system according to claim 16, further configured to issue the foreign currency transaction request at the predicted time.
A program executed on a computer system, said program being executed when:
Processing the computer system including processing of predicting when the value of the foreign currency will be a value of a specific rank within a predetermined period based on the past value change of the foreign currency whose value changes as time passes. The program to be done.
The process is
The program according to claim 18, further comprising: issuing the foreign currency transaction request at the predicted time.