Commodity price estimation system

Abstract

To provide information corresponding to price change of a prescribed commodity.SOLUTION: A commodity price estimation system (1) according to one aspect of the present invention includes an information processing apparatus (101) configured to estimate a price of a variable commodity after a predetermined time point, on the basis of a price of the variable commodity at the predetermined time point, past price variation data of the variable commodity, and data related to an event affecting the price of the price variable commodity.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a product price estimation system. [Background technology]

[0002] Information on investment products has been provided for some time. For example, information such as price information, company information, and chart information on investment products is provided. However, investment behavior varies from investor to investor, and there is no way to provide information that can be dynamically changed according to the buying and selling behavior of investors.

[0003] Patent Document 1 discloses a system for providing stock information to shareholders, investors, etc. on the Internet, which displays sales trends, operating profit trends, ordinary profit trends, and stock price trends along with past highs and lows of stock prices, and is characterized by the ability to search for information and display detailed content. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-240964 [Non-patent literature]

[0005] [Non-Patent Document 1] “Time Series Table”, [online], Statistics Bureau, Ministry of Internal Affairs and Communications, [Retrieved April 11, 2022], Internet<https: / / dashboard.e-stat.go.jp / timeSeries> Summary of the Invention [Problem to be solved by the invention]

[0006] The above-mentioned conventional investment information providing systems do not dynamically change the information they provide in response to the user's investment behavior. In other words, they are tools for evaluating and diagnosing static portfolios and stocks, or provide uniform information. They do not dynamically change the information they provide in response to the dynamic changes in each user's investment behavior and situation.

[0007] However, in reality, each investor dynamically engages in various investment activities, and there is a need for an investment product information provision system that can respond to dynamic changes in each investor's investment activities so that advice, evaluations, and necessary predictions can be made dynamically in response to these activities.

[0008] In other words, rather than providing stock price data and corporate performance data uniformly, it will be necessary to rethink the way and system for providing information on investment products, etc., from the perspective that prices of investment products change dynamically in response to dynamic changes in investment behavior and the environment for each individual investor.

[0009] One aspect of the present invention aims to provide information corresponding to price changes of a given product, more specifically, to provide information corresponding to dynamic changes in investment products. [Means for solving the problem]

[0010] In order to solve the above problem, a product price estimation system according to one embodiment of the present invention has an information processing device that estimates the price of a price-changing product after a predetermined time based on the price of the price-changing product at the predetermined time, past price fluctuation data for the price-changing product, and data regarding events that affect the price of the price-changing product.

[0011] The product price estimation system and other systems relating to each aspect of the present invention may be realized by a computer. In this case, the control program for the product price estimation system that realizes the product price estimation system on a computer by causing the computer to operate as each part (software element) of the product price estimation system, and the computer-readable recording medium on which it is recorded, also fall within the scope of the present invention. [Effects of the Invention]

[0012] According to one aspect of the present invention, it is possible to provide information corresponding to price changes of a predetermined product, particularly information corresponding to dynamic changes in investment products. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram showing a configuration of an investment product information providing system according to a first embodiment. [Figure 2] 1 is a block diagram showing a configuration of an investment product information providing system according to a first embodiment. [Figure 3] FIG. 10 is a block diagram showing the configuration of an investment product price fluctuation prediction system according to a second embodiment. [Figure 4] FIG. 10 is a block diagram showing the hardware configuration of an investment product price fluctuation prediction system according to a second embodiment. [Figure 5] FIG. 10 is a functional block diagram of an information processing device and a conversion device according to a second embodiment. [Figure 6] 10 is a flowchart showing the processing of the investment product price fluctuation prediction system according to the second embodiment. [Figure 7] FIG. 10 is a diagram illustrating time lag and year-on-year comparison data according to the second embodiment. [Figure 8] FIG. 10 is a diagram illustrating a method for acquiring event data and price fluctuation data of investment products according to the second embodiment. [Figure 9] FIG. 10 is a diagram illustrating a date linking method according to the second embodiment. [Figure 10]FIG. 10 is a diagram showing a method for linking dates and related issues according to the second embodiment. [Figure 11] FIG. 10 is a diagram showing the relationship between events and investment product prices according to the second embodiment. [Figure 12] FIG. 10 is a diagram showing classified products according to the second embodiment. [Figure 13] FIG. 10 is a diagram showing a method for evaluating correlation using a table lookup method according to the second embodiment. [Figure 14] 10 is a diagram showing time-series data and rate of change data according to the second embodiment. FIG. [Figure 15] FIG. 10 is a diagram showing a method for capturing event data after a certain period of time has elapsed according to the second embodiment. [Figure 16] FIG. 10 is a diagram illustrating a time lag determination system and a prediction system after time lag determination according to a second embodiment. [Figure 17] FIG. 10 is a diagram showing the relationship between fluctuations in event data and fluctuations in investment product prices according to the second embodiment. [Figure 18] FIG. 10 is a diagram illustrating a table lookup method according to the second embodiment. [Figure 19] FIG. 10 is a diagram showing the structure of a table lookup method according to the second embodiment. [Figure 20] FIG. 10 is a diagram showing a method for estimating the relationship between macro index change events and price fluctuations of investment products using a table reference method according to the second embodiment. [Figure 21] FIG. 10 is a diagram showing an event change input and a stock output according to the second embodiment. [Figure 22] FIG. 10 is a diagram showing the procedure of a table lookup method according to the second embodiment. [Figure 23] FIG. 10 is a diagram showing an AI-based correlation check and correlation table according to the second embodiment. [Figure 24] FIG. 10 is a diagram illustrating the concept of an AI method according to a second embodiment. [Figure 25] FIG. 10 is a diagram showing an input method according to a second embodiment. [Figure 26] FIG. 10 is a diagram showing a table relating to corporate performance events according to the second embodiment. [Figure 27]FIG. 10 is a diagram showing a table of corporate earnings forecast revisions and stock price fluctuation rates, and a table reference method according to the second embodiment. [Figure 28] FIG. 10 is a diagram showing the relationship between the rate of change in earnings revision and the rate of change of investment products according to the second embodiment. [Figure 29] FIG. 10 is a diagram illustrating machine learning regarding corporate performance events and stock price fluctuation rates after a certain period of time according to the second embodiment. [Figure 30] FIG. 10 is a diagram illustrating machine learning regarding corporate earnings revisions and stock price fluctuation rates after a certain period of time according to the second embodiment. [Figure 31] FIG. 10 is a diagram illustrating a corporate performance-adjusted stock price fluctuation pattern according to the second embodiment. [Figure 32] FIG. 10 is a diagram showing the transition of upward revisions and how to calculate the revision rate according to the second embodiment. [Figure 33] FIG. 10 is a diagram showing the configuration of a trading rule agreement table etc. according to the second embodiment. [Figure 34] FIG. 10 is a diagram showing the relationship between technical indicators and the rate of change of investment products according to the second embodiment. [Figure 35] FIG. 10 is a diagram illustrating machine learning regarding a golden cross according to the second embodiment. [Figure 36] FIG. 10 is a diagram illustrating a buy signal according to the second embodiment. [Figure 37] FIG. 10 is a diagram showing the relationship between the year-on-year conversion rate of crude oil prices and the rate of change in stock prices according to the second embodiment. [Figure 38] FIG. 10 is a diagram showing the relationship between the event increase rate, the stock price increase rate, and the time lag according to the second embodiment. [Figure 39] FIG. 10 is a diagram showing a table of macro index changes and corporate performance changes according to the second embodiment. [Figure 40] FIG. 10 is a diagram showing the relationship between the year-on-year change rate of crude oil prices and the rate of change of stock prices according to the second embodiment. [Figure 41] FIG. 10 is a diagram illustrating the processing of a conversion device and a database unit according to the second embodiment. [Figure 42] FIG. 10 is a diagram illustrating the processing of the best stock determination system according to the second embodiment. [Figure 43] FIG. 10 is a diagram illustrating a system for predicting macro index change events and stock price fluctuations of Japanese stocks using a table reference method according to a second embodiment. [Figure 44] FIG. 10 is a diagram illustrating machine learning of crude oil price changes and stock price volatility according to the second embodiment. [Figure 45] FIG. 10 is a diagram illustrating a process of classifying stocks according to correlation according to the second embodiment. [Figure 46] FIG. 10 is a diagram illustrating a process for estimating an expected stock price increase rate according to the second embodiment. [Figure 47] FIG. 10 is a diagram illustrating the processing of the best stock determination system according to the second embodiment. [Figure 48] FIG. 10 is a diagram illustrating the processing of a neural network and a recurrent neural network according to the second embodiment. [Figure 49] FIG. 10 is a diagram illustrating a method for classifying events using a classifier according to a second embodiment, and inputs and outputs using a neural network. [Figure 50] 10 is a flowchart showing a method for creating a trained model according to the second embodiment. [Figure 51] FIG. 10 is a diagram illustrating the processing of the investment product price fluctuation prediction system according to the second embodiment. [Figure 52] FIG. 10 is a diagram illustrating the process of learning and predicting the relationship between the occurrence of a rating event and the stock price fluctuation rate according to the second embodiment. [Figure 53] FIG. 10 is a diagram illustrating the features of a neural network and a recurrent neural network according to the second embodiment. [Figure 54] FIG. 2 is a diagram illustrating a hardware configuration of an advice presentation system according to embodiment A1. [Figure 55] FIG. 2 is a block diagram showing the configuration of a terminal and a server according to embodiment A1. [Figure 56] FIG. 10 is a diagram showing an outline of processing of the advice presentation system according to embodiment A1. [Figure 57] FIG. 10(a) is a diagram showing an example of trading data of an investment product according to embodiment A1, and FIG. 10(b) is a diagram showing an example of evaluation indexes of trading data according to embodiment A1. [Figure 58] 10 is a flowchart showing a diagnosis process based on a principal rotation period according to embodiment A1. [Figure 59] 10 is a flowchart showing a diagnostic process based on a winning return rate according to embodiment A1. [Figure 60] 10 is a flowchart showing a diagnostic process based on a loss rate according to embodiment A1. [Figure 61] 10 is a flowchart showing a diagnostic process based on trading profit and loss according to embodiment A1. [Figure 62] 10 is a flowchart showing a classification process of buying and selling patterns according to embodiment A1. [Figure 63] 10 is a flowchart showing a diagnostic process based on the rate of change of held stocks according to embodiment A1. [Figure 64] 10 is a flowchart showing a ranking process based on principal increase / decrease rate according to embodiment A1. [Figure 65] 10 is a flowchart showing a process of an overall profit and loss analysis according to embodiment A1. [Figure 66] FIG. 10 is a diagram showing examples of evaluation values of total profit / loss, trading profit / loss, and unrealized profit / loss according to the level of detail according to embodiment A1. [Figure 67] FIG. 10 is a diagram showing an example of evaluation indexes for owned products according to embodiment A1. [Figure 68] FIG. 10 is a diagram showing an example of a pattern of owned products according to embodiment A1. [Figure 69] FIG. 10 is a diagram showing an example of an initial screen of a stock investment simulation according to embodiment A1. [Figure 70] FIG. 10 is a diagram showing an example of a question screen for a stock investment simulation according to embodiment A1. [Figure 71] FIG. 10 is a diagram showing the transition of stock prices in a stock investment simulation according to embodiment A2. [Figure 72] FIG. 10 is a diagram showing the transition of the valuation amount for each branch of each question in the stock investment simulation according to embodiment A2. [Figure 73] FIG. 10 is a diagram showing the configuration of an information presentation system according to embodiment A4. [Figure 74] FIG. 10 is a diagram showing a comparison of evaluation process methods according to embodiment A4. [Figure 75] FIG. 10 is a diagram for explaining period-specific aggregation target purchase and sale data according to embodiment A4. [Figure 76] FIG. 10 is a diagram showing aggregated trading data by period according to embodiment A4. [Figure 77] FIG. 10 is a diagram showing the procedure for reevaluation according to embodiment A4. [Figure 78] FIG. 10 is a diagram showing an example of modifying trading profit and loss data into period-specific data according to embodiment A4. [Figure 79] FIG. 10 is a diagram showing an example of modifying trading profit and loss data into period-specific data according to embodiment A4. [Figure 80] FIG. 10 is a diagram showing the procedure for revaluing unrealized profit and loss trading data according to embodiment A4. [Figure 81] FIG. 10 is a diagram showing an example of a table of aggregated trading data by investor according to embodiment A4. [Figure 82] FIG. 10 is a diagram showing an example of a table of aggregated trading data by investment target according to embodiment A4. [Figure 83] FIG. 10 is a diagram showing the difference between the profit-and-loss-classified aggregation target trading data and the profit-and-loss level trading data according to embodiment A4. [Figure 84] FIG. 10 is a diagram showing the difference between the old and new methods of processing trading data to be aggregated by profit and loss in embodiment A4. [Figure 85] FIG. 10 is a diagram showing five methods of the evaluation method according to embodiment A4. [Figure 86] FIG. 79 is a diagram showing an example of extracted (or classified, tabulated, or processed) trading profit / loss level trading data according to embodiment A4 (processed trading profit / loss level trading data of FIG. 79). [Figure 87] FIG. 10 is a diagram showing the relationship between trading profit and loss and unrealized profit and loss (excluding cash) according to embodiment A4. [Figure 88] FIG. 10 is a diagram showing the relationship between trading profit and loss and unrealized profit and loss (including cash) according to embodiment A4. [Figure 89] FIG. 10 is a diagram showing the breakdown of the valuation amount of the period-by-period profit and loss trading data and opportunity loss according to embodiment A4. [Figure 90] FIG. 10 is a diagram showing the relationship between trading profit and loss, cash, and unrealized profit and loss according to embodiment A4. [Figure 91] FIG. 10 is a diagram showing the extraction of data on winning profit levels according to embodiment A4. [Figure 92] FIG. 10 is a diagram showing the extraction of data on winning profit levels according to embodiment A4. [Figure 93] FIG. 92 is a diagram showing the processing data (new method) of FIG. 91 in embodiment A4. [Figure 94] A diagram calculated in stages from profit and loss level trading data relating to embodiment A4. [Figure 95] A diagram calculated in stages from profit and loss level trading data relating to embodiment A4. [Figure 96] FIG. 10 is a diagram showing a specific example of calculation of a profit and loss level stage evaluation index according to embodiment A4. [Figure 97] FIG. 10 is a conceptual diagram of the second level (trading profit / loss level trading data) according to embodiment A4. [Figure 98] FIG. 10 is a diagram showing a specific example of the second level (trading profit / loss level trading data) according to embodiment A4. [Figure 99] FIG. 10 is a diagram showing a specific example of the second level (unrealized profit / loss level) according to embodiment A4. [Figure 100] FIG. 10 is a diagram showing the effects of leverage and compound interest according to embodiment A4. [Figure 101] FIG. 10 is a diagram showing a specific example of the aggregation target comparison process according to embodiment A4. [Figure 102] FIG. 10 is a diagram showing a component comparison process according to embodiment A4. [Figure 103] FIG. 10 is an explanatory diagram of a profit and loss level evaluation index comparison process according to embodiment A4. [Figure 104] FIG. 10 is a diagram showing a specific example of a ranking explanation according to embodiment A4. [Figure 105] FIG. 10 is an explanatory diagram of the component ranking according to embodiment A4 (when investors and brands are counted). [Figure 106]FIG. 10 is a diagram showing a specific example of a multi-layered ranking according to embodiment A4. [Figure 107] FIG. 10 is a diagram showing a specific example of a ranking to be counted according to embodiment A4. [Figure 108] FIG. 10 is a diagram showing a specific example of a multi-layered ranking for each aggregation target according to embodiment A4. [Figure 109] FIG. 10 is a diagram showing a specific example of ranking by profit and loss level according to embodiment A4. [Figure 110] FIG. 10 is a diagram showing a specific example of linked unrealized profit and loss level buying and selling data according to embodiment A4. [Figure 111] FIG. 10 is a diagram showing a specific example of trading data at the winning pattern 1 level according to embodiment A4. [Figure 112] FIG. 10 is a diagram showing an example of unrealized profit and loss pattern level trading data according to embodiment A4. [Figure 113] FIG. 10 is a diagram showing a specific example of linked unrealized profit and loss level buying and selling data according to embodiment A4. [Figure 114] FIG. 10 is a diagram showing a specific example of trading data at the winning pattern 1 level according to embodiment A4. [Figure 115] FIG. 10 is a diagram showing an example of a winning pattern according to embodiment A4. [Figure 116] FIG. 10 shows three comparison processes according to embodiment A4. [Figure 117] 10 is an information flow diagram of a client and a server according to embodiment A4. [Figure 118] FIG. 10 is a diagram showing that the generation of investment issues and articles according to embodiment A4 is synonymous with the results of the advice generation system. [Figure 119] FIG. 10 is a diagram showing what data is accumulated according to embodiment A4. [Figure 120] FIG. 10 is a diagram showing a process using hardware resources according to embodiment A4. [Figure 121] FIG. 10 is a diagram showing a processing method of an information processing system according to embodiment A4. [Figure 122] FIG. 10 is a diagram showing the flow of processing by a server in an information processing system according to embodiment A4. [Figure 123] FIG. 10 is a diagram showing a processing method 2 of the information processing system according to embodiment A4. [Figure 124] FIG. 10 is a diagram showing a calculation processing program of the information processing system according to embodiment A4. [Figure 125] FIG. 10 is a diagram showing a data structure of an information processing system according to embodiment A4. [Figure 126] FIG. 10 is a diagram showing a look-up table system of an information processing system according to embodiment A4. [Figure 127] FIG. 10 is a diagram showing an AI machine learning process of an information processing process according to embodiment A4. [Figure 128] FIG. 10 is a diagram showing a reference diagram of a display table according to embodiment A4. [Figure 129] FIG. 10 is a diagram showing a summary of the process of buying and selling data according to embodiment A4. [Figure 130] FIG. 10 is a diagram showing a flow up to an evaluation step of an information processing process according to embodiment A4. [Figure 131] FIG. 10 is a diagram showing an evaluation index determination step according to embodiment A4. [Figure 132] FIG. 10 is a diagram showing an evaluation index importance determination and display step according to embodiment A4. [Figure 133] FIG. 10 is a diagram showing an evaluation index importance determination process according to embodiment A4. [Figure 134] FIG. 10 is a diagram showing an evaluation index importance determination process 2 according to embodiment A4. [Figure 135] FIG. 10 is a diagram showing a machine learning model of the evaluation index importance determination process according to embodiment A4. [Figure 136] FIG. 10 is a diagram showing an evaluation index importance determination process 2 according to embodiment A4. [Figure 137] FIG. 10 is a diagram showing a generating and displaying step of ranking articles according to embodiment A4. [Figure 138] FIG. 10 is a diagram showing the process of identifying uncountering trade data and valuing market value according to embodiment A4. [Figure 139] FIG. 10 is a diagram showing a method for capturing investment product prices according to embodiment A4. [Figure 140] FIG. 10 is a diagram showing the creation of period-specific aggregation target buying and selling data according to embodiment A4. [Figure 141] FIG. 10 is a diagram showing a linked ownership evaluation according to embodiment A4. [Figure 142] FIG. 10 is a diagram showing a table reference method for an information processing process according to embodiment A4. [Figure 143] FIG. 10 is a diagram showing a network according to embodiment A4. [Figure 144] FIG. 10 is a database relation diagram according to embodiment A4. [Figure 145] FIG. 10 is a diagram showing a relational diagram of AI learning according to embodiment A4. [Figure 146] FIG. 10 is a diagram showing the relationship of table references according to embodiment A4. [Figure 147] FIG. 10 is a diagram showing an input form format (transaction data) according to embodiment A4. [Figure 148] This is a detailed diagram of the first phase of AI learning according to embodiment A4. [Figure 149] This is a detailed diagram of the second phase of AI learning according to embodiment A4. [Figure 150] This is a detailed diagram of the third phase of AI learning according to embodiment A4. [Figure 151] This is a detailed diagram of the fourth phase of AI learning according to embodiment A4. [Figure 152] 10 is a table of data to be aggregated by period according to embodiment A4. [Figure 153] FIG. 77 is a summary diagram of FIGS. 77 to 79 according to embodiment A4. [Fig. 154] FIG. 10 is an explanatory diagram of the first phase according to embodiment A4. [Figure 155] FIG. 10 is an explanatory diagram of the second to fourth phases according to embodiment A4. [Figure 156] FIG. 10 is a verification chart of stock selection according to embodiment A4. [Figure 157] FIG. 10 is a verification chart of the timing of brand purchases according to embodiment A4. [Figure 158]FIG. 10 is a chart showing the stock investment trends of other investors during the holding period according to embodiment A4. [Figure 159] FIG. 10 is a chart showing the stock investment trends of other investors during the holding period according to embodiment A4. [Figure 160] FIG. 10 is an explanatory diagram of the step of calculating the evaluation index according to embodiment A4. [Figure 161] FIG. 11 is an explanatory diagram of a combined table of purchase data and sales data according to embodiment A4. [Figure 162] FIG. 10 is a diagram showing the leverage effect and compound interest effect according to embodiment A4. [Figure 163] FIG. 10 is an explanatory diagram of a plurality of methods for calculating evaluation indexes according to embodiment A4. [Fig. 164] FIG. 10 is a calculation table of evaluation indexes according to embodiment A4. [Figure 165] FIG. 10 is a diagram illustrating a hardware configuration of an advice presentation system according to embodiment B1. [Figure 166] FIG. 10 is a block diagram showing the configuration of a terminal and a server according to embodiment B1. [Figure 167] FIG. 10 is a diagram showing a display example of a "For first-time users" screen according to embodiment B1. [Figure 168] FIG. 10 is a diagram showing a display example of a "held stocks" screen according to embodiment B1. [Figure 169] FIG. 10 is a diagram showing a display example of an “evaluation amount” screen according to embodiment B1. [Figure 170] FIG. 10 is a diagram showing an example of a display of a "trading history" screen according to embodiment B1. [Figure 171] FIG. 10 is a diagram showing an example of a display of a screen of "Evaluation amount history" according to embodiment B1. [Figure 172] FIG. 10 is a diagram showing a display example of a "detailed evaluation amount" screen according to embodiment B1. [Fig. 173] FIG. 10 is a diagram showing a display example of a "portfolio" screen according to embodiment B1. [Fig. 174] 10 is a flowchart showing the processing of the advice presentation system according to embodiment B1. [Figure 175]FIG. 10 is a diagram showing a display example of a virtual trading model curve according to embodiment B2. [Figure 176] 10 is a flowchart showing the processing of the advice presentation system according to embodiment B2. [Figure 177] FIG. 10 is a diagram showing an example of the continuity of advice according to embodiment B1. [Figure 178] FIG. 10 is a diagram showing a series of flows of advice generation processing according to embodiment B3. [Figure 179] FIG. 10 is a diagram showing a framework for generating advice according to embodiment B3. [Figure 180] FIG. 10 is a diagram showing an example of the configuration of a database of brands or commodities according to embodiment B3. [Figure 181] FIG. 10 is a diagram showing an example of the configuration of various data according to embodiment B4. [Figure 182] FIG. 10 is a diagram showing an example of an initial screen of a stock investment simulation. [Figure 183] FIG. 10 is a diagram showing an example of a question screen for a stock investment simulation. [Figure 184] FIG. 10 is a diagram showing the transition of stock prices in a stock investment simulation. [Figure 185] FIG. 10 is a diagram showing the progress of the valuation amount for each branch of each question in the stock investment simulation. [Figure 186] 10 is an example of a table showing the relationship between the percentage of correct answers to each quiz and the total amount of assets after virtual trading. [Figure 187] 10 is an example of a graph showing the relationship between the percentage of correct answers to a quiz in each session and the total amount of assets after virtual trading. [Figure 188] 10 is an example of a sequence diagram illustrating a processing flow of a buying and selling simulation. [Figure 189] 10 is an example of a screen for executing a trading simulation. [Figure 190] 10 is an example of a screen for executing a trading simulation. [Figure 191] 10 is an example of a screen for executing a trading simulation. [Figure 192] 10 is an example of a screen for executing a trading simulation. [Figure 193] 10 is an example of a general quiz execution screen. [Figure 194] 10 is an example of a screen for executing a case study quiz. [Figure 195] 10 is an example of a screen showing a case study quiz execution screen and registered account details. [Figure 196] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 197] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 198] This is a conceptual diagram showing the cooperation between the Stock Investment Learning Navigator, the Stock Investment Success Navigator, and the Stock Investment Diagnostic Navigator. [Figure 199] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 200] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 201] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 202] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 203] 10 is an example of a flow relating to a stock simulation. [Figure 204] 10 is an example of a flow for a case study quiz. [Figure 205] This is an example of a flow for past and present advice on NISA accounts. [Figure 206] This is an example of a flow for advising on selling a NISA account. [Figure 207] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 208] FIG. 2 is a block diagram illustrating an example of a configuration of a terminal and a server. [Figure 209] This is an example of a flow for supporting the purchase of a NISA account. [Figure 210] This is an example of a flow for advice on a NISA account when there is a purchase record. [Figure 211]FIG. 10 is a diagram for explaining mutual cooperation of the respective embodiments. [Figure 212] This shows the configuration of a system that allows you to ask questions in natural language and receive answers in natural language. [Figure 213] 1 is a hardware configuration diagram of an investment advice system in which questions are asked in natural language and answers are obtained in natural language. [Figure 214] FIG. 1 is an explanatory diagram illustrating a messaging API. [Figure 215] FIG. 1 is a diagram for explaining the process from when a question written in natural language on a terminal is processed in a server to when a response from the server is displayed in natural language on the terminal. [Figure 216] FIG. 1 is a diagram for explaining a machine learning model. [Figure 217] FIG. 10 is a diagram illustrating a process in which a user asks a first question, gives a first answer, and then asks a second question and receives a second answer. [Figure 218] FIG. 1 is a diagram illustrating the process from placing a buy order using natural language information to reporting the results. [Figure 219] It explains the process of monitoring price fluctuations of specific stocks and placing orders when certain conditions are met. [Figure 220] It represents a collection of investment services called the comprehensive investment service system. DETAILED DESCRIPTION OF THE INVENTION

[0014] [Embodiment 1] (Investment product information provision systems 200, 201) Fig. 1 is a block diagram showing the configuration of an investment product information provision system 200 according to the first embodiment of the present invention. Fig. 2 is a block diagram showing the configuration of an investment product information provision system 201 according to the first embodiment of the present invention.

[0015] The investment product information provision system 200 is a system that generates information on dynamic changes in investment products, and includes a terminal 2a and an information provision device 101a. The information provision device 101a may be formed by a single server computer or multiple network computer systems. The holders are not limited.

[0016] The investment product information providing system 201 generates advice, evaluation of trading data, and information on price fluctuations in response to dynamic changes resulting from investment product information and investors' trading behavior, and provides dynamic change data on investment products. As shown in FIG. 1, the conversion device and storage unit may constitute part of the information providing device 101a. As shown in FIG. 2, the conversion device 102b and storage unit 103b may be configured as devices separate from the information providing device 101b.

[0017] Furthermore, in the investment product information provision system 200 according to this embodiment, the information provision device 101a may generate price fluctuation prediction data for investment products. This processing is performed by the investment product price fluctuation prediction system 1 of embodiment 2 (see FIG. 3) or the like. The information provision device 101a may generate evaluation, comparison, ranking, diagnosis, and advice for trading data. This processing is performed by the advice presentation system 1 of embodiment A1 (see FIG. 54) or the like. The information provision device 101a may generate a virtual trading system or the like. This processing is performed by the virtual trading system (stock investment simulation) 1 of embodiment A2 or the like. The information provision device 101a may generate investment type-specific diagnosis or the like. This processing is performed by the investment type-specific diagnosis system of embodiment A3 or the like. The information provision device 101a may generate evaluation and diagnosis of trading data or the like. This processing is performed by the broad trading data evaluation system (information provision system) 1 of embodiment A4 (see FIG. 73) or the like. The information provision device 101a may generate advice data. This process is performed by the advice presentation system 1 of embodiment B1 (see FIG. 165) or the like. The information providing device 101a may generate quizzes, investment diagnosis, etc. This process is performed by the information providing system of embodiment C1 (see FIG. 215) or the like. The information providing device 101a may generate other information generation data for investment products.

[0018] In the above, the information providing device 101a in the investment product information providing system 200 may be replaced with the information providing device 101b in the investment product information providing system 201.

[0019] [Embodiment 2] The investment product price fluctuation prediction system (commodity price estimation system, investment advice system) 1 according to the second embodiment of the present invention will be described in detail below. Note that the system configuration, flowchart, advice results, prediction results, etc. shown below are merely examples and do not limit the present invention. Note that the investment product price fluctuation prediction system 1 is included in the investment product information provision systems 200 and 201. In other words, the investment product price fluctuation prediction system 1 realizes part of the functions of the investment product information provision systems 200 and 201.

[0020] (Investment product price fluctuation prediction system 1) FIG. 3 is a block diagram showing the configuration of an investment product price fluctuation prediction system 1 according to this embodiment. As shown in FIG. 3, the investment product price fluctuation prediction system 1 includes an information processing device 101, a conversion device 102, a database unit 103, and a terminal 2. The information processing device 101 and the terminal 2 can communicate with each other via the Internet 3. The information processing device 101, the conversion device 102, and the database unit 103 may be configured together on a single server, or may be distributed across multiple servers. The investment product price fluctuation prediction system 1 of this embodiment 2 handles information on investment products and, in particular, estimates the prices of investment products.

[0021] The investment product price fluctuation prediction system 1 can estimate future prices of multiple price-fluctuating products, providing very useful information for investors. Therefore, a service provider offering an investment advice service can provide users with estimated future prices of multiple price-fluctuating products obtained from the investment product price fluctuation prediction system 1 as an investment advice service, either for a fee or free of charge. Specifically, this information is provided to unspecified or specific users via a communication network such as the Internet, e-mail, a chat application (LINE), voice calls, etc. To provide useful information to specific users, the service provider needs information that can uniquely identify the user providing the service. Specifically, this information includes a phone number, e-mail address, LINE account information, etc.

[0022] For example, when creating a LINE (registered trademark) account, the phone number (mobile phone number or landline phone number) of the newly registered user is always authenticated. The phone number is an example of identification information, which is unique information for identifying an individual or user in advance, in the present disclosure. When creating an actual LINE account, the user is first asked to enter a phone number (mobile phone number or landline phone number). An "authentication number" is sent to the entered phone number via SMS (short message service) in the case of a mobile phone number, or via automated voice guidance over the landline in the case of a landline phone number, and the account can only be created by entering this number on the account creation screen. By checking that the phone number (mobile phone number or landline phone number) of the newly registered user is correct, it is possible to avoid false registrations using fictitious phone numbers (mobile phone numbers or landline phone numbers). Details of the aforementioned LINE will be provided later.

[0023] An investment service provider can provide investment services to a pre-specified user, member, or individual through two-way communication via a terminal 2 used by the unique user, member, individual, etc., who has been pre-specified by unique information such as a unique number, unique code, or telephone number (mobile phone number or landline number). Also, through a unique terminal 2 pre-specified by unique information such as a telephone number (mobile phone number or landline number) and on which a chat application such as LINE has been installed, an investment service provider can provide investment services to the pre-specified user or individual by utilizing information, data, and content stored in server 3 through almost real-time chat-style two-way communication.

[0024] Furthermore, a user or individual is identified in advance by a telephone number (mobile phone number or landline number), which is unique information, and through a terminal 2 on which a chat application such as LINE is installed, investment advice services, commodity price estimation services, information provision system 200, investment advice system 1, information provision device 101a, communication methods, commodity price estimation methods, advice provision methods, information analysis methods, information provision methods, etc. provided by an investment service provider can be provided to the user or individual whose information has been identified in advance, for a fee or free of charge, through near real-time two-way communication.

[0025] The information processing device 101, the conversion device 102, and the database unit 103 are provided in a server computer managed and operated by an investment product information provider, etc. The server computer has a web server function and a database function for storing large amounts of data.

[0026] Terminal 2 is a communication terminal with data communication capabilities that is owned and used by an individual or corporation that receives information about investment products, and is a personal computer, tablet, smartphone, mobile phone terminal, or the like.

[0027] A smartphone is a type of mobile phone and is a portable electronic device with advanced features. Smartphones can access the Internet, send and receive email, take and edit photos and videos, and use GPS to find their location. Smartphones are often operated using a touchscreen or dedicated buttons, whereas older mobile phones are operated using physical buttons. Smartphones can also be used to download and install many applications. These apps vary in function and purpose. Smartphones can use external storage in addition to their built-in storage.

[0028] Smartphones have built-in software called an OS (operating system), which controls the smartphone and enables it to run applications and provide and add functions. Typical smartphone OSes include the following: Android: An open-source smartphone operating system developed by Google, it is the most widely used worldwide. Many applications can be downloaded from the Google Play Store. Android smartphones are manufactured by manufacturers such as Google, Sony, and others. iOS: A smartphone operating system developed by Apple that is installed on the iPhone, iPad, and iPod touch. Many applications can be downloaded from the App Store. Windows Phone: A smartphone operating system developed by Microsoft, integrated into Windows 10 Mobile. Many applications can be downloaded from the Windows Store. BlackBerry OS: A smartphone operating system developed by BlackBerry, it has strong security features. Many applications can be downloaded from BlackBerry World. Tizen: A smartphone operating system developed by Samsung, primarily installed on Samsung smartphones. Many applications can be downloaded from Samsung Galaxy Apps.

[0029] Like a smartphone or a mobile phone, terminal 2 is equipped with a notification unit 26 that has an incoming call notification function. When information such as investment product advice is received from server 3 as an email or LINE chat, this incoming call notification function may be configured to notify those nearby that the investment advice or other information is present by sounding a ringtone from a built-in speaker, vibrating from a vibration generating source (motor, electromagnetic plunger) built into terminal 2, changing the brightness or blinking of a display means built into the terminal device, or blinking or lighting up a specific light-emitting source such as an LED. Furthermore, when the investment advice or other information is displayed on display unit 23, a message indicating that the information has been read is displayed on the screen.

[0030] The incoming call notification function from the notification unit 26 is important for investment support service providers and users, members, and individuals who receive investment support services. The incoming call notification function of the terminal 2 can shorten the gaps that occur between two-way communications between investment support service providers and users, members, and individuals who receive investment support services. This shortening effect is particularly important in two-way communications regarding financial products and investment products, where speed is important.

[0031] Smartphones, mobile phones, and tablet devices are portable and can be carried to various locations. Such portable and portable terminal 2 is particularly important in the invention disclosed herein. Similar to the incoming call notification function of terminal 2, it can shorten the time distance and gap between two-way communications between an investment support service provider and a user, member, or individual receiving the investment support service. This shortening effect is particularly important in the commodity price estimation system 1, information provision system 200, investment advice system 1, information provision device 101a, communication method, commodity price estimation method, advice provision method, information analysis method, and information provision method, in which speed is important for financial products and investment products.

[0032] The terminal 2 includes an operation unit 21 consisting of a mouse, keyboard, etc. and used to input various instructions, a display unit 22 consisting of an LCD (Liquid Crystal Display), etc. and displaying various instructions input from the operation unit 21 and various images acquired from the communication unit 23, and the communication unit 23 exchanging data via the Internet 3. However, the operation unit 21 and the display unit 22 may be configured as a touch panel or the like that is operated by the operator touching or moving a finger, a touch pen, etc. on the screen.

[0033] Terminal 2 is also provided with a microphone, allowing voice information to be input to control unit 24. Control unit 24 is equipped with a voice recognition function that recognizes voice information and converts it into text information. Therefore, information input by voice can be converted into text information and output to the outside. Terminal 2 is also provided with a speaker, allowing voice output.

[0034] The terminal 2 may be provided with a function for converting text information into audio information. In this case, information obtained from the outside as text information can be displayed on the display unit 22 of the terminal 2 as text information.

[0035] 4 is a block diagram showing the hardware configuration of the investment product price fluctuation prediction system 1 according to this embodiment. As shown in FIG. 4, the information processing device 101, the conversion device 102, and the database unit 103 each include at least one CPU (Central Processing Unit), a RAM (Random Access Memory) that functions as a work area for the CPU, a startup memory (RAM), and a The computer is equipped with a ROM (Read Only Memory) that stores boot programs for the computer, an auxiliary storage device such as a hard disk that stores various programs and data, and a communication interface used for sending and receiving data. The auxiliary storage device stores programs for the OS (Operating System), various application programs, database data, etc. These programs and data work in cooperation with hardware resources through the calculations of the CPU to realize various functions.

[0036] As shown in Figure 3, the investment product price fluctuation prediction system 1 has an information processing device 101 and a conversion device 102 as functional means realized based on the various programs and hardware resources described above, and a database unit 103 in which various data processed by the information processing device 101 and the conversion device 102 are recorded.

[0037] 5 is a functional block diagram of the information processing device 101 and the conversion device 102 according to this embodiment. As shown in FIG. 5, the information processing device 101 and the conversion device 102 perform processes such as generating and processing various data used in the investment product price fluctuation prediction system 1.

[0038] The conversion device 102 includes an event input reception unit 12 as an "event information reception unit", an investment product price input reception unit (price data acquisition unit) 13, a conversion unit (price fluctuation data generation unit) 14, a database connection unit 15, an authentication unit 24, a billing unit 25, an other input reception unit 26, a correlation unit 27, a time lag unit (delay time generation unit) 28, an optimization unit 29, and a relationship table creation unit 30.

[0039] The information processing device 101 includes a learning model generation unit 16 as a "learning generation unit", a learning unit 17 as a "learning unit of the learning model", an estimation unit (price estimation unit) 18, a parameter update unit 19, a program generation unit 20, an other input acceptance unit 26, a database connection unit 31, a distribution unit 32, and a table reference unit 33.

[0040] The event input receiving unit 12 receives data related to various events, such as statistical data input or distributed from a server, a terminal, the Internet, a storage medium, etc., and performs various processes required to process the relationship with price fluctuations of investment products. In this embodiment, the event input receiving unit 12 periodically acquires information related to events.

[0041] (Event definition) An event refers to something that is occurring at a given time, particularly an event that affects the price of an investment product. The date indicates the date on which the event occurred. Redefining an event, it includes all indicators mentioned in the embodiments, such as technical indicators and performance indicators. It includes all factors that affect the price fluctuations of an investment target. Investment actions such as buying and selling are also considered events.

[0042] The price input receiving unit 13 receives price information of investment products from a server, a terminal, the Internet, a storage medium, etc. The price input receiving unit 13 acquires information about the prices of financial products handled by the investment product price fluctuation prediction system 1. The acquired information is used by the information processing device 101, the conversion device 102, the database unit 103, and the terminal 2. In this embodiment, the price input receiving unit 13 periodically acquires price data of investment products at a predetermined time point.

[0043] The conversion unit 14 performs processes such as converting various time series data into change data in order to store the information received by the event input reception unit 12 and the information received by the investment product price input reception unit 13 in the database unit 103. The conversion unit 14 also performs processes such as processing input information received from the user by the other input reception unit 26.

[0044] The conversion device 102, as an event identification unit, performs processing to identify at least one of a technical indicator, a corporate earnings announcement, a macro indicator announcement, a rating change, and a milestone event based on the past price of an investment product. The conversion unit 14 generates price fluctuation data of the investment product up to a predetermined point in time for the identified event.

[0045] The correlation unit 27 calculates an index or the like that represents the correlation between various combinations of the change data processed by the conversion unit 14. For example, in the case of investment product price fluctuation data, there are data on year-on-year comparisons, month-on-month comparisons, day-on-day comparisons, monthly, weekly, and daily charts, and calculates the correlation coefficient for each of the event change data. This allows for a multifaceted view.

[0046] The time lag unit 28 optimizes the time lag in order to determine the time lag (delay time) between the change data combined in the correlation unit 27. The time lag unit 28 generates the time lag from the occurrence of an event to the change in the price of the investment product by, for example, keeping the date of the event change data as it is and shifting the date of the price fluctuation data of the investment product later to find the optimal correlation coefficient.

[0047] The optimization unit 29 optimizes the combination of event changes and investment product price fluctuation data. In the correlation unit 27, even when a certain event and a certain investment product are combined, there are various elements such as change rate data, time series data, daily data, monthly data, and year-on-year comparison, and the correlation coefficient changes depending on the combination of these elements. The optimization unit 29 finds the combination with the correlation coefficient closest to 1. The optimization unit 29 may also optimize the time lag.

[0048] The relational table creation unit 30 stores the results of processing by the conversion device 102 in an event change data table, an investment product price fluctuation data table, a join table, a teacher data table, a correlation table, a date conversion table, and the like.

[0049] The database connection unit 15 is connected to the database unit 103 in order to store the data processed by the conversion unit 14, correlation unit 27, and time lag unit 28 in the database.

[0050] The learning model generation unit 16 generates a learning model for estimating the price of an investment product.

[0051] The learning unit 17, which serves as a "learning unit for the learning model," further learns the relationship between each event and investment product price fluctuation data, etc., based on the correlation and time lag between each event and investment product price fluctuation data, etc., which are training data stored in the database unit 103. The learning unit 17 trains the estimation model by using, for example, change data related to the event and price fluctuation data of the investment product as training data. Events handled here include corporate performance change events, technical indicator events, macro indicator change events, milestone events, rating events, etc. The learning unit 17 learns the relationship between each event and investment product price fluctuation data, etc. Note that the learning unit 17 may also learn the relationship between events.

[0052] The estimation unit 18 performs estimation processing based on the learning model generated by the learning unit 17 and the input data. The estimation unit 18 transmits the completed learned model to the terminal 2 and causes the terminal 2 to execute a program for making estimation. The estimation unit 18 estimates the price of the investment product from a predetermined point in time onward, for example, based on price data at a predetermined point in time and price fluctuation data.

[0053] The parameter update unit 19 updates the parameters based on the changing information and performs processing to update the trained model. This may include processing to send update data for the trained model (program) to the terminal 2 or the like.

[0054] The program generation unit 20 generates the learning model generated by the learning unit 17 or a completed learning model, and performs processing to send it to the terminal 2.

[0055] The table reference unit 33 references the relationship between the event change data and the investment product price fluctuation data based on the database unit 103, performs processing such as extraction, and performs processing to reference and output the necessary information. The table reference unit 33 extracts the necessary information from the database unit 103 based on the input information from the terminal 2 received by the other input receiving unit 26. The events handled here include corporate performance change events, technical indicator events, macro indicator change events, milestone events, rating events, etc., and reference tables such as a binding table between event changes and investment product price fluctuations, a correlation table, and a time lag table play an important role.

[0056] The other input accepting unit 26 accepts other information from the server, the terminal 2, the Internet, a storage medium, etc. The other input accepting unit 26 acquires input information from users handled by the investment product price fluctuation prediction system 1 (including membership information, trading data information, form input information, product purchase information, etc.). The acquired information is used by the information processing device 101, the conversion device 102, the database unit 103, and the terminal 2. When the input information is to be formatted before being sent to the database unit 103, the conversion device 102 accepts the input. In this embodiment, the other input accepting unit 26 periodically acquires user information, etc.

[0057] The database connection unit 31 converts data generated and processed in the information processing device 101 and data stored in the database unit 103 (for example, converts between a logical data structure and a physical data structure). The database connection unit 31 also performs processing required for mutual communication of data between the information processing device 101 and the database unit 103.

[0058] The database unit 103 stores data used by the information processing device 101 and the conversion device 102. In this embodiment, the database unit 103 is configured as a relational database, but any format suitable for storing and rewriting large amounts of data, such as an object database, may be used. The database unit 103 stores an "event change table 181" as an "event information recording unit," an "investment product price fluctuation table 182" as an "investment product price information recording unit," a "correlation table 183" as a "correlation recording unit between the two," a date conversion table 184, and the like. Details of the event change table 181, the investment product price fluctuation table 182, the correlation table 183, and the date conversion table 184 will be described later.

[0059] The distribution unit 32 creates output data to be output to the display unit 22 of the terminal 2 , and transmits the created output data to the terminal 2 .

[0060] The authentication unit 24 checks the "product purchase information" and "ID" contained in the customer table. With the authentication result, the authentication unit 24 enables operation of the My Page on the terminal 2. Paid information data is sent to the terminal 2 to be displayed on the display unit 22 of the terminal 2.

[0061] The accounting unit 25 associates user identification information corresponding to a user with one or more purchase item identifiers that identify purchase items purchased by the user. The user purchase information resulting from the association indicates purchase items previously purchased by the user.

[0062] The information processing device 101, conversion device 102, database unit 103, terminal 2, event input reception unit 12 as an "event information reception unit", investment product price input reception unit 13, conversion unit 14, database connection unit 15, learning model generation unit 16 as a "learning generation unit", learning unit 17 as a "learning unit of the learning model", estimation unit 18, parameter update unit 19, program generation unit 20, table reference unit 33, database connection unit 31, distribution unit 32, authentication unit 24, billing unit 25, other input reception unit 26, correlation unit 27, time lag unit 28, optimization unit 29, and relationship table creation unit 30 in the investment product price fluctuation prediction system may be configured in any form.

[0063] For example, if the investment product price fluctuation prediction system 1 is configured as a network computer system consisting of multiple server computers, each component may be distributed across multiple server computers. Furthermore, these server computers may be distributed across computer systems owned by multiple traders or server administrators (e.g., computer systems owned or managed by financial securities exchanges, other financial product transaction management companies, providers, hosting system providers, etc.). The server computers may be configured as a so-called cloud computer system.

[0064] Furthermore, at least some of the components of the information processing device 101, conversion device 102, database unit 103, terminal 2, event input reception unit 12 as an "event information reception unit", investment product price input reception unit 13, conversion unit 14, database connection unit 15, learning model generation unit 16 as a "learning generation unit", learning unit 17 as a "learning model learning unit", estimation unit 18, parameter update unit 19, program generation unit 20, table reference unit 33, database connection unit 31, distribution unit 32, authentication unit 24, billing unit 25, other input reception unit 26, correlation unit 27, time lag unit 28, optimization unit 29, and relationship table creation unit 30 may be provided in the terminal 2 that constitutes the investment product price fluctuation prediction system 1, rather than in the investment product price fluctuation prediction system 1.

[0065] (Processing of investment product price fluctuation prediction system 1) FIG. 6 is a flowchart showing the processing of the investment product price fluctuation prediction system 1 according to this embodiment.

[0066] Logging in is a procedure required to access various computer systems, websites, services, web communications, etc. described in this specification. When a user or individual enters correct unique authentication information (account information, ID information, password information), the system operator, service provider, or website compares the authentication information with information previously held by the system operator, service provider, or website. If the information matches, the system operator, service provider, or website recognizes that the correct user or individual is accessing the system, officially authenticates the user or individual, and grants them access rights. Logging in has the following functions: Security: Logins are used to verify the identity of users and individuals and protect confidential and personal information. Account management: By using login, the system and various service providers manage user and personal account information, access history information, service billing information, profile information, etc. Personalized experience: By using login, users can access personalized services, personalized information, and use their favorite features.

[0067] Generally, various system operators, service providers, or websites can provide services and grant access to logged-in users and individuals from the time they log in until they log out, as well as manage the user's or individual's behavioral history and activity time, charge users or individuals, and make system improvements.

[0068] The login and logout steps are processes that are always performed in the commodity price estimation system 1, information provision system 200, investment advice system 1, information provision device 101a, communication method, commodity price estimation method, advice provision method, information analysis method, and information provision method disclosed in this specification.

[0069] 6, "start" can be read as logging in and "end" as logging out. The processing of the investment product price fluctuation prediction system 1 will be described according to the flow of FIG.

[0070] (S1001) The terminal 2 acquires input items such as question items of the target user, and transmits data indicating the input items to the conversion device 102 (or the information processing device 101).

[0071] (S1002) The conversion device 102 receives various data from the terminal 2 and other companies' servers. The conversion device 102 processes the received data in the conversion unit 14, and performs processing to store the processed data in the database unit 103.

[0072] (S1003) The database unit 103 receives various data from the conversion device 102 or the information processing device 101 and transmits the received data to the information processing device 101 .

[0073] (S1004) The information processing device 101 processes various data and transmits the processed data to the terminal 2.

[0074] (S1005) The terminal 2 receives various data from the information processing device 101 and displays the data on the display unit 22.

[0075] (Relationship between event changes and event occurrences) For example, if an announcement is made that crude oil prices will rise 30% year-on-year, this means that an "event change" has been announced. An event change refers to a state that has changed, such as when compared to the previous state. On the other hand, an event occurrence is an event such as a 30% increase in crude oil prices year-on-year, or point-in-time information such as crude oil prices being $50 this month (not including change information). Event occurrence also includes events that do not include change information. Change information is the most important information when predicting price fluctuations of investment products. For this reason, the two are defined separately.

[0076] (assignment) The challenge is how to obtain information such as which stocks an event change will have a positive effect on, how much the price will fluctuate three months from now, and what will happen one month from now.

[0077] (Means for solving the problem) There are various methods, such as a table reference method in which event change information and price fluctuations of investment products are recorded in the database unit 103, and an AI method that uses machine learning to understand the relationship between event changes and price fluctuations of investment products. However, this also includes cases in which the event is not change information. The change period may be three months or one week, and may be changed depending on the event or investment product. Multiple periods such as one week, one month, or three months may always be used, or a period may be flexibly selected by incorporating time series data.

[0078] In either case, the event change information and investment product price fluctuation information contain change information, and this information is stored in the event change table and investment product price fluctuation table. Correlation tables and the like play a role in organizing the relationship between the two.

[0079] (effect) The relationship between event changes and price fluctuations of investment products becomes clear, which is expected to have various benefits for investors. For example, if a six-month high is updated (a type of information on price changes from six months ago), predictions will be made on how to act when the event occurs and how much price fluctuation there will be. If figures such as an 80% probability that the price will fluctuate by -5% in three months' time are obtained, this can be very useful for trading.

[0080] (Definition of Investment Product Price Fluctuation Forecasting System 1 (Table Reference Method)) If we break down embodiment A2 and embodiment A4 in more detail, it is the relationship between event changes and price fluctuations of investment products. This is embodiment 2, and the aggregation of these is embodiment 3. Here, aggregation refers to the flow in which a first event spreads to a second event, which then spreads to a third event, ultimately leading to price fluctuations of investment products. First of all, the relationship between event changes and price fluctuations of investment products is an important item. By storing, organizing, and learning this, it becomes possible to predict price fluctuations of investment products when an event occurs.

[0081] The estimation unit 18 estimates the price of the investment product after a predetermined time point based on the change data related to the event. In particular, the estimation unit 18 associates the change data related to the event with the price fluctuation data of the investment product to estimate the price of the investment product. This can be done by managing the data in a table and referencing the table, or by using AI through machine learning. First, the table reference method will be explained.

[0082] (Relationship table creation unit 30) (assignment) The occurrence of events is announced through regular news, press releases, statistical publications, corporate earnings announcements, and rating announcements. Events are dynamic data that change dynamically, and are primarily time-series data, but there is also single-point data and year-on-year data. Price fluctuation data for investment products is also published as time-series data by stock exchanges and other institutions. Technical indicator values and milestones can also be calculated based on this data.

[0083] Event data includes both point-in-time data and time-series data. Event changes, along with fluctuations in investment product prices, are essentially dynamic data, but static data also exists. To measure the impact of event changes on investment product price fluctuations, at a minimum, the event details, event change information (or the price of the event at two points in time), and investment product price fluctuation information (or the price of the investment product at two points in time) are required. Because these are managed separately, there is an issue in that it is not possible to measure the impact of event changes on investment product price fluctuations.

[0084] Fig. 7 is a diagram illustrating the time lag and year-on-year comparison data according to this embodiment. Fig. 8 is a diagram illustrating a method for incorporating event data and investment product price fluctuation data according to this embodiment.

[0085] For example, even if stock price data is available, data on crude oil price changes and data on investment product price fluctuations are managed separately, making it impossible to build a relationship. Furthermore, in the case of year-on-year comparison data, it is necessary to convert it into data for two points in time, as shown in Figure 7, and then perform the appropriate processing, as shown in Figure 8.

[0086] (Means for solving the problem) To clarify the relationship between event changes and price fluctuations of investment products, the relationship table creation unit 30 creates a relationship table that shows these relationships. There are several ways to create a relationship table, but it is necessary to link the relationship between event changes and investment product price fluctuations.

[0087] There are methods for linking by date, or by date and stock, and a relational table can be created by associating an event change table with an investment product price fluctuation table. This is possible by linking the event table with the investment product price data by date and stock (including stock code, etc.), or by date and related stock, or by date, and preparing a combined table of event changes and investment product price fluctuations, which is a table that incorporates information about the date and time of the event occurrence and change, the price of the investment product on that date, and the price of the investment product a certain date and time after the event occurrence. Combined tables of event changes and investment product price fluctuations can be divided into several types depending on the method of linking.

[0088] (Linking date and stock) As shown in the upper diagram of Figure 8, linking a date and a stock means linking the event occurrence date and investment product in the event table with the date and stock of the investment product in the investment product price data table. The stock code and its date are linked to the event content (crude oil prices rising 30% year-on-year) and the stock price on that date. For example, the stock price from one year ago is linked.

[0089] As shown in the middle diagram of Figure 8, the data structure includes date, event content, event data, comparison target, brand, etc. This makes it easy to manage by brand.

[0090] As shown in the bottom diagram of Figure 8, the correlation between event changes and stock price fluctuations is also easy to manage. As explained in the middle diagram of Figure 8, information on changes in crude oil prices is event data and year-on-year comparisons. On October 1, 2021, crude oil prices rose 30% year-on-year, while stock A's rate of change increased 11%. The format of this table does not matter, as it can be used to build a relationship between these two changes. This is just one specific example, and event changes can also be captured using prices at two points in time, rather than year-on-year comparisons. This includes all methods that can build a relationship between event changes and stock price fluctuations for each stock.

[0091] (Date link method) Figure 9 is a diagram explaining the date linking method. As shown in the upper diagram of Figure 9, linking dates means linking the event occurrence date in the event table with the date in the investment product price data. The date, event content (crude oil prices rise 20% compared to the previous month), and stock price on that date are linked. Because the stock name is not specified, information on other stocks is also included. To compare with the previous month, the stock price from one month ago is added.

[0092] The bottom diagram in Figure 9 shows a table linking event changes with investment product price fluctuations. As shown in the bottom diagram in Figure 9, the data structure includes the date, event content, event data, comparison target, and stock. The event change (a 20% increase in crude oil prices compared to the previous month) is linked to an 11% increase in stock price for stock A and a 29% increase in stock price for stock B, and multiple stocks are included, including stocks with no correlation. Linking is only by date, making it easy to compare stocks and products for a certain event change. This is expected to have the effect of making clear the correlations between each.

[0093] As shown in the upper diagram of Figure 9, for example, it is expected that changes in crude oil prices can be easily compared between brands A, B, C, and D. The relationship between brand A and crude oil price data is managed in a correlation table, so it can be easily extracted.

[0094] (Linking date and related stocks) Figure 10 is a diagram showing how to link dates with related stocks. As shown in the upper diagram of Figure 10, the date, the event content (crude oil prices rose 20% the previous month), and the stock price on that date are linked. Because related stocks are specified, only related stocks are included. Linking dates with related stocks means linking the date in the event table with the date in the investment product price data table, and linking the related stocks in the event table with multiple stocks in the investment product price data table in a one-to-many relationship. An event-related stock table may also be created.

[0095] The lower diagram in Figure 10 shows the structure of a join table between event changes and investment product price fluctuations. As shown in the lower diagram in Figure 10, the join table has the following data structure items: event content, related code, event data, comparison target, stock, stock price one month before the event occurred, stock price on the day the event occurred, and rate of change. Specifically, the join table contains an event change (a 20% increase in crude oil prices compared to the previous month) and multiple stocks, with stock A's stock price increasing by 11% and stock B's stock price increasing by 25%.

[0096] For events such as corporate earnings events, technical indicators, ratings, and milestones, the link between stocks and dates is the primary method, clarifying the relationship between event changes and investment product price fluctuations for each stock, date, and event. For macro indicators, since there are typically multiple related stocks, the link between related stocks is the primary method. When a macro indicator changes, related stocks can be quickly pulled in, providing a special benefit in that it becomes easier to select the best stock. Furthermore, in cases with a large overall impact, such as President Trump's victory (also a type of event change information), linking by date allows for comparison and ranking of the impact on the price fluctuations of each investment product. It is used to identify related stocks or stocks with strong correlations. When it comes to investment product price fluctuation data, how to incorporate price data a certain period after the event occurs is important. A typical method involves determining the date and stock, then incorporating the daily price of the investment product starting from that date. However, this method includes all methods for examining the relationship between event change information and investment product price fluctuation information.

[0097] (effect) What can be done by creating a join table? By accumulating data in a join table of event changes and investment product price fluctuations, such as those shown in Figures 8 to 10, we can expect to see a special effect: we can always extract how the price of each investment product has changed in response to event changes such as changes in macro indicators or corporate performance.

[0098] For example, the stock price change rate of stock A corresponding to event changes such as the crude oil price rise event and grain price rise event of stock A in Figure 8 is stored in the database unit 103. This clarifies the relationship between the event change and stock price increase rate of stock A, and is expected to have the special effect of making it possible to clarify correlations using correlation coefficients and least squares methods. It also makes it easier to calculate time lags. It is expected to have the effect of clarifying that event changes closely related to stock price increase rate of stock A have a strong correlation with grain events in addition to crude oil, and it is expected that information on events closely related to stock price of stock A will be dramatically increased. Displaying this information on terminal 2 is expected to improve the quality of advice and strengthen support capabilities.

[0099] As shown in Figure 9, by linking various stock information to dates, we can expect to see a clear correlation between the information and event changes. For example, large differences can be seen between stocks in response to rising crude oil prices. In this example, stocks B and C appear to have a strong correlation, stock A appears to have a weak correlation, and stock D appears to have a negative correlation. Of course, it is difficult to make a judgment based on just this fixed period of time, but the more data we collect, the more statistically significant the data becomes. In this case, it becomes clear that each stock responds differently to event changes, which can be expected to have special benefits, such as providing basic data for advising holders of stock D to switch to stock B, for example.

[0100] Figure 10 extracts only related stocks, which has the special effect of allowing comparison of crude oil-related stocks. Figure 9 requires filtering, as it also includes negatively correlated and unrelated stocks. On the other hand, Figure 10 is already filtered from the start, extracting only stocks that are closely related to crude oil price change events, making it possible to compare them side by side. In particular, for stocks with a high degree of correlation, a special effect can be expected, in that it can present top-picked stocks (best-selected stocks).

[0101] In addition, determining the time lag is a very important process, just like calculating correlation. It is best to calculate the relationship between each event change and stock price fluctuation, but the importance of technical indicators, which do not have a time lag, changes depending on the event change, so the correlation calculation process is not essential.

[0102] (Event data definition) Event data refers to various numerical values that occur when an event occurs. Event data includes investment product prices such as stock prices, technical indicators such as RSI, profit growth rates, macroeconomic indicators such as semiconductor production volume, milestone prices such as year-to-date highs, and rating prices. Event occurrences and changes in event data can be paired (e.g., the RSI goes from 89% to 90%, causing the RSI to reach 90%, resulting in a year-on-year increase in semiconductor production). Others are unpaired (e.g., the voting date of the presidential midterm elections), and time series data (e.g., crude oil price time series). Some kind of event is always occurring, and when an investor makes a purchase (e.g., buying 100 shares of stock A for 1,000 yen), if it coincides with another event, then the purchase event coincides with, for example, an upward revision of earnings, resulting in multiple events. If an investor's purchase events always coincide with upward revision events, then it can be concluded that the investor's purchasing decisions are closely related to the upward revision events. Adding to this the event data of an upward revision rate of 30% or more on an operating profit basis provides even more useful information.

[0103] (Relationship between changes in events and fluctuations in investment product prices) FIG. 11 is a diagram showing the relationship between events and investment product prices.

[0104] There is a close relationship between event changes and investment product price fluctuations. However, depending on the content of the event, there are many changes that do not affect investment product price fluctuations, and the relationship is complex, with changes depending on the brand, the degree of impact on price fluctuations, and the relationship differing for each brand. The relationship between temporary event information such as crude oil prices reaching $50 and investment product price fluctuations is not fixed, but event change information such as crude oil prices rising 50% year-on-year and investment product price fluctuation information is closely related, depending on the brand. Therefore, it is important to clarify the relationship between event changes and investment product price fluctuations, rather than simply the relationship between events and investment product prices.

[0105] The top four examples in Figure 11 show the relationship between events and investment product prices. As shown in Figure 11, time series data of course includes information such as a crude oil price of $30 in June 2020 and a crude oil price of $60 in June 2021, and also contains information about changes. While information about changes is undoubtedly important, this does not mean that the other three relationships (the top three relationships in Figure 8) should be ignored. In some cases, these three relationships can also become important.

[0106] (Issues with stock price prediction systems) For example, an upward revision of a company's earnings is well known to investors. Such events attract significant attention, and when announced on the same day as financial results, they often result in large monetary movements. This event is an event resulting from a change in a company's forecast. Other examples include technical indicator events. A notable change is when the RSI indicator goes from 11% to 10% and then falls below 10%. A breakthrough in a high price is also a change event, surpassing a previous high. Ratings also see many change events, such as changes in target stock prices, from sell to buy, and forecast revisions. Despite the large number of events, there is a lack of research into how these events ultimately affect investment product prices. Prices do not rise simply because of favorable news; various factors are involved, and the mechanisms of price fluctuations are extremely complex. To clarify the relationship between these two, a challenge remains: how to manage databases and other tools to predict the date of an event.

[0107] (Means for solving the problem) As shown by the ◎ in Figure 11, it is necessary to compare changes with changes. As in the top example in Figure 8, comparing a single event with the price of an investment product at a single point in time is meaningless in most cases. Furthermore, rather than focusing on comparing a single event with the price fluctuations of an investment product, or a change in an event with the price of an investment product at a single point in time, it is important to contrast the change in an event with the price fluctuations of an investment product. This means that predictions become possible. For example, if there is a relationship between an event at time A, an event at time B, the price of an investment product at time C, and the price of an investment product at time D, this means that changes in events from time A will affect the price of the investment product. This relationship is particularly important when there is a time lag. Figure 11 shows cases with and without a time lag.

[0108] (effect) Explaining Figure 11, we can understand why this prediction is possible. At point A in July 2020, the price of crude oil was $50 and stock A was 500 yen. In October 2020, a change event occurred in which the price of crude oil rose significantly to $80 (a 60% increase). Perhaps in response, stock A also rose to 600 yen (a 20% increase).

[0109] If we try to derive the relationship between the two from this relationship, we will end up lost in a maze. This is because we try to grasp the relationship between the two from the top three relationships in Figure 11. The key point of this invention is to understand the relationship between the two in terms of the relationship between changes in both, including a time lag of three months.

[0110] For example, in the case of Figure 8, where there is a time lag, the price of crude oil at point A in July 2020 was $50, and stock A was priced at ¥500. In October 2020, a significant change occurred: the price of crude oil rose to $80 (a 60% increase). Perhaps in response, stock A also rose to ¥600 (a 20% increase). However, the rise in crude oil prices from July to October 2020 actually had a three-month time lag, and this relationship—which is related to the price rise of stock A from October 2020 to January 2021—unravels the fourth relationship from the top of Figure 8. This is a surprising blind spot, but it is a unique benefit of this invention, which clarifies the relationship between the two changes and the time lag (which can even be zero). Understanding this allows us to predict and infer that stock A would rise as of January 2021, even in October 2020.

[0111] (How to import event change information and investment product price fluctuation data) (Creating a data table of event change information and investment product price fluctuations) So, how can we utilize the relationship between event change information and investment product price fluctuation information? Here, we will explain two methods, one for referencing the information by creating a table and the other for machine learning using AI, as examples.

[0112] First, we will explain how to create and reference a table. For example, Figure 8 illustrates one method for creating a table that combines event changes and investment product price fluctuations. First, investment product price data is imported from the investment product price data and event table using the product code and the event occurrence date as keys. Next, the date one month before the event occurrence date is identified in the date conversion table, the price data for the investment product on that date is imported, and the one-month rate of change or the price data for both dates is imported into the table that combines event changes and investment product price fluctuations.

[0113] This is just one method, and there are many other methods, as described below. The important thing is to incorporate not only price data on the day the event occurs, but also previous price data or information on the rate of change. Furthermore, event data should also incorporate information on changes between two or more points in time (including year-on-year or quarter-on-quarter comparisons). This makes it possible to manage event changes and the associated fluctuations in the price of investment products. If there is a time lag, the time lag between the event change and the price fluctuations of the investment product should be included in the model. Additionally, information that assumes change, such as technical indicators, should be converted into change information (see Figure 7).

[0114] (1) A one-time method that links the event table with the investment product price table and references the price, for example, one month prior to the event occurrence. (2) Once the date and issue are determined, the time series method is used to incorporate the price of the investment product, including past data, starting from that date. (3) An unspecified point-in-time method in which, once the date and issue are determined, the price of the investment product on that date and at that point in time is taken into account. (4) A reference method that references the time-series information in the investment product price data table and the event data table. This includes all methods of capturing price data of investment products at dates and times other than when an event occurs, such as:

[0115] (effect) There are methods for linking stocks with the occurrence of events, but this method is expected to clarify the relationship between event changes, investment products that are affected by price fluctuations due to event changes, and the price fluctuations of those investment products. There are many specific forms that can be created by combining the above event types, linking methods, and methods for importing price fluctuation data. The following are representative examples, and by changing the wording, the combination will change, the issues will also change, and the means to solve them and the effects will also change. For example, specific form 1 is a corporate performance event, but if it is replaced with a rating event, it becomes a rating event (stock and date linking method) and investment product price fluctuations (single point in time method).

[0116] A specific example of this is the "clarification of the relationship between the two" example shown in FIG.

[0117] (Type of event and type of investment product) There are many types of event changes, such as ratings, changes in corporate performance, technical indicators, changes in macro indicators, reaching milestones, etc. The aim is to clarify the relationship between these event changes and price fluctuations of investment products. There are also types of investment products, as listed below.

[0118] (Regarding the conversion device 102, the teacher data generation unit, and the join table generation device) The conversion device 102 is a device that accepts time-series data and processes the data for storage in the database unit 103, and is called by different names depending on whether the method is AI or table lookup. In the case of the AI method, it is called a teacher data generation device, and in the case of the table lookup method, it is called a binding table generation device, and both are defined by different names for the conversion device. Each device includes the conversion unit 102 (which converts time-series data and the like into change data), a correlation unit 27 (which calculates the correlation between event change data and investment product price fluctuation data), a time lag unit 28 (which calculates the time lag between event change data and investment product price fluctuation data), an optimization unit 29 (which determines the optimal combination of correlations), and a relationship table creation unit 30 (which creates correlation tables, binding tables, teacher data tables, etc., that show the relationships between data), and the data is recorded in the database unit 103 after processing by these units.

[0119] (Type of investment product) Not only in this embodiment, but in all descriptions in the specification, the terms can be appropriately replaced with all floating products such as stocks, FX, investment trusts, virtual currencies, ETFs, commodities, crude oil, etc. Here, floating products refer to products whose value fluctuates.

[0120] (Definition of classified investment products) The important thing is whether to do it by stock or by category. This is expected to differ depending on the stock. For example, an increase in semiconductor production volume will not affect resource stocks, and conversely, a rise in crude oil prices will not have much of an impact on semiconductor-related stocks. Event weighting varies depending on the investment product. For this reason, it is expected that the most accurate prediction of when an event occurs will be achieved by learning both by stock. However, this has the disadvantage of reducing the amount of data, making it difficult to advance machine learning.

[0121] Therefore, a realistic solution requires categorizing investment products with similar trends based on past stock price data, and then using machine learning to determine the relationship between events and the prices of investment products in that category for each categorized investment product. The target is each investment product within that categorized investment product. Categorization can be achieved using AI, either through clustering from unsupervised data or through supervised machine learning. Therefore, price fluctuation data for investment products is clustered as unsupervised data, and stock price data with similar movements is grouped together. The relationship between data and events is then determined according to the classification criteria. This is expected to improve accuracy. This clustering method can be used not only with unsupervised data, but also with supervised data classification. It can also be used for standard classifications such as resource stocks, semiconductor-related stocks, and construction stocks. It can also be used to group stocks by theme.

[0122] These clustered and categorized investment products are called categorized investment products. Figure 12 is a diagram showing categorized products according to this embodiment. For example, as shown in Figure 12, stocks C and D can be classified as condominium sales-related stocks. When correlation coefficients and the like are calculated for condominium sales-related stocks, it becomes clear that stock C has the strongest correlation among them. The most detailed classification of categorized investment products is stock, and the broadest classifications are stocks, FX, investment trusts, etc., and categories such as resource stocks are also called categorized investment products.

[0123] (Issues regarding classified investment products) If investment products are simply categorized, there is a concern that the relationship between event changes and price fluctuations of investment products will be less accurate. For example, if crude oil prices rise, resource stocks will have a positive impact but airline stocks will have a negative impact. The impact on foreign exchange and investment trusts will also vary depending on the stocks included. Therefore, if machine learning is used to predict the relationship between event changes and investment product price fluctuations and the impact of events on investment product price fluctuations, it is necessary to process the data by stock or by classification of investment products with similar price movements. The more detailed the classification, the less data there is, making machine learning more difficult. Conversely, the more general the classification, the less accurate the data but the greater the amount of data. By categorizing investment products by category, the issue can be resolved by using different categories depending on the desired objective. Furthermore, information on thematic stocks and related stocks, such as resource stocks and semiconductor-related stocks, is generally available. However, simply engaging in related businesses can also fall into that category. In other words, for example, if crude oil prices rise, it is difficult to provide information such as how much price change is expected, the degree of correlation, and which stock is the best.

[0124] (Methods for solving problems with classified investment products) The more detailed the relationship between events and price fluctuations of classified investment products, the more accurate the relationship will be, provided there is enough data. Of course, if there is no problem with the amount of data, there is no problem with doing it by stock. With the table reference method, related stocks can be grouped together by inserting a related stock table, as shown in Figure 10. The resource-related stock table, capital investment-related stock table, upward revision-related stock table, etc. can be divided depending on the situation, and can also be easily extracted. The relationship between event changes and price fluctuations of related stocks becomes clear. The relationship between price fluctuations of classified investment products as a group and event changes can be clarified.

[0125] The estimation unit 18 also estimates the price of investment products using a trained estimation model for estimating the price of investment products. AI technology uses deep learning to advance machine learning on the relationship between the two (the relationship between event changes and investment product price fluctuations), and the relationship between event changes and price fluctuations of classified investment products is used as a trained model to improve accuracy, thereby resolving the conflicting issues of accuracy and data volume. For example, a trained model for resource-related stocks is conceivable. The same is true for trained models for capital investment-related stocks. In terms of technical indicators, a trained model that can make highly accurate predictions using stochastics is conceivable. Trained models specialized in predicting upward price fluctuations are also possible. Furthermore, it becomes clear which stocks among classified investment products have the strongest correlation, and which stocks should be purchased or sold when an event changes.

[0126] (effect) Whether using the table lookup method or the AI method, price fluctuations of multiple stocks are associated with the occurrence of events in the resource stock category. This adds information on both the category and individual stocks, such as how the impact of high crude oil prices on resource stocks is when grouped together as resource stocks, and how it affects individual stocks. This is expected to have a special effect in that it will be possible to select stocks that are best to buy when crude oil prices rise, such as stock A among resource stocks.

[0127] Figure 13 illustrates a method for evaluating correlation using a table lookup method. For example, it is possible to extract only crude oil-related stocks and provide information on correlation, including correlation coefficients and correlation degrees, as shown in the upper panel of Figure 13. Correlations can be calculated daily and are highly reliable because they are based on accumulated data. This provides immeasurable benefits by adding information such as correlations, correlation degrees, and identification of the most sensitive stocks to information on thematic stocks and related stocks. While there are various ways to utilize this information, for example, as shown in the second and third panels from the top of Figure 13, it is possible to display the results on Terminal 2, which can have special effects. The results and process of the judgments can also be included. It is also possible to incorporate time lag judgments, as shown in the third panel from the top of Figure 13.

[0128] (How to collect price data for classified investment products and the correlation coefficient between price fluctuations of classified investment products and changes in events) If it has been indexed as an index or ETF, then you can simply take that price data. If it is composed of multiple stocks, such as thematic stocks, and is not indexed, then you can use standard indexing techniques to index it and create time series data. You can simply take the average, or if the rates of change are the same, it is easy to calculate the correlation coefficient. All of these methods are included. Calculating the correlation coefficient of related stocks makes it possible to calculate the overall correlation coefficient as well as individual correlation coefficients, further expanding the information. It is also a good idea to include a time lag assessment.

[0129] (Method of manufacturing a teacher data generation device and a table creation device) (Data acquisition process) (Regarding investment product price data) Price data for investment products usually changes with the date and time, so it is often managed as time series data. It can be monthly or minute-by-minute. The appropriate time interval is determined by the type of event (an event that affects the price of an investment product, such as an oil crisis) and the price of the investment product.

[0130] For example, in the case of scalping, price data down to the second or minute is important, and change information at such intervals is also required. On the other hand, in the case of investment trusts, monthly intervals may be sufficient. Since it is important to compare change data rather than price data at a single point in time, it is important to incorporate event data (price data for products subject to the event) between two points in time (this can also be rate of change data compared to the previous year, etc.) and investment product price data between two points in time (this can also be rate of change data compared to the previous year, etc.). Methods for incorporating time series data will be discussed later.

[0131] (How to obtain and format event data) For example, Non-Patent Document 1 listed in [Prior Art Documents] discloses time series data. It discloses a wide variety of monthly, quarterly, and annual data items (e.g., public works order amounts, machinery orders, housing starts, new car sales, etc.). For example, monthly new car sales are defined as the number of new car sales of light vehicles and vehicles registered with the Motor Vehicle Inspection and Registration Office, and the Japan Automobile Dealers Association publishes the number of registered vehicles for the previous month on the first working day. The number of registered vehicles for February is disclosed on March 1st. Data can be easily downloaded and can be fully used as basic data. This is just one example, but event data can be obtained from a variety of sources. However, most of this is time series data.

[0132] (Handling of time series data) (Time series data and point-in-time data of event data) Here, time series data is defined as a plurality of pieces of data at different times, and single-point-in-time data is defined as data at a single point in time.

[0133] (assignment) With only point-in-time data at the time of an event rather than time-series data, some events show a continuous improvement after the event occurs, while others show a darker outcome within a week, making it difficult to continuously reflect subsequent developments. For example, time-series data on crude oil prices compared to the previous period has items for multiple dates, and there are cases where an event of a 10% increase compared to the previous period occurs, followed by a 20% increase and a 30% increase, and cases where an event of a 10% increase occurs, followed by a 5% increase and a 20% decrease. The impact on subsequent price fluctuations of investment products will be different (see Figure 15, bottom).

[0134] As shown in the lower diagram of Figure 15, even if the crude oil price is rising on the starting date, there are cases where it starts to fall three months later, and there are also cases where it rises further. There is a challenge in managing time-series data that changes over time depending on the content of events and investment products.

[0135] (Means for solving the problem) This problem can be solved by acquiring event data as time series data (multiple data points at different times) rather than as single-point data.

[0136] (effect) By using time-series data for both the event data and the price data of the investment product, it becomes easier to manage and a special effect can be expected in that it becomes easier to obtain data at any two points in time.

[0137] By utilizing time series data, for example, if you want to measure short-term overheating over a week or so, you can take both prices one week after the event and calculate the volatility. If you want to measure the medium-term impact of a surge in crude oil prices, you can calculate the volatility of both prices over a period of about three months. Since the period of impact varies depending on the event, you cannot set the period flexibly unless you manage it using time series data. Time series data offers the special benefit of being able to change the period flexibly. Setting the period depending on the event enables even more accurate predictions.

[0138] (How to link investment product price data with time series data) (assignment) With conventional technology, stock price data is not only taken into account on the day an event occurs, but also taken in after a certain period of time has passed, making it extremely difficult to manage the period, as some events produce results three months after the event occurs, while others produce results within a week and then lose their effect.The issue is that the period to be taken varies greatly depending on the content of the event and the investment product.

[0139] (Means for solving the problem) This issue can be resolved by converting the price data of linked investment products into time-series data. For example, time-series data for stock prices contains multiple stock price and date fields for each stock. This includes cases with records of multiple dates and stock prices, such as (Stock (Stock A) Date (2021 / 1 / 21), Closing Price (1000), Date (2021 / 1 / 22), Closing Price (1020)). By joining these two tables using two composite keys (stock name and date), stock price data for a stock before the event can be retrieved as needed. This allows for retrieval of data from one week or one year before the event, depending on the situation. For example, if an event occurs for a stock code of 1111, the details of the event, the stock price of the stock code 1111 on the day of the event, and subsequent stock price trends can be retrieved as time-series data. Data can be retrieved three days before, three days after, one week after, or three months after the event. As shown in Figure 15, creating a date conversion table will further increase convenience.

[0140] (Cases where comparison is not between two points in time) Even with technical indicator events, rating events, corporate earnings events, macro indicator events, and milestone events, there are cases where comparisons between two points in time are not made. For example, the event of President Trump's landslide victory in the election is not, when viewed alone, an event between two points in time. Similarly, the event of the announcement of corporate earnings for the fiscal year ending March 2021 is not, when viewed alone, an event between two points in time. The same is true for technical indicators, such as the RSI falling below 10%, and rating events.

[0141] However, if corporate earnings for the fiscal year ending March 2021 and forecasts for the fiscal year ending March 2022 are announced, it is possible to capture changes between the two points in time: the actual results for the fiscal year ending March 2021 and the forecasts for the fiscal year ending March 2022. In the event that the RSI falls below 10%, if it is the first time it has fallen below this level this year, it becomes a comparison between more than two points in time, and changes can be captured. Comparing the rating to the previous rating is also a comparison between two points in time.

[0142] In this way, depending on how the data is collected, it can be a comparison between two points in time or an event at one point in time. The event change and investment product price fluctuation prediction system 1 basically targets changes between two or more points in time, but it is not limited to this.

[0143] (Conversion device 102) The conversion device 102 converts raw data such as event data and investment product price data into a format suitable for storage in the database unit 103 .

[0144] (Definition of change data) Change data includes time series data and refers to data that captures event data and price fluctuations of investment products. Daily, weekly, and monthly data are also change data. Data showing the rate of change, such as year-on-year, month-on-month, and day-on-day data, is defined as change data.

[0145] (Definition of time series data) Here, time series data is defined as multiple pieces of data at different points in time, and one-time point data is defined as data at a single point in time.

[0146] (Definition of rate of change data) Data that shows the rate of change between two points in time, such as year-on-year data, month-on-month data, or day-on-day data, are defined as rate of change data. The rate of upward revision when comparing an upward revision to an initial forecast, or a 30% increase in crude oil prices compared to the previous year, are also defined as rate of change data. For example, when looking at the rate of upward revision, if the rate of change is between the initial sales forecast for May 2021 and the upwardly revised sales for September 2021, the rate of change in stock price for that period is used as the basis. If the time lag is not taken into account, the two can be compared directly, but if there is a time lag, the time must be adjusted.

[0147] The conversion device 102 does not refer to or learn time series data and rate of change data, but performs processing to align the periods, start dates, end dates, and comparison targets (compared to the previous year or the previous month) between rate of change data and between time series data.

[0148] (Definition of conversion device 102) The conversion device 102 performs various conversions to store raw data such as time-series data in teacher data or tables. The conversion device 102 includes a conversion unit 14, a correlation unit 27, a time lag unit 28, an optimization unit 29, and a relational table creation unit 30. The conversion device 102 serves as a lookup table creation device in the table lookup method, and as a teacher data generation device in the AI method.

[0149] (Conversion unit 14) The conversion unit 14 is responsible for converting time series data and the like into change data and the like. As shown in FIG. 7, depending on the data, the correlation unit 27 may calculate correlations from the raw data. What is important here is that in the case of daily (weekly or monthly) time series data, correlations are derived between daily (weekly or monthly) time series data (between time series months in FIG. 7), and in the case of change data, correlations are derived between change data (between year-on-year comparisons in the case of year-on-year comparisons) (between month-on-month comparisons in FIG. 7). As shown in FIG. 7, the relationship between event changes and investment product price fluctuation data is clarified by determining the time lag and the correlation at that time.

[0150] Preparing a conversion table makes it clear at a glance what type of data is being collected. It clarifies whether it's time series data or change data, whether the period is monthly, weekly, daily, or hourly, whether it's compared to the previous month or day, and the data spans a certain period. This process ensures that data of the same nature is used when comparing event data with investment product price fluctuations and training. Even when comparing time series data, it's important to derive them using the same standards. The numbers vary significantly depending on the type of change data (year-on-year, month-on-month, or day-on-day). For short-term changes, you'll need to use day-on-day or week-on-week comparisons, while for medium-term changes, you'll need to use month-on-month comparisons, and for long-term changes, you'll need to use year-on-year comparisons. Time series data can be measured in daily, weekly, monthly, or yearly time frames. While daily-updated data like stock prices allows for year-on-month or month-on-month comparisons, macro indicator data often only requires year-on-month or monthly data.

[0151] For this reason, the conversion unit 14 must process the data to match the accepted data (monthly or year-on-year comparisons). By properly distinguishing between change data, time-series data, month-on-month or year-on-year comparisons, daily, or monthly, it is important to compare the same data when creating training data and tables, and to use data with the same standards as training data. Care must be taken when using time-series data as training data. For example, it is better to train rate-of-change data than to train a crude oil price of $50 and a stock A price of ¥500. In the case of technical indicators, the training method varies depending on the event content, such as training an RSI of 10% and the subsequent rate of change in the price of an investment product.

[0152] (Two-point data conversion process (process of converting time series data into two-point data)) (assignment) For example, if the event change data is rate-of-change data compared to the previous year, the investment product price change data must also be aligned with the rate-of-change data. An upward revision rate is a typical example. Various cases, such as year-on-year data, are possible. Here, we discuss how to align time series data with rate-of-change data. By incorporating stock price data from other points in time rather than just the date of the event, stock price data can be integrated. However, this method makes it extremely difficult to manage the period, as some events produce results three months after the event, while others produce results one week and then lose their effect. Rules must be established for whether data should be kept three weeks later or one month later, which can be extremely cumbersome. The required period varies significantly depending on the nature of the event and the investment product. For example, it is necessary to capture the rate of change between the initial forecast in May 2021, when the upward revision was made, and the upward revision in November 2021. In this case, the question arises of how to capture the stock price volatility.

[0153] (Means for solving the problem) Figure 15 shows how to import event data after a certain period of time has passed. As shown in the upper diagram of Figure 15, the problem can be solved by treating price data as time-series data and using a date conversion table or similar to obtain price data from two or more points in time. The event table can be made so that data from the day the event occurred and event data from a certain period of time before the event occurred can be extracted depending on the situation. This makes it possible to extract data from one week or three months before the event occurred, depending on the situation.

[0154] For example, time-series data for stock prices may contain multiple stock price and date fields for each stock. For example, for stock (Stock A), the fields might be date (2021 / 1 / 21), closing price (1000), date (2021 / 1 / 22), and closing price (1020). As shown in the upper diagram of Figure 9, this time-series data table can be linked with a date conversion table and two composite keys (stock and date) to enable stock price data for a stock at a time different from the event occurrence date to be retrieved as needed (Figure 8 shows a one-year date interval). It is possible to retrieve data from one week before the event occurred or three months before, depending on the situation (by simply changing the date interval). As shown in Figure 9, inserting a date conversion table increases convenience and allows price data from two or more points in time to be freely imported, such as price data from 30 days ago or three months later.

[0155] For example, consider the case where an event occurs for a stock code of 1111. The details of the event on the day it occurred, as well as the stock price of the stock code 1111 on the day the event occurred, and subsequent stock price trends can be pulled as time series data. Data can be extracted one week before, three days after, one week after, or three months after the event date, as long as it is linked to time series data. If there is a time lag, it can be inserted between the date of the event and the price data of the investment product. The date conversion table is just one example, and any method that can import data from two points in time from a time series data table or the like will suffice.

[0156] (effect) For example, if you want to measure short-term overheating over a week or so, you can take the price one week after the event and calculate the volatility. If you want to measure the medium-term impact of a surge in crude oil prices, you can calculate the volatility over a period of about three months.

[0157] (Cases where correlation of time series data is sufficient) For example, when it comes to changes in technical indicator events and stock prices, there is no time lag, and in most cases the technical indicator itself is an indicator that reflects the change. For example, with the RSI, the total increase (compared to the previous day) over a certain period of time is calculated by adding the total increase and decrease over the same period, dividing the result by 100, and the indicator itself reflects the change over a certain period. Therefore, there is no problem in calculating the correlation with time series data on stock prices as is. However, for example, if data shows crude oil at $100 and a 30% increase in crude oil compared to the previous month, it is the change in crude oil that is most closely related to changes in stock prices. Therefore, there are cases where a process of converting the data into change data is required.

[0158] (Cases where a process to convert to change data is required) (Conversion of time series data or data between two points in time into rate of change data) (assignment) Conventional technology, which simply incorporates event data and investment product prices into data on the day the event occurs, is unable to accurately capture changes in events or fluctuations in investment product prices. Even stock price data, for example, is typically incorporated as time-series data, and even if machine learning is used, tables are created using a table lookup method, or time-series data is compared or combined, ultimately there is the issue that it is very inconvenient and difficult to understand when observing changes.

[0159] (Means for solving the problem) Figure 14 shows time series data and rate of change data. As shown in Figure 14, time series data represents price fluctuations at a single point in time. In some cases, converting the data into a rate of change between two points in time provides better correlation. In this case, it is necessary to compare the rates of change, including converting the data into a rate of change after combining the data. There are differences between time series data and rate of change data. Rate of change data is measured as a change rate compared to the previous month. There are various methods, such as year-over-year and three-month comparisons, but these are typical examples of conversion to rate of change data. For example, the bottom left figure in Figure 14 is an example of time series data for event prices. The item on the left rose from $70 on June 1, 2020, to $100 in September 2020, and then returned to $70 in March 2021. The item on the right rose from $70 on June 1, 2020, to $100 in September 2020, and then continued to rise to $220 in March 2021. The bottom right figure in Figure 14 converts this time series data into month-on-month data. In other words, July 2020 saw a change from $70 in June to $80, which is 114% compared to the previous month. The same procedure is used to plot the rate of change, as shown in the bottom right figure in Figure 14.

[0160] The same applies to price fluctuation data of investment products. Figure 14 is one example. Time series data is converted into data on the rate of change compared to the previous month. However, this method is merely a typical example, and the method for capturing the rate of change is not limited to this example. The investment product price fluctuation prediction system 1, which predicts price fluctuations of investment products from event changes, may require a process of converting time series data, etc., into rate of change data, and there are several ways to do this besides this typical example, but the method is not important.

[0161] Rather than a single price point, price data from two or more points in time is required, or rate of change data, such as year-on-year or month-on-month change data, can be used as the base data from the start. However, for example, an event such as a new all-time high is also data on the change between two points in time, comparing the previous high and the current price. It is also possible to create a separate table and change the rate of change from the previous month, as shown in the example in Figure 14, to year-on-year data. For example, if the price in July 2020 was $80 but $50 in July 2019, the year-on-year data would be 160%, and the month-on-month data would be 114%, as in the previous example. The same applies below.

[0162] Depending on the event content and desired results, it is possible to change to year-on-year or month-on-month comparison data. The data to be compared (for example, data from the previous year or month) will change depending on the forecast figures you are using and the data with different correlations. You can manage month-on-month and year-on-year comparisons in the same table, or you can manage them in different tables. In either case, it is best to be able to use them flexibly. In addition to the event content on the day the event occurred, you can also pull in the stock price of stock code 1111 on the day the event occurred, as well as subsequent event data and stock price trends, as time series data. Data can be pulled out three days, one week, or three months after the event occurred, as long as it is linked to time series data.

[0163] There are four typical combinations: (1) the relationship between time series data such as event data at time A, event data at time B, A-price data, and B-price data; (2) the relationship between change data and time series data such as the rate of change in event data at time B and price data at time A and price data at time B; (3) the relationship between the rate of change in rate B and the rate of change in rate B; and (4) the relationship between event data at time A, event data at time B, and the rate of change in rate B. It is also important to calculate the time lag and align the time axis (year-on-year or month-on-month, etc.).

[0164] (effect) Change begets change; for example, when crude oil prices change, the stock price of Stock A also changes; the change itself affects the price fluctuations of investment products. Rather than the fact that crude oil prices have reached $100, what is important is how the price has changed compared to the previous month or the previous year, and how this information is linked to price fluctuations of investment products. A drop from $120 to $100 is likely to have the opposite impact on the price fluctuations of investment products as a rise from $80 to $100, so it is important to understand this relationship.

[0165] To explain using the example of Figure 14, when looking at the time series data alone, the item to the right of the time series data on the bottom left appears to have grown significantly, but when compared to the previous month, it is stable at around 10% compared to the previous month. When looking at the rate of change, the same numerical change looks different. In other words, it is clear at a glance that the momentum in the case on the left of the time series data has clearly slowed since September 2020 and entered negative territory. On the other hand, the momentum in the case on the right of the time series data has not slowed down, exceeding 100% compared to the previous month every month, making it clear that the pace of increase has not slowed.

[0166] As shown in the lower diagram of Figure 17, the relationship between event changes and investment product price fluctuations is basically determined by looking at the rates of change, such as the event change rate (year-on-year change in crude oil prices in %) and the investment product price fluctuation rate (year-on-year change in stock price of stock A in %). This is true for both machine learning and the table lookup method. This method has a special effect of producing results that cannot be obtained by simply comparing time-series data. For example, one important aspect of this invention is that the event of crude oil prices reaching $100 and the event of a 30% increase in crude oil prices compared to the previous year are closely related to investment product price fluctuations. The optimization unit 29 makes this point even clearer. The conversion unit 14 prepares data in various formats.

[0167] The relationship between event changes and investment product price fluctuations varies depending on the event content. In some cases, it is better to compare time series data, in other cases it is better to find correlations between rate of change data, in other cases daily data is better, or in other cases month-over-month data is better. The change section may include a process for determining which combination is optimal, but this is adjusted by the optimization section 29.

[0168] (Correlation between event change data and investment product price fluctuation data) FIG. 17 is a diagram showing the relationship between fluctuations in event data and fluctuations in investment product prices.

[0169] (Correlation Section 27) The correlation unit 27 calculates the correlation between various event changes and price fluctuations of the investment target product. In the case of time-series data, it is important to calculate the correlation between the time-series data itself, or between the change data itself, by matching the date unit (for example, daily data for daily data, or year-on-year comparisons for year-on-year comparisons). This process performs the matching. As will be described later, this also includes correlations between event changes, such as between macro index changes or between macro index changes and corporate performance changes. In this process, for example, the correlation coefficient between event changes and price fluctuations of the investment target product is calculated.

[0170] In the example of Figure 37, the time series data is converted to year-on-year change data, and the correlation coefficient is 0.58. This process also calculates whether the correlation is stronger in the daily data of the time series data or in the year-on-year change rate data. The correlation unit 27 may classify data into those for which a positive correlation is observed, those for which no correlation is observed, and those for which a negative correlation is observed. However, this process may be implemented in the correlation process or in a later process. The advantage of implementing this process is that it allows these three classification items (such as positive correlation) to be established in the database unit 103, which has the special effect of making it easier to extract and learn only those relationships with strong correlations when learning training data and creating tables.

[0171] (assignment) The relationship between changes in events and fluctuations in investment product prices is complex and difficult to understand. There is no clear understanding of the relationship, the extent of the impact, or past trends, so there is no clear rule. Linking unrelated data is meaningless and may actually reduce accuracy.

[0172] (Means for solving the problem) The problem is solved by linking the event change data and investment product price fluctuation data with a correlation table that clarifies the relationship between them. A typical method is to use a correlation coefficient, which creates a foundation by calculating the degree of correlation between the event change data and investment product price fluctuation data, and then using the correlated change data as training data or creating a combined table. The correlation coefficient is used to calculate which data is correlated with which other data, such as the event change data and the investment product price fluctuation data. The problem is solved by preparing a correlation table and classifying the data into those with a strong positive correlation, those with no correlation, and those with an inverse correlation (see Figure 18). Figure 18 is a diagram showing the table reference method.

[0173] (effect) This method is expected to have special effects such as identifying highly correlated investment products when an event change occurs, identifying groups of correlated investment products from past data, displaying the impact of past event changes on investment product prices, and predicting price fluctuations of investment products that can be derived from past relationships. This is true for both machine learning and the table lookup method.

[0174] (Adjusting the time lag between event change data and investment product price fluctuation data) (Cases where there is a time lag or discrepancy when comparing two points in time) (Time lag process) Price fluctuations of investment products typically involve a time lag or time difference relative to event changes. The role of the time lag section 28 is to adjust for this time lag. The correlation section 27 and the time lag section 28 are interrelated. For example, first, the correlation coefficient indicating the correlation between change data is calculated without considering the time lag (0.58 in the case of Figure 37). Using this correlation coefficient as a reference, correlation coefficients are calculated by shifting the time axis, and the time lag at which the correlation coefficient is closest to 1 (in the case of a positive correlation) is calculated to identify the time lag (0.66 in the case of Figure 40). Of course, it is not necessary to be precise; if a one-month shift brings the correlation closer to 1 than without a shift, or if a two-month shift brings it further away, then one month can be considered the time lag. Any method is acceptable; trial and error or AI can be used to determine this.

[0175] (assignment) However, as will be explained later, there are cases where a time lag occurs, which makes date calculations even more complicated. For example, if an event such as a rise in crude oil prices occurs after the market closes on Friday, September 1st, Japanese stocks will be priced at 9:00 on September 4th, the day after the holiday. A date adjustment is also required for such discrepancies. This is called an adjustment date. A time lag is a time difference in which the impact on the price of an investment is delayed. An adjustment date refers to an adjustment made after a holiday, etc.

[0176] (Means for solving the problem) For this reason, it is easier to understand by preparing a date conversion table (time lag). For example, let's assume that a crude oil price hike event occurred on September 1, 2020, for a stock code 1111, increasing 30% year-on-year. If there is a one-and-a-half-month time lag between the rise in crude oil prices and fluctuations in investment product prices in the upper diagram of Figure 15, if the time lag is one and a half months, the target date is one and a half months later than the event occurrence date (i.e., the crude oil price hike), so the base date for investment product price fluctuations is October 15, 2020. From this, it is determined that there is a strong correlation between the price fluctuations of investment products over the one year period from October 15, 2019, one year ago (see Figure 16). Figure 16 illustrates the time lag determination system and the prediction system after time lag determination.

[0177] The third figure from the top of Figure 13 shows an example of a 15-day time lag, the first figure from the top of Figure 7 shows an example of a case with no time lag, and the second figure from the top of Figure 7 shows an example of a case with a 2-month time lag. For year-on-year comparisons, see the third figure from the top of Figure 7.

[0178] (effect) While the details of how to determine the time lag will be left to the time lag determination device, once the time lag period is determined, there are several other ways to resolve the issue. The important thing is to reflect the difference between the event occurrence date and the starting date of the investment product price fluctuation, but the method is not important. Understanding and reflecting the time lag and adjustment date can have the unique effect of clarifying the relationship between event changes that cause time lags and adjustment dates and investment product price fluctuations. For example, an event change such as a Middle Eastern crude oil supply crisis requires a certain amount of time to be reflected in the stock price fluctuations of stock A, which has a high correlation with crude oil prices. Incorporating this time lag into the model improves the accuracy of predictions and results, whether using machine learning or table lookup. This unique effect can be expected, as it enables more accurate predictions.

[0179] (Optimization section 29) (assignment) There are many combinations of change data. For example, should you use year-on-year comparisons, month-on-month comparisons, monthly comparisons, or daily comparisons? Depending on the changes in each event and the price fluctuations of the investment target product, it is not possible to make a uniform decision. (Means for solving the problem) The optimization unit 29 optimizes the time lag by, for example, shifting the date of the investment target commodity price fluctuation data so as to bring the correlation coefficient closer to 1. One way to combine data is to adopt a combination with a correlation coefficient close to 1 if the data is the same. Alternatively, a combination table is created and, in the case of technical indicators, the technical indicator data, the price at the time, and the price after a certain period of time are adopted. For upward revisions, the prices on the previous forecast date and the current forecast date are adopted. For macro indicator data, the year-on-year comparisons are used. For a record high price event, the record high price date and the price after a certain period of time are adopted. In this way, the optimal combination is adopted depending on the type of event change.

[0180] (effect) This process plays a very important role, as it both improves learning and optimizes table creation, whether it's the time lag or the combination of event change data and investment product price fluctuation data. In other words, even when looking at the correlation between crude oil prices and the price fluctuations of Stock A, the correlation coefficient will be different when comparing monthly time series data and data showing year-on-year change rates. The correlation will be stronger depending on the event change, with a stronger correlation being observed between a 50% year-on-year increase in crude oil prices and a 50% year-on-year increase in Stock A's stock price than between a crude oil price of $50 and a stock price of Stock A of ¥500.

[0181] However, this cannot be generalized, and by inserting a process called correlation coefficient optimization, it becomes possible to respond by changing the combination, such that even when the same event is combined with the same investment product, in this case the year-on-year comparison will match, in this case the previous month comparison will match, and in this case the daily charts will match.

[0182] The same applies to time lags; by inserting an optimization process, the optimal time lag, whether 30 days or 60 days, can be determined. In the case of technical indicators, the relationship between the investment product price fluctuation rate after a certain period of time has passed since the purchase conditions were met is important, and in the case of upward revisions, the relationship between the rate of change from the previous forecast to the current forecast is important. In the case of macro indicator data, the rate of change is the basis. In the case of ratings, the combination of rates of change from the previous forecast is important. In this way, it is the role of the optimization unit 29 to determine the optimal combination depending on the event content.

[0183] (Relationship table creation unit 30) (assignment) There are issues such as it being difficult to understand what kind of impact an event change will have on which investment products and what kind of price fluctuations it will have. There are also issues such as it being difficult to know what to set as input and output data for training purposes.

[0184] (Method to solve the problem) The relationship table creation unit 30 issues instructions to store each piece of data in the appropriate location by determining the storage location of each piece of data in a combined table, correlation table, teacher data table, etc. The relationship table creation unit 30 combines event change data and investment product price fluctuation data into a single combined table, or summarizes the relationship between the two data sets using a correlation table that shows the correlation between the event change table and the investment product price fluctuation table. The combined table can display a list of event changes and investment product price fluctuations, making it easy to use as a reference table. If it is a one-to-one relationship, it can also be used as a teacher data table. A many-to-one teacher data table can be created using a correlation table to combine an event change data table and an investment product price fluctuation data table. For example, when predicting investment product price fluctuations, highly correlated target events are determined using a correlation table, and the relationship between the multiple event changes and the investment product price fluctuations is learned.

[0185] (effect) In the case of the table reference method, the data is stored in a join table or correlation table, and in the case of the AI method, it is stored in a teacher data table or correlation table, which can be expected to have special effects such as making it possible to refer to the data or to learn using the optimal combination.

[0186] (Database records of training data or change data, correlations, time lags, etc.) The relationships between the manufactured change data are established in the process up to the creation of the relational table. From this point, the process splits into table reference method and AI method.

[0187] (Examples of inputs and outputs of event changes and investment product price fluctuations) (Event Change and Investment Product Price Prediction System (Table Reference Method)) (Process of creating a table linking event changes and investment product price fluctuations) (Table lookup method) The relationship table creating unit 30 stores the calculated time lag calculation figure (for example, 90 days), the correlation coefficient, and the relationship between the change data in the database unit 103. This is defined as a correlation table.

[0188] If the correlation table is the relationship between event change data and investment product price fluctuation data, the table linking the two data is defined as a combined table of event changes and investment product price fluctuations. Combinations of event changes and investment product price fluctuations are also possible.

[0189] The relational table creation unit 30 creates a combined table of event changes and investment product price fluctuations based on the results obtained from creating the relational table. However, the joining method is not important, and it can be a single table or multiple tables linked by a key. As long as there is a connection between event change data and price change data of related investment products, the form is not important.

[0190] (assignment) When it comes to the relationship between events and investment product prices, both are point-in-time information, so the data often does not provide meaningful insight. In terms of corporate performance, for example, trying to unravel the relationship between Stock A's performance for the fiscal year ending March 2021 and Stock A's stock price is extremely difficult. However, this is the majority of information presentations up until now. While information on changes is certainly included, there is virtually no information linking these changes. Even if there is, it is almost meaningless when the information is fragmented, with one showing year-on-year changes and the other showing month-on-month changes. The challenge is the need to create unified rules. Because price fluctuations in investment products are caused by changes, it is necessary to link these changes and clarify the relationships, but this presents a major challenge.

[0191] (Means for solving the problem) The relationship table creation unit 30 can solve the problem by creating a table that clarifies the relationships between changes and storing it in the database unit 103. As an example, when information on crude oil price changes is linked with information on stock price changes and year-on-year increases or decreases are displayed, various changes in information are generated (see FIG. 14). Although information on crude oil prices and information on stock prices exist separately, the problem is solved by linking information on crude oil price changes with information on stock price changes. The relationship between changes, namely, information on crude oil price changes and information on stock price changes, is important. Of course, when comparing year-on-year, it is usually important to include both the year-on-year comparisons and the period between the two points in time.

[0192] (effect) By creating a combined table (multiple tables are possible) of event changes and investment product price fluctuations, a special effect can be expected, in that it can be used in a system for predicting investment product price fluctuations when events change.

[0193] (Prediction process using a table linking event changes and investment product price fluctuations (table reference method)) (assignment) Investors are faced with a wide variety of information every day. There is an endless amount of information available, including corporate performance, technical indicators, economic news, and rating information. For investors, there is information that is important and information that is not, and the current situation is that it is difficult to distinguish between the two. The information that each user is looking for is different and changes from time to time. The information that users are looking for is diverse, but there is a need to provide information that outputs the necessary information according to the user's own needs.

[0194] (Means for solving the problem) For example, questions such as "Of the many technical indicators, which one is most reliable for the stocks I currently own?", "Prices are rising. Which stocks are likely to rise?", and "Can I trust the buys in the daily rating information?" are issues that need to be derived from the relationship between event changes and investment product price fluctuations. As listed below, these issues can be solved using the question list method, selection input method, text input method, etc. of the investment product price fluctuation prediction system 1 (table reference method). Events can be entered by the user in a variety of ways, including selection method, natural search method, and question format. Based on the event input or selection item by the user, the investment product price fluctuation prediction system 1 references and extracts a binding table or correlation table between event changes and investment product price fluctuations, and derives various judgment results (see Figure 18).

[0195] 19 is a diagram showing the structure of the table reference method. As shown in Fig. 19, the information processing device 101 reads out investment product price fluctuation data corresponding to event change data from the database unit 103, performs judgment processing, and outputs the judgment result to the terminal 2.

[0196] The conversion device 102 acquires event data and investment product price data (S1), converts each into change data, etc., checks the correlation between the two change data, adjusts for time differences such as time lag (S2), records the data in a connection table or the like in the database unit 103, and stores the relationship between the two in a correlation table (S3).

[0197] The information processing device 101 retrieves necessary information whenever necessary by referring to the association table and correlation table between event changes and investment product price fluctuations in the database unit 103 (S5).

[0198] For example, the information processing device 101 or the conversion device 102 converts a query entered as text into the terminal 2 into event change data (S4). The information processing device 101 acquires the event change data, reads out investment product price fluctuation data corresponding to the event change data from the database unit 103 (S5), performs a judgment process, and outputs the judgment result to the terminal 2 (S6).

[0199] The steps S1 to S5 are repeated as needed, and the connection table and correlation table in the database unit 103 are updated as needed.

[0200] According to the above, by unraveling the relationship between changes in events and fluctuations in investment product prices, it is possible to solve various worries and problems that users have regarding investments.

[0201] (effect) It has the special effect of being able to solve various problems, as shown in the examples below. Of course, the tables exist separately, and the same thing happens when you query them using database technology. Below, we will explain the behavior for each input method, referring to Figure 22.

[0202] (Specific processing example 1) How does a user actually use this system (table reference method)? First, there is the question item method, in which a user selects question items prepared in advance in terminal 2, the form input method, in which a user selects form items to enter input items, and the text input method, in which a user inputs text. Terminal 2 receives what the user wants to request (S11). Note that FIG. 25 is a diagram showing the input methods.

[0203] (Question item method) In response to the question, the information processing device 101 queries the database unit 103, and returns output data that matches the current situation (question item method: S14).

[0204] (explanation) For example, in response to questions such as "Which stocks are likely to rise in value due to high crude oil prices?", "Which stocks are likely to rise in value due to high US interest rates?", and "Which notable stocks are expected to rise after upward revisions in the past month?", the answer will be "Stock A." The question item method can be solved by preparing possible questions. In response to the questions, a query is made to the database unit 103, and an answer that matches the current situation is returned.

[0205] For example, a list of questions such as "Which stocks are likely to rise in value due to high crude oil prices?" is prepared in advance, and an inquiry is made to the database unit 103 with the event = crude oil and correlation coefficient > 0.8, and the stock with the highest correlation coefficient is extracted and output as the answer. Since fluctuations in crude oil prices and fluctuations in investment product prices are updated daily, the correlation coefficient changes daily. Therefore, the special effect of this answer being updated daily can be expected. The output data is displayed on the terminal (S15).

[0206] (Form input method) Depending on the content selected on the form, an inquiry is made to the database unit 103, and output data suited to the current situation is returned (form input method: S14).

[0207] (explanation) FIG. 21 shows event change input and stock output. As shown in FIG. 21, an example of a selection input method, when a form is selected, for example, the items are displayed in a pull-down menu. For example, the event details, such as crude oil prices, machine orders, and department store sales, are displayed, and the form is assumed to allow the user to select whether the increase data is 10% or 30%. Furthermore, the question content can be selected using a pull-down menu. For example, if the form selection is, "Machine orders increased by 30%, and how many times has this happened so far?", the event = machine orders, and the event data reaching 30% are extracted from the event table, and the number of times is output as output data. As the machine order data is updated, the number of times the event occurs changes. Therefore, the answer can be expected to have a special effect of being updated daily. The output data is displayed on terminal 2 (S15).

[0208] (Output method) The average price fluctuation rate of investment products is output and displayed, sorted and displayed in order of price fluctuation rate (see the combined table), or stocks with high price fluctuation rate (see the correlation table) are selected. Here, the judgment results, judgment process, etc. may also be output and displayed.

[0209] (Text input method) In the case of the text input method, the following procedure is followed: It is determined whether the data in question can be resolved based on the relationship between the event change information and the investment product price fluctuation information (S12). If the issue can be resolved, an inquiry is made as to whether the event change data is in the database unit 103 (S13). If it is present, processing is carried out as follows: A natural search is carried out in the form, for example, by inputting text such as "Which stocks are likely to rise in price due to high crude oil prices?", and the current output data is returned (text input method: S14). The text input items can be resolved by referring to the question list method described above.

[0210] As shown in the lower diagram of Figure 21, in order to be able to pull out a list of questions, multiple questions with the same answer can be registered, such as "What stocks are likely to rise in price when crude oil prices rise?", "What stocks are likely to rise in price when crude oil prices rise?", and "What stocks are likely to rise in price when crude oil prices rise?", and then pulled out by matching them with the text input. For example, if the text input is "What stocks are likely to rise in price when crude oil prices rise?", it matches "What stocks are likely to rise in price when crude oil prices rise?" in the bottom right column, so the question "What stocks are likely to rise in price when crude oil prices rise?" in the bottom left column will be asked.

[0211] (simple formula) For even more advanced processing, AI could break down sentences like "crude oil / high / soaring / what / stocks / are likely to rise?" and analyze the question to determine what the user is looking for, under what circumstances, and what price-varying product the user is asking about. It's also possible to identify the event affecting the price of a price-varying product as crude oil, derive the change / increase / decrease as a price surge, and the investment product as stocks. Furthermore, as AI advances every day, it could identify the event affecting the price of a price-varying product as "crude oil," the direction of the event as "high," identify the change in the price of a price-varying product due to the event as "rising," and determine that the investment product the user is interested in is "stocks."

[0212] The information processing device 101 (AI) in FIG. 24 queries the database unit 103 to find out what the user wants to buy and what kind of price-varying product the user is asking about under what circumstances. The information processing device 101 (AI) then reads AI parameters from the database unit 103 (S26), performs a judgment process, and outputs the judgment result to the terminal 2 (S28). Crude oil price fluctuations and investment product price fluctuations are constantly updated every day, so the correlation coefficient between changes in events and price fluctuations of price-varying products affected by such changes in events, as well as the AI parameter data, dynamically changes and updates every day. Therefore, the answer (the judgment result of the judgment process performed by the information processing device 101 (AI)) can also be expected to have a special effect of dynamically changing and updating every day, and the output data is displayed on the terminal (S15). In FIG. 19, "Brand A" is output to the terminal 2, but different judgment results, such as "Brand B" or "Brand C and Brand D," may be output to the terminal 2 as text information. In addition to "crude oil," events that affect the prices of price-volatile commodities identified by AI include "rising interest rates in the United States" and "stocks with upward price revisions in the past month," as shown in Figure 21.

[0213] In response to a question regarding the possible future price fluctuation range of a specific stock "stock A" input by the user as text information into terminal 2, "how much will stock A rise three months from now due to high crude oil prices?" regarding the possible future price fluctuation range after a predetermined period of "three months" or a specific period of time has passed, due to the influence of a specific event, "high crude oil prices," the information processing device 101 (AI) in Figure 24 expresses multiple (specifically, three) different possible price fluctuation ranges as numerical indicators and percentages, such as "10% possibility is 70%, 20% possibility is 10%, and 5% possibility is 3%, and outputs this as a judgment result as text information to terminal 2. Of these three judgment results by the information processing device 101 (AI), the top judgment result, "10% possibility is 70%," indicates that the degree or range of price increase of specific stock A is "10%" when expressed as a numerical indicator and percentage, and the feasibility of the price of stock A changing by a specific degree or range of increase is output as text information to terminal 2 as a numerical indicator of 70% and a percentage.

[0214] That is, the information processing device 101 (AI) in Figure 24 outputs multiple different judgment results in response to a question entered by a user, and for each judgment result, it outputs to terminal 2, in the form of text information, first judgment result information that expresses the degree or range of price increase (expected rate of change, here expected rate of increase) of a specific stock, namely, stock A, using a numerical index and a percentage, and second judgment result information that also specifies the possibility (expectation) that the first judgment result will occur using text information consisting of a numerical index and a percentage.

[0215] (Event Change and Investment Product Price Prediction System (AI Method)) As shown in FIG. 24, the information processing device 101 reads out AI parameters from the database unit 103, performs a determination process, and outputs the determination result to the terminal 2.

[0216] The conversion device 102 acquires the event data and the investment product price data (S21), converts each into change data, etc., checks the correlation between the two change data, adjusts for time differences such as time lag (S22), records them in a connection table or the like in the database unit 103, and stores the relationship between the two in a correlation table (S23).

[0217] The information processing device 101 reads highly correlated teacher data from the database unit 103 and performs learning (S24). After completing learning, the information processing device 101 stores the AI parameter data in the database unit 103 (S25). By repeating this process, the correlations and AI parameter data in the database unit 103 are updated as needed.

[0218] For example, a query entered as text into terminal 2 is sent to information processing device 101 (S27). In response to the query, information processing device 101 reads AI parameters from database unit 103 (S26), performs a determination process, and outputs the determination result to terminal 2 (S28).

[0219] Here, the information processing device 101 (AI) which is an artificial intelligence performs a judgment process, and outputs the judgment result of the information processing device 101 (AI) which is an artificial intelligence to the terminal 2 as text information as a response of the artificial intelligence.

[0220] Figure 24 shows example answers such as "There is a 70% chance that (Stock A's rate of increase) is 10%," "There is a 10% chance that (Stock A's rate of increase) is 20%, and there is a 3% chance that (Stock A's rate of increase) is 5%." The degree of increase is shown as a numerical indicator (percentage), and its feasibility is also shown as a numerical indicator (percentage). The degree of increase is then divided into three cases: 10%, 20%, and 5%. The three cases are high increase, medium increase, and low increase. The answers also include a numerical indicator (percentage) of the likelihood of each occurring. By providing numerical indicators of the likelihood (feasibility) of multiple possible outcomes, users and investors can make investment decisions tailored to their actual circumstances. In other words, users and investors are provided with multiple options for profiting that suit their own situation, along with numerical indicators of the feasibility of each option.

[0221] The output of the information processing device 101 (AI) which is an artificial intelligence to the terminal 2 may be more detailed information. For example, the following may be an example of the output of the information processing device 101 (AI) which is an artificial intelligence to the terminal 2.

[0222] (Specific examples of investment advice regarding the buying and selling of investment products) Reasons for recommending ENEOS Holdings (5020) "ENEOS has a 50% share of the domestic market for gasoline, kerosene, and diesel, and gas stations can be found in every town, so it's a familiar company. The stock price is around 500 yen, and trading is possible in units of 100 shares, with recommended purchases starting from 50,000 yen. The trading value is over 5 billion yen, so there is a sense of security, liquidity, and ease of trading. The forecast PER for this fiscal year is 9.3 times, and the dividend yield is 4.4%. However, it is also a perennially undervalued stock, so you shouldn't just take all the indicators at face value. Crude oil prices remain high, and there are many factors to consider, such as the impact of the weak yen and policy support. The demand for gasoline itself is on a downward trend due to the trend toward hybrid and electrified vehicles and young people's move away from cars. However, ENEOS Holdings not only operates gas stations, but also owns interests in paraxylene, propylene, natural gas, and copper mines, and is engaged in a wide range of energy businesses. In order to move away from fossil fuels such as gasoline, major changes are needed. While this is highly desirable, it is not easy. It is safe to say that the company will undergo significant change over the next five to ten years. We are currently in a transitional period. Aiming to realize a decarbonized society, the company is also involved in mega-solar and offshore wind power generation projects, and this year acquired Japan Renewal Energy as a subsidiary. Many future projects, including an offshore wind power generation project in Taiwan, are in the planning stages. However, as the company has relied primarily on fossil fuels for decades, excessive expectations should not be expected. The key to success will be whether the company can shift away from its fossil fuel dependency. The company is also engaged in resource development. Its crude oil and natural gas development activities are in Qatar, Indonesia, Malaysia, New Guinea, Vietnam, and the United States. Its advanced technology in functional materials is also noteworthy, with a 60% global market share in semiconductor targets, a 40% global market share in indium phosphide compound semiconductors, and a 60% market share in titanium copper. We recommend starting with 300 shares for around 150,000 yen. In this way, it is desirable that the output to terminal 2 of information processing device 101 (AI), which is an artificial intelligence, includes one or more of the following: the name of a specific volatile product, general information about that name, stock price information for that name at a specified time, information on the size of the trading volume of that name, comments on the liquidity of that name, information about the PER of that name, information on the dividend yield of that name, crude oil price information and information on its impact on the name, technological trends related to that name and an explanation of them, business information about that name, information on affiliated companies such as subsidiaries of that name, overseas business information about that name, share information for each product of that company, information on notable technologies held by that company, information on the number of shares to purchase, information on the number of shares to sell, the expected rate of price increase, the expected rate of price decrease, the likelihood of the prediction being correct, the expected rate of increase for that name after a specified period, the expected rate of decrease for that name after a specified period, the product price forecast for that name and the basis for it, information after a specified time has passed, information on the degree of price increase or decrease, a numerical indicator of the likelihood of price fluctuations, etc.

[0223] <Dynamic changes in information processing device output> Furthermore, the output of the information processing device 101 (AI), which is an artificial intelligence, to the terminal 2 changes dynamically. For example, suppose that at a certain point in time, a user or investor inputs a question A into the terminal 2, and the information processing device 101 (AI), which is an artificial intelligence, outputs an answer A to the terminal 2. If the same user or investor then asks the same question A to the information processing device 101 (AI), which is an artificial intelligence, via the terminal 2 after a predetermined time has passed, the answer from the information processing device 101 (AI), which is an artificial intelligence, may be the same as the answer A, or it may have changed to an answer B, which is different. One reason for this is that, for example, during the passage of a predetermined time, some event occurs, causing changes in the AI parameters in the database unit 103. In this way, the AI parameters in the database unit 103 change dynamically due to the passage of time, updates to event information, changes in investment product price data, etc. As a result, the output information of the information processing device 101 (AI), which is an artificial intelligence, also changes dynamically. Furthermore, the assets of the user or investor may change depending on the buying and selling behavior of the user or investor during the passage of a predetermined time. The personal assets of the user or investor change dynamically. In this way, the information processing device such as artificial intelligence dynamically predicts information about fluctuating products over a predetermined time period based on the dynamically changing personal asset status of the user or investor and investment product price fluctuation data that changes dynamically due to dynamically occurring events in response to a question input by the user or investor as text information, and outputs the predicted information to the terminal 2. More preferably, the information about fluctuating products after a predetermined time has elapsed is predicted and output to the terminal 2.

[0224] Furthermore, a user or individual can ask multiple questions using text information from the terminal 2. After asking the first question and receiving an answer from the information processing device 101 (AI), which is an artificial intelligence, the user can use the answer as a reference to ask a next question that is more narrowed down. By asking a question multiple times, more necessary information can be obtained from the information processing device 101 (AI), which is an artificial intelligence. The terminal 2 can store in its storage means a history of questions asked by the user or individual (text information of all questions) and a history of answers obtained from the information processing device 101 (AI), which is an artificial intelligence (text information of all answers). The user or individual can check all questions and answers using text information at a given time, thereby preventing misunderstandings and misconceptions.

[0225] <Combining two-way communication with chat apps> The interaction in which a user or individual asks a question from the terminal 2 and the information processing device 101 (AI), which is an artificial intelligence, answers is two-way communication carried out via the Internet. In addition, it is desirable that this is real-time communication that is carried out in almost real time.

[0226] For example, by using a real-time chat function (instant messaging application) such as LINE, two-way communication is possible, allowing for the mutual exchange of messages such as text messages, images, and voice messages. Using a real-time chat tool such as LINE allows for the exchange of message information consisting of text, images, voice messages, and the like in near real time between the terminal 2 used by a user or individual and the information processing device 101 (AI), which is an artificial intelligence used by an investment support service provider or the like, enabling two-way communication. For this purpose, the LINE application is installed on the terminal 2 and the information processing device 101 (AI), which is an artificial intelligence. Furthermore, the terminal 2 and the information processing device 101 (AI), which is an artificial intelligence, are configured with a LINE official account (described below) and a chat tool, which is a tool for expanding the functionality of the LINE official account. Using the LINE environment, real-time two-way chat communication between the terminal 2 used by a user or individual and the information processing device 101 (AI), which is an artificial intelligence used by an investment support service provider or the like, allows for one-to-one communication in a closed environment.

[0227] <Text information, audio information, hybrid information> So far, we have explained that the questions that users or individuals input into the terminal 2 are text information. However, questions can be posed not only as text information but also as voice information, or as hybrid information that combines both text and voice information. Furthermore, answers output by the information processing device 101 (AI), which is an artificial intelligence, can also be posed not only as text information but also as voice information, or as both text and voice information. Text information is more accurate than voice information, but takes longer to input than voice information. Therefore, it is possible to utilize the advantages of each. For example, a question can be posed by voice, and the content of the voice information can be converted into text information and displayed on the display unit of the terminal 2. In this case, the user or individual can input a question to the information processing device 101 (AI), which is an artificial intelligence, after checking the display on the terminal 2. It is also preferable that the answer from the information processing device 101 (AI), which is an artificial intelligence, is posed by voice, and that the text information be displayed on the display unit of the terminal 2 and stored in a storage means of the terminal 2.

[0228] <Warning to users and investors> The output information (answer) of the information processing device 101 (AI), which is an artificial intelligence, is very useful information for users and individuals. However, reality does not necessarily move in accordance with the answer information output from the information processing device 101 (AI), which is an artificial intelligence, to the terminal 2. In addition, there is a possibility that it may have a significant economic impact on users and investors. Therefore, it is desirable to also output the following warning information and limit indication information to the terminal 2.

[0229] (Examples of investment warnings) Specific examples of the above-mentioned warning information are given below. ◇The reports posted on this site are for the purpose of providing information, and are not intended to solicit investments such as the sale or purchase of individual stocks. Final decisions must be made at the customer's own discretion and responsibility. ◇Recommended stocks and follow-up comments are not intended to force any trading or trading method. Target stock prices and stop losses are merely guidelines. Please use them as a reference when investing. ◇Recommended stocks do not necessarily have to be bought or sold as suggested. If you prioritize taking profits, please feel free to buy or sell as you wish, even if you take profits in small increments. ◇We have taken every precaution to ensure the accuracy of the information provided, but we cannot guarantee its accuracy. This information is provided solely to provide reference information for stock investment decisions, and is not intended as a solicitation for investment. Furthermore, we will not be held responsible for any damages incurred as a result of this information. Please make any investment decisions at your own discretion and responsibility. ◇Stock investments may result in a loss of investment capital due to fluctuations in stock prices. In addition, when investing in foreign stocks, exchange rates fluctuate and the investment may fall below the original investment amount converted into yen. ◇The information posted has been created and displayed by our company based on sources that we have determined to be reliable, but we do not guarantee or assume any responsibility for the accuracy, completeness or timeliness of the content or information. ◇The contents of the published materials were created at the time of their creation and are subject to change without notice. ◇ Intellectual property rights, including copyrights of the text and data, belong to Rising Bull Investment Advisors, Inc., and it is strictly prohibited to modify or alter this document or any copies thereof without the prior written consent of Rising Bull Investment Advisors, Inc. It is also strictly prohibited to transmit, copy, distribute or transfer this document or any copies thereof. ◇The stocks and foreign stocks listed in the materials may increase or decrease in value, or may even lose value, depending on the activities of the companies, economic policies, and world affairs. ◇The published materials do not guarantee that the value of the funds invested based on those materials will be maintained or increased, and Rising Bull Investment Advisors, Inc. will not be held liable for any damages suffered by customers as a result of investing based on those materials, regardless of the reason. ◇Directors, officers and employees of Rising Bull Investment Advisors, Inc. may hold positions in the financial products mentioned in the materials. Past performance recommendations do not guarantee future performance. ◇The simulation is not based on actual operations. Please use it for reference only. Transaction costs, taxes, etc. are not taken into consideration. ◇The small stock simulations we conduct are based on actual operations and take into account transaction costs, taxes, etc. When selling, we also take into account the associated transaction costs, taxes, etc.

[0230] In other words, the information to be used to warn users / investors should preferably include the following: (1) the investment advice information regarding investment products provided by the investment advice service provider is for reference only; (2) the results may not necessarily be as predicted by the service provider based on the investment advice information; (3) the final investment decision must be made by the user / investor using the investment advice service at the user / investor's own risk; (4) the service provider shall not be held responsible for the results of any investment product purchases or sales made by the user based on the investment advice; (5) intellectual property rights, such as copyright, related to the investment advice information provided by the service provider belong to the service provider; (6) it is prohibited to transmit, copy, distribute or transfer copies of the information provided by the service provider; and (7) the official name of the provider of the investment advice service.

[0231] The alert information for users and investors from (1) to (7) may be any one of them, or may be a combination of multiple alert information.

[0232] (The relationship between input and output in machine learning (learning process)) (In the case of AI method) The conversion device 102 records the time lag calculation figure (for example, 90 days) calculated by the time lag unit 28, the correlation coefficient, and the relationship between the change data in the database unit 103. This is defined as a correlation table. Alternatively, event change data a and investment target commodity price fluctuation data a may be stored as training data, and the correlation coefficient and time lag may be recorded in the correlation table. Any format is possible. The relationship is not limited to the relationship between event change data and investment target commodity price fluctuation data, but combinations of event changes and investment target commodity price fluctuations may also be considered.

[0233] (In the case of a single event change) For example, when predicting price fluctuations of stock A, the learning unit 17 performs machine learning using event change data correlated with price fluctuations of stock A as input data and price fluctuation data of stock A as output data.

[0234] (in case of multiple event changes) For example, when predicting price fluctuations of stock A, the learning unit 17 performs machine learning using a plurality of event change data correlated with the price fluctuations of stock A as input data and the price fluctuation data of stock A as output data. One method is to use the plurality of event change data as input data for a neural network, but another method is to use a recurrent neural network. Details will be described later.

[0235] (How to reflect time lag in machine learning?) The time lag calculated by the time lag unit 28 is used to determine the temporal relationship of event changes related to price fluctuations of stock A. For example, if it is determined that the time lag between crude oil price changes and changes in Iraqi crude oil production volume is shorter than that between changes in Iraqi crude oil production volume, and that the time lag between crude oil price changes and changes in Iraqi crude oil production volume is longer, this information is incorporated into the recurrent neural network model. By inputting these time flows into the recurrent neural network, more precise predictions of price fluctuations of stock A are possible. Alternatively, it may be derived from the relationship between event changes. In other words, the learning unit 17 may separately learn the relationship between changes in Iraqi crude oil production volume and crude oil price fluctuations and the relationship between crude oil price fluctuations and price fluctuations of stock A.

[0236] (Teacher data generation method for a teacher data generation device that converts time-series data into change data) (Teacher data generation process) In order to perform machine learning on the relationship between changes in events and fluctuations in investment product prices, training data is required, and a device for generating this training data is required.

[0237] (assignment) The relationship between the occurrence of an event and the price of an investment product is often not meaningful because both are point-in-time data. For example, in the case of corporate performance, it is extremely difficult to unravel the relationship between Stock A's performance for the fiscal year ending March 2021 and its stock price using machine learning. However, most information display methods are like this, and while information on changes is certainly included, information on the connections between these changes is almost nonexistent. This is even more true with machine learning. Because price fluctuations of investment products are caused by changes in events, there is a challenge in learning how events change with each other. To achieve this, a training data generation device is needed that converts time-series data—investment product price data and time-series event data—into change data.

[0238] (Means for solving the problem) As described above, the conversion unit 14 converts event data (time-series data) into change data, and converts investment product price data into change data. This change data is generated as training data by the training data generation device.

[0239] (effect) As mentioned above, price changes in investment products are brought about by changes in events. Therefore, it is important to learn data on changes in both types of events. The training data generation device can achieve this goal, bringing about a unique effect.

[0240] (Correlation process) (assignment) The relationship between event change data and investment product price fluctuation data varies widely. In some cases there is an inverse correlation (for example, high crude oil prices and airline stocks), in other cases there is a positive correlation, and in other cases no correlation can be detected. There is no point in using uncorrelated data as training data for learning. A prediction system that goes through a learning process without clarifying the correlation is likely to experience a decline in accuracy. The challenge is that it is necessary to clarify the correlation between this event change data and investment product price fluctuation data before providing it as training data for the learning process.

[0241] (Means for solving the problem) Figure 49 is a diagram showing a method for classifying events using a classifier and input / output using a neural network. As shown in the upper diagram of Figure 49, correlations are grasped using each event change data and each investment product's price fluctuation data. The correlation unit 27, for example, calculates the correlation coefficient between the two. The relationship table creation unit 30, for example, stores and saves the correlation coefficients in the database unit 103 as a correlation table. By updating the correlation table, the latest correlations can always be extracted from the correlation table.

[0242] (effect) By sending correlated event change data and investment product price fluctuation data to the learning process, the accuracy of machine learning will increase, and ultimately, the accuracy of the event change and investment product price prediction system will improve, which is expected to have a significant effect in terms of reducing the time lag.

[0243] (Process) (assignment) However, as will be discussed later, there are cases where a time lag occurs, making date calculations even more complicated. Furthermore, for example, if an event change such as a rise in crude oil prices occurs after the market closes on Friday, September 1st, Japanese stock prices will be set at 9:00 AM on September 4th, the day after the holiday. Such a discrepancy also necessitates a date adjustment. This is called an adjustment date. A time lag refers to a time difference in the impact on the price of an investment target. An adjustment date is a period during which the stock market is closed, and is used to adjust the dates between the event change data and the price fluctuation data of investment products. For example, if a company's earnings announcement is made at 4:00 PM on Friday, March 17th, the event change date is March 17th, but trading of that company's stock begins at 9:00 AM on Monday, March 19th, so the two holiday days of March 18th and 19th are used as adjustment dates.

[0244] (Means for solving the problem) For this reason, it is easier to understand by preparing a date conversion table (time lag table). For example, assume that a crude oil price hike event occurred on September 1, 2020, resulting in a 30% year-on-year increase for the stock code 1111. If there is a one-and-a-half-month time lag between the crude oil price hike and the fluctuation in the price of an investment product, as shown in the upper graph of Figure 15, with a one-and-a-half-month time lag, the target date (i.e., the crude oil price hike) is one-and-a-half months later. Therefore, the base date for the price fluctuation of the investment product is October 15, 2020. From this, it is determined that there is a strong correlation between the price fluctuation of the investment product over the one-year period from October 15, 2019, one year prior (see Figure 16). The fourth graph from the top of Figure 13 illustrates a case with a 15-day time lag, the first graph from the top of Figure 7 illustrates a case with no time lag, and the second graph from the top of Figure 7 illustrates a case with a two-month time lag. For a year-on-year comparison, see the third graph from the top of Figure 7.

[0245] (effect) While the details of the process for determining the time lag will be explained in the time lag section 28, once the time lag period is determined, the issue can be resolved in several ways, including this one. The important thing is to reflect the time lag period, since there is a difference between the event occurrence date and the starting date of the investment product price fluctuation. The method for doing so is not important. By understanding and reflecting the time lag and adjustment date, a special effect can be expected: clarifying the relationship between the event change that causes the time lag and adjustment date and the price fluctuation of the investment product. For example, it takes a certain amount of time for an event change such as a Middle Eastern crude oil supply crisis to be reflected in the stock price fluctuation of stock A, which has a high correlation with crude oil prices. Incorporating this time lag into the model improves the accuracy of machine learning predictions and the accuracy of the prediction results. A special effect can be expected: more accurate predictions.

[0246] (Learning process) (assignment) Regarding the relationship between crude oil prices and Stock A, what problems arise when machine learning is performed using crude oil prices as the explanatory variable and Stock A's stock price as the target variable? Learning the relationship between a crude oil price of $100 and Stock A's price of ¥1,000 is not very meaningful. This is because the learning process will not distinguish between a crude oil price rising from $80 to $100 and a crude oil price falling from $120 to $100. In reality, the former has a positive impact on Stock A's stock price, while the latter is likely to have the opposite negative impact, as is the mechanism behind price fluctuations in investment products such as stocks. Furthermore, no amount of machine learning will be meaningful for an investment product that has no correlation with changes in crude oil prices.

[0247] (Means for solving the problem) Therefore, it is necessary to learn correlated change data. The training data generated by the training data generation device is sifted for correlation in a correlation process and then sent to the learning unit 17, whereby the learning proceeds and the problem is solved (see the upper diagram of Figure 23).

[0248] To give one example, the learning unit 17 proceeds with learning by linking information on crude oil price changes with information on stock price changes (see the upper diagram of FIG. 17). Although information on crude oil prices and information on stock prices exist separately, the problem can be solved by learning information on crude oil price changes and information on stock price changes. It is also important to use information on crude oil price changes as input information and information on stock price changes as output information, and to further learn changes over the same period.

[0249] (effect) As shown in the lower diagram of Figure 17, the input items are the event date and the year-on-year crude oil price change, and the output item is the year-on-year price change rate of stock A, and learning is performed. Of course, this assumes that there is a correlation, but by doing so, we can expect the special effect of being able to predict the year-on-year price change rate of stock A by inputting the year-on-year increase rate of crude oil prices into the prediction system as soon as it becomes known.

[0250] This is particularly evident in cases where a time lag exists. For Stock A, the year-on-year crude oil price change from October 1, 2021, is based on the August 2021 average, i.e., mid-August, resulting in a one-and-a-half-month time lag (see bottom of Figure 17). Naturally, the relationship between crude oil price changes and Stock A's price fluctuations is merely an example. Learning all the event changes and investment product price fluctuations discussed in this specification promises unprecedented, special benefits (see top of Figure 17). As an AI system, it goes without saying that it can naturally learn not only one-to-one relationships but also multiple event changes. In this case, too, training event change data and investment product price fluctuation data, which have multiple correlations, promises the special benefit of creating highly accurate trained data.

[0251] (Prediction process) (assignment) For example, when crude oil prices rise, which stocks are likely to rise? Of the many technical indicators available, which are effective for this stock? An upward revision of earnings has been announced, but when is the best time to buy? There are many issues regarding price fluctuations of investment products.

[0252] (Means for solving the problem) The trained model has learned the relationship between various event changes and investment product price fluctuations. Event change data that is likely to contribute to price prediction is input into the trained model that has learned the relationship between event changes and investment product price fluctuations. In this way, by utilizing the parameters stored in the trained model, it is possible to output prediction results for investment product price fluctuations.

[0253] (effect) It can respond to various event changes, and the more you learn, the more special effects you can obtain that cannot be obtained anywhere else.

[0254] (Parameter update process) (assignment) The world of investment product price fluctuations is constantly changing. Rules that were valid until recently quickly become outdated, and new rules are created. Previously trained models are merely past trained models, and there is a challenge in that they must be updated to new rules on a daily basis. Trained models also need to change over time.

[0255] (Means for solving the problem) A trained model is not complete once it has been created. New events occur every day, and the prices of investment products fluctuate daily. This new data is also generated daily, and the training data generation device must store the training data, update correlations, and continue training. During this process, prediction errors occur. Parameters must be updated to compensate for these errors. There are various methods for optimization, such as backpropagation and gradient descent, and these parameter update methods can solve the problem.

[0256] (effect) The rules governing the price fluctuations of investment products change every day. It is extremely important to constantly update parameters to keep up with these new price movements. Constantly updating parameters brings about different effects.

[0257] Here, a specific example of the actual input method and display method shown in FIG. 25 will be described.

[0258] (Specific examples of actual input and display methods in cases where there is a time lag) The top group in Figure 25 shows a first question entered by a user or investor as text from Terminal 2: "Which stocks are likely to rise in value due to high crude oil prices?" Also shown is a second question entered by a user or investor as text from Terminal 2: "Which stocks are likely to rise in value due to high US interest rates?"

[0259] In response to the first question, the information processing device 101 (AI), which is an artificial intelligence, performs a judgment process and gives a first answer, "(Specific stock) Stock A forecast in 3 months (later), (expected) increase rate 10%," and the judgment result of the information processing device 101 (AI) is output as text information to the terminal 2 as an answer from the artificial intelligence. In response to the second question, the information processing device 101 (AI), which is an artificial intelligence, performs a judgment process and gives a second answer, "(Specific stock) Stock B forecast in 1 month (later), (expected) increase rate 5% (probability of this happening) 70%," and the judgment result of the information processing device 101 (AI) is output as text information to the terminal 2 as an answer from the artificial intelligence. In these first and second answers, the specific stock information is presented as a recommended stock to buy, and a predetermined period, such as a forecast 3 months later or a forecast 1 month later, is specified, and the degree of increase in the specific stock, i.e., the degree of price fluctuation, is predicted as a numerical indicator, as a percentage, as a prediction result after the predetermined period has elapsed. In this way, in the first and second answers, the information processing device 101 (AI), which is an artificial intelligence, performs a decision process taking into account the predicted time lag in response to the question information entered by the user or investor via the terminal 2. The AI uses machine learning of time lag information, which is the time lag between the occurrence of an event and the fluctuation of a product price, using actual past time lag data as training data, and uses the resulting AI parameter data related to the time lag to perform more accurate decision processing that takes into account the predicted time lag. The AI predicts the information and situation after a predetermined time has elapsed for a volatile product whose price fluctuates, and outputs the predicted results after the predetermined time has elapsed to the terminal. It also predicts recommended stocks to buy and sell after the predetermined time has elapsed. In other words, the system is characterized by presenting information (predicted growth rate) after a predetermined period of time has elapsed to recommend buying and selling a specific stock. The second answer further predicts the feasibility (expected value) of the predicted growth rate of the specific stock as a numerical indicator (percentage). Furthermore, these answers may also include an explanation of the decision-making process.

[0260] These answers are provided by the artificial intelligence information processing device 101 (AI), which dynamically updates data on event changes at any time or every day, and also obtains AI parameter data from a database unit that dynamically updates data on fluctuations in investment product prices at any time or every day, processes the data, and outputs it to the terminal 2 via the communication unit.

[0261] In the first and second responses shown in Figure 25, the information processing device 101 (AI), which is an artificial intelligence, responds by specifying a predetermined period, such as a prediction three months later or a prediction one month later, and providing a prediction result after the predetermined period has elapsed. However, these specified periods and prediction results after the predetermined period have been obtained by taking into account the time lag between the occurrence of an event such as a rise in crude oil prices and the rise in the stock prices of each stock. Figure 23 shows that the information processing device 101 (AI), which is an artificial intelligence, analyzes the correlation between the event and the price fluctuation data of investment products to determine the time lag period from the occurrence of an event until the impact on the stock prices of each stock appears, as well as the degree of impact. Furthermore, the method of machine learning the relationship between event changes and investment product price fluctuations has been described above in the section (Relationship between Machine Learning Input and Output (Learning Process)) and subsequent sections.

[0262] (Specific examples of actual input and display methods) The user inputs an event from the terminal 2, queries a binding table between event changes and investment product price fluctuations, and obtains output results. Events can be input in a variety of ways, including selection, natural search, and question format. Based on the user's event input, the information processing device 101 derives investment product price fluctuation data and judgment results from the relationship table between event changes and investment product price fluctuations in the learned data.

[0263] (Corporate performance change events and their impact on investment product price fluctuations (multiple companies)) The information processing device 101 has a method of learning and estimating the relationship between corporate performance change events and price fluctuations of investment products using a table reference method.

[0264] (assignment) How do stock prices react to events such as a company announcing a forecast of a 20% increase in profits compared to the previous fiscal year? This can be done by stock, by investment product category, or even the entire market. For example, one method is to use a database table to store and manage data on the occurrence of a forecast event, such as a 30% increase in sales next fiscal year, along with the announcement date and time, the price of the investment product at the time of the announcement, and the price of that investment product a certain period after the announcement. Since data for a single stock is limited, data can be accumulated based on the past 20 years of data for all listed companies or representative stocks. For a company whose fiscal year ends in March, this could be the current fiscal year's sales forecast announced in May, the first actual figures announced for the first quarter, or even when figures are revised upward during the fiscal year. Separating these figures will result in greater accuracy.

[0265] (Means for solving the problem) The relational table creation unit 30 creates a table with the following items: stock (stock A), event occurrence date (2021 / 1 / 21), time lag, event category (corporate performance event), event target (operating profit), target (year-on-year comparison) event data (30%), event content (profit increase), and stock price fluctuation rate over the same period (20%). This provides a solution by making it possible to extract information useful for investment at any time.

[0266] (effect) As can be seen in the table that displays the earnings revision rate and the three-month fluctuation rate of investment products, extracting only the events with increased earnings and calculating the average, or extracting and displaying only the events with high stock price fluctuation rates, can bring special benefits to investors and managers that are useful for investment.

[0267] For example, if you are looking for an indicator to measure short-term overheating, such as a week, you can take the price one week before the event and calculate the volatility. If you are looking to measure the medium-term impact of a surge in crude oil prices, you can calculate the volatility over a period of about three months. For events whose duration is difficult to measure, this has the distinct effect of making it possible to obtain high prices after the event occurs. This method has similar effects not only for forecasting systems for corporate performance change events and stock price volatility, but also for all forms of this specification.

[0268] We will explain corporate performance change events using an upward revision event as an example.

[0269] (Importance of upward revision events) While there are many events that can change a company's performance, upward revisions to earnings are known to attract significant investor attention and have a significant impact on stock prices. As announcements made by the company, they are more credible than market, analyst, or user forecasts. They are common among Japanese companies and carry special significance. Companies are the most sensitive to changes in corporate performance, are the most likely to detect changes early, and are announced by the company itself, giving them special significance compared to other forecasts. Quarterly earnings announcement dates are typically announced in advance, and upward revisions are often announced at the same time. Capital inflows increase around the announcement date, attracting significant investor attention, making upward revisions a significantly different and more special forecast compared to other forecasts. Deriving expected stock price volatility from company forecast changes has a heterogeneous effect.

[0270] (Regarding the correction rate) Figure 26 is a diagram showing a table related to corporate performance events. As shown in the lower diagram of Figure 26, revisions were made from the forecast figures for May 2021 (sales forecast of 10 billion yen) to July (sales forecast of 11 billion yen), November (sales forecast of 12 billion yen), and May 2022 (sales forecast of 13 billion yen) until actual figures were determined. Revision rates include forecast (current forecast) / forecast (previous forecast), forecast (current forecast) / actual (previous actual), and forecast (current forecast) / forecast (initial forecast), and forecast figures include sales, operating profit / loss, ordinary profit / loss, net profit / loss, and dividends. Figure 32 shows the trend in upward revisions and how the revision rate is calculated.

[0271] (A system that predicts stock price fluctuations by managing the relationship between upward revision events and stock price fluctuation rates in a table) (assignment) (Issues common to both table lookup and AI methods) An upward revision event refers to a revision of forecast performance figures for the current period (and for the next period and beyond, including by the company, the Nikkei Financial Yearbook, analysts, forecasting agencies, etc.) during the period. For example, results for the fiscal year ending March 2021 are often announced in May 2021, and forecasts for the current period ending March 2022 are announced at the same time. This is explained as a typical example, but it also includes analyst forecasts, Nikkei Financial Yearbook forecasts, and forecasting agencies' forecasts. In particular, company announcements, as mentioned above, have special significance.

[0272] There are revision rules, such as a 10% increase in sales and a 30% increase in operating profit, and companies are obligated to disclose them as soon as they become aware of them. Because market confusion would result if actual results were suddenly released without a year of revisions, revisions are typically made as needed. Company forecasts for the fiscal year ending March 2022 are revised over the course of one year, from May 2021 to May 2022, when the actual results for the fiscal year ending March 2022 are announced (see Figure 26, top). In the event of an upward revision, the first forecast figures announced in May 2021 are the first point of comparison between the fiscal year ending March 2021 results and the forecast for the fiscal year ending March 2022 (e.g., a 20% increase in profits compared to actual results), and the forecast begins from there (see Figure 26, top). This is truly a type of change event. However, how to act when an upward or downward revision actually occurs varies greatly depending on the stock and the timing, making it difficult to establish general rules.

[0273] (Means for solving the problem (table lookup method)) There is a relationship between this upward revision rate (part of the change information on upward revision events, comparison of sales with the most recent forecast, etc.) and the stock price fluctuation rate, and one way to do this is to prepare a table (table of corporate earnings forecast revisions and stock price fluctuation rates) and create rules (see the top diagram of Figure 27). Figure 27 shows a table of corporate earnings forecast revisions and stock price fluctuation rates, and a table reference method.

[0274] The information processing device 101 and the conversion device 102 store data in the database unit 103. The data includes the number of upward revisions, the upward revision rate (one of current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, or net profit), revision date, forecast target (full-year forecast for the current fiscal year or interim forecast for the current fiscal year), growth drivers, preceding events, the degree of revision for the previous fiscal year, the source of the forecast (company, analyst, etc.), stock price before revision, stock price on the revision date, stock price after the revision date (or rate of change after a certain period has passed), trends in the most recent upward revision event, stock code, time lag, etc.

[0275] The information processing device 101 and the conversion device 102 may manage the data in separate tables (see the lower diagram of FIG. 27). The information processing device 101 and the conversion device 102 may extract past data from the tables and clearly display average values, rate of change, brand names, etc.

[0276] Figure 28 is a diagram showing the relationship between the rate of change in earnings revisions and the rate of change in investment products. By graphing or tabulating the upward revision events for the fiscal year ending March 2021 and the subsequent stock price trends, it is possible to calculate the average of the first upward revision event, the average of the second upward revision event, etc., making it easy to refer to. Therefore, the relationship table creation unit 30 creates a table (e.g., the first diagram from the top of Figure 28: a table that displays the rate of change in earnings revisions and the three-month rate of change in investment products; the second diagram from the top of Figure 28: a table that summarizes the rate of change in earnings revisions and the two-month rate of change in investment products).

[0277] By referencing this table, you can make a forecast by taking the average of announcements from the most recent month (extracted by revision date), or by taking the average of the previous announcements for the stock in question (extracted by stock).For example, it is possible to output the average three-month stock price change rate when the first revision in the past three years was 20% or more (extracted using AND conditions) (see the fourth diagram from the top in Figure 28).

[0278] (effect) It is very easy to understand because it can be displayed in a list. As shown in the table that displays the change rate of earnings revisions and the two-month fluctuation rate of investment products (first figure from the top of Figure 28), extracting only the first revision events and calculating the average, or extracting and displaying only events with high stock price volatility (fourth figure from the top of Figure 28), can provide special benefits that are useful for investors and managers in their investments. The display can also include the decision results and decision process.

[0279] (Display method) Terminal 2 displays average values and reference values for these three-month and two-month rate of change by stock, industry, date of occurrence, and as text data, as shown in the fourth diagram from the top of Figure 28. This is expected to have the special effect of enabling investors to make predictions.

[0280] This table reference method is quite useful, but going a step further, further knowledge can be gained by having the learning unit 17 perform machine learning on the relationship between past increase revision events and stock price fluctuation rates.

[0281] (Supervised machine learning method) (A system for predicting the rate of change in stock prices and revisions to corporate earnings forecasts by analysts and users) (Specific Example 1) (assignment) This problem is solved by the fact that the impact of each analyst's earnings forecast on stock prices varies greatly depending on the analyst.

[0282] (Means to resolve the issue) Fig. 29 is a diagram illustrating machine learning regarding corporate performance events and stock price fluctuation rates after a certain period of time has passed. Fig. 30 is a diagram illustrating machine learning regarding corporate performance revisions and stock price fluctuation rates after a certain period of time has passed.

[0283] The trained program is completed by machine learning the relationship between past upward revision rates (analyst, upward revision rate, stock code) and stock price volatility. The input items for the corporate earnings revision (AI method) in Figure 30 and corporate earnings event 1-2 in Figure 29 are analyst, upward revision rate, and stock code, and the output item is the stock price change rate after a certain period of time has passed. The learning unit 17 performs machine learning on the output items corresponding to the input items. The learning unit 17 inputs the upward revision rate (current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, or net profit), revision, stock code, etc., and proceeds with machine learning using the stock price change rate after a certain period of time has passed since the revision announcement date (or around that date) as output. Stock price volatility varies significantly depending on the size of the upward revision rate, whether it is sales or profit, who made the forecast, etc., and the learning unit 17 machine-learns this. This makes it possible to predict future stock price volatility when an analyst makes a forecast for stock A. This can greatly contribute to determining the investment strategy to take when an event occurs. The display of the stock price fluctuation rate may also include the judgment result, the judgment process, etc.

[0284] (effect) Supervised machine learning allows for multifaceted evaluations, such as determining which analyst's announcements and to what extent the revision rate will increase stock price volatility. For example, by entering figures such as the upward revision rate (current forecast / previous forecast, sales, operating income / loss, ordinary income / loss, or net income), revision date, and forecasting analyst into a form, it is possible to display the stock price volatility rate one week, one month, and three months from now, along with probabilities. This is extremely effective (see Figure 29). Furthermore, combining this data with other events can further improve accuracy. For example, new insights can be gained, such as what happens when upward revision events occur simultaneously when the RSI is below 10%. Deep learning, a supervised machine learning method where weighting and layering are important, is one solution. Downward revision events are similar, simply reversed.

[0285] (Supervised Machine Learning) (Prediction system for revisions to Japanese corporate earnings forecasts and the rate of change in Japanese stock prices) (Specific Example 2) (assignment) Revisions to company forecasts are unique to Japan, and there is an issue of how many revisions there are, how investors' reactions to the current period's forecasts tend to be, and how the reactions vary greatly depending on the period, causing confusion among investors. For this reason, input fields are provided for which period the forecast is for, how many revisions there are, etc. To briefly touch on investor behavior, as will be discussed later, the dates on which companies announce their quarterly financial results are fixed, and Japanese stocks have a unique circumstance in which there is a strong tendency for extreme capital movements to occur around the time of these announcements. Even when good announcements are made, the influx of speculative capital causes market confusion, making predictions very difficult and creating a breeding ground for speculative capital inflows. This issue will be resolved.

[0286] (Means to resolve the issue) (Machine Learning Generation) The learning unit 17 completes the trained program by machine learning the relationship between past upward revision rates (number of upward revisions, upward revision rate, revision date, forecast target, degree of revision in the previous period, stock code) and stock price volatility (see Figures 28 and 29). The input items are as shown in Figure 28, and the output is the stock price fluctuation rate after a certain period of time has passed, which is machine-learned. The number of upward revisions, upward revision rate (one of current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, or net profit), revision date, forecast target (full-year forecast for the current period or interim forecast for the current period), degree of revision in the previous period, forecast source (company, analyst, etc.), stock code, etc. are input, and machine learning is performed using the stock price fluctuation rate from the forecast before revision (or its vicinity) to the revision date as the output. Stock price volatility varies significantly depending on factors such as the first or last upward revision, the magnitude of the upward revision rate, whether it is sales or profit, and the timing. The learning unit 17 machine-learns these. This allows the estimation unit 18 to estimate the future rate of change in stock price when an event of an increase in stock price revision occurs for the A stock. This can greatly contribute to determining the guidelines for investment actions to be taken when an event occurs. However, as the time lag has been explained elsewhere, we will not repeat it here.

[0287] (effect) Supervised machine learning allows for multifaceted evaluations, such as when the first upward revision was announced and how much of a revision rate it will increase stock price volatility. For example, by inputting the number of upward revisions, the upward revision rate (current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, or net profit), revision date, forecast target (full-year forecast or interim forecast), revision level for the previous period, and forecast source (company, analyst, etc.), it becomes possible to present stock price volatility with probability one week, one month, and three months from now, which is extremely effective (see Figure 29). Furthermore, by combining this with other events, further accuracy can be improved. For example, new insights can be gained, such as what happens when upward revision events occur at the same time when the RSI is below 10%. The deep learning of supervised machine learning, where weighting and layering are important, is a key factor. Learning is one solution: reverse the downward adjustment events, but keep everything else the same.

[0288] (Issues with methods that do not take into account time lags and concentration of funds) (assignment) There is a high possibility that there will be a time lag or concentration of funds between the upward revision events mentioned above and the rate of rise or fall of stock prices, so it is important to take this into consideration and incorporate it into the model. This applies to both the table lookup method and the AI method.

[0289] Typically, the process for Stock A is as follows: the announcement of the company's forecast, the announcement of the first quarter (with simultaneous announcement of upward revisions), the announcement of interim financial results (with simultaneous announcement of upward revisions), the announcement of the third quarter (with simultaneous announcement of upward revisions), the announcement of end-of-period releases (with simultaneous announcement of upward revisions), and the announcement of the forecast for the next period. Of course, there are cases where announcements are made outside of the financial results announcement, but in most cases, they are announced simultaneously. This tendency is particularly strong for upward revisions.

[0290] As a result, the earnings announcement date is fixed, and there are many cases where expectations are ahead of time. There are also many cases where stocks rise in anticipation of an upward revision before the earnings announcement, only to plummet when the announcement actually comes. Machine learning and recognition of upward revision events have progressed, and capital concentration may be one of the causes of market volatility, disrupting the market. The present invention aims to take advantage of this time lag and capital concentration tendency to enable early investment decisions.

[0291] (Means for solving the problem) Figure 31 is a diagram that explains the pattern of fluctuations in stock prices adjusted for corporate earnings. Figure 31 shows a method for grasping trends by converting the trends from the previous period into tables and graphs. The recent average trend is that there is a concentration of funds before the announcement of financial results, stock price fluctuations peak, and then the stock price tends to fall for a while after the announcement, and then after a while it starts to rise gradually, peaking before the announcement of financial results. This type of trend is often seen in the market as a whole. This is because a rule-based method is used in the market as a whole. It can be addressed by capturing the trends and patterns before and after the announcement of financial results and reflecting them.

[0292] (effect) This makes it possible to take into account the time lag between upward revisions and stock price fluctuations, as well as capital concentration. It is expected to have the special effect of providing a guide for how to act when an upward revision event occurs.

[0293] There are various ways to understand the time lag, including adjusting for the discrepancy according to the procedure described above. This method includes all methods that incorporate time lag adjustment. There are many cases where the degree of correlation can be recognized for the first time by taking the time lag into consideration.

[0294] The same applies to other indicators other than upward revision events of corporate earnings events, and other events. The same is true for both the table lookup method and the AI method. However, there are also events for which time lags do not need to be taken into account, such as technical indicator events.

[0295] In the second embodiment, many of the methods are used without taking time lag into consideration, but all of the methods can be switched to methods that take time lag into consideration.

[0296] (Issues regarding time lag) There is a time lag between the occurrence of an event and the price fluctuation of an investment product, but this changes depending on the content of the event. The challenge is to clarify this rule. One method is to set the time lag as shown in Figure 15, and this is the theme of this issue: if the time lag is set like this, there will be a time lag and concentration of funds before the change is reflected in the stock price. The challenge this time is to solve the problem of how much of this time lag and concentration of funds will occur. Time adjustment has been touched upon elsewhere, so I will not repeat it here, but time adjustment is naturally an important item.

[0297] (Means for solving the problem) When it comes to upward revisions, the important thing is not the time lag, but the logic at work. In other words, in the case of a surprise announcement with a large upward revision, funds will be concentrated all at once, and the stock price fluctuation will end on the same day, and there may be cases where it will fall thereafter. Therefore, it is necessary to quickly grasp the trends before, on the day of, and after the event announcement, and take investment actions in line with those trends.

[0298] The learning unit 17 is required to learn these. To do this, a period before and after the event announcement is determined based on the stock price on the day the event is announced, the stock price before the event announcement, the stock price after the event announcement, and the trading value of each. For example, the period before the event announcement is three months, and the period after the event announcement is three months. To capture the characteristics of price movements during that period, for example, the day the event occurs (the next day if it is after 3:00 PM) is set as time 0, and since the movement on that day is the most important, the opening price (e.g., time 0:00), the highest price (e.g., time 0:00), and the lowest price (e.g., time 0:00) are taken, and one day later is set as time 1, two days later as time 2, 30 days later as time 30, one day before as time -1, and three days before as time -3. The stock price, trading value, and trading volume one day before are set as 100, and trends are grasped.

[0299] As shown in Figure 31, it becomes possible to grasp the trend. Pattern B is a pattern in which the trend is reflected gradually, and it is best to notice the upward revision event early and take action. Pattern C is a pattern in which funds are concentrated at time 0 and are all folded in, and then the trend tends to decline, so you should either stay away or decide to sell your holdings. If you are going to buy, the correct answer is to buy in advance and sell at time 0. Pattern A is a pattern in which the selling period occurs for several days after the announcement.

[0300] There are many possible patterns beyond these three, but one thing is certain: there is a high possibility that stocks will move in a similar manner at certain times (for example, the quarterly announcement of financial results for the fiscal year ending March 2022). It is important to quickly grasp this current trend (recent trends, such as previous trends) and link it to investment behavior. For example, this includes separating patterns using the zero-point base method or supervised machine learning classification methods mentioned above, or using unsupervised machine learning clustering techniques. Many other methods are possible, but all methods for identifying patterns when the event in question occurs are included.

[0301] (effect) By recognizing patterns, it becomes easier to predict stock price fluctuations in advance, which has a significant effect on investment behavior.

[0302] (Display method) Terminal 2 displays the probability of the current pattern occurring, notifies the user of the trend in price movements before and after the most recent upward revision event, and informs the user of the announcement date and whether or not there was a previous revision.

[0303] (Time lag determination) Usually, when an event such as an upward revision has a large impact on the price fluctuations of an investment product, there is a time lag between the occurrence of the event and the price fluctuation of the investment product. In some cases, the price movement occurs beforehand, and the price movements before and after the event are relevant. In any case, surprise events tend to attract a large amount of capital, and are undoubtedly an opportunity for investment.

[0304] Typically, investor sentiment does not change all at once; rather, it takes time for information to permeate, and the impact on the price fluctuations of investment products is reflected over time. However, when a surprise event is announced, there are various patterns, such as a sudden shift of funds occurring on the day or a shift of funds occurring in advance. This time lag varies in length depending on the event, and even for the same event, it changes depending on the situation. By taking this time lag into account, managing the time lag that changes depending on the event, and reflecting it in learning and predictions, it is expected that accuracy will improve.

[0305] (Means for solving the problem) For example, the process of determining a time lag will be explained using the example of price fluctuations of stock A before and after the announcement of an upward revision event. The correlation unit 27 calculates a correlation coefficient based on the upward revision event occurrence data and its date, and the price data of the announcing stock and its date. Next, the correlation unit 27 calculates the correlation coefficient by shifting the date of the price data of the announcing stock forward or backward while leaving the date of the event occurrence unchanged. If the time lag unit 28 determines that the shift with the strongest degree of correlation (a number closest to 1) is two days before the event announcement, it determines this to be a time lag.

[0306] This allows us to derive a time lag of two days from past data. In other words, in this case, it is important to act two days in advance to start moving in advance. However, this derivation of the correlation coefficient is merely a representative example, and includes all methods for determining the time lag between the two.

[0307] (effect) If we can determine that there is a two-day time lag, then purchasing stock A two days before the upward revision event will have a significant effect on increasing the expected rate of increase. This can be used not only for upward revision events, but also for the relationship between all events with a time lag and the price fluctuations of all investment products.

[0308] (Display method) It will be possible to show the planned event announcement date and the timing for buying, as well as the timing for selling and the expected stock price fluctuation rate with probability.

[0309] (Best stock judgement) (assignment) When an upward revision event is announced, it is unclear which investment products will have a significant impact on price fluctuations, and when this impact will occur. This is because, in addition to the time lag, it becomes difficult to understand and is complicated when multiple other events are occurring. If the best investment stocks were identified at the time of the upward revision, it would be possible to narrow down the target, but this is a difficult issue.

[0310] Logically speaking, the greater the degree of upward revision and the earlier it is made (for example, for the fiscal year ending March 2022, July 2021 will have a greater impact than December 2021), the greater the impact it will have on subsequent stock price fluctuations, and there is some merit in making it the best stock.

[0311] However, there is a possibility that the reaction will be extreme, such as a sudden rise in stock prices, immediately after the announcement, so it is unclear whether buying immediately is the right decision. There may also be a selling opportunity after the event announcement. In many cases, even if the announcement for the first quarter was good, the forecast is left unchanged without any upward revision, and this may result in the stock becoming the best. There is an issue in that there are no clear rules for determining whether this announcement is good or bad (whether the stock price fluctuations will be in a good or bad direction).

[0312] (Means for solving the problem) The first step is to sort out the relationship between upward revision events and stock price fluctuations. The relationship table creation unit 30 creates a table showing the results of the previous period and the current period. Items shown in the upper diagram of Figure 26, such as the announcement date, are stored in a database of the relationship between the items and stock price fluctuations. In addition, by adding trends such as the upward revision rate (current forecast / previous forecast, sales, operating profit, net profit, dividends, etc.), the number of upward revisions, the deviation of the current period forecast from the previous period's results, and the number of upward and downward revisions in the past, it is possible to compile a summary of how the market reacted to the size of the upward revision rate.

[0313] For example, if the larger the sales revision rate, the greater the stock price fluctuation rate, then stocks with a large sales revision rate can be determined to be the best stocks. Conversely, one method would be to extract the characteristics of upward revisions with a large three-month stock price change rate and determine stocks that fit those characteristics as the best stocks.

[0314] (effect) Being able to predict the best stock when an event occurs can be expected to be extremely effective. By transmitting information to the systems of the following embodiments A1 to A4 and B1 to B5, even greater effectiveness can be expected. However, it is important to note that it is also important to simultaneously utilize the time lag judgment mentioned above. This is because the stock price may continue to fall at the same time as the event is announced.

[0315] (Display method) Terminal 2 displays the best stock candidates, and also displays the reasons (increase revision rate and expected stock price fluctuation rate) and process.

[0316] (Extract only a part of the period) (assignment) While it is often possible to derive best stock determinations and time lag determinations from the relationship between upward revision data and its dates and the price data and dates of investment products, there are periods when the data is susceptible to the influence of other events and periods when it is not. For example, if there are no upward revisions, does that mean that the stock in question will not move much? It is not the case, and there is a high possibility that other factors will influence its movements. Conversely, during periods of drastic upward revision fluctuations, correlations are expected to tend to become stronger. There is a challenge in being unable to distinguish between periods of strong correlation and periods of weak correlation. Therefore, by recognizing periods of strong correlation and making predictions during those periods, greater accuracy is expected.

[0317] (Means for solving the problem) For example, the problem can be solved by dividing the trends in earnings revision data into periods of upward revision trends, periods of downward revision trends, and periods with few revisions, and then comparing the data with price data for investment products only for periods of upward revision trends or periods of downward revision trends.

[0318] It is expected that this period will have a stronger correlation than a normal period. There are various methods for determining a period with a strong correlation, and this includes all of these methods. One method is to determine the upward revision trend period in some way, for example, by using the date on the x-axis and the sales forecast on the y-axis, and starting from a certain point when the slope starts to go from 0 to 1. This includes all of the methods for determining a period with a strong correlation.

[0319] (effect) By setting the learning period to a period with a strong correlation, it is expected that accuracy will increase. Furthermore, by combining it with methods such as the aforementioned time lag judgment and best stock judgment, it is expected that the special effect of even higher accuracy will be achieved. These are all representative examples of upward revision data and stock price fluctuations, but they also apply to the relationship between all other events and the price fluctuations of all other investment products. They also apply to both table lookup methods and AI methods.

[0320] (Display method) Terminal 2 is then able to display the information during the increase adjustment period.

[0321] (Identify further leading indicator events) (assignment) While it is certainly important to capture stock price fluctuations before and after upward revision events, there are always preceding events before upward revisions of earnings. For example, before an upward revision event of an oil development company, there are preceding events such as those mentioned in the section on macro indicator events.

[0322] It is best to take investment action based on leading events, which will increase your profit margin. Upward revision events tend to be easy to understand and have simple rules (buy for upward revision events, sell for downward revision events, etc.), but they tend to be short-term trades and have small price swings. It is important to capture leading events, but they vary greatly depending on the investment product, making them difficult to understand.

[0323] (Means for solving the problem) The conversion device 102 can solve the problem by identifying the background of the upward revision event for each investment product and by identifying the leading indicator event. As mentioned in the macro indicator event, the leading event can be the crude oil price for a crude oil development company, the trend in the order amount of semiconductor manufacturing equipment for a semiconductor manufacturing equipment manufacturer, the semiconductor production volume for a semiconductor manufacturer, the office capital investment for office equipment, the order amount of machine tools for a machine tool manufacturer, etc.

[0324] The conversion device 102 finds the correlation between sales and leading indicators and identifies leading events that show a correlation. Naturally, the larger the company and the more diverse its business operations, the greater the number and complexity of leading events. One solution is to use supervised deep learning to learn which events are closely related. For example, in the case of a large company with a wide range of businesses, including the film, music, and game industries, the growth drivers of its recent rapid growth are its image sensor semiconductor business and its accumulated game content business.

[0325] This relationship is compiled into a database, and by registering it as a preceding event of an upward revision event and capturing changes in the preceding event, it becomes possible to catch an upward revision in advance.

[0326] (effect) Since investment actions can be taken before the upward revision event, there is a high possibility of further improving investment results. To do this, it is important to capture preceding events and growth drivers along with the upward revision event.

[0327] (Display method on terminal 2) Terminal 2 displays the correlation between sales and each event. Terminal 2 displays preceding events and important growth drivers. Terminal 2 conveys the trends of preceding events.

[0328] (Linkage with trading data in embodiment 2) This allows for cooperation by picking out stocks with high stock price volatility and presenting information on them as stock replacement candidates, presenting them as stocks for which advice is provided, and presenting advice on selling held stocks. The data structure is stored in the database unit 103 and includes a date ID indicating the date on which an event occurred, investment products affected by the event, the number of upward revisions, and the upward revision rate (sales, operating profit, or net profit). The data structure is characterized by its use in processing to calculate and display the price volatility rate of investment products affected by the event occurrence when the event occurs.

[0329] (Program that generates learning models) A program that generates a learning model for each investment product, with the input set to an upward revision event (such as the increase revision rate) and the output set to the price fluctuation rate of the investment product (stock price fluctuation rate over a certain period of time).

[0330] (A program that uses learning models to make predictions) Once the learning model is complete, it is used to make predictions when an upward revision event occurs. For example, when an event occurs in which Stock A's operating profit forecast for the current fiscal year is revised upward by 20%, the stock price volatility rate of Stock A at the time of the same event in the past is used to make predictions. Alternatively, although the same event for the same stock is expected to be more accurate, since there are only a few of these, the price volatility rate of investment products at the time of the event, including other stocks, is used to make predictions. It is also possible to use the model for predictions by providing an input screen or list format, inputting the input information into the learning model, and outputting it.

[0331] (Learning model) A learning model that has the above-mentioned data structure, has input and output, and displays a predicted value of the price fluctuation rate of an investment product after a certain period of time has passed since the occurrence of an event.

[0332] (Manufacturing method of equipment with learning model) A learning model that displays the probability of stock price fluctuations after a certain period of time has passed for a stock when an upward revision event occurs, or that displays the probability of stock price fluctuations after a certain period of time for a stock that has experienced an upward revision event, is installed in a device, thereby generating a device that implements the learning model.A method for manufacturing a device that has the learning model.

[0333] (Learning model delivery method) This is a distribution method in which the learning model, which has been trained using a huge amount of training data, is distributed via the Internet, added as a function to a securities company's personal page, or predicted stock price growth rates and advice are presented on a paid membership site for those who pay a specified fee.

[0334] (How to update the parameters of the learning model) The learning unit 17 also updates the parameters of the neurons that make up the neural network as needed. Since upward revision events occur daily and investment product prices fluctuate, accumulating new training data further improves accuracy. This parameter update information can be updated via the app, program, or My Page.

[0335] (Pre-trained model) For each upward revision event, the learning unit 17 generates a learned model by deep reinforcement learning, with the upward revision event including the items described above as input and the price fluctuation rate of the investment product (for a certain period) as output.

[0336] (Pre-trained model) A trained model that causes a computer to function so as to output a price fluctuation rate of an investment product based on the occurrence of an upward revision event, the trained model comprising a first neural network and a second neural network connected to receive the output from the first neural network, the first neural network having at least one intermediate layer with a smaller number of neurons than the input layer, the input layer and the output layer with the same number of neurons, and comprising an input layer to an intermediate layer of a feature extraction neural network in which weighting coefficients have been trained so that the input value of each input layer is equal to the output value from each output layer corresponding to each input layer, the weighting coefficients of the second neural network being trained without changing the weighting coefficients of the first neural network, the trained model causing a computer to function so as to perform calculations based on the trained weighting coefficients of the first and second neural networks in response to an upward revision event input to the input layer of the first neural network, and output a value relating to the price fluctuation of the investment product from the output layer of the second neural network.

[0337] The above example is merely one example, and applies equally to all corporate performance events, including not only upward revision events, but also events such as a return from red to black, dividend increases or resumption, and record-high profits. It is also possible to create an estimation system using similar procedures for other events, such as technical indicator triple gaps and triple tops, and macro indicator events such as changes in government bond ratings. Below are some specific examples, but these are not exhaustive.

[0338] (Invention related to pattern recognition between upward revision events and stock price fluctuations) (assignment) Upward revision events, which revise corporate earnings forecasts, are becoming more widely known among investors, and are being utilized, including through program trading and automated trading. However, the current situation is that extreme capital concentration occurs around the time of financial results announcements, often resulting in volatile price movements. There are occasional cases of prices moving in advance, or crashing or soaring after the announcement. Short-term trading is also encouraged, causing confusion among investors, and there is the issue of a lack of clear rules.

[0339] (Means for solving the problem) One way to solve this problem is to understand the criteria by which the market moves and recognize patterns in those movements. One method is a table lookup method that uses the following items: revision date, revision rate, revision count, forecast target (such as full-year forecast for the current fiscal year ending March 2022), stock code, and a stock price table. This is used to match stock price trends around the revision date to several patterns. As shown in Figure 31, the problem can be solved by dividing the patterns into multiple patterns, such as the opening peak pattern on the announcement date with pre-investment, the positive revision surprise pattern, and the surprise peak pattern on the announcement date. This problem can be solved by identifying patterns such as which patterns have been most common recently and the differences between high and low revision rates, as well as the characteristics of upward revision events that fit into the patterns.

[0340] (effect) Price movements before and after upward revision events are becoming increasingly intense, and short-term funds are also flowing in at a rapid pace. By creating patterns in price movements before and after upward revision events and predicting which pattern this upward revision fits, the effect of being able to grasp price movements before and after the event is immeasurable.

[0341] (A system that predicts stock price fluctuations after an upward revision is announced) (assignment) The pattern recognition system (table reference method) mentioned above is a system that inputs the upward revision rate, revision date, etc., and recognizes patterns in price movements before and after the revision. However, there is an issue in that the best it can do is predict stock prices one week or three months after the announcement of the revision.

[0342] (Means for solving the problem) (AI method (supervised machine learning)) Supervised machine learning (classification) is one method that allows for stock price prediction after upward revision events.

[0343] The learning model generation unit 16 and the learning unit 17 create a learning model by using the stock code, the number of upward revisions, the upward revision rate (current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, net profit, etc.), revision date, forecast target (full-year forecast for this fiscal year or interim forecast for this fiscal year), growth drivers, preceding events, degree of revision for the previous fiscal year, and forecast medium (company, analyst, etc.) as input data, and learning is performed using the opening price on the revision date, the closing price three months later (stock price after a certain period has passed), or the three-month rate of change (stock price rate of change after a certain period has passed) as output data.

[0344] (effect) As the learning process progresses, input information such as the revision date and upward revision rate into the prediction system, and it will output the stock price rise / fall rate after a certain period of time has passed since the revision date, along with a probability.

[0345] (Display method) By inputting information such as the revision date, the stock price change rate for the stock in question after a certain period of time will be output along with the probability.

[0346] (AI method (supervised machine learning)) Supervised machine learning (classification) is one technique that makes it possible to identify patterns.

[0347] The learning model generation unit 16 and the learning unit 17 create a learning model using, as input data, the stock code, the number of upward revisions, the upward revision rate (current forecast / previous forecast, sales, operating profit / loss, ordinary profit / loss, net profit, etc.), revision date, forecast target (full-year forecast for this fiscal year or interim forecast for this fiscal year), growth drivers, preceding events, degree of revision for the previous fiscal year, forecast medium (company, analyst, etc.), etc., and learning is performed using, as output data, the closing price on the day before the revision date as 0, the closing price one week ago, the closing price one day ago, the opening price on the revision date, the high price on the revision date, the low price on the revision date, the closing price on the revision date, the closing price three days later, the closing price one week later, the closing price one month later, etc.

[0348] (effect) By inputting information such as the revision date and upward revision rate, the stock price trend pattern before and after the revision date will be output.

[0349] (Display method) When you enter input information such as the modification date, pattern predictions are output.

[0350] (About the procedure for generating a learning model) Figure 50 is a flowchart showing a method for creating a trained model to be applied to predicting investment products based on the occurrence of a change event in a macro indicator.

[0351] A method for creating a trained model to be applied to predicting stock price fluctuations due to the occurrence of upward revision events, A feature map generation step (S21) of extracting features from each of a plurality of sample data sets to which training data on the relationship between the occurrence of an upward revision event and stock price fluctuations is added, using a learning device to generate a feature map; a distance map generation step (S22) of generating a distance map by taking a difference between the feature map when there is a discrepancy between the prediction and the actual results for a combination of an upward revision event and a stock price volatility rate among a plurality of sample data; a distance value calculation step (S23) for calculating a distance value for each combination from the distance map; a distance value discrimination step (S24) of comparing the distance value with a predetermined threshold to determine whether the combination is a combination of an upward revision event and a stock price fluctuation that is less than the threshold or an upward revision event and a stock price fluctuation that is equal to or greater than the threshold; a parameter correction step (S25) of correcting the calculation parameters of the learning device and performing learning so that the discrimination result in the distance value discrimination step matches the combination of teacher events in the sample data; A method for generating trained models, including:

[0352] (Learning model estimation stage) Obtain the stock name and stock code, Obtain the fiscal year for which you want to forecast, Obtain the upward revision rate, Enter the name of each stock, the fiscal year for each forecast, and the upward revision rate. A program that causes a computer to execute a process in which the acquired stock name, the fiscal year for the forecast, and the upward revision rate are input into a learning model that has been trained using training data that outputs the stock price fluctuation rate of the stock in question, and a value related to the stock price fluctuation is output.

[0353] (classifier) 1. A method using machine learning to predict stock price volatility based on upward revision events, comprising: Identify the brand, Identifying the fiscal year for which the stock is forecasted; Identifying an upward revision rate for said stock; A method using a machine learning classifier based on predictions for a stock, its target fiscal period, and its upward revision rate to generate a prediction of stock price volatility from an upward revision event.

[0354] (Learning model estimation stage) An input layer where stocks, target fiscal periods, and upward revision rates are input; an output layer that outputs a value related to the stock price volatility of the stock; An intermediate layer in which parameters are trained using training data that inputs stock data, target accounting period data, and upward revision rates, and outputs stock price volatility rates; Equipped with Obtain stock data, target accounting period data, and upward revision rate data, A trained model that causes a computer to function by inputting stock data, target fiscal year data, and upward revision rate data into the input layer, performing calculations in the intermediate layer, and outputting values related to stock price fluctuations of the stock from the output layer.

[0355] (Content data structure) A data structure that enables a computer to process information by acquiring stock data, target fiscal year data, and upward revision rate data, and then calculating or processing, storing, and displaying the stock price fluctuation rate of the stock in question.

[0356] (This is the data of the target content) an analysis unit of the server analyzing the relationship between the revision event data for revising the forecast data for the fiscal year announced by the company and the stock price fluctuation rate based on past performance; a receiving unit of the server receiving correction event data of the stock for a predetermined period; a step in which a prediction unit of the server predicts and outputs a future stock price fluctuation rate based on the analyzed relationship, using the received stock data and correction event data for the predetermined period as input; A method for predicting stock price volatility based on earnings revisions for stocks including:

[0357] (Upward revision event occurrence and investment product price fluctuation forecasting system based on deep learning with recurrent neural networks) (assignment) As explained using the earnings revision table in Figure 26, upward revision events typically occur in chronological order, with the first forecast for March 2022 announced at the same time as the actual results for March 2021 are announced in May 2021, followed by the first revision in July 2021 and the second revision in November 2021.

[0358] Changes in corporate performance are time-series information, and they change over time. The company forecasts published by companies also change over time. While there are methods to solve this problem using conventional neural networks, such as incorporating time lags, this becomes complicated. Even if a system for predicting stock price fluctuations due to upward revision events is created without incorporating this time-series relationship, there is still the issue of poor accuracy.

[0359] (Means for solving the problem) This problem can be solved by incorporating time series data into machine learning using recurrent neural networks. Methods for incorporating time series data into machine learning include recurrent neural networks, BPTT (Backpropagation Through Time), LSTM (Long Short Term Memory) networks, transformers, future technologies, and recurrent use of intermediate layers. This includes all methods, such as propagating past information through

[0360] Typically, the relationship between upward revision events and stock price volatility is not simply a matter of the rate of upward revisions; if the company has tended to overestimate its forecasts in the past, the market will not react because it lacks credibility. This is evidence that there is a relationship between the company's past forecast announcements and its current announcements. Furthermore, the reaction to stock prices will be significantly different if the company's performance in the previous period was very good and continues to be good in the current period, compared to if the company's performance in the previous period fell sharply and is recovering in the current period. This also suggests that incorporating past corporate performance into the model is likely to result in more accurate results.

[0361] Therefore, for the investment product price fluctuation prediction system 1 due to the occurrence of an upward revision event, machine learning that can process time-series information, such as a recurrent neural network, will enable more accurate predictions.

[0362] (effect) Even if past events of corporate earnings revisions are simultaneously input into a neural network, the neural network cannot process time series data well. However, the above-mentioned methods, such as recurrent neural networks, make it possible to process time series data, thereby realizing a highly accurate investment product price fluctuation prediction system 1 based on the occurrence of upward revision events.

[0363] The relationship between the occurrence of upward revision events and stock price fluctuations is complex. In particular, there are causal relationships between each event and the time period in which they occur. Unless these are incorporated into the deep learning model, the prediction system will not function properly. Using the deep learning method described above, such as recurrent neural networks, to create an investment product price fluctuation prediction system1 based on the occurrence of events will bring immeasurable benefits and effects.

[0364] (Display method) The learning device is characterized in that the learning unit inputs each event included in the event group into a recurrent neural network that serves as the learner in the order in which the events appear, and causes the network to learn the order in which each event appears in the event group as well as the characteristics of the event group.

[0365] (How to generate a trained model) (About the procedure for generating a learning model) A method for creating a trained model to be applied to predicting stock price fluctuations due to the occurrence of events that change corporate performance. A recurrent neural network generation procedure that inputs events in order according to the order of occurrence of corporate performance change events; a feature map generation step of extracting features from each of a plurality of sample data sets to which training data on the relationship between the occurrence of correction events and stock price fluctuations is added, using a learning device to generate a feature map; a distance map generation step of generating a distance map by taking a difference between the feature maps when there is a discrepancy between the predicted and actual results for a combination of an event and a stock price fluctuation rate among a plurality of sample data; a distance value calculation procedure for calculating a distance value for each combination from the distance map; a distance value discrimination step of comparing the distance value with a predetermined threshold to determine whether the combination is a combination of an event and stock price fluctuation below the threshold or an event and stock price fluctuation above the threshold; a parameter correction step of correcting the operation parameters of the learning device and performing learning so that the determination result in the distance value determination step matches the combination of teacher events in the sample data; A method for generating trained models, including:

[0366] (Learning model estimation stage) Obtain the stock name and stock code, Obtain corporate performance change events and enter information about each stock name and each company's performance change. A program that causes a computer to execute a process of inputting the acquired stock name and corporate performance change information into a learning model trained using training data that outputs the stock price fluctuation rate of the stock in question, and outputting a value related to the stock price fluctuation.

[0367] (Classifier method) 1. A method of using machine learning to predict stock price volatility based on events, comprising: Identify the brand, Identifying information on changes in corporate performance of the stock; A method for generating a prediction of stock price volatility from an event, comprising using a machine learning classifier based on predictions regarding the stock and company performance change information.

[0368] (Learning model estimation stage) an input layer to which stock and corporate performance change information is input; an output layer that outputs a value related to the stock price volatility of the stock; An intermediate layer in which parameters are trained using training data that takes stock data and corporate performance change information as input and stock price volatility as output; Equipped with Obtain stock data and corporate performance change data, A trained model that causes a computer to function by inputting stock data and corporate performance change data into the input layer, performing calculations in the intermediate layer, and outputting values related to stock price fluctuations of the stock from the output layer.

[0369] (Content data structure) A data structure that enables a computer to process information by acquiring stock data and data on changes in corporate performance, and then calculating or processing, storing, and displaying the stock price fluctuation rate of the stock in question.

[0370] (This is the data of the target content) an analysis unit of the server analyzing the relationship between the event data and the stock price fluctuation rate based on past performance; a receiving unit of the server receiving event data of the stock for a predetermined period of time; a step in which a prediction unit of the server receives the received stock data and correction event data for the predetermined period based on the analyzed relationship, and predicts and outputs a future stock price fluctuation rate; A method for predicting stock price volatility based on stock events, including:

[0371] (Technical indicators and investment product price fluctuations) (table lookup method) (assignment) There are many technical indicators, and it is difficult to know which ones to trust, as many of them are misleading. Depending on the stock and the time, they can be hit or miss, making them difficult to use.

[0372] (Means for solving the problem) Each type of investment product, that is, each stock, has its own price movement habits. Depending on the stock, buying at the golden cross may work well, or buying when the RSI is below 10%, or stochastics may be correct, and the probability of success varies depending on the stock.

[0373] Figure 33 shows the structure of a trading rule definition table, etc. For each stock, a database is created that records the relationship between the buying rule and the stock price volatility (the stock price volatility after a certain period of time has passed since the conditions of the buying rule were met). It has items such as the buying rule, the stock price volatility (1 month, 3 months, etc.), and the date the buying rule conditions were reached. For example, the rule of RSI below 5% was met 20 times over 20 years, with an average 15% three-month return and an 80% success rate (a three-month return exceeding 0). Meanwhile, there were 35 golden crosses (25 and 75 days), with an average 5% three-month return and a 50% success rate. Such buying rules can now be verified.

[0374] (effect) The effect of being able to adopt technical indicators with a high probability of success and high expected stock price volatility for each stock brings immeasurable benefits to investors.

[0375] (Display method) The terminal can display technical indicators with a high probability of success. It can also display the past probability of success and the average price fluctuation rate of the investment product after the purchase conditions are met. It can also display technical indicators that are prone to failure.

[0376] (AI-based solution) (assignment) There are limitations to selecting the optimal technical indicator for each stock using the table lookup method. For example, in the example above, if the RSI were 3% instead of 10%, the success rate would increase to 90%, and the 3-month return rate would increase to 20%. The table lookup method has the problem of not being able to know that this would increase accuracy.

[0377] (Means for solving the problem) The learning unit 17 uses supervised machine learning to learn each technical indicator (input) and the price fluctuation rate of the investment product (for example, three-month change rate) (output) for each stock. In this way, learning progresses using combinations of change rates over three months from the achievement of technical indicators, whether the RSI is 3%, 5%, or 7%, or the Stochastics is 5%. It is also possible to increase the output to include one-week change rates or one-month change rates.

[0378] (effect) By using supervised machine learning with the conditions of technical indicators as input and the price volatility of an investment product as output, it becomes possible to calculate the optimal technical indicators for that investment product and the price volatility after a certain period of time has passed when the conditions of those technical indicators are met. Furthermore, by inputting the technical indicators of that investment product, the trained model can present the price volatility rate from the present to the future with a probability, which has an immeasurable effect on investors.

[0379] (Display method) When the stock and the technical indicator to be adopted are selected as input, terminal 2 displays the price fluctuation rate after the next fixed period has elapsed, along with the probability.

[0380] (A system that uses machine learning to predict the relationship between golden cross events and stock price volatility when the event occurs) (assignment) Figure 36 is a diagram explaining a buy signal. A golden cross event, for example, occurs when the 25-day moving average crosses the 75-day moving average. It's generally known as a buy signal, and the subsequent stock price volatility varies depending on factors such as whether the price has just updated or not, the previous price movement, and the trading volume. Golden cross stocks are circulating daily. However, when a golden cross event actually occurs, how to act varies greatly depending on the stock and the timing, making it difficult to establish a general rule. Even after a golden cross, it's not uncommon for the price to suddenly drop afterward.

[0381] (Means for solving the problem) There is a relationship between golden cross events and stock price volatility, and one way to solve this problem is to prepare a table and create rules. One method would be to graph or tabulate the 2021 golden cross events and subsequent stock price trends, calculate the average rate of increase after the golden cross event and the rate of increase in trading volume during the golden cross event, compile statistics on what types of golden cross events increase stock price volatility, and use this to predict golden cross events in 2022.

[0382] The relational table creation unit 30 creates a table with a data structure of golden cross date, trading value increase rate, lowest price date, lowest price, highest price date, highest price, golden cross date stock price, type of moving average, stock price after the golden cross date (stock price one month later, stock price three months later, etc.) The estimation unit 18 estimates stock price fluctuations after the current golden cross from past data.

[0383] (effect) It will be possible to verify the golden cross, which will provide investors with various benefits, such as information on how many times it occurred in 2021, what the results were after a certain period of time, what the success rate was, and whether the price rose or fell on average.

[0384] FIG. 34 is a diagram showing the relationship between technical indicators and the rate of change of investment products.

[0385] The first figure from the top of Figure 34 is a table listing the manifestations of the buy rules of each technical indicator for one stock and the subsequent price fluctuation rate. This makes it clear which technical indicators are effective and which cannot be used, and allows you to see the peculiar price movements of each stock from the perspective of technical indicators.

[0386] The second figure from the top in Figure 34 is a table that includes other stocks, which has the special effect of making it easy to check the compatibility and effectiveness of combinations of stocks and technical indicators.

[0387] The third diagram from the top in Figure 34 can be used to create a report based on the results of these calculations. It is also possible to automatically generate a report by pulling figures that convey the latest trends and tendencies from the database unit 103.

[0388] The fourth diagram from the top in Figure 34 shows that when a user makes various inquiries, a query is returned from the database unit 103 and various answers are displayed. This has a significant impact on investors. This table reference method is quite useful, but going a step further, machine learning the relationship between past golden cross events and stock price volatility can provide even greater insight. Note that the displayed answers may also include the judgment results, the judgment process, etc.

[0389] A system that uses supervised machine learning to learn the relationship between golden cross events and stock price volatility, and estimates the expected stock price volatility when the event occurs. The learning unit 17 completes the trained program by machine learning the relationship between past golden crosses and stock price volatility.

[0390] (assignment) The table reference method mentioned above is quite useful, but the results will vary greatly depending on, for example, whether the golden cross is based on the 25-day moving average and the 7-day moving average, or the 25-day moving average and the 75-day moving average, or whether a golden cross that occurred shortly after a low is credible. There are limits to rule-based learning, which involves deciding on rules and acting according to those rules.

[0391] The rate of stock price fluctuation varies greatly depending on the results of the first golden cross, the difference in results from the second golden cross, when the cross occurred, the lowest and highest prices, the rate of increase in trading volume, and the price movements prior to that. The learning unit 17 performs machine learning on this information, allowing the prediction unit 18 to predict the rate of future stock price change when a golden cross event occurs for stock A. This can greatly contribute to determining what investment behavior should be taken when an event occurs. It is expected that learning will progress, such as when a golden cross appears not many days after the lowest price, the probability of success is high and the rate of subsequent price fluctuation is also high (see the second diagram from the top in Figure 35). Figure 35 is a diagram explaining machine learning related to golden crosses.

[0392] (effect) Supervised machine learning makes it possible to make a multifaceted evaluation, such as the golden cross date, when it is announced, and what kind of golden cross event increases stock price volatility.

[0393] For example, by entering the golden cross date, trading volume increase rate, lowest price date, lowest price, highest price, type of moving average, etc. into the form, it becomes possible to display the stock price fluctuation rate one week, one month, and three months from now with probability, which is a very effective feature.

[0394] Using techniques such as recurrent neural networks (supervised machine learning), we can learn the relationship between golden cross events and stock price volatility, and use a system that can process time series to estimate the expected stock price volatility when the event occurs.By using such a system, we can expect to achieve even more accurate predictions (see the third figure from the top in Figure 35).

[0395] Furthermore, by combining it with other events, accuracy can be further improved. For example, new knowledge can be gained, such as what happens when a significant upward revision occurs and a golden cross event occurs at the same time. In this case, weighting and layering become important. Deep learning, a type of supervised machine learning, is one solution. A dead cross event can be handled in the same way as a golden cross event, simply by reversing the case.

[0396] (Linkage with trading data such as in embodiment 2) We can collaborate by picking out stocks that are likely to have high stock price fluctuations and presenting information about them as candidates for stock replacement, presenting them as candidates for advice, and providing advice on selling held stocks.

[0397] The memory unit stores a data structure including a date ID indicating the date on which an event occurs, an investment product affected by the event, a moving average type 1, a moving average type 2, the period from the previous golden cross date, and a volume increase rate (trading value increase rate), and the data structure is characterized by being used for processing to calculate and display the price fluctuation rate of an investment product affected by the event occurrence upon the occurrence of the event.

[0398] (Pre-trained model) A program that generates a trained model for each golden cross event using deep reinforcement learning, with the golden cross event set as input and the price fluctuation rate of the investment product (for a certain period) set as output.

[0399] A program that generates a learning model for each investment product, with the golden cross event as input and the price fluctuation rate of that investment product (over a certain period) as output.

[0400] (A program that uses learning models to make predictions) Once the learning model is completed, the learning model is used for prediction when a golden cross event occurs. For example, when a golden cross event occurs for stock A, the prediction unit 18 uses the stock price fluctuation rate of stock A at the time of the same event in the past for prediction. Alternatively, if the same event for the same stock is expected to have a higher probability of occurring but there are only a few of them, the prediction unit 18 uses the price fluctuation rate of the investment product at the time of the event, including other stocks, for prediction.

[0401] (Learning model) A learning model that has the above-mentioned data structure, has input and output, and includes a predicted value for the price fluctuation rate of an investment product after a certain period of time has passed since the occurrence of an event.

[0402] (Manufacturing method of equipment with learning model) A method for manufacturing a device having a learning model, in which a device having the learning model implemented is generated by installing a learning model into the device that presents the probability of price fluctuation rate of the investment product after a certain period of time has passed when a golden cross event occurs, and presents the probability of price fluctuation rate of the investment product after a certain period of time for the stock in which the golden cross event occurs.

[0403] (Learning model delivery method) Methods for distributing learning models include distributing the learning models, which have been trained using a huge amount of training data, over the Internet, adding them as a function to a securities company's personal page, or presenting them on a member's site.

[0404] (How to update the parameters of the learning model) The learning unit 17 also updates the parameters of the neurons that make up the neural network as needed. Golden cross events occur daily, and prices of investment products fluctuate, so accuracy will be further improved by accumulating new training data.

[0405] (Pre-trained model) A trained model that causes a computer to function to output a price volatility rate of an investment product based on the occurrence of a golden cross event, a first neural network and a second neural network connected to receive an output from the first neural network; the first neural network is configured from the input layer to the intermediate layer of a feature extraction neural network in which the number of neurons in at least one intermediate layer is smaller than the number of neurons in the input layer, the number of neurons in the input layer and the number of neurons in the output layer are the same, and weighting coefficients are trained so that the input value of each input layer is equal to the output value from each output layer corresponding to each input layer; the weighting coefficients of the second neural network are trained without changing the weighting coefficients of the first neural network; performing a calculation based on the learned weighting coefficients in the first and second neural networks for the upward revision event input to the input layer of the first neural network, and outputting a value relating to the price fluctuation of the investment product from the output layer of the second neural network; A trained model that makes a computer function like this.

[0406] (Macro indicator change event) (A method of estimating the relationship between macro index change events and investment product price fluctuations using a table reference method) (Conversion device 102) (Conversion process) (Correlation process) (assignment) There is a close relationship between macroeconomic indicator changes and price fluctuations of investment products, such as rising crude oil prices, rising US interest rates, worsening conditions in the Middle East, and the US presidential election. However, there are complex and multiple factors involved, some investment products are uncorrelated, there is a time lag, and there are differences depending on the type of event, so there are currently no rules.

[0407] In the stock market, it is common to see thematic investments such as oil-related stocks, and mutual funds offering thematic investment products such as robot-related investments. However, complex and wide-ranging analysis is required to understand how corporate performance changes with changes in macroeconomic indicators, and what changes this will bring to certain investment products.

[0408] To give a simple example, there is the issue of not being able to identify stocks that tend to rise most when crude oil prices rise, stocks that are most likely to fall, or stocks that are most susceptible to a decline when US interest rates rise. While an event such as a rise in crude oil prices is an easy-to-understand example, there are many different types of events that can occur and the degree of impact varies widely, such as the impact of a drug price revision on pharmaceutical companies, or the impact of a change event such as a decrease in investment by European companies on export-related companies, so the issue is that the system is complex, diverse, and confusing.

[0409] (Means for solving the problem) Past changes in macro indexes and past price fluctuation data for investment products are accumulated in the database unit 103, tables are created, and the database unit 103 manages the data so that it can be used for predictions whenever an event occurs. Necessary information can be provided at any time. Using past crude oil price data and, for example, past stock price data for brand A, the relationship between crude oil prices and stock prices during crude oil price increases over the past 20 years can be plotted in graphs or tables, and the average calculated, making it possible to identify past relationships.

[0410] For example, based on the stock price change rate of a stock when crude oil prices rose 40% in 20xx and the stock price change rate of a stock when crude oil prices rose 60% in 201x, the relationship between the two can be expressed in a formula or averaged, and the relationship between the two can be stored in the database unit 103 so that it can be retrieved at any time.

[0411] Figure 37 shows the relationship between the year-on-year conversion rate of crude oil prices and the rate of change in stock prices. The key here is to calculate the rate of change for both. For crude oil prices, the year-on-year percentage change is graphed or tabulated, while for stock prices, the year-on-year percentage change is similarly plotted. It is changes in macroeconomic indicators that cause stock price fluctuations, not the absolute value. In other words, even if crude oil is at $100, if it remains stable at a high price, it will not cause any changes in stock prices. By comparing the rates of change between the two and calculating the correlation, a relationship can be established. Of course, month-on-month comparisons can also be used (see the first graph from the top of Figure 37). Examples include regression curves using the least squares method and correlation coefficients, and these figures are used to measure the impact of changes in macroeconomic indicators.

[0412] (effect) This has the effect of identifying stocks that are predicted to be most sensitive in the event of a rise in crude oil prices, allowing for a grasp of the extent of the impact in advance and the ability to reflect this in advice and trading early on. Looking at the graph and table data in Figure 37, a correlation between the two is evident, with a correlation degree of 0.58. As shown in the fourth figure from the top of Figure 37, a positive correlation is evident.

[0413] (Time lag process) (assignment) There is a high possibility that a time lag will occur between the aforementioned crude oil price fluctuation rate and stock price fluctuation rate, and it is important to take this into consideration and incorporate it into the model. Normally, if we trace the history of crude oil price increases, increases in monthly sales of crude oil development companies, increases in expected sales of crude oil development companies, improvements in the quarterly financial results of crude oil development companies, and their reflection in stock prices, there will be a time lag. Events such as drug price revisions are also difficult to understand, so their impact on stock prices usually appears gradually. The present invention aims to take advantage of this time lag and enable earlier investment decisions.

[0414] (Means for solving the problem) Figure 40 shows the relationship between the year-on-year rate of change in crude oil prices and the rate of change in stock prices. The first figure from the top of Figure 40 shows a graph that reflects a time lag of one and a half months. The correlation coefficient between the rate of change in crude oil prices and the rate of change in stock prices is 0.66, which is higher than the 0.58 obtained without taking the time lag into account. The time lag unit 28 calculates the optimal time lag based on how much of a shift is needed to maximize the correlation coefficient (get it closest to 1).

[0415] (effect) The above makes it possible to take into account the time lag between rising crude oil prices and fluctuations in investment product prices. This has the special effect of enabling more accurate predictions of stock prices one and a half months later when an event such as a rise in crude oil prices occurs. In other words, rising crude oil prices can be used as a leading indicator of rising stock prices, and taking the time lag into account has the special effect of enabling more accurate predictions.

[0416] Of course, depending on the event, there may be no time lag, or the time lag may be longer or variable, or the correlation coefficient may be lower due to the strong involvement of other events. There are various ways to understand the time lag, in addition to the method described above that adjusts for the discrepancy. This method includes all methods that incorporate time lag adjustment.

[0417] There are many cases where the degree of correlation can only be recognized by taking time lag into account. The same is true for other macro indicators besides crude oil prices, and for other events. This is true for both table lookup methods and AI methods. However, there are also events that do not require time lag consideration, such as technical indicator events. While many of the methods presented in this specification do not take time lag into account, all methods can be switched to methods that take time lag into account.

[0418] (Time lag calculation) (Issues regarding time lag) There is a time lag between the occurrence of an event and the price fluctuation of an investment product, but the time lag varies depending on the content of the event. There is an issue of whether this rule should be clarified.

[0419] One method is to create a relationship diagram of preceding events and grasp the time lag of events, as shown in the third diagram from the top in Figure 40. To explain this in the case of a sudden rise in crude oil prices, events that affect the supply and demand of crude oil prices often occur before the sudden rise in crude oil prices. By considering these events as preceding events that occur before the sudden rise in crude oil prices, it is expected that the longer the preceding event, the longer the time lag will tend to be. The way to grasp the time lag of occurrence here is to go through a time lag determination process.

[0420] Figure 39 shows a table of macro indicator changes and corporate performance changes. As shown in Figure 39, by measuring the rise in crude oil prices and the degree of impact on the company INPEX, the impact on INPEX's stock price can be broken down into high crude oil prices, changes in corporate performance, and changes in stock price. This can be divided into two parts: the relationship between high crude oil prices and changes in corporate performance, and the impact of changes in corporate performance on stock price changes. The relationship between the Middle East's crude oil supply shortage and crude oil prices can be determined by the relationship between changes in Middle Eastern crude oil production and changes in crude oil prices. While the previous example focused on the relationship between changes in crude oil prices and changes in stock prices, the relationship between changes in macro indicators and changes in corporate performance can also be used in a similar way. An example would be an event where supply issues arose in the Middle East or the United States, both of which have large crude oil reserves.

[0421] On the demand side, events that could occur before a crude oil price spike include the onset of winter, increased energy consumption due to COVID-19 countermeasures, and a surge in energy consumption due to China's economic stimulus measures. Even after a crude oil price spike occurs, events like those mentioned above will occur. Potential events include an increase in monthly sales at crude oil development companies, an increase in expected sales at crude oil development companies, and improvements in the quarterly financial results of crude oil development companies, all of which could be reflected in stock prices. While this creates a time lag before these events are reflected in stock prices, the question here is whether it is possible to specify the order to some extent. The challenge here is to solve the problem of this order and the extent of the time lag.

[0422] (Means for solving the problem) One method is to create a relationship diagram of preceding events like the one mentioned above (see the third diagram from the top in Figure 40), and for stocks that are highly correlated with each event, find the time lag with the strongest correlation between the occurrence of the event and the stock's stock price. This can be done by finding the correlation when there is no time lag, and the correlation when the time is adjusted by one month, three months, etc. (shifting the price fluctuations of the investment product later), and then identifying the time when the correlation is strong as the time lag. This method can be used for any data that has a time series. There are many other possible methods, but all of these include methods for identifying time lags.

[0423] (effect) For example, if an event has a time lag of one and a half months, it is possible to more accurately determine the price fluctuation of an investment product one and a half months after the event occurs. As shown in the second figure from the top of Figure 40, due to the one and a half month time lag, the crude oil price surge event in January 2022 has already occurred, and is 160% compared to the previous year, but the stock price on March 1, 2022 is a future figure that has not yet been determined. How much will it rise compared to the level in March 2021? It is precisely because of the time lag that predictions are possible. If a later time lag event occurs, such as an increase in monthly sales at an oil development company, it is expected that prices will fluctuate even more in the near future. If an earlier event such as supply instability in the Middle East occurs, there may be a time lag of around six months. The longer the time lag, the earlier you can make investments, which has the special effect of increasing the expected price volatility.

[0424] (Time lag judgment system) (assignment) Typically, there is a time lag between the occurrence of various events such as macro indicators, ratings, and business performance, and the price fluctuations of investment products. This is because investor sentiment does not generally change all at once; rather, it takes time for information to permeate, and the impact on investment product price fluctuations is reflected over time. The duration of this time lag varies depending on the event, and even for the same event, it changes depending on the situation. Taking this time lag into account, managing the time lag that changes depending on the event, and reflecting this time lag in learning and predictions is expected to improve the accuracy of investment product price predictions. However, there is a challenge in how to measure this time lag.

[0425] (Means for solving the problem) For example, to explain the time lag determination system using the example of price fluctuations of stock A, which has a strong correlation with rising crude oil prices, the correlation unit 27 calculates a correlation coefficient based on past crude oil price data and its date, and past price data of stock A and its date (see the third diagram from the top in Figure 37). Next, the correlation unit 27 calculates the correlation coefficient by shifting the date of the price data of stock A backward while leaving the date of the crude oil price unchanged (see the second diagram from the top in Figure 40). If the difference with the strongest correlation (the correlation coefficient closest to 1) is one and a half months (in this example, the correlation coefficient increases from 0.58 to 0.66), the time lag unit 28 determines this as a time lag (see the top diagram in Figure 16). As a result, a time lag of one and a half months can be derived from past data. However, this derivation using the correlation coefficient is merely a representative example, and all methods for determining the time lag between the two are included.

[0426] (effect) If it can be determined that the time lag is one and a half months, it will be very effective to be able to predict the expected rate of increase one and a half months later by purchasing stock A when a crude oil price rise event occurs. This system can be applied not only to crude oil price rises, but also to the relationship between all events with a time lag and the price fluctuations of all investment products.

[0427] (Database recording process) Figure 38 is a diagram showing the relationship between the event increase rate, the stock price increase rate, and the time lag. Recording is performed in the database unit 103 in a format similar to the table shown in Figure 38 (event (crude oil price) increase rate 11%, stock A increase rate 33%, time lag 1 month). The relationship table creation unit 30 creates a combined table that can display the relationship between event changes and investment product price fluctuations at a glance. Furthermore, the relationship table creation unit 30 records a correlation table that records the correlation between the two, and a date adjustment table that records the time lag, in the database unit 103. As long as the relationship between the two is recorded, the format is not important.

[0428] According to the above, the relationship table is stored in the database unit 103 and can be read out at any time. For example, when searching by semiconductor production volume, related stocks are extracted, and a relationship table between increases in semiconductor production volume and the rate of increase in stock prices is extracted. This has the effect of clarifying correlations and stocks that are likely to rise when semiconductor production volume increases.

[0429] (Best stock determination system) (assignment) It is unclear which investment products will have the greatest impact on price fluctuations when crude oil prices rise. This is because there is a time lag and multiple other events are occurring, making it difficult to understand and complex. If the best investment products were identified when crude oil prices rise, it would be possible to narrow down the target, but this is not possible.

[0430] (Means for solving the problem) Fig. 41 is a diagram showing the processing of the conversion device 102 and the database unit 103. Fig. 42 is a diagram explaining the processing of the best stock determination system.

[0431] As shown in Figure 41, in the event selection step, the conversion device 102 identifies an event with a large current change. If the event is crude oil prices, the conversion device 102 calculates the correlation coefficient of each stock based on the crude oil price data change rate and date, and the investment product price data change rate and date. The conversion device 102 determines the top stock (the stock with a correlation coefficient close to 1) as the best stock.

[0432] Taking time lag into account is expected to result in greater accuracy. For example, determine the top 10 stocks without time lag, then use the time lag determination system described above to determine the correlation coefficient for the top 10 stocks, determine the time lag, and compare the correlation coefficients taking the time lag into account. One example would be to determine the stock with the strongest correlation as the best stock. However, before making this determination, the magnitude of the rate of change of the event becomes important. Even better results can be achieved by determining the best stock when an event with a large rate of change occurs.

[0433] (effect) If the best stocks can be predicted when an event occurs, a very high effect can be expected. By transmitting the information to the systems of the embodiments A1 to A4 and B1 to B5 described later, an even greater effect can be expected.

[0434] (Period determination system (cleansing)) (assignment) While it is often possible to derive correlations between crude oil price data and its dates and investment product price data and its dates through best stock and time lag determinations, there are periods when the price is susceptible to the influence of other events and periods when it is not. For example, in the case of crude oil prices, when there is little movement in crude oil prices, does that mean that there is little movement in the relevant stock? It is not the case, and there is a high possibility that other factors will influence the price. Conversely, when there are drastic changes in crude oil prices, the correlation is expected to become stronger. There is a challenge in being unable to distinguish between periods with strong correlation and periods with weak correlation. Recognizing periods with strong correlation and forecasting during those periods is expected to improve accuracy.

[0435] (Means for solving the problem) For example, the problem can be solved by dividing crude oil price data into periods of upward trend, downward trend, and slow movement, and then comparing only the periods of upward trend and downward trend with investment product price data. This is expected to result in periods with a stronger correlation than normal periods. There are various methods for determining periods with a stronger correlation, and all of these methods are included. Another method, known as a cleansing process, is to remove data with special factors and strengthen the correlation. One method involves defining the period of an upward trend in some way, such as using the date on the x-axis and the crude oil price on the y-axis, and starting at a point when the slope begins to change from 0 to 1. This method includes all methods for determining periods with a stronger correlation.

[0436] (effect) By setting the learning period to a period with a strong correlation, it is expected that the accuracy of predictions will increase. Furthermore, by combining it with methods such as the time lag determination and best stock determination mentioned above, it is expected that the special effect of even higher accuracy will be achieved. All of these examples use crude oil price data and stock price fluctuations as representative examples, but they also apply to the relationship between all other events and the price fluctuations of all other investment products. They also apply to both table lookup methods and AI methods.

[0437] (A method of estimating the relationship between macro index change events and investment product price fluctuations using a table lookup method) FIG. 20 is a diagram showing a method for estimating the relationship between macro index change events and price fluctuations of investment products using a table reference method.

[0438] (Specific processing examp...

Claims

1. A terminal device used by a user identified by unique information and having an operation unit, a display unit, and a communication unit; and the price of a variable commodity at a given time; Past price fluctuation data of the volatile product; and an information processing device that estimates the price of a price-varying commodity after the predetermined time point based on data on events that affect the price of the price-varying commodity; The information processing device performs a judgment process in response to a question about stocks input by the user through a text input unit of the terminal device, and outputs a result of the judgment process as a text to the terminal device used by the user. A product price estimation system comprising:

2. The commodity price estimation system described in claim 1, characterized in that in response to the user's question regarding stocks entered as text from the operation unit of the terminal device, the information processing device outputs the result as text to the terminal device and displays it on the display unit of the terminal device almost in real time.

3. 2. The product price estimation system according to claim 1, wherein the question is a question in a natural language, and the result output by the information processing device to the terminal device is a result created in the natural language.

4. The product price estimation system according to claim 1, characterized in that the question includes conditions related to the company, performance, or stock, or a combination of these conditions, and is written in natural language.

Images