JP6870020B2 - Market Prediction Management System, Market Prediction Management Method and Market Prediction Management Program - Google Patents

Description

The present invention relates to a market precursor management system, a market precursor management method, and a market precursor management program for detecting market precursors.

In the stock market, it is important to predict fluctuations in transaction prices. For example, a technique for predicting market trends using machine learning such as deep learning is being studied (see, for example, Patent Document 1). In the technique disclosed in Patent Document 1, the forecast control unit of the transaction management system uses a data set including a stock price fluctuation history for each combination of a forecast period candidate and a stock price fluctuation range candidate for a plurality of issues. Then, for the stock price after the forecast period candidate, the forecast fluctuation range is calculated by machine learning consisting of two or more hidden layers. Next, the actual stock price fluctuation range candidate after the forecast period candidate is compared with the predicted fluctuation range, and the combination of the forecast period and the fluctuation range with high prediction accuracy is specified. Then, when the order information is acquired, the relationship between the forecast stock price after the forecast period of this combination and the fluctuation range is calculated, and the transaction execution timing is determined based on the relationship with the fluctuation range.

Further, a technique for predicting the transition of events is being studied based on the transition of causal relationships between events (see, for example, Patent Document 2). In the technique disclosed in Patent Document 2, the time-series predictor is used for time-series data related to each of a plurality of events including the event to be predicted, and time-series data related to the causal relationship between the events. Based on this, the degree of relevance, which is an index showing the strength of the causal relationship between the above events, is calculated, and the transition of the time series data related to the above events is predicted based on the calculated degree of relevance. The time-series predictor calculates the degree of relevance using the co-occurrence frequency of terms related to the above-mentioned events in the time-series data related to the causal relationship between the above-mentioned events. The time-series prediction device builds a plurality of prediction models that predict the transition of the time-series data related to the event to be predicted based on the time-series data related to the event having a causal relationship with the event to be predicted. By weighting each prediction model according to the degree of relevance, the prediction results of each prediction model are integrated.

Japanese Unexamined Patent Publication No. 2018-28945 International Publication No. 2016/063341

However, even when a prediction model is used, it is not always possible to accurately capture a sign with one prediction model. Moreover, when the trend is directly predicted by the prediction model, the basis of the prediction is not clear.

The market predictive management system for solving the above problems includes a history information storage unit that records the feature amount of each past day calculated by a plurality of prediction models using the market data related to the past day for each past day. It is equipped with a control unit that detects market trends using a plurality of forecast models. Then, the control unit calculates the feature amount of the prediction target day by each prediction model using the market data related to the prediction target date, and records the calculated feature amount and the history information storage unit. By comparing with the feature amount of each past day, the similar day is specified according to the similarity degree of the feature amount on the past day, the market data of the similar day is acquired, and based on the market data. Calculate predictive information.

According to the present invention, it is possible to efficiently manage the signs in the market by utilizing a plurality of prediction models in a cross-cutting manner.

Explanatory drawing of the system of this embodiment. The explanatory view of the hardware configuration of this embodiment. It is explanatory drawing of the data recorded in the storage part of this embodiment, (a) is the model storage part, (b) is the evaluation information storage part, (c) is the explanatory view of the history storage part. The explanatory view of the processing procedure of this embodiment. The explanatory view of the processing procedure of this embodiment.

Hereinafter, an embodiment of a market predictive management system, a market predictive management method, and a market predictive management program will be described with reference to FIGS. 1 to 5. In the present embodiment, a case will be described in which output values using a plurality of prediction models are calculated, similar days in the past history are specified for each prediction result, and prediction is performed using the actual distribution of the similar days.
As shown in FIG. 1, in the market predictive management method of the present embodiment, the user terminal 10, the predictive management server 20, and the market information management server 30 are used.

(Hardware configuration example)
FIG. 2 is a hardware configuration example of the information processing device H10 that functions as a user terminal 10 to a market information management server 30 and the like.

The information processing device H10 includes a communication device H11, an input device H12, a display device H13, a storage unit H14, and a processor H15. Note that this hardware configuration is an example, and other hardware may be included.

The communication device H11 is an interface that establishes a communication path with another device and executes data transmission / reception, such as a network interface card or a wireless interface.

The input device H12 is a device that receives input from a user or the like, and is, for example, a mouse, a keyboard, or the like. The display device H13 is a display, a touch panel, or the like that displays various information.

The storage unit H14 is a storage device that stores data and various programs for executing various functions of the user terminal 10 and the sign management server 20. For example, the storage unit H14 stores various information used for predictive management. An example of the storage unit H14 is a ROM, a RAM, a hard disk, or the like.

The processor H15 controls each process (for example, predictive management process) in the user terminal 10 to the market information management server 30 by using the program or data stored in the storage unit H14. Examples of the processor H15 include a CPU, an MPU, and the like. The processor H15 expands a program stored in a ROM or the like into a RAM and executes various processes corresponding to various processes. For example, when the application program of the user terminal 10 and the sign management server 20 is started, the processor H15 operates a process for executing each process shown in FIGS. 3 and 4 described later.

The processor H15 is not limited to the one that performs software processing for all the processing executed by itself. For example, the processor H15 may include a dedicated hardware circuit (for example, an integrated circuit for a specific application: ASIC) that performs hardware processing for at least a part of the processing executed by the processor H15. That is, the processor H15 is (1) one or more processors that operate according to a computer program (software), (2) one or more dedicated hardware circuits that execute at least a part of various processes, or ( 3) It can be configured as a circuitry including a combination thereof. The processor includes a CPU and a memory such as a RAM and a ROM, and the memory stores a program code or an instruction configured to cause the CPU to execute a process. Memory or computer-readable media includes any available medium accessible by a general purpose or dedicated computer.

(Function of predictive management system)
Next, the functions of the user terminal 10 and the sign management server 20 as the sign management system will be described.
The user terminal 10 is a computer terminal used by a person in charge of predictive management. The user terminal 10 includes an input unit (keyboard, pointing device, etc.) for inputting various information and an output unit (display, printer, etc.) for outputting various information.

The sign management server 20 is a computer system that manages signs of the market. The predictive management server 20 includes a control unit 21, a model storage unit 22, an evaluation information storage unit 23, and a history storage unit 24.

The control unit 21 stores the predictive management program, and uses the information recorded in the model storage unit 22 to the history storage unit 24 to perform predictive management processing (learning stage, individual prediction stage, integrated prediction stage, and evaluation according to the predictive management program. Execute each process of the stage). Therefore, by executing the predictive management program, the control unit 21 functions as a learning unit 210, an individual prediction unit 211, an integrated prediction unit 212, and an evaluation unit 213.

The learning unit 210 uses the market data to generate and update each prediction model.
The individual prediction unit 211 calculates a plurality of prediction results using a plurality of prediction models. In the present embodiment, as a prediction result, the feature amount of the latest market trend is calculated.

The integrated prediction unit 212 makes an integrated prediction using a plurality of prediction results calculated by the individual prediction unit 211. In the present embodiment, past days (similar day candidates) having similar market features are specified based on the prediction result. Then, the similar date candidates are weighted to identify the similar dates, and the trends in the market are predicted based on the similar dates.
The evaluation unit 213 optimizes the prediction model and weighting based on the comparison with the actual market result.

As shown in FIG. 3A, a plurality of prediction models 220 for predicting market trends are recorded in the model storage unit 22. An identifier (model ID) for identifying each prediction model 220 is assigned. This prediction model 220 is a prediction method (pattern recognition means, function, network, etc.) for calculating an index (feature amount) representing a future market trend using each index value in the market. For example, as a pattern recognition means, a method of indexing the similarity of the correlation of each index of the market as a market structure and supplementing the change of this index by pattern recognition can be used (market structural change prediction model). Further, as another pattern recognition means, it is possible to use a method of generating a phase in which change patterns (structural indexes) of market data are clustered and analyzing the characteristics of the phase and the characteristics of the phase transition (market phase prediction model). ). Further, as the function, for example, a method of calculating the feature amount of the market data by using a regression analysis model can be used. As the network, an intermediate layer (network) generated by deep learning or the like can be used.

As shown in FIG. 3B, the model evaluation table 230 is recorded in the evaluation information storage unit 23. In this model evaluation table 230, a weighting value (model evaluation information) for weighting the model ID for specifying each prediction model is recorded. This weighted value is information that evaluates the certainty of the output value in each prediction model. This weighted value is updated by the evaluation unit 213 when the market value is acquired.

As shown in FIG. 3C, the history management record 240 is recorded in the history storage unit 24. This history management record 240 is recorded when the market result of each day is acquired. The history management record 240 records data on past dates, output values associated with each model ID, and market results.

In the past date data area, data related to the date (past date) of the forecast is recorded.
An identifier for identifying each prediction model is recorded in the model ID data area.
In the output value data area, data related to the output value calculated using each prediction model is recorded. Here, the output value (feature amount) that predicts the market trend of the past date is recorded by using the market data of the date before the past date. In this embodiment, since a plurality of prediction models are used, a plurality of sets of model IDs and output values are recorded.
The actual market data of the past date is recorded in the market result data area.

The sign management server 20 is connected to the market information management server 30. The market information management server 30 is a computer system that manages various market data. This market data includes various indicators of the market (for example, price-weighted stock index, free-float-based stock index, market value-weighted stock index, etc.). The predictive management server 20 executes the market predictive management process using the market data acquired from the market information management server 30.

(Forecast model management process)
Next, the processing procedure of the prediction model management process will be described with reference to FIG. 4 in the predictive management server 20 configured as described above. This predictive model management process is executed periodically.

First, the control unit 21 of the predictive management server 20 executes the market data acquisition process (step S1-1). Specifically, the learning unit 210 of the control unit 21 acquires market data from the market information management server 30. Here, market data for a predetermined period (for example, 20 years) in the past is acquired.

Then, the control unit 21 of the predictive management server 20 executes a prediction model generation process for each prediction method (step S1-2). Specifically, the learning unit 210 of the control unit 21 generates each prediction model by the learning method of each prediction method. Then, the learning unit 210 updates the prediction model 220 of the model storage unit 22 by using each of the generated prediction models. Further, when the prediction model 220 is updated, the learning unit 210 calculates the output value of each past date using the updated prediction model 220, and updates the history management record 240 of the history storage unit 24.

(Market predictive management processing)
Next, with reference to FIG. 5, a processing procedure in the case of managing market signs in the sign management server 20 configured as described above will be described. This process is performed when the execution instruction of the market predictive management process on the forecast target day (for example, the current day) is acquired from the user terminal 10.

First, the control unit 21 of the predictive management server 20 executes the market data acquisition process (step S2-1). Specifically, the learning unit 210 of the control unit 21 acquires the market data used for each prediction model from the market information management server 30 in order to predict the market trend on the forecast target date from the market information management server 30. ..

Next, the control unit 21 of the predictive management server 20 executes an output value acquisition process for each prediction model (step S2-2). Specifically, the individual prediction unit 211 of the control unit 21 applies each prediction model 220 recorded in the model storage unit 22 to the acquired market data, and calculates the output value of each prediction model 220. Then, the individual prediction unit 211 temporarily stores the output value in association with the model ID.

Next, the control unit 21 of the sign management server 20 executes the extraction process of the similar day candidate using each output value (step S2-3). Specifically, the integrated prediction unit 212 of the control unit 21 extracts similar days in which the calculated output value is included in the predetermined range of the output value recorded in the history storage unit 24 for each prediction model 220. Then, the integrated prediction unit 212 temporarily stores the similar date candidate in association with the model ID.

Next, the control unit 21 of the predictive management server 20 executes the identification of similar dates using the weighted value of the prediction model (step S2-4). Specifically, the integrated prediction unit 212 of the control unit 21 acquires the weighted value of each prediction model 220 from the model evaluation table 230 of the evaluation information storage unit 23. Next, the integrated prediction unit 212 narrows down the extracted similar day candidates to a plurality of similar day candidates having a high degree of similarity by using the weighted values of each prediction model 220. For example, the integrated prediction unit 212 prioritizes the similar day candidates associated with the model ID having a high weighting value over the similar day candidates associated with the model ID having a low weighting value, so that the similar days having a high degree of similarity can be selected. Identify. Then, the integrated prediction unit 212 specifies similar day candidates narrowed down to a plurality of similar days using the weighted value as similar days.

Next, the control unit 21 of the predictive management server 20 executes the acquisition process of the market results of a plurality of similar days (step S2-5). Specifically, the integrated prediction unit 212 of the control unit 21 acquires the market results on the specified plurality of similar days from the history storage unit 24.

Next, the control unit 21 of the predictive management server 20 executes a calculation process of statistical values of market results of a plurality of similar days (step S2-6). Specifically, the integrated forecasting unit 212 of the control unit 21 calculates the statistical values of the market results on a plurality of acquired similar days. Here, as the statistical value, the similar day actual distribution (first statistical value) of the market results of a plurality of similar days is calculated. As the similar day actual distribution, for example, a frequency distribution for each price movement value is used.

Next, the control unit 21 of the predictive management server 20 executes a comparison process of the statistical values of similar days and the overall statistical values (step S2-7). Specifically, the integrated prediction unit 212 of the control unit 21 acquires the market results of all the past dates of a predetermined predetermined period (for example, 20 years of the entire period) from the history storage unit 24. Next, the integrated forecasting unit 212 calculates the statistical value of the market result for the entire period (the actual distribution for the entire period as the second statistical value) in the same manner as in step S2-6. Then, the actual distribution for the entire period and the actual distribution for similar days are compared.

Next, the control unit 21 of the predictive management server 20 executes the market trend prediction process based on the comparison result (step S2-8). Specifically, the integrated prediction unit 212 of the control unit 21 calculates the prediction value according to the difference between the total period actual distribution and the similar day actual distribution. For example, when the frequency of the similar day actual distribution on the price increase tendency side is high with respect to the entire period actual distribution, the strength of the "price increase tendency" is given to the user terminal 10 as a predicted value according to the difference between the two distributions. Output.

After that, when the actual market data is calculated for the forecast target date, the control unit 21 of the predictive management server 20 executes the acquisition process of the market result (step S2-9). Specifically, the evaluation unit 213 of the control unit 21 acquires the actual market result of the forecast target date on which the forecast is made from the market information management server 30.

Next, the control unit 21 of the predictive management server 20 executes the weighting optimization process (step S2-10). Specifically, the evaluation unit 213 of the control unit 21 acquires the market result of the similar day candidate temporarily stored in the memory in association with the model ID from the history storage unit 24. Next, the evaluation unit 213 adjusts the weighting value so that the degree of similarity between the similar day actual distribution calculated from the market results of the similar day candidates and the actual market results is maximized. For example, based on the weighted value of the model evaluation table 230 recorded in the evaluation information storage unit 23, the similar day actual distribution approaches the actual market result (high degree of similarity) within the adjustable range of the weighted value. The weighting value is adjusted so as to become). Then, the evaluation unit 213 updates the model evaluation table 230 of the evaluation information storage unit 23 by using the adjusted weighting value.

According to the market predictive management system of the present embodiment, the following effects can be obtained.
(1) In the present embodiment, the control unit 21 of the predictive management server 20 executes the predictive model management process. This makes it possible to generate multiple forecast models for evaluating market data.

(2) In the present embodiment, the control unit 21 of the predictive management server 20 executes an output value acquisition process for each prediction model (step S2-2). As a result, market data can be evaluated by a plurality of models.

(3) In the present embodiment, the control unit 21 of the predictive management server 20 uses each output value to extract similar day candidates (step S2-3), and uses the weighted value of the prediction model to identify similar days. The process (step S2-4) is executed. This makes it possible to identify past dates that are similar to the current market conditions, based on the certainty of each forecast model.

(4) In the present embodiment, the control unit 21 of the predictive management server 20 acquires market results of a plurality of similar days (step S2-5) and calculates statistical values of market results of a plurality of similar days (step). Execute S2-6). This makes it possible to predict future market trends using trends on similar days.

(5) In the present embodiment, the control unit 21 of the predictive management server 20 compares the statistical values of similar days with the overall statistical values (step S2-7), and predicts the market trend based on the comparison results (step). Execute S2-8). This makes it possible to grasp the signs of market trends based on the difference between the overall situation and the current situation.

(6) In the present embodiment, the control unit 21 of the predictive management server 20 executes the market result acquisition process (step S2-9) and the weighting optimization process (step S2-10). This makes it possible to review the evaluation of each prediction model.

This embodiment can be modified and implemented as follows. The present embodiment and the following modified examples can be implemented in combination with each other within a technically consistent range.
-In the above embodiment, a plurality of prediction models 220 for predicting market signs are recorded in the model storage unit 22. As the prediction model 220, a market structural change prediction model, a market phase prediction model, a regression analysis model, a network, and the like are used. A plurality of prediction models 220 may be used, and each prediction model 220 is not limited to the above method.

-In the above embodiment, acquisition processing of market results of a plurality of similar days (step S2-5), comparison processing of statistical values of similar days and overall statistical values (step S2-7), and weighting optimization processing (step S2). In -10), the market result is acquired from the history storage unit 24. The market result may be acquired from the market information management server 30. In this case, the market information management server 30 stores the past market results.

-In the above embodiment, the control unit 21 of the predictive management server 20 executes the market result acquisition process (step S2-9) and the weighting optimization process (step S2-10). Here, the weighting value is adjusted using the output value of each prediction model 220 on the prediction target date. The weighting optimization method is not limited to this. For example, the output value of the similar day candidate in a predetermined period may be used to adjust the weighting value of each prediction model 220 so as to match the actual market result.

-In the above embodiment, the current day is used as the forecast target date, but it may be a future date. For example, past market data may be used to predict market trends in a predetermined number of days ahead.
-In the above embodiment, the control unit 21 of the predictive management server 20 executes a comparison process of the statistical value of the similar day and the overall statistical value (step S2-7). Here, the integrated forecasting unit 212 of the control unit 21 acquires the market results for the entire period from the history storage unit 24, and calculates the statistical value of the market results for the entire period (actual distribution for the entire period). Instead of this, the actual distribution for the entire period may be calculated in advance and recorded in the storage unit of the predictive management server 20. In this case, every time the market data is updated, the actual distribution for the entire period is updated.

-In the above embodiment, the control unit 21 of the predictive management server 20 executes the market trend prediction process based on the comparison result (step S2-8). In this case, the integrated prediction unit 212 of the control unit 21 outputs the prediction result to the user terminal 10. Here, the output destination of the prediction result is not limited to the user terminal 10. For example, it may be output to a transaction execution system for conducting a transaction, and the transaction execution system may execute the transaction based on the prediction result.

10 ... user terminal, 20 ... predictive management server, 21 ... control unit, 210 ... learning unit, 211 ... individual prediction unit, 212 ... integrated prediction unit, 213 ... evaluation unit, 22 ... model storage unit, 23 ... evaluation information storage unit , 24 ... History storage unit, 30 ... Market information management server.

Claims (5)

For each past day, a history information storage unit that records the features of each past day calculated by multiple forecast models using market data related to the past day,
It is a market predictive management system equipped with a control unit that detects market trends using multiple forecast models.
The control unit
Using the market data related to the forecast target date, the feature amount of the forecast target date is calculated by each forecast model.
The calculated feature amount is compared with the feature amount of each past day recorded in the history information storage unit, and on the past day, similar days are specified according to the similarity degree of the feature amount.
The market data of the similar day is acquired , and the first statistical value is calculated using the market data of the similar day.
Calculate the second statistic using the market data of the past day of the specified period,
A market predictive management system characterized in that a difference between the first statistical value and the second statistical value is calculated, and predictive information corresponding to the difference is calculated. For each prediction model, an evaluation information storage unit that records model evaluation information according to the certainty of the prediction is further provided.
The control unit
The market predictive management system according to claim 1 , wherein the model evaluation information recorded in the evaluation information storage unit is used to identify the similar dates on which the feature quantities calculated by the prediction models are similar. .. The control unit
Acquire the actual value of the market data of the forecast target date,
In the model evaluation information recorded in the evaluation information storage unit, the model evaluation information adjusted so that the similarity between the actual value of the market data on the forecast target date and the first statistical value is maximized is calculated. Execute the optimization process and
The market predictive management system according to claim 2, wherein the calculated model evaluation information is updated and recorded in the evaluation information storage unit. For each past day, a history information storage unit that records the features of each past day calculated by multiple forecast models using market data related to the past day,
It is a method of managing market signs using a market sign management system equipped with a control unit that detects market trends using multiple forecast models.
The control unit
Using the market data related to the forecast target date, the feature amount of the forecast target date is calculated by each forecast model.
The calculated feature amount is compared with the feature amount of each past day recorded in the history information storage unit, and on the past day, similar days are specified according to the similarity degree of the feature amount.
The market data of the similar day is acquired , and the first statistical value is calculated using the market data of the similar day.
Calculate the second statistic using the market data of the past day of the specified period,
A market predictive management method characterized in that a difference between the first statistical value and the second statistical value is calculated, and predictive information corresponding to the difference is calculated. For each past day, a history information storage unit that records the features of each past day calculated by multiple forecast models using market data related to the past day,
It is a program for managing market forecasts using a market forecast management system equipped with a control unit that detects market trends using multiple forecast models.
The control unit
Using the market data related to the forecast target date, the feature amount of the forecast target date is calculated by each forecast model.
The calculated feature amount is compared with the feature amount of each past day recorded in the history information storage unit, and on the past day, similar days are specified according to the similarity degree of the feature amount.
The market data of the similar day is acquired , and the first statistical value is calculated using the market data of the similar day.
Calculate the second statistic using the market data of the past day of the specified period,
A market predictive management program characterized in that a difference between the first statistical value and the second statistical value is calculated and used as a means for calculating predictive information according to the difference.

Images