JP2017151695A - Information processor, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processor, a method and a program capable of accurately estimating the total deposit balance of various types of deposits.SOLUTION: An information processor 1 includes: a database 2 storing time series data of a balance of each past deposit and past data of a mutual fund flow of each deposit and a fund transfer flow of reception/payment from/to outside; a coefficient identification section 3 which identifies a coefficient of regression analysis so that the total of differences between a model estimation value of a fund transfer flow of each deposit, a model estimation value of balance data of each deposit and a model estimation value of the total balance data, and actual values of each of the model estimation values becomes the smallest, on the basis of data in a period during which the past data of the fund transfer flow can be referred and data in a period during which the past data cannot be referred; and an estimation section 4 that estimates the total deposit balance in the future on the basis of a regression expression using the identified coefficient.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing apparatus, method, and program.

  Various methods have been proposed for predicting the amount of cash prepared in bank deposits and ATMs (see, for example, Patent Document 1-2).

Japanese Patent No. 5667731 JP 2012-141765 A

  When estimating the total deposit balance, even if the balance estimation models for various deposits such as time deposits and ordinary deposits are optimally designed, the sum of these models is not necessarily the optimal total deposit balance model. In addition, there are many cases in which the period in which funds can be referred retroactively differs between the funds transfer flow data on the network composed of the funds transfer between the deposits and the external input / output and the mere balance data. In some cases, selection of explanatory variables that influence the accuracy of each balance estimation model when the balance is estimated by regression analysis is not appropriate. For several reasons, it is difficult to accurately predict the future total bank deposits.

  Therefore, the present invention provides a technique capable of accurately estimating the total deposit balance of various deposits.

  In order to solve the above problems, in the present invention, the difference between the estimated value of the funds transfer flow of each deposit, the estimated value of the balance data of each deposit, and the actual value of the estimated model of the total balance data. The regression analysis coefficient that minimizes the total is identified based on the data for the period in which the historical data of the funds transfer flow can be traced and the data for the period that cannot be traced, and the future total based on the regression formula using the identified coefficient. We decided to estimate the deposit balance.

  More specifically, the present invention is an information processing apparatus that stores time-series data of each past deposit balance, and past data of a fund transfer flow between each deposit and a deposit / withdrawal funds transfer flow. Regression analysis that minimizes the sum of the difference between the database and the model estimate of the funds transfer flow of each deposit, the model estimate of the balance data of each deposit, and the actual value of the model estimate of the total balance data Coefficient identification means to identify the coefficient based on the period data that can refer to the past data of the funds transfer flow and the data that cannot be referred to, and the future total deposit balance based on the regression equation using the identified coefficient And estimating means.

  With such an information processing apparatus, each fund transfer flow, the balance of each deposit, and the total balance are modeled in a well-balanced manner. Then, an estimation model is constructed that effectively uses past time-series data including a period during which funds transfer flow data can be referred to and a period during which reference cannot be made. Therefore, it is possible to improve the accuracy of the calculated model estimated value of the total deposit balance.

The coefficient specifying means may use each parameter multiplied by the difference, and specify a coefficient for regression analysis that minimizes the sum of the differences multiplied by each parameter. If the information processing apparatus includes the coefficient specifying unit, the user can easily grasp the relationship between each parameter and the model estimated value of the total deposit balance.

  The coefficient specifying means may further calculate a validity index of the model of the fund transfer flow, balance data, and total balance data while sequentially changing each parameter. If each parameter is sequentially changed while calculating the validity index, an optimal estimation model can be generated while viewing the validity index.

  Further, the present invention may be one in which the information processing apparatus is captured from the aspect of a method or a program.

  With the information processing system, method and program, it is possible to accurately estimate the total deposit balance of various deposits.

FIG. 1 is a diagram illustrating a configuration of an information processing apparatus according to the embodiment. FIG. 2 is an image diagram showing an example of data stored in the database. FIG. 3 is a flowchart showing the processing contents realized in the information processing apparatus. FIG. 4 is an image diagram of a model estimated value of the future total deposit balance calculated by the estimation unit.

  Hereinafter, embodiments of the present invention will be described. The following embodiments are merely examples, and the technical scope of the present invention is not limited to the following modes.

FIG. 1 is a diagram illustrating a configuration of the information processing apparatus. The information processing apparatus 1 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory),
A communication interface is provided, and various functional units are realized by executing computer programs. The information processing apparatus 1 realizes the coefficient specifying unit 3 and the estimation unit 4 that perform various arithmetic processes using data in the database 2 by executing a balance estimation program that estimates the total deposit balance of the bank, for example.

  The database 2 stores time series data D1 of each fund transfer flow, time series data D2 of each deposit balance, time series data D3 of total deposit balance, and time series data D4 of explanatory variables. FIG. 2 is an image diagram showing an example of data stored in the database 2. The time series data D1 of each money transfer flow is the past time series data of the money transfer flow between the deposits and the outside. Further, the time series data D2 of each deposit balance is past time series data of the balance of each deposit. The total deposit balance time-series data D3 is past time-series data of the total deposit balance obtained by totaling the deposits. The explanatory variable time-series data D4 is past time-series data relating to matters relating to the fund transfer flow, and includes various data that affects the fund transfer flow such as deposit interest rate and market interest rate. is there.

  In FIG. 2, three pieces of time series data D1 of each money transfer flow and three pieces of time series data D2 of each deposit balance are shown. However, the information processing apparatus 1 of the present embodiment uses the number of deposits handled by a bank. It is possible to process an appropriate amount of data according to. The database 2 may be stored in a storage device provided separately from the information processing device 1, or may be stored in a storage device in the information processing device 1.

  The coefficient specifying unit 3 is a functional unit mainly responsible for the process of specifying a regression analysis coefficient used for estimating the total deposit balance of the bank. Moreover, the estimation part 4 is a function part which mainly bears the process which estimates the future total deposit balance based on a regression formula.

  Hereinafter, processing contents realized in the information processing apparatus 1 will be described. FIG. 3 is a flowchart showing the processing contents realized in the information processing apparatus 1. When the information processing apparatus 1 executes the above balance estimation program, the coefficient specifying unit 3 performs each of the model estimated value of each fund transfer flow, the model estimated value of the balance data of each deposit, and the model estimated value of the total balance data. The coefficient of regression analysis that minimizes the total difference from the actual value (hereinafter referred to as “regression coefficient”) is identified based on the data in the period in which the historical data of the funds transfer flow can be traced and the data in the period that cannot be traced (S101). And the estimation part 4 estimates a future total deposit balance based on the regression equation using the coefficient which the coefficient specific | specification part 3 specified (S102).

  That is, in the above step S101, the coefficient specifying unit 3 performs the difference between the model estimated value of each fund transfer flow and the time series data D1, the difference between the model estimated value of each deposit balance and the time series data D2, the total deposit balance A regression coefficient that minimizes the objective function is calculated on the basis of a function (hereinafter referred to as “objective function”) that represents the sum of differences between the model estimated value and the time series data D3. In addition, since the recording period of the time series data D1 of the money transfer flow is shorter than the other time series data D2, D3, D4, in the objective function, these differences are recorded as the time series data D1 of the money transfer flow. It is prepared separately for a certain period and other periods. The regression coefficient is added to the time series data D4 and the coefficient multiplied to the time series data D4 used when calculating the model estimate value of each money transfer flow from the time series data D4 of the explanatory variable. Value. The model estimated value of each deposit balance is calculated based on the model estimated value of the fund transfer flow. The model estimated value of the total deposit balance is calculated based on the model estimated value of each deposit balance.

  In step S102, the estimation unit 4 calculates a model estimated value of the future total deposit balance using the regression coefficient calculated in step S101. The model estimate of the future total deposit balance is the model estimate of the funds transfer flow obtained by multiplying the appropriate time-series data of explanatory variables such as future deposit interest rate and market interest rate by the above regression coefficient. After calculating the model estimated value of each deposit balance using the value, it is obtained by adding the calculated model estimated value of each deposit balance.

  FIG. 4 is an image diagram of a model estimated value of the future total deposit balance calculated by the estimating unit 4. When the above-described series of processing is executed, the information processing apparatus 1 calculates a model estimated value of the future total deposit balance based on explanatory variables such as a future deposit interest rate and a market interest rate that are appropriately set. The information processing apparatus 1 outputs the calculated model estimated value of the future total deposit balance to a display device such as a display. The information processing apparatus 1 may output the calculated model estimated value of each future deposit balance and the model estimated value of each fund transfer flow to a display device such as a display.

  In general, the total deposit balance fluctuates depending on the deposit interest rate, market interest rate, and the like. Therefore, for example, it may be possible to create a model that fits the past total balance data by regression analysis using these as explanatory variables. However, it is difficult to construct a model with high accuracy by regression analysis using deposit interest rates and market interest rates as explanatory variables. Therefore, it is conceivable to construct a model with higher accuracy by decomposing the total deposit into smaller parts, constructing a balance estimation model for each part, and summing them. However, when the estimated value of each deposit balance is summed to make the total deposit balance, even if the estimation model of each deposit balance is optimally designed (difference minimized), the estimation model of the total deposit balance as the sum of them Is not necessarily the optimum (minimum difference). In other words, even if partial optimization is accumulated, it is not necessarily the overall optimization. As the cause, for example, it is conceivable that, as a result of partial optimization, there is a result of having a positive correlation between partial differences due to chance or a common difference factor, and the sum increases when summed.

  In this regard, if the processing is executed by the information processing apparatus 1 of the present embodiment, the degree of optimization of the estimated value of each money transfer flow and the estimated value of each deposit balance, and the overall (accumulation to each deposit balance, Since the estimation model of the total deposit balance in which the degree of suppression of the deterioration in the optimality in the total sum of the total deposit balance) is balanced is established, the accuracy of the model estimate value of the calculated total deposit balance is relatively high.

  Further, in the data held by the financial institution, for example, as illustrated in FIG. 2, there are cases where the recording period of the fund transfer flow data is different from the data recording period of each deposit balance data or total deposit balance. In this case, if the estimation model is built for data with a short data recording period, the data acquired in the long term is not used effectively. It may exceed the allowable range.

  In this regard, if the process is executed by the information processing apparatus 1 of the present embodiment, for example, regarding the money transfer flow, all the past time-series data is obtained regardless of the period in which the data can be referred to and the period in which the data cannot be referred to. Since the estimation model used effectively is constructed, the possibility that the difference exceeds the allowable range can be reduced even in the back test using long-term data, for example.

<Modification 1>
By the way, a parameter set appropriately may be multiplied by each of the differences constituting each term of the objective function. For example, in step S101 described above, the coefficient specifying unit 3 calculates the regression coefficient that minimizes the objective function by using the sum of the appropriately set parameters multiplied by the respective differences as the objective function. You may do it. Here, the parameter multiplied by each difference is a parameter for controlling the degree of optimization in the regression analysis, and the following settings are conceivable.

  (Setting example 1) For example, if for some reason, only the time series data D1 of the funds transfer flow is to be subjected to regression analysis, the difference between the model estimated value of each fund transfer flow and the time series data D1 is “ A regression coefficient that minimizes the objective function is calculated by multiplying the difference by 1 and multiplying the difference by "0" as the objective function. When the regression coefficient calculated from such an objective function is used, a model estimate value of the total deposit balance excluding the influence of data other than the time series data D1 of the funds transfer flow is calculated.

  (Setting example 2) For example, if for some reason, only the time series data D1 of the fund transfer flow and the time series data D2 of each deposit balance are to be subjected to regression analysis, Multiply the difference from the time series data D1 by “0.5”, multiply the difference between the model estimated value of each deposit balance and the time series data D2 by “0.5”, Then, the product of “0” is used as the above objective function, and a regression coefficient that minimizes the objective function is calculated. When the regression coefficient calculated from such an objective function is used, a model estimate value of the total deposit balance excluding the influence of data other than the time series data D1 and D2 of the funds transfer flow and each deposit balance is calculated. Become.

  (Setting example 3) For example, for some reason, when it is desired to increase the influence level of the time series data D1 of the funds transfer flow in the setting example 2 and reduce the influence level of the time series data D2 of each deposit balance, Multiply the difference between the model estimated value of each money transfer flow and the time series data D1 by “1.0”, and set “0.5” to the difference between the model estimated value of each deposit balance and the time series data D2. A regression coefficient that minimizes the objective function is calculated by multiplying by the above objective function. When the regression coefficient calculated from such an objective function is used, a model estimate value of the total deposit balance is calculated, in which the influence of the time series data D1 of the funds transfer flow is higher than in the case of setting example 2 above. Become.

  For example, if the parameter set for each difference in the first modification changes automatically as follows, the user can easily enjoy the desired balance estimation model. For example, the coefficient specifying unit 3 changes each parameter multiplied by each difference in small increments within a range of variation width of each parameter set appropriately. For example, the range of the fluctuation range of each parameter may be set by the user, or may be set in advance in the program. Further, the amount of change of each parameter may be, for example, a value obtained by dividing the fluctuation range of each parameter by an appropriate number of changes, or may be a value of an appropriately set amount of change.

  The coefficient specifying unit 3 calculates the regression coefficient each time while changing the parameter set for each difference in increments, and the estimation unit 4 calculates various model estimated values of the total deposit balance according to each parameter. The user can easily grasp the relationship between each parameter multiplied by each difference and the estimated model value of the calculated future total deposit balance.

<Modification 2>
Further, the explanatory variable used from the time series data D4 in the calculation of the model estimated value of the total balance data may be selected according to a predetermined constraint condition, for example. Here, the predetermined constraint condition is a condition that is inevitably determined in the relationship between the explanatory variable and the money transfer flow, and includes, for example, conditions from various viewpoints such as a necessary condition, a prohibition condition, and a range condition.

  For example, in the case of funds transfer flow from regular deposits to time deposits, it is natural for customers to think that they have moved funds to time deposits because of the interest rate of time deposits. In the funds transfer flow, it is conceivable to include at least the interest rate of time deposits as an explanatory variable. Also, for example, in the case of funds transfer flow from savings to checking account, it is natural that the customer who transferred funds is unlikely to consider the interest rate of time deposits, so from savings account to checking account. In the fund transfer flow, it is possible to make it a prohibition condition not to include the interest rate of time deposits as an explanatory variable. In addition, for example, in the case of funds transfer flow from ordinary deposits to time deposits, it is natural to think that time deposits can be expected to increase deposit balances, so the regression coefficient of time deposit interest rates always takes a positive value. It is possible to make this a range condition.

  If the explanatory variable used from the time series data D4 in the calculation of the model estimate value of the total balance data is selected, for example, in accordance with the constraint conditions as described above and using the validity index, the user can A highly accurate model estimate of the total deposit balance can be obtained. The validity index is an index indicating how effectively the combination of the selected variables can contribute to the difference reduction of the model. For example, various indexes such as the Akaike Information Criterion (AIC) can be applied.

  In general, the accuracy of the model depends on the validity of the selection of the explanatory variables. The method of selecting explanatory variables may differ depending on how to fit by regression analysis. Therefore, it is not easy to manually test a combination of explanatory variables with high validity every time fitting is performed by regression analysis. In this regard, if the above modification 1 and modification 2 are appropriately combined, for example, the selection of the explanatory variable can be changed according to the constraint condition, and the validity index can be grasped for each change. It is possible to generate an optimal estimation model while looking at the validity index.

1. Information processing device: 2. Database: 3. Coefficient identification unit: 4. Estimation unit: D1 to D4 Time series data

Claims (5)

A database that stores time-series data of the balance of each deposit in the past, and past data of funds transfer flow between each deposit and deposit / withdrawal between the deposits,
The coefficient of regression analysis that minimizes the sum of the difference between the model estimate value of the funds transfer flow of each deposit, the model estimate value of the balance data of each deposit, and the actual value of the model estimate value of the total balance data, A coefficient specifying means for specifying based on data in a period in which past data of the funds transfer flow can be referred to and data in a period in which it cannot be referred;
An estimation means for estimating a future total deposit balance based on a regression equation using the identified coefficient,
Information processing device. The coefficient specifying means uses each parameter multiplied by the difference, and specifies a coefficient for regression analysis that minimizes the sum of the differences multiplied by the parameters.
The information processing apparatus according to claim 1. The coefficient specifying means further calculates a validity index of a model of the funds transfer flow, balance data, and total balance data while sequentially changing the parameters.
The information processing apparatus according to claim 2. Computer
Model estimation of funds transfer flow of each deposit by referring to time series data of the balance of each deposit stored in the database and past data of funds transfer flow between each deposit and deposit / withdrawal with each other You can refer to the historical data of the funds transfer flow for the coefficient of regression analysis that minimizes the sum of the difference between the actual value of each value of the model, estimated value of the balance data of each deposit, and the estimated value of the total balance data Based on date range data and non-referenceable date data,
Estimate future total deposit balance based on regression equation using specified coefficient,
Information processing method. On the computer,
Model estimation of funds transfer flow of each deposit by referring to time series data of the balance of each deposit stored in the database and past data of funds transfer flow between each deposit and deposit / withdrawal with each other You can refer to the historical data of the funds transfer flow for the coefficient of regression analysis that minimizes the sum of the difference between the actual value of each value of the model, estimated value of the balance data of each deposit, and the estimated value of the total balance data Identifying based on period data and non-referenceable period data,
Performing a step of estimating a future total deposit balance based on a regression equation using the identified coefficient,
Information processing program.

Images