JP6354059B2 - Financial information analysis system and program - Google Patents

Description

The present invention relates to a financial information analysis system and program.

Conventionally, when making a corporate loan or investment decision, a default rate indicating the default of the company is usually calculated and a decision is made based on that value. Therefore, the default rate is required to be calculated with a highly accurate value that reflects the actual situation of the company and the market. Until now, financial institutions have made their own calculations, or roughly grasped the probability of bankruptcy based on credit information from rating agencies. .
However, default rates calculated by financial institutions and credit information from rating agencies have the following problems.
(1) In the case of commonly used regression models, the accuracy is low for SMEs.
(2) There is no real-time property because the calculation is based on the financial statement data.
(3) Credit information of rating agencies is slow to be updated due to the time required for interviews with the target company.
(4) Evaluation criteria and information sources are unclear and poor in objectivity. As a result, there are many cases where evaluations are divided among rating agencies.

  On the other hand, in Patent Document 1, the price of a bond, which is a borrowing document of a company, is credited with the price and attributes of the bond on the assumption that a market evaluation regarding the probability of bankruptcy of the issuing company is incorporated. A technique for calculating a company's creditworthiness by comparing with the price and attributes of a government bond whose risk is almost zero is disclosed.

In addition, since there are many companies that have not issued publicly offered bonds, the number of companies that can estimate creditworthiness is limited, and for most companies, an objective measure that represents creditworthiness such as market interest rates. The fact is that does not exist. For this reason, the technique of Patent Document 1 has a problem that it cannot be applied to the purpose of estimating the creditworthiness of companies that have not issued publicly offered bonds. There is also room for reconsideration of the fact that the bankruptcy data of actual results is not used at all, despite the intention of estimating the creditworthiness of the company.
In recent years, corporate debts, such as corporate loans, that do not have market interest rates are attracting attention as transactions, and there is a need for a corporate loan market interest rate estimation method (pricing) and a consistent credit risk management method. Therefore, the establishment of technology to objectively calculate the numerical value indicating the creditworthiness of companies that have not issued publicly offered bonds is recognized as an urgent need by financial institutions.

  Therefore, in Patent Document 2, spreads are calculated based on the actual spread of each publicly issued bond issuer company at the first time point and the second time point and various financial data indicating the safety and profitability of each publicly issued bond issuer company. A regression model showing correlation with financial data is generated at each time point, and the financial data of the companies that have not issued publicly issued bonds is applied to the regression model at the first time point, thereby spreading the debt of each company at the first time point , The regression formula showing the correlation between the spread and default is calculated based on this spread and the default performance data of companies that have not issued publicly issued bonds between the two points in time. Apply the financial data and remaining years at the time to the regression model at the second time to estimate the debt spread for the company, and use this By substituting in, calculates the probability of default in the future of the public issue unissued companies

JP 2001-125953 A JP 2008-46835 A

  In the above-mentioned conventional technology, when calculating the default rate, a regression model for two time periods is generated and the default rate is calculated based on the regression model. However, in order to adopt an arbitrary point When the time to adopt was not appropriate, the accuracy of the default rate calculation was not necessarily high. In calculating the default rate, it is desirable to calculate based on the latest company data as much as possible. However, when calculating conventional loans, etc., it is calculated based on annual data based on financial statement data. In many cases, there is no choice but to calculate the default rate immediately based on real-time accounting information as much as possible, and it has been necessary to make it easier and more accurate.

  The present invention has been made in view of the above problems, and an object of the present invention is to make it possible to easily calculate the default rate of a target company with higher accuracy.

  In order to solve the above-described problem, the embodiment of the invention described in claim 1 includes default information that is managed in association with company identification information uniquely set for each company and indicates whether there is a default. A company information storage unit that stores company information; and an accounting information storage unit that stores company accounting information that is associated with the company identification information and includes at least monthly income statement data and balance sheet data. , By searching for an item included in the company information, a company selection unit that selects the company accounting information to be used for the learning process, and an item to be used for the learning process among the company accounting information is selected. The value of the item selected by the accounting item selection unit of the accounting item selection unit and the company accounting information selected by using the company selection unit is input as an explanatory variable, and the recursive type A learning model setting unit that generates a learning model to learn using a global network; and a default rate for each company selected by the company selection unit based on the learning model generated by the learning model setting unit. A learning execution unit that repeats learning of the optimal parameters of the learning model with reference to the calculated company default rate and the corresponding company default information, and the learning execution unit completes the learning process. A learning model storage unit that stores a learning model, a model determination unit that determines which of the stored learning models to use, and the learning model determined by the model determination unit, A default rate calculation unit that calculates the default rate in the target month.

  In addition, the learning execution unit calculates a loss value indicating the certainty of the default rate, calculates an average value of the loss values of the companies that are learning targets, and sets the learning model based on the loss value You may make it change the said parameter of a part.

  In addition, the change of the loss value every time learning is executed by the learning execution unit is visually displayed in a graph, and for each condition selected by the company selection unit and the accounting item selection unit, And a display unit that generates a graph and displays the graph on the same screen.

  In addition, the learning execution unit calculates the difference between the calculated default rate for each company and the actual value that is 1 when there is a history of actual default of the company and 0 when there is no loss. May be calculated as

  The learning execution unit may further include a learning model setting unit that sets the number of learning executions, and the learning execution unit may repeat the learning process for the number of times of learning set by the learning model setting unit. Good.

  The learning model setting unit may be configured to set the number of hidden layers and the number of nodes.

  The learning model setting unit may be able to select an activation function to be used for each hidden layer.

  The learning model setting unit may set a learning rate.

  The learning model setting unit may be able to set a batch size.

  Further, the learning model setting unit may be configured to set the number of divisions of cross validation.

  The learning model setting unit may set a dropout rate.

  The learning model setting unit may be configured to set a regularization rate.

  The company selection unit may search and select a company that satisfies at least one of the set maximum value and minimum value with reference to the sales scale of the company to be selected.

  The company selecting unit may search and select a company corresponding to the selected industry by referring to the industry information of the company to be selected.

  A financial data storage unit including financial index information calculated based on the profit and loss statement stored in the accounting information storage unit and a numerical value of a balance sheet; and the learning process among the financial data storage unit And a financial item selection unit for selecting items to be used.

  In the embodiment of the present invention, the default rate of the target company can be easily calculated with higher accuracy.

It is a block diagram which shows the hardware constitutions of the financial information analysis system which concerns on embodiment. It is a block diagram which shows the function structure of the financial information analysis system which concerns on embodiment. It is a figure which shows an example of the operation screen of the financial information analysis system which concerns on embodiment. It is a figure which shows an example of the screen which displays the change of the loss value calculated in the financial information analysis system which concerns on embodiment by a graph. It is a figure which shows an example of the screen which shows the calculation result of the default rate of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the company information storage part of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the accounting information storage part of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the accounting information storage part of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the accounting information storage part of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the financial index information of the financial information analysis system which concerns on embodiment. It is a figure which shows the data structure of the economic index information of the financial information analysis system which concerns on embodiment. It is a flowchart which shows the flow of the process concerning the production | generation of the learning model of the financial information analysis system which concerns on embodiment. It is a flowchart which shows the flow of the process concerning calculation of the default rate of the financial information analysis system which concerns on embodiment.

(Example)
Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a financial information analysis system. As shown in FIG. 1, the financial information analysis system 10 includes a CPU / GPU 11, an HDD 12, a memory 13, an input device 14, an output device 15, and a network I / O 16, and exists on the cloud through the network 20. It is connected to a database 30 of company information and accounting information. These databases 30 may be configured as a collection of a plurality of databases instead of a single database. For example, although accounting information of a large number of companies is stored in a cloud-type accounting service, it is desirable to use an API provided from such a service to acquire data. In this case, the latest company information and company accounting information can always be acquired in real time. On the other hand, instead of such a mode in which the database 30 and the network 20 are connected, a format in which company information or the like is directly taken in through a storage medium may be used.

  FIG. 2 is a block diagram showing a functional configuration of the financial information analysis system. The financial information analysis system 10 includes a company information storage unit 21, an accounting information storage unit 22, a financial data storage unit 23, a learning model storage unit 24, an input information registration unit 25, a company selection unit 26, an accounting item selection unit 27, a financial item. A selection unit 28, a learning model setting unit 29, a learning execution unit 30, a model determination unit 31, a default rate calculation unit 32, and an output unit 33 are provided.

  The company information storage unit 21 stores company information for each company, and FIG. 6 shows the data structure of the company information. Examples of information items include a company code, a Kanji trade name, a Kana trade name, an address code, a Kanji location, a Kana location, a bankruptcy code, a bankruptcy date, a designated company flag, a listing division flag, and an industry code. The bankruptcy code is a code that is set when there is an actual default, and the bankruptcy date describes the date when the default occurred. The default information may not be stored in the same table. For example, a default history table may be provided as another table and associated with the company information.

  The accounting information storage unit 22 stores accounting information for each company, and FIGS. 7A, 7B, and 7C are diagrams illustrating the data structure of the company accounting information. The company accounting information stores monthly values of accounting items included in the balance sheet, manufacturing cost report, and the like, here, monthly total values. That is, data is input in the format of “July 2016” at the data year and month, and data for each month is stored for each company. It is desirable that corporate accounting information can be acquired in real time from data such as accounting software as well as corporate information.

  The financial data storage unit 23 is a financial index calculated based on the corporate accounting information stored in the accounting information storage unit 22. 8 and 9 correspond to the financial index information. Each item in FIG. 8 will be described. The current ratio is a value calculated by the calculation formula of total current assets / total current liabilities × 100 in the corporate accounting information. The fixed long-term conformance rate is a value calculated by the formula of total fixed assets / (total net assets + total fixed liabilities) × 100.

  The fixed long-term conformance rate is a value calculated by the formula of total fixed assets / (total net assets + total fixed liabilities) × 100. The capital adequacy ratio is a value calculated using the formula: Total net assets / total assets × 100. The gearing ratio is a value calculated by a formula of total debt / total net assets × 100. The gross profit margin on sales is a value calculated by the following formula: gross profit / loss total ÷ total sales × 100. The recurring profit ratio is a value calculated by the formula of recurring profit / loss ÷ total sales × 100. Total capital turnover is a value calculated by the formula: total sales / total assets × 100. The total return on capital is a value calculated as: ordinary profit / loss amount / total assets × 100.

  Net income is the amount of net income, and is the total value of net income for the current fiscal year for each month in the corporate accounting information. Operating cash flow is: Ordinary profit / loss + Depreciation and amortization-Corporate tax, Resident tax and Business tax-(Last allowance for doubtful accounts-Allowance for doubtful accounts 12 months ago)-(Recent total trade receivables-12 months ago Total trade receivables) + (total recent trade receivables-total 12 months ago payables)-(total recent inventory totals-12 months previous total inventory).

  The sales increase rate is a value calculated by the formula: (total sales for the last 12 months-total sales for the previous year) / total sales for the previous year x 100. The rate of increase in ordinary profit is a value calculated by the formula: (Recurring profit / loss for the last 12 months-Recurring profit / loss for the previous year) / Recurring profit / loss for the previous year x 100. The forecasted sales growth rate is calculated by (total forecast sales for the current term-total sales for the previous term) / total sales for the previous term x 100. If this term is in the middle of the year, sales for the previous term are also calculated by the total value up to the same month.

  The forecasted ordinary profit margin is calculated by the following formula: (Current Ordinary Profit / Loss Amount-Previous Ordinary Profit / Loss Amount) / Previous Ordinary Profit / Loss Amount x 100. Debt repayment period is ((Short-term debt + Long-term debt + Long-term debt repaid within 1 year + Corporate bonds + Discount bills-Total cash and deposits)-(Total trade receivables-Total trade payables + Total inventory)) ÷ Net income It is a value calculated by the following formula: + Depreciation expense + Long-term prepaid expense amortization + Deferred asset amortization (sales). The interest coverage ratio is a value calculated by the value of (operating income + interest income + dividend income) / (interest expense + discount fee). The actual default rate is the RDB company default rate. Based on the historical database of anonymous companies contributed by Japan Risk Data Bank, Ltd. (RDB) from member financial institutions nationwide, It is a new economic indicator that provides monthly default rates for companies calculated. The above values may be automatically updated every time the corporate accounting information is updated, or may be updated by monthly batch processing, for example. The items in FIGS. 8 and 9 are examples, and other index data may be included in addition to this.

  Further, FIG. 9 includes macroeconomic indicators included in the financial indicator information, and this is where the user directly inputs data. The items in FIG. 9 are merely examples, and other data may be included depending on the type of business. The data shown in FIGS. 7 to 9 may have different data structures for each service when the data is input from an external cloud accounting service or the like. In this case, in order to automatically convert the data structure at the time of data input / output, if a correspondence table of data for each database is provided and data can be automatically converted according to the correspondence table, Even users who have no knowledge of programs can work easily.

  The data stored in each of the storage units 21 to 23 is input to the input information registration unit 25 based on the operation of the input device 14. The input information registration unit 25 acquires data from each of the storage units 21 to 23 based on the conditions selected by the selection units 26 to 28 described below, and processes the data into a predetermined format. The company selection unit 26 determines which company information is used when generating the learning model.

  FIG. 3 shows an operation screen of the financial information analysis system 10, and the company selection unit 26 performs processing based on the operation of the company selection operation unit 26a. In the present embodiment, the company selection operation unit 26a performs monthly / yearly switching, setting of a target year / month period, presence / absence of business type consideration, setting of the maximum value and minimum value of sales, and the number of employees The items such as the maximum value, the minimum value, and the prefecture can be selected, and the company information narrowed down by the condition of the result of these operations is used for selecting the learning model.

  The accounting item selection unit 27 selects which accounting item is used to generate the learning model from the corporate accounting information associated with the selected corporate information. In FIG. 3, the accounting item selection unit 27 performs processing based on the operation of the accounting item selection operation unit 27a. Each item of the corporate accounting information shown in FIGS. 7-1, 7-2, and 7-3 is displayed in the accounting item selection operation unit 27a, and the item checked here generates a learning model. Used for

  The financial item selection unit 28 selects which item is used to generate the learning model from the financial index information associated with the selected company information. In FIG. 3, the financial item selection unit 28 performs processing based on the operation of the financial item selection operation units 28a and 28b. Each item of the financial index shown in FIG. 8 is displayed in the financial item selection operation unit 28a, and the item checked here is used for generating a learning model. Each item of the financial index shown in FIG. 9 is displayed in the financial item selection operation unit 28b, and the item checked here is used for generating a learning model.

  The learning model setting unit 29 can set various parameters when generating a learning model. In FIG. 3, the learning model setting unit 29 performs processing for setting each parameter based on the operation of the learning model setting operation unit 40. The learning operation unit 40a is a part that displays the number of nodes in the input layer, and the number of items checked by the accounting item selection unit 27a and the financial item selection operation units 28a and 28b is displayed as the number of nodes in the input layer. The learning operation unit 40i can set the number of hidden layers of the neural network. The learning operation unit 40j can set the type of the activation function. The learning operation unit 40b displays the number of output layers, and is 1 because it is a learning model in which only the default rate is calculated here. The learning setting operation unit 40c can set a batch size, and the batch size is a size for dividing data in order to perform learning at high speed. In other words, when there are a large number of selected companies, instead of targeting all of them in one learning, a sample of companies for a set batch size of N companies is extracted from the selected group of companies and batched. Learning is performed for size N companies, and in the next learning, N companies are again extracted at random and learning is performed. If the number of selected companies is large, the learning process tends to take time, so the learning process can be speeded up by setting the batch size.

  The learning setting operation unit 40d can set an optimization function. The optimization function is a function for updating a parameter (weight) to be optimal from an error between a predicted value of a model and an actual value. In this embodiment, it is possible to select from, for example, Gradient Desc (gradient descent method), Ada Delta, ADAGRAD, Adam, and the like.

  The learning operation setting unit 40e can set a learning rate, and the learning rate is the degree when the next value is calculated along the gradient when the gradient calculation is performed by the method selected by the learning setting operation unit 40d. It is a value that specifies the degree to which the gradient is moved. . The learning rate can be set to a value between 0 and 1, and a slight weight correction is performed when close to 0, and a larger weight correction is performed when close to 1.

  The learning operation setting unit 40f can set the number of iterations, and the number of iterations is the number of times learning is repeated. The learning operation setting unit 40g can set the name of the learning process, and can name and save the learning model that has been executed and learned. The learning operation setting unit 40h can set the number of cross-validation divisions (cross-validation). Cross-validation is performed by dividing into K sample groups at random when verifying with a loss value. .

  The learning model setting unit 29 generates a learning model based on the conditions set in the selection units 26, 27, and 28 described above. The learning model is generated using a recursive neural network model, and each selected parameter is set as an explanatory variable. The recursive neural network can be constructed using, for example, TENSORFLOW (registered trademark) of google. A learning model is generated by passing each explanatory variable input as an input layer to the generated learning model. In the company selection operation unit 26a, when the period of the target data is not set, the default setting that the data for the latest 12 months is used is applied and the process is executed. Also, the default rate calculated by the learning model is set so that the default rate for March 2016 is predicted when the monthly target sample is March 2014 to March 2015, for example. The relationship value between the period of the sample to be calculated and the timing of the default rate to be calculated can be adjusted as appropriate. Further, by using a recursive neural network, it is possible to make a determination that incorporates changes in numerical values such as monthly sales, and it is possible to calculate a default rate with higher accuracy.

  The learning execution unit 30 repeats the learning process for the generated learning model a predetermined number of times. Each time the learning execution unit 30 performs the learning process, the learning execution unit 30 also calculates a loss value derived from the calculated default rate value. The loss value is a value indicating the certainty of the default rate, and is calculated using a cross entropy function and a softmax function. The total number of input variables used for generating the learning model, the total number of classification classes (here, only 0 and 1 indicating whether there was a default, 2), weight variables, input variable values, implementation defaults Correct data indicating the presence or absence is a variable. Therefore, the loss value is a variable that returns an error between the calculated default rate and the actual default value, that is, a numerical value of “0” or “1” by repeating learning using this calculation method. By this process, the default rate and the loss value for all the companies that are the calculation target are calculated, and the average value of all the companies that are the target becomes the value of the loss value in one learning.

  FIG. 10 is a flowchart showing a flow of processing relating to generation of a learning model. First, the company selection unit 26 selects a company to be used as an explanatory variable according to the conditions input by the user (step S101). Next, the accounting item selection unit 27 acquires the value of the selected accounting item from the corporate accounting information associated with the selected corporate information (step S102). Further, the financial item selection unit 28 acquires the value of the selected financial data from the financial index information associated with the selected company information (step S103).

  Next, the learning model setting unit 29 generates a learning model based on the explanatory variable that is the input value selected by each selection unit and the setting of the learning model. At this time, a learning model is generated based on the above calculation formula for calculating a loss value indicating a difference between the above-described default value and a value indicating the presence or absence of an actual default.

  The learning execution unit 30 repeats the learning process based on the learning model and the input value to calculate a default rate and a loss value, and outputs a graph of the loss value as a learning result to the output device 15 through the output unit 33 (step) S105). FIG. 4 is a graph showing the learning results, and the learning results for each learning model are displayed from 50a to 50i. In FIG. 4, the value of the loss value is shown on the Y-axis, and the number of learnings is shown on the X-axis, and it is shown that the value of the loss value decreases every time learning is repeated.

  By viewing this figure, the user can visually determine which learning model has the highest accuracy. In this figure, the learning model indicated by 50h has the highest accuracy. In the learning set information 51 and the model information 52, the names of learning models 50a to 50i and information on various settings are displayed as the selected learning result. The learning execution unit 30 stores information on the learning model in the learning model storage unit 24 every time the learning process is executed.

  The learning model for calculating the default rate generated as described above is stored in the learning model storage unit 24 as a set with explanatory variables. The model determining unit 31 determines a learning model used for calculating the default rate of the target company from the learning model stored in the learning model storage unit 24 (step S201). Next, the default rate calculation unit 32 selects a target company for calculating the default rate, and selects information corresponding to the item selected in the selected learning model from the accounting information and financial index information of the target company. Obtain (step S202). At this time, it is desirable that the period of information to be acquired is the same as the period selected in the learning model. Then, the default rate calculation unit 32 calculates a default rate according to the learning model calculation formula based on the acquired information (step S203). The default rate calculation unit 32 outputs the calculated default rate through the output unit 33 in a manner as shown in FIG.

  According to the financial information analysis system 10 shown above, learning is performed using a recursive neural network at the time of learning processing. Therefore, when monthly accounting information is provided as an input value, sales for that month are performed. Since the default rate can be calculated in consideration of changes such as increase and decrease, it is possible to calculate with higher accuracy.

  Further, in order to calculate such a default rate, it is usually necessary for a data analysis expert to formulate a calculation model and process the data, and thus requires advanced operations. On the other hand, in the financial information analysis system 10 of the present embodiment, selection of company information, selection of accounting items, etc. can be performed with a simple operation screen, and operations such as selection of items and setting of learning can be performed. All you have to do is generate a learning model for automatic default rate calculations. And since the results of the default rate calculation by the learning model can be grasped by the value of the loss value and the AR value, even users who do not have specialized knowledge themselves can determine the default rate of the company whenever they need it. It can be calculated. Therefore, the process of calculating the default rate at the time of financing or investment can be simplified, and the efficiency of operations can be improved. Note that the AR values are arranged in descending order of the default rate when the default rates are, for example, Company A: 1%, Company B: 2%, Company C 3%. In this case, if company C actually has a default and B and A do not have a default, the AR value is 100%. On the other hand, when company A is the default and C and B are not the default, the AR value is calculated to be lower than 100%.

  In order to improve the efficiency of learning, it is necessary to tune various learning settings. In this embodiment, the number of executions of learning, the number of hierarchies and the number of nodes, and each hidden layer Since the learning can be executed by freely changing parameters such as the activation function to be used, the learning rate, and the batch size, the adjustment of the learning is very easy.

  In addition, in calculating the default rate at the time of normal financing, it was difficult to calculate the default rate reflecting the data in the middle of the fiscal year because it was processing for each annual settlement, but in this embodiment, The default rate can be calculated by referring to the data in monthly units.

  In addition, in the learning setting, for example, a setting such as a dropout rate and a regularization rate can be set by the learning model setting unit 29 in addition to the above setting items.

DESCRIPTION OF SYMBOLS 10 ... Financial information analysis system 21 ... Company information storage part 22 ... Accounting information storage part 23 ... Financial data storage part 24 ... Learning model storage part 26 ... Company selection part 27 ... Accounting item selection part 28 ... Financial item selection part 29 ... Learning Model setting unit 30 ... learning execution unit 31 ... model determination unit 32 ... default rate calculation unit

Claims (17)

A company information storage unit that stores company information including default information that is managed in association with company identification information uniquely set for each company and indicates whether there is a default,
An accounting information storage unit that stores corporate accounting information associated with the company identification information and includes at least monthly income statement data for each company and balance sheet data;
By searching for items included in the company information, a company selection unit that selects the company accounting information used for learning processing;
Among the corporate accounting information, an accounting item selection unit for selecting an item to be used for the learning process,
A learning model for learning using a recursive neural network by inputting, as an explanatory variable, the value of an item selected by the accounting item selection unit of the corporate accounting information selected using the company selection unit. A learning model setting unit to be generated;
Based on the learning model generated by the learning model setting unit, the default rate for each company selected by the company selecting unit is calculated, and the default information of the company corresponding to the calculated company default rate and A learning execution unit that repeats learning of the optimum parameters of the learning model,
A learning model storage unit for storing a learning model for which learning processing has been completed by the learning execution unit;
A model determining unit that determines which of the stored learning models to use,
Using the learning model determined by the model determination unit, a default rate calculation unit for calculating the default rate in a target month of a specific company;
A financial information analysis system characterized by comprising: The learning execution unit calculates a loss value indicating the certainty of the default rate, calculates an average value of the loss values of the companies that are learning targets, and calculates the learning model setting unit from the value of the loss value. 2. The financial information analysis system according to claim 1, wherein the parameter is changed. The change of the loss value every time learning is executed by the learning execution unit is visually displayed in a graph, and the graph is displayed for each condition selected by the company selection unit and the accounting item selection unit. A display unit that is generated and displayed on the same screen;
The financial information analysis system according to claim 2, further comprising: The learning execution unit uses a loss value as an index to compare the calculated default rate for each of the companies with an actual value that is 1 when the company actually has defaulted and 0 when there is no default. The financial information analysis system according to claim 2, wherein the financial information analysis system is calculated as follows. The learning model setting unit can set the number of executions of learning in the learning execution unit, and the learning execution unit repeats learning processing for the number of times of learning set by the learning model setting unit. The financial information analysis system according to any one of claims 1 to 4. The financial information analysis system according to any one of claims 1 to 5, wherein the learning model setting unit can set the number of hidden layers and the number of nodes. The financial information analysis system according to any one of claims 1 to 6, wherein the learning model setting unit can select an activation function to be used for each hidden layer. The financial information analysis system according to any one of claims 1 to 7, wherein the learning model setting unit can set a learning rate. The financial information analysis system according to claim 1, wherein the learning model setting unit is capable of setting a batch size. The financial information analysis system according to claim 1, wherein the learning model setting unit can set the number of divisions of cross validation. The financial information analysis system according to claim 1, wherein the learning model setting unit can set a dropout rate. The financial information analysis system according to any one of claims 1 to 11, wherein the learning model setting unit can set a regularization rate. The said company selection part searches and selects the company which satisfy | fills at least 1 condition of the set maximum value and the minimum value with reference to the sales scale of the company to select. The financial information analysis system according to any one of the preceding claims. The company selection unit searches for and selects a company that satisfies at least one of a set maximum value and a minimum value with reference to the number of employees of the company to be selected. The financial information analysis system according to any one of the above. The financial information according to any one of claims 1 to 14, wherein the company selection unit searches and selects a company corresponding to the selected industry with reference to the industry information of the company to be selected. Analysis system. A financial data storage unit including financial index information calculated based on the profit and loss statement stored in the accounting information storage unit and the numerical values of the balance sheet;
Of the financial data storage unit, a financial item selection unit for selecting an item to be used for the learning process;
The financial information analysis system according to claim 1, further comprising: From a company information storage unit that stores company information including default information indicating whether there is default or not, managed in association with company identification information uniquely set for each company on the computer, for items included in the company information To select a company accounting information to be used for the learning process,
Accounting item selection step for selecting an item to be used for the learning process from among company accounting information associated with the company identification information and including at least monthly income statement data for each company and balance sheet data;
Of the corporate accounting information selected in the enterprise selection step, the value of the item selected in the accounting item selection step is input as an explanatory variable, and a learning model for learning using a recursive neural network is generated. A learning model setting step;
Based on the learning model generated by the learning model setting step, the default rate for each company selected by the company selection step is calculated, and the default information of the company corresponding to the calculated company default rate and The learning execution step that repeats learning of the optimum parameters with reference to
A model storage step for storing the learning model for which the learning process has been completed;
A model determination step for determining which of the stored learning models to use;
A default rate calculating step of calculating the default rate in a target month of a specific company using the learning model determined by the model determining step;
A program for running