TWI776370B - Investment risk scoring method and system for fund commodities - Google Patents

Abstract

In an investment risk scoring system for fund commodities, the data server collects the historical financial behavior data related to at least one of the customer's fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior, as well as personal basic information. Reference risk attribute data related to data, financial situation and investment experience; the scoring server uses machine learning algorithms to analyze and calculate a number of sample data sets collected by the data server for model training to establish parameters corresponding to the target The historical financial behavior data is fed into the scoring model to obtain a model score, and according to the model score and the reference score corresponding to the reference risk attribute data, the customer's risk score corresponding to the target parameter is generated.

Description

Investment risk scoring method and system for fund commodities

The present invention relates to investment risk assessment, in particular to an investment risk scoring method and system for fund commodities.

With the explosion of investment information and the introduction of new investment and wealth management products, the investment behavior of investors has become more diversified and more elusive. Therefore, if investors are only evaluated based on existing basic information such as gender, age, annual income, etc. Especially for the investment risk level or preference of fund commodities, the evaluation results obtained in this way are often difficult to fully reflect the investor's investment risk level or preference for fund commodities.

Therefore, in order to effectively promote financial services such as fund commodities, how to develop a risk scoring method that can effectively reflect customers' investment in fund commodities has become one of the urgent issues that financial institutions need to solve.

Therefore, the purpose of the present invention is to provide an investment risk scoring method and system for fund commodities, which can overcome at least one disadvantage of the prior art.

Therefore, the present invention provides an investment risk assessment for fund commodities. The sub-method is used to evaluate a client's risk when investing in fund commodities and is implemented by a computer system, and includes the following steps: (A) collecting the client's historical financial behavior data and reference risk attribute data, and a plurality of A sample data set for model training. The historical financial behavior data is associated with at least one of fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior, and credit card consumption behavior. The reference risk attribute data is related to personal basic data, financial The situation is related to investment experience, and each sample data set contains a plurality of parameter values corresponding to M characteristic parameters related to a target parameter. The M characteristic parameters are related to fund transaction behavior, fund profit and loss records, other asset status, Digital browsing behavior and credit card consumption behavior are correlated; (B) use machine learning algorithms to analyze and calculate the sample data sets to establish a scoring model corresponding to the target parameter, wherein the scoring model includes a plurality of decision trees, each A decision tree is trained by analyzing a training data set consisting of a predetermined number of sample data sets according to m characteristic parameters randomly selected from the M characteristic parameters, and is operable to generate a classification result in response to the input data; (C) feeding the historical financial behavior data into the scoring model, and estimating a model score based on the classification results produced by each decision tree in response to the historical financial behavior data; and (D) at least based on the model score, and Corresponding to the reference score of the reference risk attribute data, a risk score of the client corresponding to the target parameter is generated.

The investment risk scoring method for fund commodities of the present invention further includes the following steps: (E) outputting a risk assessment result corresponding to the client and containing the risk score.

In the investment risk scoring method for fund commodities of the present invention, in step (B), the machine learning algorithm includes a random forest algorithm.

In the investment risk scoring method for fund commodities of the present invention, in step (B), the target parameter is the proportion of assets allocated to fund commodities.

In the investment risk scoring method for fund commodities of the present invention, in step (D), the computer system further generates the risk according to a predetermined first weight and a predetermined second weight assigned to the reference score and the model score respectively Score, so that the risk score=(the reference score×the predetermined first weight)+(the model score×the predetermined second weight).

Therefore, an investment risk scoring system for fund commodities provided by the present invention is used for risk assessment when a client invests in fund commodities, and includes a data server and a scoring server.

The data server operates to collect the client's historical financial behavior data and reference risk attribute data, as well as a plurality of sample data sets for model training. The historical financial behavior data is associated with at least one of fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card spending behavior. The reference risk attribute information is associated with personal basic information, financial situation and investment experience. Each sample data set includes a plurality of parameter values respectively corresponding to M characteristic parameters related to a target parameter, and these characteristic parameters are related to fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior. link.

The scoring server is connected to the data server to receive the historical financial activity data, the reference risk attribute data, and the sample data sets, and include a modeling module, and a scoring module. The modeling module operates to analyze and calculate the sample data sets using machine learning algorithms to establish a scoring model corresponding to the target parameter. The scoring model includes a plurality of decision trees, each decision tree is based on the M The m feature parameters randomly selected from the feature parameters are trained by analyzing a training data set consisting of a predetermined number of sample data sets, and are operable to generate a classification result in response to the input data. The scoring module is connected to the modeling module and operates to feed the historical financial behavior data into the scoring model to estimate a model score based on at least the model score and a reference score corresponding to the reference risk attribute data , which generates the risk score for the client corresponding to the target parameter.

In the investment risk scoring system for fund commodities of the present invention, the scoring server further includes an output module. The input module is connected to the scoring module and operates to output a risk assessment result corresponding to the client and containing the risk score.

In the investment risk scoring system for fund commodities of the present invention, the machine learning algorithm includes a random forest algorithm.

In the investment risk scoring system for fund commodities of the present invention, the target parameter is the proportion of assets allocated to fund commodities.

In the investment risk scoring system for fund commodities of the present invention, the scoring module further generates the risk score according to a predetermined first weight and a predetermined second weight respectively assigned to the reference score and the model score, so that the risk Rating = (The reference rating score × the predetermined first weight) + (the model score × the predetermined second weight).

The effect of the present invention is: according to the historical financial behavior data associated with the customer's fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and/or credit card consumption behavior, and use the machine learning algorithm to establish corresponding targets. The model score estimated by the scoring model of the parameters; then the risk score generated according to the model score and the reference score corresponding to the reference risk attribute data of the customer, compared with the score obtained by the existing risk scoring method, can fully and truly reflect the customer As for the risk appetite that fund commodities can bear, this risk score can provide customers with follow-up fund management and/or investment planning services on the premise of effectively avoiding investment risks.

100: Investment Risk Scoring System

1: Data server

2: Scoring Server

21: Modeling Modules

22: Scoring Module

23: Output module

S21-S25: Steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: FIG. 1 is a block diagram illustrating the structure of an investment risk scoring system for fund commodities according to an embodiment of the present invention; FIG. 2 is a flow chart illustrating how the embodiment performs an investment risk scoring procedure for a client who wants to invest in fund commodities; and FIG. 3 is a schematic diagram illustrating an example of a scoring model established by an embodiment of the present invention. Architecture.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are designated by the same reference numerals.

Referring to FIG. 1 , an investment risk scoring system 100 for fund commodities according to an embodiment of the present invention is used for risk assessment when a client invests in fund commodities, and includes a data server 1 and a scoring server 2 .

In this embodiment, the data server 1 can be implemented by, for example, a computer device, and can be connected to an external database (not shown), such as a database of a financial service institution, and used for data collection, especially for collecting data such as Information about the (to be scored) customers of the banking institution's financial services organization, and all the information needed for modeling.

In this embodiment, the scoring server 2 can also be implemented as a computer device, which is connected to the data server 1 , and can include, for example, a modeling module 21 and a scoring module connected to the modeling module 21 . 22, and an output module 23 connected to the scoring module 22. Each of the modeling module 21 and the scoring module 22 may be implemented in hardware, software, firmware, or a combination thereof, and the output module 23 may be implemented as, but not limited to, a display, for example.

Hereinafter, referring to FIG. 1 and FIG. 2 , how the investment risk scoring system 100 performs an investment risk scoring procedure for a customer who belongs to a banking institution and wants to invest in fund commodities, for example, will be described in detail by way of example. It should be noted that, the external database connected to the data server 1 is, for example, a database provided by the banking institution. Preferably, This database provides richer and more diverse financial information about all of the banking institution's customers. The client personal information management program may include, for example, the following steps S21 to S25.

First, in step S21, the data server 1 searches, for example, the database of the banking institution to collect historical financial behavior data and reference risk attribute data of the customer, as well as multiple sample data sets for model training, and store The collected historical financial behavior data, the reference risk attribute data and the sample data sets are sent to the scoring server 2 . It is worth noting that the historical financial behavior information is associated with at least one of fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior, and the reference risk attribute data is related to personal basic information, financial situation and The investment experience is related, and each sample data set contains a plurality of parameter values corresponding to M characteristic parameters related to a target parameter. These characteristic parameters are related to fund transaction behavior, fund profit and loss records, other asset status, and digital browsing behavior. associated with credit card spending.

More specifically, the historical financial behavior data may include information about the fund commodities that the client has purchased and their profits and losses (such as the average holding days of the fund commodities, the type of the preferred fund commodities, the maximum loss amount, and the degree of volatility of the return). ), other financial products issued by the banking institution and the relevant information of the account balance of the financial account held by the customer, the customer's browsing footprint on the service website provided by the banking institution such as fund wealth management (for example, long-term and short-term browsing information about the number of times and the time of each visit), and information about the customer's use of the credit card issued by the banking institution (e.g. such as card swiping and whether revolving rates are used). The reference risk attribute data is, for example, a risk attribute questionnaire answered by the customer through manual filling or input before the customer first subscribes for fund commodities. The content of the risk attribute questionnaire may include, for example, basic information (such as age, occupation), Financial background information (such as income, source of funds) and investment experience data (such as purchased financial products), and according to the content of the risk attribute questionnaire through a predetermined scoring method, a reference score of the customer can be obtained. In this embodiment, the target parameter is, for example, "the proportion of assets allocated to fund commodities", and for example, M=35, but it is not limited to this example. Therefore, each sample data set can include, for example, 35 parameter values, including: 16 “number of bond funds held in the past two years” associated with fund trading behavior, “stock funds held in the past two years” Quantity”, “The number of multi-asset funds held in the past two years”, “The number of currency funds held in the past two years”, “The number of balanced funds held in the past two years”, “The portfolio held in the past two years” number of ETF funds, number of ETF funds held in the past two years, average subscription amount, average subscription amount as a percentage of total assets, average holding days, and subscription times as a percentage of all transactions ", "Number of purchases as a percentage of all transactions", "Number of purchases/redemptions", "Maximum profit amount", "Maximum profit amount - Maximum loss amount", "Assets under management (AUM) ranking" Parameter values of such characteristic parameters; 7 "average rate of return of historical exchange-traded funds", "volatility of average rate of return of historical exchange-traded funds", "number of fund products held in history", " Volatility of historically largest loss return”, “Historical The parameter values of characteristic parameters such as the largest loss in history, the volatility of the largest loss in history, and the profit factor; 5 "TWD living balance" and "TWD fixed deposit balance" associated with other asset status , "check deposit balance", "foreign currency active deposit balance", "foreign currency fixed deposit balance" and other characteristic parameters parameter values; 4 parameters associated with digital browsing behaviors are "number of visits to the fund wealth management network in the last month", "the most recent one" The parameter values of characteristic parameters such as the average stay time on the fund wealth management network in the last five months, the number of visits to the fund wealth management network in the last five months, and the average stay time on the fund wealth management network in the last five months; and three parameters related to credit card consumption behavior The parameter values of the characteristic parameters such as "Credit Limit/Annual Income", "Whether to Use Revolving Interest Rate", "Average Credit Limit Utilization Rate" and so on.

When the scoring server 2 receives the historical financial behavior data, the reference risk attribute data and the sample data sets from the data server 2, in step S22, the modeling module 21 operates to A sample data set, and a scoring model corresponding to the target parameter is established by using a machine learning algorithm. More specifically, in this embodiment, the machine learning algorithm includes a random forest algorithm, the scoring model includes a plurality of decision trees, and each decision tree is based on m corresponding characteristic parameters selected from the M characteristic parameters. Do the decision calculus and m<M. It is worth noting that the random forest algorithm is a kind of "supervised learning" algorithm. Compared with the neural network algorithm of deep learning, it usually has a high model interpretation ability, and the target parameters have been defined before modeling. (ie, "the proportion of assets allocated to fund commodities"), so that a relatively accurate scoring model can be developed by using these sample data sets to be continuously revised during the learning process.

More specifically, referring to FIG. 3 , the modeling module 21 uses the bagging algorithm to randomly select a predetermined number of sample data sets from the sample data sets each time as the training data set for this time, and a total of Execute n times, so that n training data sets can be obtained. For each training data set, in order to generate different random vectors θ i, i=1, . . . , n, node segmentation is performed on each of the m feature parameters randomly selected from the M feature parameters to select The division method that achieves the smallest Gini coefficient is divided and finally a decision tree is generated. In this case, each decision tree can be used as a classifier and can generate a classification result in response to the input data; then, the modeling module 21 also uses a classification or regression algorithm to assemble and calculate the number of n The n classification results generated by a decision tree are used to determine a target classification level from a plurality of predetermined classification levels, and output a score corresponding to the target classification level. It is worth noting that because the scoring model combines the bagging algorithm and the classification or regression algorithm, the accuracy and stability of the model can be improved, and the occurrence of overfitting can be avoided while maintaining the model ensemble (Ensemble). characteristic.

Then, in step S23, the scoring module 22 operates to feed the historical financial behavior data into the scoring model as input data to estimate a model score corresponding to the customer. Since the target parameter of the scoring model is "proportion of assets allocated to fund commodities", if the ratio is higher, it means that the client has a higher preference for fund commodities and also represents a higher risk of holding fund commodities. will be higher, and vice versa.

Then, in step S24, the scoring module 22 generates the customer's score corresponding to the target parameter according to the model score, the reference score corresponding to the reference risk attribute data, a predetermined first weight and a predetermined second weight Risk Score. More specifically, the predetermined first weight and the predetermined second weight formula are respectively assigned to the reference score and the model score, so that the risk score=(the reference score×the predetermined first weight)+(the model score× the predetermined second weight).

Finally, in step S25, the output module 23 outputs a risk score result including the risk score. For example, the output module 23 can display the risk score result, or in other embodiments, the output module 23 can be a communication module to transmit the risk score result to a specific user terminal (such as a bank a computer terminal of the institution or a mobile phone held by the client).

Therefore, the banking institution can provide the customer with suggestions on the allocation of fund commodities (such as the ratio of stocks and debts) in the subsequent fund management or investment planning according to the risk score of the customer.

To sum up, since the scoring model is trained using the sample data set associated with fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior, the obtained model score can be realistic. It reflects the client's risk appetite for fund commodities. Therefore, the risk score obtained by combining the (objective) model score with the (subjective) reference score, compared with the score obtained by the existing risk scoring method, can provide customers with after-sales service under the premise of effectively avoiding investment risks. Continued fund management and/or investment planning services. Therefore, the investment risk scoring system 100 of the present invention can indeed achieve the purpose of the present invention.

However, the above are only examples of the present invention, and should not limit the scope of implementation of the present invention. Any simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the patent specification are still included in the scope of the present invention. within the scope of the invention patent.

100: Investment Risk Scoring System 1: Data Server 2: Scoring Server 21: Modeling Module 22: Scoring Module 23: Output Module

Claims (10)

An investment risk scoring method for fund commodities, which is used to evaluate a client's risk when investing in fund commodities and is executed by a computer system. The investment risk scoring method comprises the following steps: (A) Collecting the client's historical financial behavior Data and reference risk attribute data and multiple sample data sets used for model training, the historical financial behavior data is associated with at least one of fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior, The reference risk attribute data is associated with personal basic data, financial situation and investment experience. Each sample data set includes a plurality of parameter values respectively corresponding to M characteristic parameters related to a target parameter. The M characteristic parameters are related to the fund Transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior are correlated; (B) use machine learning algorithms to analyze and calculate these sample data sets to establish a scoring model corresponding to the target parameters, The scoring model includes a plurality of decision trees, and each decision tree is obtained by analyzing a training data set composed of a predetermined number of sample data sets from the sample data sets according to m characteristic parameters randomly selected from the M characteristic parameters. trained and operable to generate a classification result in response to input data; (C) feeding the historical financial behavior data into the scoring model, and generating classification results in response to the historical financial behavior data according to each decision tree estimating a model score; and (D) generating a risk of the client corresponding to the target parameter based on at least the model score and the reference score corresponding to the reference risk attribute data score. The investment risk scoring method for fund commodities as described in claim 1, further comprising the following steps: (E) outputting a risk assessment result corresponding to the client and containing the risk score. The investment risk scoring method for fund commodities according to claim 1, wherein, in step (B), the machine learning algorithm includes a random forest algorithm. The investment risk scoring method for fund commodities according to claim 3, wherein, in step (B), the target parameter is the proportion of assets allocated to fund commodities. The investment risk scoring method for fund commodities as described in claim 1, wherein, in step (D), the computer system is further based on a predetermined first weight and a predetermined second weight respectively assigned to the reference score and the model score weights to generate the risk score such that the risk score=(the reference score×the predetermined first weight)+(the model score×the predetermined second weight). An investment risk scoring system for fund commodities, which is used for risk assessment of a client when investing in fund commodities, and includes: a data server, which operates to collect the client's historical financial behavior data and reference risk attribute data, and uses Multiple sample data sets used in model training, the historical financial behavior data is associated with at least one of fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior, and the reference risk attribute data is related to personal basic Information, financial situation and investment experience are related, and each sample data set contains a plurality of parameter values respectively corresponding to M characteristic parameters related to a target parameter, the M characteristic parameters being associated with fund transaction behavior, fund profit and loss records, other asset status, digital browsing behavior and credit card consumption behavior; and a score a server, connected to the data server to receive the historical financial behavior data, the reference risk attribute data, and the sample data sets, and includes a modeling module operative to analyze and calculate the sample data using machine learning algorithms set, to establish a scoring model corresponding to the target parameter, the scoring model includes a plurality of decision trees, each decision tree is based on the m feature parameters randomly selected from the M feature parameters to analyze the predetermined number of sample data is trained and operable to generate a classification result in response to input data, and a scoring module connected to the modeling module and operative to feed the historical financial behavior data into the scoring model , and estimate a model score according to the classification result generated by each decision tree in response to the historical financial behavior data, and at least according to the model score and the reference score corresponding to the reference risk attribute data, generate the customer corresponding to the Risk score for the target parameter. The investment risk scoring system for fund commodities according to claim 6, wherein the scoring server further comprises: an output module, connected to the scoring module, and operated to output a risk assessment corresponding to the client and containing the risk score result. For the investment risk scoring system for fund commodities as described in claim 6, the , the machine learning algorithm includes a random forest algorithm. The investment risk scoring system for fund commodities according to claim 8, wherein the target parameter is the proportion of assets allocated to fund commodities. The investment risk scoring system for fund commodities according to claim 6, wherein the scoring module further generates the risk score according to a predetermined first weight and a predetermined second weight assigned to the reference score and the model score respectively , so that the risk score=(the reference score×the predetermined first weight)+(the model score×the predetermined second weight).

Images