Risk control variable self-adaptive box separation and modeling method

Abstract

The invention discloses a risk control variable self-adaptive binning and modeling method, and particularly relates to the technical field of binning algorithms, and the method specifically comprises the steps: analyzing the distribution characteristics of continuous variables and the relation between the continuous variables and target variables, verifying the chaos of a risk control system, and verifying the chaos of the risk control system according to a verification result. The effectiveness of the box separation algorithm is evaluated according to the box separation effect; the state classification is performed on the box separation model according to the evaluation result; and the decision-making assistance is performed on the working personnel according to the state type of the box separation model. Therefore, the problem that the binning logic based on the linear law loses practicability in a disordered state is solved.

Description

Technical Field

[0001] The present invention relates to the field of binning algorithm technology, and more specifically, to an adaptive binning and modeling method for risk control variables. Background Art

[0002] In risk control, binning risk variables can effectively convert continuous variables into discrete variables to improve the performance of the model. However, the relationship between risk variables and target variables is often not linear. Single equal-width binning or equal-frequency binning is difficult to capture the nonlinear relationship between risk variables and target variables. On the other hand, in the risk control system, binning modeling is based on various data of the risk control system. The basis of binning comes from the reliability of the risk control system itself, but the stability of the risk control system is not always constant. The risk control system may enter a chaotic state. In the chaotic state, slight changes in initial conditions will cause huge differences in the evolutionary path. The underlying binning logic of the risk control variables will also collapse with the disorder of nonlinear changes in the chaotic state.

[0003] In order to solve the above defects, an adaptive binning and modeling method of risk control variables is proposed. Summary of the invention

[0004] The purpose of the present invention is to provide an adaptive binning and modeling method for risk control variables to solve the problems in the above-mentioned background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an adaptive binning and modeling method for risk control variables, the specific steps of which include analyzing the distribution characteristics of continuous variables and the relationship between continuous variables and target variables, and verifying the chaos of the risk control system; According to the verification results, the chaos of the risk control system is tested, and the binning algorithm is used to calculate and determine the binning boundaries of the risk control variables; Evaluate the effectiveness of the binning algorithm based on the binning effect, and classify the binning model based on the evaluation results; Provide decision-making assistance to staff based on the status type of the binning model.

[0006] Preferably, the method for verifying the chaos of the risk control system is: Combined with the mathematical method model of the risk control system, the Lyapunov index is calculated according to the state trajectory of the phase space of the risk control system to evaluate the sensitivity of the risk control system to the initial conditions. The specific evaluation method is: Selecting an initial point in the phase space of the risk control system as a reference point; At the initial point Some small perturbation points are selected nearby as adjacent initial points, and the adjacent initial points meet the small changes required for sensitivity; Use the mathematical method model of the risk control system to perform time evolution on the initial point and the adjacent initial points, generate a reference trajectory based on the initial point, and generate a disturbance trajectory based on the adjacent initial points; The difference between the reference trajectory and the perturbation trajectory is evaluated by calculating the Euclidean distance between the reference trajectory and each perturbation trajectory; The Lyapunov exponent is calculated from the growth rate of the difference between the reference trajectory and the perturbation trajectory.

[0007] Preferably, the phase space construction method of the risk control system is: Construct a phase space model of the risk control system. The risk factors in the risk control system include market risk, credit risk, operational risk and liquidity risk. Each risk factor is a dimension in the phase space model. Different risk factors correspond to different dimensions and together constitute the phase space of the risk control system.

[0008] Preferably, the state trajectory of the phase space is formed by: In the phase space of the risk control system, each possible risk state is represented by a point in the phase space, which is defined as the state point. The position of this point is determined by the values ​​of various risk factors. As time goes by, the risk state of the risk control system changes, which is manifested as the movement of the point in the phase space to form a trajectory, and the trajectory is defined as the state trajectory.

[0009] Preferably, the mathematical method model of the risk control system is: Taking time as the independent variable, the random process model is used to simulate the evolution of the state trajectory, and the mathematical method model is used to describe the risk control system, including using the variance-covariance method to use existing historical data to evaluate the expected return and risk level of risky assets, using the risk value method to evaluate the degree of loss by limiting the time interval and confidence level, and using the Monte Carlo simulation to evaluate the risk-free value model.

[0010] Preferably, the calculated Lyapunov index is analyzed, and the state of the risk control system is classified according to the analysis result, and the analysis method is: When the calculated Lyapunov exponent is less than or equal to 0, the risk control system is insensitive to the initial conditions and the marker is in a non-chaotic state; When the calculated Lyapunov exponent is greater than 0, the risk control system is sensitive to the initial conditions and is marked as being in a chaotic state.

[0011] Preferably, the method of using the binning algorithm to calculate and determine the binning boundaries of the risk control variables is: When the risk control system is marked as non-chaotic, the data are sorted by the values ​​of the continuous variables and the continuous variables are arranged corresponding to the target variables; Use equal-width binning or equal-frequency binning to get an initial binning scheme based on the data distribution; For each bin, calculate the statistics of the target variable, including default rate and mean; Merge adjacent bins and keep the merged bins monotonic on the target variable; Adjust the binning according to the merge results so that the final binning meets the monotonicity requirement.

[0012] Preferably, when the risk control system is marked as a chaotic state, the specific method for optimizing and improving the binning algorithm is: Use rolling window techniques to periodically recalculate bin boundaries to adapt to data changes; Evaluate the effects of different binning algorithms through cross-validation method and select the best binning algorithm; Use machine learning algorithms to clean risk control variables, identify and process noise, and remove outliers; Implement feature engineering on risk control variables through normalization, feature selection and construction to improve the feature expression ability of risk control data; Methods based on information gain, chi-square test, and decision tree were used to determine the optimal binning boundaries.

[0013] Preferably, the method for evaluating the effectiveness of the binning algorithm according to the binning effect is: The calibration bin effectiveness index is Eb, and the calculation expression is , where PSI is the population stability index, expressed as , where is the actual proportion within the binning interval, is the expected proportion within the binning interval, is the weight of evidence, expressed as , where i is the bin number and i is a positive integer, is the proportion of non-default quantity in the i-th bin, is the proportion of defaults in the i-th bin, and the expression of KS value is , where is the cumulative bad account ratio in the i-th bin, is the cumulative good account ratio in the i-th bin, are the weight coefficients of PSI and KS respectively, and are all positive integers; The binning validity threshold Et is preset. When the calculated binning validity index Eb is greater than or equal to the binning validity threshold Et, the robustness, predictive ability and distinguishing ability of the binning algorithm can meet the validity requirements, and the marked binning modeling is effective. When the calculated binning validity index Eb is less than the binning validity threshold Et, the robustness, predictive ability, and discrimination ability of the binning algorithm cannot meet the validity requirements, and the marked binning modeling is invalid.

[0014] Preferably, the logic for assisting staff in making decisions according to the status type of the binning model is: When the binning modeling is marked as valid, it reminds the staff that the binning model has the ability to identify fluctuations in the stability of the risk control system and does not need to be inspected and maintained; When the binning modeling is marked as valid, it reminds the staff that the binning model does not have the ability to distinguish across the stability fluctuations of the risk control system, and requires the staff to optimize and improve the binning algorithm.

[0015] In the above technical solution, the technical effects and advantages provided by the present invention are: This application incorporates the chaos of the risk control system into the evaluation scope by combining it with the risk control stability state indicator, identifies the chaotic system by verifying the stability state of the risk control system, and incorporates the chaotic state of the risk control system into the optimization method of bin modeling to ensure that when the risk control system enters a chaotic state or changes between a chaotic state and a stable state, the feasibility of bin modeling is verified to ensure the applicability of bin modeling, avoid major failures caused by directional errors in the bin algorithm when the stability of the risk control system is unbalanced, and effectively improve the stability and scope of applicability of bin modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0017] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1: Please refer to Figure 1 As shown, the present invention is an adaptive binning and modeling method for risk control variables, and the specific steps include analyzing the distribution characteristics of continuous variables and the relationship between continuous variables and target variables, and verifying the chaos of the risk control system; According to the verification results, the chaos of the risk control system is tested, and the binning algorithm is used to calculate and determine the binning boundaries of the risk control variables; Evaluate the effectiveness of the binning algorithm based on the binning effect, and classify the binning model based on the evaluation results; Provide decision-making assistance to staff based on the status type of the binning model.

[0020] It should be noted that the risk control variables in this application are all continuous variables.

[0021] Analyze the distribution characteristics of continuous variables and the relationship between continuous variables and target variables, calculate the basic statistics of continuous variables to understand the overall distribution characteristics of continuous variables, including mean, median, standard deviation, minimum and maximum values, and quartiles, visualize the data through charts, intuitively display the distribution characteristics of continuous variables and the relationship between continuous variables and target variables, and evaluate the linear relationship between continuous variables and target variables by calculating the correlation coefficient between continuous variables and target variables; It should be noted that quartiles are a measure used in statistics to describe the characteristics of data distribution. A set of data is divided into three equal parts through four equally divided points, and the data set is divided into four equal subsets. The four points are called the first quartile Q1, the second quartile Q2, and the third quartile Q3. Quartiles are very effective in analyzing data because they provide information about the shape of the data distribution, not just the central trend or extreme values. Quartiles can be used to identify outliers in the data and perform more detailed group analysis on the data. The interquartile range is also an effective indicator for measuring the degree of data dispersion when dealing with skewed distribution data. Commonly used visualization icons include histograms, box plots, scatter plots, etc. Histograms can display the distribution of continuous variables and identify the central tendency and dispersion of data. Box plots can display the distribution and outliers of data. Scatter plots are suitable for displaying the relationship between continuous variables and target variables, especially for continuous target variables. Commonly used correlation coefficients include the Pearson correlation coefficient and the Spearman correlation coefficient. The Pearson correlation coefficient is used to measure linear relationships, while the Spearman correlation coefficient is used to measure ranking relationships and is suitable for nonlinear but monotonic relationships.

[0022] The chaos of the risk control system is verified and the phase space model of the risk control system is constructed. The risk factors in the risk control system include market risk, credit risk, operational risk and liquidity risk. Each risk factor is a dimension in the phase space model. Different risk factors correspond to different dimensions and together constitute the phase space of the risk control system. In the phase space of the risk control system, each possible risk state is represented by a point in the phase space, which is defined as a state point. The position of this point is determined by the values ​​of various risk factors. As time goes by, the risk state of the risk control system changes, which is manifested as a trajectory formed by the movement of the point in the phase space. The trajectory is defined as a state trajectory. By analyzing the state trajectory, the changing trend and dynamic characteristics of the risk state of the risk control system can be evaluated. Using time as the independent variable, the random process model is used to simulate the evolution of the state trajectory, and the mathematical method model is used to describe the risk control system, including using the variance-covariance method to use existing historical data to evaluate the expected return and risk level of risky assets, using the risk value method to evaluate the degree of loss by limiting the time interval and confidence level, and using the Monte Carlo simulation to evaluate the risk-free value model; Combined with the mathematical method model of the risk control system, the Lyapunov index is calculated according to the state trajectory of the phase space to evaluate the sensitivity of the risk control system to the initial conditions. The specific evaluation method is: Selecting an initial point in the phase space of the risk control system as a reference point; At the initial point Some small perturbation points are selected nearby as adjacent initial points, and the adjacent initial points meet the small changes required for sensitivity; Use the mathematical method model of the risk control system to perform time evolution on the initial point and the adjacent initial points, generate a reference trajectory based on the initial point, and generate a disturbance trajectory based on the adjacent initial points; The difference between the reference trajectory and the perturbation trajectory is evaluated by calculating the Euclidean distance between the reference trajectory and each perturbation trajectory; The Lyapunov exponent is calculated from the growth rate of the difference between the reference trajectory and the perturbation trajectory.

[0023] It should be noted that market risk is the risk of investment losses due to market behavior, which is manifested as the potential losses to the investment portfolio caused by fluctuations in financial asset prices and commodity prices. It is the result of the combined effects of multiple factors such as market supply and demand, economic environment, and policy changes. Credit risk is the risk of economic losses caused by the counterparty's failure to perform its obligations in the agreed contract, which is related to the borrower's repayment ability, repayment willingness, and changes in the debtor's credit quality. Operational risk is the risk of losses caused by the imperfection or failure of internal procedures, personnel, systems, or external events, which is related to factors such as defects in internal control, human errors, and system failures. Liquidity risk is the risk that commercial banks cannot obtain sufficient funds in a timely manner or at a reasonable cost to repay matured debts or other payment obligations, meet asset growth or other business development needs, and is related to factors such as the bank's asset-liability structure and market liquidity conditions. Commonly used stochastic process models include Brownian motion, diffusion process, and jump diffusion process; It should be pointed out that the adjacent initial points meet the small changes in sensitivity requirements. The adjacent initial points should be close enough to simulate the small changes in the initial conditions. The adjacent initial points are close enough to simulate the small changes in the initial conditions more accurately. This small change is the basis for quantitative system sensitivity analysis and helps to reveal the system's sensitivity to changes in initial conditions. Different systems have different sensitivities to initial conditions. Therefore, when selecting adjacent initial points, it is necessary to consider the specific characteristics of the system, such as stability, nonlinearity, etc.; depending on the purpose of the analysis, the requirements for the proximity of adjacent initial points are also different. For example, when evaluating the stability of the system, it is necessary to select closer initial points to reveal the small changes in the system, while when studying the long-term behavior of the system, the requirements for the proximity of the initial points can be appropriately relaxed.

[0024] The calculated Lyapunov index is analyzed, and the state of the risk control system is classified according to the analysis results. The analysis method is as follows: When the calculated Lyapunov exponent is less than or equal to 0, the risk control system is insensitive to the initial conditions and the marker is in a non-chaotic state; When the calculated Lyapunov exponent is greater than 0, the risk control system is sensitive to the initial conditions and is marked as being in a chaotic state.

[0025] System chaos refers to the uncertain or unpredictable random phenomenon presented by a certain macroscopic nonlinear system under certain conditions. It is a phenomenon in which certainty and uncertainty, regularity and irregularity, or order and disorder are integrated. Chaotic behavior is not completely disordered, but a seemingly random uncertain behavior generated by a simple deterministic system. The sensitive dependence on initial conditions is the most prominent feature of a chaotic system, which is manifested as applying an infinitesimal disturbance to a chaotic orbit. In the process of time evolution, the chaotic orbit will deviate from the original orbit in the form of exponential divergence. This sensitivity makes the long-term behavior of the chaotic system unpredictable.

[0026] The binning algorithm is used to calculate and determine the binning boundaries of the risk control variables. When the risk control system is marked as a non-chaotic state, the specific method is as follows: Sort the data by the values ​​of the continuous variable and arrange the continuous variable against the target variable; Use equal-width binning or equal-frequency binning to get an initial binning scheme based on the data distribution; For each bin, calculate the statistics of the target variable, including default rate and mean; Merge adjacent bins and keep the merged bins monotonic on the target variable; Adjust the binning according to the merge results so that the final binning meets the monotonicity requirement.

[0027] The binning method for stable linear systems is difficult to adapt to chaotic risk control systems. Under chaotic risk control systems, data distribution may change frequently, including fluctuations in financial markets and changes in the economic environment, which will cause significant changes in data distribution. Fixed binning boundaries cannot accurately reflect the distribution of current data, thereby reducing the prediction accuracy and stability of the model. In a chaotic state, the proportion of noise and outliers in the data will increase significantly. Noise data will interfere with the determination of bin boundaries, resulting in unstable bin results. Noise and outliers will introduce bias, resulting in the inability of binned variables to effectively distinguish different risk levels, thus affecting the performance of the model. The amount of data may fluctuate significantly. Sometimes the amount of data is too large, and sometimes it is insufficient. The significant fluctuation of the amount of data will affect the stability and effect of the binning algorithm. When the amount of data is insufficient, the number of samples in the bin is too small to ensure the effectiveness of the binning. When the amount of data is too large, the computational complexity and time cost will increase significantly. In a chaotic state, the relationship between variables and target variables is more complex and nonlinear. Traditional binning methods, including equal-width binning and equal-frequency binning, are unable to capture this complex relationship and effectively reflect the true relationship between variables and target variables, which will also lead to a decrease in the prediction ability of the model. In a chaotic state, the importance of different features may change dynamically. Some features have a significant impact on the target variable in a certain period of time, but have less impact in other periods of time. The fixed binning method cannot adapt to the changes in feature importance in time, resulting in the model being unable to reflect new risk factors in a timely manner. In a chaotic state, the stability and generalization ability of the model are also challenged. The binning method must maintain a stable effect in different time periods and different data distributions. The instability of the binning boundary may cause the performance of the model to fluctuate greatly in different time periods and the generalization ability is insufficient. Complex binning methods such as binning based on information gain or decision trees significantly increase computational complexity and time cost when processing large amounts of data. When binning boundaries need to be adjusted frequently, high computational complexity will affect the real-time and efficiency of the binning method.

[0028] When the risk control system is marked as chaotic, the specific method to optimize and improve the binning algorithm is as follows: Use rolling window techniques to periodically recalculate bin boundaries to adapt to data changes; Evaluate the effects of different binning algorithms through cross-validation method and select the best binning algorithm; Use machine learning algorithms to clean risk control variables, identify and process noise, and remove outliers; Implement feature engineering on risk control variables through normalization, feature selection and construction to improve the feature expression ability of risk control data; Methods based on information gain, chi-square test, and decision tree were used to determine the optimal binning boundaries.

[0029] The effectiveness of the binning algorithm was evaluated based on the binning effect. The binning effect was evaluated by combining the population stability index PSI, weight of evidence WoE and KS value. The specific method is as follows: The calibration bin effectiveness index is Eb, and the calculation expression is , where PSI is the population stability index, expressed as , where is the actual proportion within the binning interval, is the expected proportion within the binning interval, is the weight of evidence, expressed as , where i is the bin number and i is a positive integer, is the proportion of non-default quantity in the i-th bin, is the proportion of defaults in the i-th bin, and the expression of KS value is , where is the cumulative bad account ratio in the i-th bin, is the cumulative good account ratio in the i-th bin, are the weight coefficients of PSI and KS respectively, and All are positive integers.

[0030] The state of the binning model is classified according to the evaluation results. The specific method is as follows: The binning validity threshold Et is preset. When the calculated binning validity index Eb is greater than or equal to the binning validity threshold Et, the robustness, predictive ability and distinguishing ability of the binning algorithm can meet the validity requirements, and the marked binning modeling is effective. When the calculated binning validity index Eb is less than the binning validity threshold Et, the robustness, predictive ability, and discrimination ability of the binning algorithm cannot meet the validity requirements, and the marked binning modeling is invalid.

[0031] Assist staff with decision-making based on the status type of the binning model. When binning modeling is marked as valid, it reminds staff that the binning model has the ability to identify fluctuations in the stability of the risk control system and does not require inspection and maintenance. When the binning modeling is marked as valid, it reminds the staff that the binning model does not have the ability to distinguish across the stability fluctuations of the risk control system, and requires the staff to optimize and improve the binning algorithm.

[0032] This application incorporates the chaos of the risk control system into the evaluation scope by combining it with the risk control stability state indicator, identifies the chaotic system by verifying the stability state of the risk control system, and incorporates the chaotic state of the risk control system into the optimization method of bin modeling to ensure that when the risk control system enters a chaotic state or changes between a chaotic state and a stable state, the feasibility of bin modeling is verified to ensure the applicability of bin modeling, avoid major failures caused by directional errors in the bin algorithm when the stability of the risk control system is unbalanced, and effectively improve the stability and scope of applicability of bin modeling.

[0033] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0034] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of computer program goods. The computer program goods include one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0035] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0036] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0037] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0038] If the functions are implemented in the form of software functional units and sold or used as independent goods, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of software goods, and the computer software goods are stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0039] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An adaptive binning and modeling method for risk control variables, characterized in that: The specific steps include analyzing the distribution characteristics of continuous variables and the relationship between continuous variables and target variables, and verifying the chaos of the risk control system; According to the verification results, the chaos of the risk control system is tested, and the binning algorithm is used to calculate and determine the binning boundaries of the risk control variables; Evaluate the effectiveness of the binning algorithm based on the binning effect, and classify the binning model based on the evaluation results; Provide decision-making assistance to staff based on the status type of the binning model; The method to verify the chaos of risk control system is: Combined with the mathematical method model of the risk control system, the Lyapunov index is calculated according to the state trajectory of the phase space of the risk control system to evaluate the sensitivity of the risk control system to the initial conditions. The specific evaluation method is: Selecting an initial point in the phase space of the risk control system as a reference point; At the initial point Some small perturbation points are selected nearby as adjacent initial points, and the adjacent initial points meet the small changes required for sensitivity; Use the mathematical method model of the risk control system to perform time evolution on the initial point and the adjacent initial points, generate a reference trajectory based on the initial point, and generate a disturbance trajectory based on the adjacent initial points; The difference between the reference trajectory and the perturbation trajectory is evaluated by calculating the Euclidean distance between the reference trajectory and each perturbation trajectory; The Lyapunov exponent is calculated from the growth rate of the difference between the reference trajectory and the perturbation trajectory.

2. The adaptive binning and modeling method of risk control variables according to claim 1, characterized in that: The phase space construction method of the risk control system is: Construct a phase space model of the risk control system. The risk factors in the risk control system include market risk, credit risk, operational risk and liquidity risk. Each risk factor is a dimension in the phase space model. Different risk factors correspond to different dimensions and together constitute the phase space of the risk control system.

3. The adaptive binning and modeling method of risk control variables according to claim 1, characterized in that: The state trajectory of the phase space is formed as follows: In the phase space of the risk control system, each possible risk state is represented by a point in the phase space, which is defined as the state point. The position of this point is determined by the values ​​of various risk factors. As time goes by, the risk state of the risk control system changes, which is manifested as the movement of the point in the phase space to form a trajectory, and the trajectory is defined as the state trajectory.

4. The adaptive binning and modeling method of risk control variables according to claim 1, characterized in that: The mathematical model of risk control system is: Taking time as the independent variable, the random process model is used to simulate the evolution of the state trajectory, and the mathematical method model is used to describe the risk control system, including using the variance-covariance method to use existing historical data to evaluate the expected return and risk level of risky assets, using the risk value method to evaluate the degree of loss by limiting the time interval and confidence level, and using the Monte Carlo simulation to evaluate the risk-free value model.

5. The adaptive binning and modeling method of risk control variables according to claim 1, characterized in that: The calculated Lyapunov index is analyzed, and the state of the risk control system is classified according to the analysis results. The analysis method is as follows: When the calculated Lyapunov exponent is less than or equal to 0, the risk control system is insensitive to the initial conditions and the marker is in a non-chaotic state; When the calculated Lyapunov exponent is greater than 0, the risk control system is sensitive to the initial conditions and is marked as being in a chaotic state.

6. The adaptive binning and modeling method of risk control variables according to claim 5, characterized in that: The method of using the binning algorithm to calculate and determine the binning boundaries of risk control variables is: When the risk control system is marked as non-chaotic, the data are sorted by the values ​​of the continuous variables and the continuous variables are arranged corresponding to the target variables; Use equal-width binning or equal-frequency binning to get an initial binning scheme based on the data distribution; For each bin, calculate the statistics of the target variable, including default rate and mean; Merge adjacent bins and keep the merged bins monotonic on the target variable; Adjust the binning according to the merge results so that the final binning meets the monotonicity requirement.

7. The adaptive binning and modeling method of risk control variables according to claim 6, characterized in that: When the risk control system is marked as chaotic, the specific method to optimize and improve the binning algorithm is as follows: Use rolling window techniques to periodically recalculate bin boundaries to adapt to data changes; Evaluate the effects of different binning algorithms through cross-validation method and select the best binning algorithm; Use machine learning algorithms to clean risk control variables, identify and process noise, and remove outliers; Implement feature engineering on risk control variables through normalization, feature selection and construction to improve the feature expression ability of risk control data; Methods based on information gain, chi-square test, and decision tree were used to determine the optimal binning boundaries.

8. The adaptive binning and modeling method of risk control variables according to claim 7, characterized in that: The method to evaluate the effectiveness of the binning algorithm based on the binning effect is: The calibration bin effectiveness index is Eb, and the calculation expression is , where PSI is the population stability index, expressed as , where is the actual proportion within the binning interval, is the expected proportion within the binning interval, is the weight of evidence, expressed as , where i is the bin number and i is a positive integer, is the proportion of non-default quantity in the i-th bin, is the proportion of defaults in the i-th bin, and the expression of KS value is , where is the cumulative bad account ratio in the i-th bin, is the cumulative good account ratio in the i-th bin, are the weight coefficients of PSI and KS respectively, and All are positive integers; The binning validity threshold Et is preset. When the calculated binning validity index Eb is greater than or equal to the binning validity threshold Et, the robustness, predictive ability and distinguishing ability of the binning algorithm can meet the validity requirements, and the marked binning modeling is effective; When the calculated binning validity index Eb is less than the binning validity threshold Et, the robustness, predictive ability, and discrimination ability of the binning algorithm cannot meet the validity requirements, and the marked binning modeling is invalid.

9. The adaptive binning and modeling method of risk control variables according to claim 8, characterized in that: The logic of assisting staff in making decisions based on the status type of the binning model is as follows: When the binning modeling is marked as valid, it reminds the staff that the binning model has the ability to identify fluctuations in the stability of the risk control system and does not need to be inspected and maintained; When the binning modeling is marked as valid, it reminds the staff that the binning model does not have the ability to distinguish across the stability fluctuations of the risk control system, and requires the staff to optimize and improve the binning algorithm.

Images