Popular insurance loss ratio prediction method based on ARIMA (autoregressive integrated moving average) model
By adopting the ARIMA model-based insured insurance compensation rate prediction method in the insurance industry, the problems of low accuracy of compensation rate prediction and difficulty in considering complex risk factors in the existing technology are solved, and higher prediction accuracy and stability are achieved, providing better support for insurance companies' risk management and business planning.
Patent Information
- Application Number
- CN202411558389.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-04
- Publication Date
- 2025-03-07
AI Technical Summary
The prior art methods used in the insurance industry for payout rate prediction have low prediction accuracy, are susceptible to extreme values or data fluctuations, and are difficult to fully consider the complex relationship between various risk factors and dynamic changes in the market environment.
The general insurance compensation rate prediction method based on the ARIMA model is adopted. By collecting historical actual compensation data and historical compensation data, pre-processing and characteristic factor construction, the basic ARIMA model is built, and the model is verified to build a regionally exclusive compensation prediction model.
It improves the accuracy and stability of the compensation rate prediction, can more accurately grasp the compensation risks in each region, and provides strong support for risk prevention and control and business layout.
Smart Images
Figure CN119579324A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of insurance business data processing, and in particular to a method for predicting the claim ratio of inclusive insurance based on an ARIMA model. Background Art
[0002] In today's insurance industry, accurate prediction of insurance claims ratio is of great significance to the risk management, business planning and decision-making of insurance companies. In particular, the existing inclusive insurance is large in scale, so it is necessary to predict the claims of inclusive insurance in different regions, so as to more accurately grasp the claims risks in each region and provide strong support for risk prevention and control and business layout.
[0003] In the insurance industry, traditional loss ratio prediction methods mainly rely on historical claims data and the experience and judgment of industry experts. It uses simple and intuitive analysis methods, such as the moving average method, to predict the future by calculating the average value of the loss ratio over a period of time. This method is simple to operate, easy to understand and implement, and is a viable option, especially in the early stages when data is insufficient or model building is complex. However, its prediction accuracy is relatively low, and it is easily affected by extreme values or data fluctuations, and it is difficult to fully consider the complex relationship between various risk factors and the dynamic changes in the market environment.
[0004] It can be said that in traditional methods, the prediction of loss ratio often relies on the experience and judgment of experts, simple statistical models and limited historical data. Although this method can meet basic prediction needs to a certain extent, its limitations are becoming increasingly prominent with the changes in the market environment and the complexity of data. The prediction results often lag behind market changes, and it is difficult to fully capture the various factors that affect the loss ratio, and it is impossible to ensure the accuracy and stability of the loss ratio prediction results. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for predicting the claim ratio of inclusive insurance based on the ARIMA model, which can improve the accuracy and stability of the claim ratio prediction.
[0006] The purpose of the present invention can be achieved by the following technical solution: A method for predicting the claim ratio of inclusive insurance based on the ARIMA model, comprising the following steps:
[0007] S1. Collect historical actual compensation data and historical compensation data;
[0008] S2. Preprocess the collected data to obtain the corresponding actual payment time series data and payable payment time series data;
[0009] S3. Based on the actual payment time series data and the payable payment time series data, the model parameters are determined by drawing the autocorrelation function (ACF) graph and the partial autocorrelation function (PACF) graph to build the basic ARIMA model;
[0010] S4. Based on the actual payment time series data and the payable payment time series data, characteristic factors are determined through regular statistics;
[0011] S5. Put the characteristic factors into the basic ARIMA model and verify the model;
[0012] S6. Input the historical data of claim amounts and the characteristic factors of the forecast period into the verified ARIMA model, and output the claim ratio corresponding to the forecast period.
[0013] Furthermore, the historical actual compensation data in step S1 is derived from the medical insurance settlement data of various regions;
[0014] Historical claims payable data comes from business collation data.
[0015] Furthermore, the preprocessing process in step S2 includes:
[0016] Mark the wrong values and outliers, make corrections for them, and then verify the data integrity;
[0017] Perform time series alignment and consistency check processing.
[0018] Furthermore, the model parameters determined in step S3 include the number of autoregressive terms P, the difference order D and the number of moving average terms Q.
[0019] Furthermore, the step S3 specifically identifies and determines the number of autoregressive terms P of the model from the autocorrelation function diagram, and identifies and determines the number of moving average terms Q of the model from the partial autocorrelation function diagram.
[0020] Furthermore, the process of determining the differential order D of the model in step S3 includes:
[0021] 1) Draw a time series diagram: First, draw a time series diagram of the time series. If the data in the time series diagram is not stable, differential processing is required and step 2) is executed. Otherwise, the current process ends.
[0022] 2) Difference processing: Perform difference processing on non-stationary time series data to obtain the difference order D.
[0023] Furthermore, the specific process of step 2) is as follows:
[0024] Through the first-order difference, second-order difference, and finally N-order difference, we gradually try them in sequence until we get a stable time series. During the difference processing, when the time series data after difference meets the stability requirement, the difference order D is determined. If the data is stable after the first-order difference, then D = 1; if the data is still not stable after the first-order difference, but the data is stable after the second-order difference, then D = 2, and so on.
[0025] Furthermore, the specific process of step S4 is: based on the payable compensation data and actual compensation data of each region in each month of previous years, combined with the time dimension, construct an effective characteristic factor.
[0026] Furthermore, the characteristic factors include beginning-of-year factors, end-of-year factors, and holiday factors.
[0027] Furthermore, the verification of the model in step S5 includes two verifications. The first verification is to input historical data into the model, predict the full-year forecast compensation data for the most recent year, and then compare the prediction result with the full-year actual compensation data for the most recent year;
[0028] The second verification is to input the historical data of the first half of the most recent year into the model, predict the predicted claims data for the second half of the most recent year, and then compare the predicted results with the actual claims data for the second half of the most recent year.
[0029] Compared with the prior art, the present invention has the following advantages:
[0030] The present invention collects historical actual payment data and historical payable payment data, pre-processes the collected data, and then based on the actual payment time series data and payable payment time series data, draws autocorrelation function graphs and partial autocorrelation function graphs to determine model parameters for building a basic ARIMA model; in addition, characteristic factors are determined through regular statistics; finally, the characteristic factors are placed in the basic ARIMA model, and the model is verified to construct a regional-specific payment prediction model. In this way, the regional factors of insurance and the time factors of previous years' data are deeply mined to obtain a prediction model matching different regions, which can greatly improve the accuracy and stability of the claim rate prediction.
[0031] The present invention preprocesses the collected data. On the one hand, it corrects the erroneous data. The present invention does not adopt a simple deletion method, but corrects it according to established business rules or the relationship between the previous and subsequent sequences. This approach aims to retain the integrity of the data while ensuring that the corrected data can truly reflect the actual business situation; on the other hand, through time series alignment and consistency verification, the logical coherence and accuracy of the data source are ensured, providing strong data support for the subsequent construction of an efficient and accurate prediction model.
[0032] The present invention identifies and determines the number of autoregressive terms P of the model from the drawn autocorrelation function diagram, identifies and determines the number of moving average terms Q of the model from the drawn partial autocorrelation function diagram, and adopts a differential processing method for possible non-stationary problems. When eliminating non-stationarity, one or more order differential processing is tried, and after repeated verification, the final differential order D is confirmed. In this way, the optimality of the model parameters can be guaranteed, which is conducive to building an accurate basic ARIMA model.
[0033] The present invention collects the data on payable and actual payments in each region on a monthly basis in previous years, combines the time dimension, constructs effective features, such as: beginning of the year factors, end of the year factors, holiday factors, etc., sorts out the variables that may have an impact, and uses these variables as important input information for optimizing the model effect, further improving the reliability and robustness of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION
[0035] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0036] Example
[0037] like Figure 1 As shown, a method for predicting the claim ratio of inclusive insurance based on the ARIMA model includes the following steps:
[0038] S1. Collect historical actual compensation data and historical payable compensation data, where the historical actual compensation data comes from the medical insurance settlement data of various regions;
[0039] The historical data on claims payable comes from the business collation data;
[0040] S2. Preprocess the collected data to obtain the corresponding actual payment time series data and payable payment time series data, wherein the preprocessing process includes:
[0041] Mark the wrong values and outliers, make corrections for them, and then verify the data integrity;
[0042] Perform time series alignment and consistency verification;
[0043] S3. Based on the actual payment time series data and the payable payment time series data, the model parameters are determined by drawing the autocorrelation function ACF diagram and the partial autocorrelation function PACF diagram to build the basic ARIMA model;
[0044] The model parameters include the number of autoregressive terms P, the number of difference orders D, and the number of moving average terms Q. The number of autoregressive terms P of the model is identified and determined from the autocorrelation function graph, the number of moving average terms Q of the model is identified and determined from the partial autocorrelation function graph, and the process of determining the number of difference orders D of the model includes:
[0045] 1) Draw a time series diagram: First, draw a time series diagram of the time series. If the data in the time series diagram is not stable, differential processing is required and step 2) is executed. Otherwise, the current process ends.
[0046] 2) Difference processing: Perform difference processing on non-stationary time series data to obtain the difference order D. Specifically, the first-order difference, second-order difference, and N-order difference are gradually tried in sequence until a stable time series is obtained. In the difference processing process, when the time series data after difference meets the stability requirements, the difference order D is determined. If the data is stable after the first-order difference, D = 1; if the data is still not stable after the first-order difference, but the data is stable after the second-order difference, D = 2, and so on;
[0047] S4. Based on the actual payment time series data and the payable payment time series data, characteristic factors are determined through regular statistics. Specifically, effective characteristic factors are constructed based on the payable payment data and actual payment data of each region in previous years and in combination with the time dimension, including the beginning of the year factor, the end of the year factor, and the holiday factor.
[0048] S5. Put the characteristic factors into the basic ARIMA model and verify the model;
[0049] This scheme is designed to verify the model twice. The first verification is to input historical data into the model, predict the full-year forecast compensation data for the most recent year, and then compare the prediction results with the full-year actual compensation data for the most recent year;
[0050] The second verification is to input the historical data of the first half of the most recent year into the model, predict the predicted compensation data of the second half of the most recent year, and then compare the predicted results with the actual compensation data of the second half of the most recent year;
[0051] S6. Input the historical data of claim amounts and the characteristic factors of the forecast period into the verified ARIMA model, and output the claim ratio corresponding to the forecast period.
[0052] This embodiment applies the above solution, and the main contents are:
[0053] 1. Data Acquisition
[0054] The data sources of this embodiment are Dolphin data (i.e. actual compensation data) and payable compensation data. The following is an introduction to the characteristics of the two sets of data and the reasons for using them simultaneously.
[0055] Dolphin data comes from medical insurance settlement data from various regions. The biggest advantage of this part of data is its high accuracy, which provides a stable and high-quality basic information framework for the model. Although its timeliness is relatively weak, it effectively improves the reliability and stability of model predictions. As one of the main data sources for the prediction model, Dolphin data is crucial to the long-term effect and accuracy of the model.
[0056] The data on payable claims are collected and organized business data. This part of the data is characterized by good timeliness and can quickly reflect market changes and claims trends. As a data source complementary to Dolphin Data, it can enhance the effectiveness of claims prediction and early warning.
[0057] Although there may be some differences between the payable data and the actual amount of compensation in actual applications, they play an important role in improving the timeliness of model prediction and verifying the accuracy of the model. By combining the stability of Dolphin data and the timeliness of payable data, a robust and sensitive prediction model can be constructed to better cope with the complex and changing compensation environment.
[0058] 2. Data Governance
[0059] During the initial data inspection, we focused on marking obvious errors and outliers in the compensation amount sequence. These errors may be caused by system reasons or negligence during manual entry, which may affect the accuracy and reliability of data analysis. At the same time, this solution checks the integrity of the data to ensure that all key fields are fully saturated and without omissions, laying a solid foundation for subsequent analysis.
[0060] Cautious governance measures were taken for the problematic data identified in the initial inspection: 1) Correction of erroneous data. For the erroneous data found, this solution did not simply delete it, but corrected it based on established business rules or the relationship between the previous and next sequences. This approach aims to preserve the integrity of the data while ensuring that the corrected data can truly reflect the actual business situation. 2) Data comparison and analysis. As one of the core contents of data governance, the sequence differences between Dolphin data and payable data were compared. Through time series alignment and consistency verification, the logical coherence and accuracy of the two major data sources are ensured, providing strong data support for the subsequent construction of efficient and accurate prediction models.
[0061] 3. Parameter fitting and tuning
[0062] First, we deeply observed the overall trend and seasonal characteristics of the time series data, and focused on evaluating the stationarity of the data series by combining intuitive judgment with statistical tests. In order to solve the possible non-stationary problem, we adopted the difference processing method. When eliminating non-stationarity, we tried one-order or multi-order difference processing, and after repeated verification, we confirmed the final difference order D. In order to ensure that the sequence after difference meets the stationarity condition, this embodiment further uses ADF test, KPSS test and PP test for verification.
[0063] After that, by drawing the autocorrelation function (ACF) diagram and the partial autocorrelation function (PACF) diagram, the autocorrelation and partial autocorrelation characteristics in the sequence were identified by graphical judgment, which served as the main basis for subsequent model order determination.
[0064] After confirming the important parameters of the model, such as P (number of autoregressive terms), D (order of difference), and Q (number of moving average terms), the modeling results were considered using the AIC and BIC criteria, and the grid search algorithm was used to confirm the final P, D, and Q combination through automated search and evaluation.
[0065] Based on the prediction results, this embodiment makes full use of the experts' understanding and experience of the business. Combining the experts' professional opinions with the actual business situation, the model prediction results are adjusted and optimized in a targeted manner, especially for special factors such as policy changes, in-depth discussions and analysis are carried out, and the model is improved and tuned accordingly.
[0066] IV. Model Effect Evaluation
[0067] The model prediction results were compared with the actual claims data in detail, and the accuracy and degree of deviation of the model prediction were quantified by calculating key indicators such as absolute error and cumulative error. At the same time, the distribution and trend of the deviation were analyzed in depth to identify potential deficiencies and sources of error in the model prediction.
[0068] Finally, the prediction performance of the model was comprehensively evaluated by combining the prediction and actual deviation comparison results with the guidance of expert experience. By comprehensively considering the accuracy, stability, timeliness and other aspects of the model, a clear conclusion was drawn that the model can meet business needs. At the same time, subsequent optimization strategies and adjustment plans were formulated based on the evaluation results to ensure that the model can continue to provide strong support for business decisions.
[0069] In summary, this solution innovatively introduces technologies such as big data and time series analysis for the prediction of inclusive insurance claims ratio, making the prediction of claims ratio more data-driven. Through the construction of mathematical models, model prediction can timely discover and identify hidden laws and trends in data, thereby achieving accurate prediction of claims ratio.
[0070] The innovations of this transformation are mainly reflected in the following aspects: first, model prediction can update data in real time to ensure that the prediction results are always synchronized with market changes; second, it introduces diversified indicators, comprehensively considers multiple factors such as market, competition, and customers, making the prediction more comprehensive and in-depth; third, model prediction improves the reliability and robustness of the prediction by establishing a risk-sensitive prediction model, modeling and evaluating uncertainty factors; finally, algorithm combination and system structure innovation also bring greater flexibility and adaptability to model prediction.
[0071] These innovations together give model prediction significant advantages. On the one hand, model prediction significantly improves the accuracy of loss ratio prediction, making users' decisions more scientific and reasonable; secondly, it enhances the real-time and comprehensiveness of prediction, enabling users to better respond to market changes; thirdly, the automation and intelligence of model prediction reduces the burden of manual prediction and improves work efficiency.
[0072] By building a region-specific claims ratio prediction model, users can more accurately grasp the claims risks in each region and provide strong support for risk prevention and control and business layout. In the regional dimension, the claims ratio model prediction will achieve real-time monitoring and dynamic adjustment. This means that users can quickly respond to market changes, policy adjustments and emergencies in the region, and make instant predictions and assessments of claims ratios. By dynamically adjusting insurance strategies and service plans, users can better adapt to regional market needs, reduce claims costs, and improve profitability.
Claims
1. A method for predicting the claim ratio of inclusive insurance based on the ARIMA model, characterized in that: The following steps are involved: S1. Collect historical actual compensation data and historical compensation data; S2. Preprocess the collected data to obtain the corresponding actual payment time series data and payable payment time series data; S3. Based on the actual payment time series data and the payable payment time series data, the model parameters are determined by drawing the autocorrelation function ACF diagram and the partial autocorrelation function PACF diagram to build the basic ARIMA model; S4. Based on the actual payment time series data and the payable payment time series data, characteristic factors are determined through regular statistics; S5. Put the characteristic factors into the basic ARIMA model and verify the model; S6. Input the historical data of claim amounts and the characteristic factors of the forecast period into the verified ARIMA model, and output the claim ratio corresponding to the forecast period.
2. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 1, characterized in that: The historical actual compensation data in step S1 is derived from the medical insurance settlement data of various regions; Historical claims payable data comes from business collation data.
3. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 1, characterized in that: The pre-processing process in step S2 includes: Mark the error values and outliers, make corrections for the error values and outliers, and then verify the data integrity; Perform time series alignment and consistency check processing.
4. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 1, characterized in that: The model parameters determined in step S3 include the number of autoregressive terms P, the number of difference orders D and the number of moving average terms Q.
5. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 4, characterized in that: The step S3 specifically identifies and determines the number of autoregressive terms P of the model from the autocorrelation function diagram, and identifies and determines the number of moving average terms Q of the model from the partial autocorrelation function diagram.
6. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 4, characterized in that: The process of determining the difference order D of the model in step S3 includes: 1) Draw a time series diagram: First, draw a time series diagram of the time series. If the data in the time series diagram is not stable, differential processing is required and step 2) is executed. Otherwise, the current process ends. 2) Difference processing: Perform difference processing on non-stationary time series data to obtain the difference order D.
7. The method for predicting the loss ratio of inclusive insurance based on the ARIMA model according to claim 6, characterized in that: The specific process of step 2) is as follows: Through the first-order difference, second-order difference, and finally N-order difference, we gradually try them in sequence until we get a stable time series. During the difference processing, when the time series data after difference meets the stability requirement, the difference order D is determined. If the data is stable after the first-order difference, then D = 1; if the data is still not stable after the first-order difference, but the data is stable after the second-order difference, then D = 2, and so on.
8. The method for predicting the loss ratio of inclusive insurance based on the ARIMA model according to claim 1, characterized in that: The specific process of step S4 is: based on the payable compensation data and actual compensation data of each region in previous years and in combination with the time dimension, construct an effective characteristic factor.
9. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 8, characterized in that: The characteristic factors include beginning-of-year factors, end-of-year factors, and holiday factors.
10. The method for predicting the claim ratio of inclusive insurance based on the ARIMA model according to claim 1, characterized in that: The verification of the model in step S5 includes two verifications. The first verification is to input historical data into the model, predict the full-year forecast compensation data for the most recent year, and then compare the prediction result with the full-year actual compensation data for the most recent year; The second verification is to input the historical data of the first half of the most recent year into the model, predict the predicted claims data for the second half of the most recent year, and then compare the predicted results with the actual claims data for the second half of the most recent year.