Quantitative and objective meteorological service benefit evaluation method for insurance industry

Abstract

The application belongs to the technical field of meteorological services, and specifically relates to a quantitative and objective meteorological service benefit evaluation method for the insurance industry, which comprises the following steps: aligning the insurance policy with the meteorological grid point space; quantifying each space-time unit to generate a meteorological service intervention degree index, and extracting its exogenous component; generating a conditional probability loss matrix, and assigning an expected loss risk score to each space-time unit; estimating the marginal damage reduction effect; estimating the additional damage reduction of the intervention group; constructing a causal regularization model; predicting the counterfactual expected claims under the counterfactual state without service in the high intervention period by using the counterfactual claims predictor; and constructing a double-accounting framework of social benefit accounts and insurance industry accounts; in the application, the evaluation defects of lacking disaster loss data and neglecting indirect benefits are solved by replacing qualitative analysis with policy process data, integrating entropy intervention, causal inference and counterfactual model, and realizing the monetization and quantitative evaluation of the damage reduction benefits in the whole value chain.

Description

Technical Field

[0001] This invention belongs to the field of meteorological service technology, specifically a quantitative and objective meteorological service benefit evaluation method for the insurance industry. Background Technology

[0002] With the intensification of global climate change and the increasing frequency of extreme weather events, the insurance industry's reliance on meteorological data has significantly increased. Modern insurance businesses, particularly in property insurance, agricultural insurance, and catastrophic insurance, have gradually built a technical support system based on multi-source meteorological data. At the data acquisition level, meteorological satellite remote sensing technology can provide large-scale cloud top temperature, vegetation index, and sea surface temperature data; Doppler weather radar networks enable three-dimensional monitoring of precipitation and wind fields; and ground-based automatic weather station networks, through high-density deployment, acquire basic elements such as temperature, humidity, air pressure, wind speed, and precipitation. After quality control, these raw observation data are input into numerical weather prediction models to generate gridded weather forecast products for the next 0-15 days. Simultaneously, historical meteorological reanalysis datasets integrate observation records from the past decades through assimilation techniques, forming long-series climate baseline data. In the data processing stage, geographic information system (GIS) technology overlays meteorological raster data with the spatial location of the insured object for analysis, while database technology supports the storage and retrieval of massive amounts of meteorological time-series data. In addition, weather index generation technology has begun to be applied to specific insurance products. By using preset algorithms, continuous weather variables are transformed into quantifiable index values, providing an objective basis for the design of insurance products.

[0003] Meteorological service benefit assessment involves the integration of interdisciplinary technical methods, and its core lies in establishing a quantitative link between meteorological service inputs and economic and social outputs.

[0004] General meteorological service benefit assessment methods tend to neglect subjective and indirect benefits to some extent, thus underestimating the effectiveness of meteorological services; moreover, they mostly use qualitative analysis methods, resulting in insufficient characterization of the effectiveness of insurance meteorological services. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a quantitative objective meteorological service benefit evaluation method for the insurance industry, which is mainly used to address the problem that general meteorological service benefit evaluation methods to some extent neglect subjective and indirect benefits.

[0006] The technical solution adopted by this invention to solve its technical problem is: Quantitative objective meteorological service benefit evaluation methods for the insurance industry include: S1: Collect meteorological monitoring data, insurance institution business operation logs and control variables, and align the policy location with meteorological grid points through a geographic information system to form a spatiotemporal matching dataset; S2: Based on the spatiotemporal matching dataset, the meteorological service intervention index is generated by quantifying each spatiotemporal unit and its exogenous components are extracted; S3: Using exogenous components as weights, the marginal distribution estimation of disaster intensity and insurance loss sequence and the Copula joint distribution construction process are weighted and corrected to generate a conditional probability loss matrix and assign expected loss risk scores to each spatiotemporal unit; S4: Input the expected loss risk score and meteorological service intervention index into the preset panel fixed effects model to estimate the marginal loss reduction effect; take regional sudden meteorological disasters as the breakpoint and use the staggered difference-in-differences model based on propensity score to estimate the additional loss reduction of the intervention group; perform consistency calibration on the marginal loss reduction effect and the additional loss reduction and merge them into a candidate quantity of monetizable objective direct benefits; S5: Based on the marginal loss reduction effect and the additional loss reduction of the intervention group, a causal regularization model is constructed; the causal regularization model is trained as a counterfactual payout predictor using samples from periods with intervention levels below a preset threshold to predict the counterfactual expected payout in the absence of service during periods of high intervention. S6: After passing the counterfactual expected compensation execution test, based on the candidate quantity of objective direct benefits in monetization, construct a dual-account accounting framework of social benefit account and insurance industry account, decompose the evaluation results according to account affiliation and insurance business links, and output the evaluation results.

[0007] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by the present invention, the step of aligning the policy location with the meteorological grid space includes: S11: Convert the geographic coordinates of the policy target into a point vector layer, perform spatial intersection calculation with the meteorological grid layer, so that each policy target obtains its own meteorological grid number; S12: Perform nearest neighbor matching on disaster events and insurance business records based on timestamps to form a standard data structure; S13: Perform outlier removal and multiple imputation of missing values ​​on the matched dataset to output a spatiotemporal matching dataset.

[0008] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by this invention, the steps for quantifying meteorological service intervention indicators include: S21: Based on the system instruction change record, calculate the trigger ratio of loss prevention inspection work orders or material allocation instructions within a preset time after the early warning is issued, and generate an active loss prevention action index. S22: Based on the statistics of underwriting rule trigger records, the proportion of policies whose underwriting conclusions were changed due to the meteorological risk field is generated as an underwriting intervention sub-dimension; S23: Based on the pricing factor loading record statistical quotation system, retrieve the call rate of meteorological risk factors to generate the actuarial pricing penetration rate; S24: Based on resource pre-deployment records, obtain the timeliness score of claims personnel and rescue vehicles arriving after the warning; S25: The proactive loss prevention action index, underwriting intervention sub-dimension, actuarial pricing penetration rate and timeliness score are weighted and condensed using the entropy weight method to generate a meteorological service intervention degree index that is continuously distributed in the [0,1] interval; S26: Using meteorological supply-side variables as exogenous tools, extract the exogenous components of intervention degree through instrumental variable method or orthogonalization to eliminate reverse causal effects; prioritize the use of exogenous components in weighting, matching and causal identification.

[0009] The quantitative objective meteorological service benefit evaluation method for the insurance industry provided by this invention includes the following steps for weighted correction: S31: Fit a generalized extreme value distribution to the meteorological disaster intensity sequence, and a gamma distribution or a log-normal distribution to the insurance loss sequence. When estimating the distribution parameters using the maximum likelihood method, introduce sample weights constructed from the exogenous component of the meteorological service intervention index into the likelihood function. S32: Determine the goodness of fit using a weighted KS test, and select the optimal marginal distribution that passes the significance test; S33: When selecting the optimal Copula, the empirical Copula adjusted by the exogenous component of the intervention degree is used as the fitting target to generate the conditional probability loss matrix.

[0010] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by this invention, the conditional probability loss matrix is ​​expressed as follows: , In the formula, M is the disaster loss matrix. The level of disaster intensity. According to the level of insured loss, In order to achieve a disaster intensity of Under the conditions, loss level occurs The probability of.

[0011] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by this invention, the formula for the panel fixed effects model is expressed as follows: , In the formula, The earned loss ratio or combined ratio, where α is a constant term; The expected loss risk score, for The main effect coefficient, As an indicator of meteorological service intervention, for The main effect coefficient, For interactive items, capture how the degree of intervention plays a mitigating role as the risk level changes; This refers to the interaction term coefficient, i.e., the marginal loss effect; For individual fixed effects, For time-fixed effects, This is the kth control variable; for coefficient, For random error term, i is the cross-sectional individual identifier, t is the time identifier, and k is the control variable number.

[0012] According to the quantitative objective meteorological service benefit assessment method for the insurance industry provided by this invention, the steps for estimating the additional loss reduction of the intervention group using a propensity score-based difference-in-differences model include: S41: Individuals whose meteorological service intervention index is higher than the first threshold are assigned to the treatment group, and individuals whose index is lower than the second threshold are assigned to the control group; the first threshold and the second threshold are determined by data-driven quantile method or Youden index, and sensitivity analysis is performed on the thresholds; S42: Estimate the propensity score using covariates, perform caliper nearest neighbor matching, and generate a balanced panel. S43: Using a single, highly disastrous meteorological event as the time breakpoint, a time breakpoint indicator variable is defined. For cases involving multiple overlapping regions, a difference-in-differences model is established using the Callaway-Sant'Anna or Sun-Abraham estimators, expressed as follows: , In the formula, For unit loss indicators, The coefficient of variation between the treatment group and the control group is the intrinsic difference coefficient. This is the time effect coefficient. To handle group indicator variables, δ is a time discontinuity indicator variable, and δ is a difference-in-differences estimator that measures the additional damage reduction effect of intervention under disaster impact; This is the random error term.

[0013] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by this invention, the steps for constructing a causal regularization model include: S51: Select spatiotemporal samples with meteorological service intervention indicators below the preset low threshold as the non-intervention sample set, with disaster intensity level, building structure type, target geomorphological code, regional GDP and insurance penetration rate as input features, and actual compensation amount as output label; S52: In the loss function of LightGBM, a monotonic derivative penalty term based on the marginal loss reduction effect is added, and the additional loss amount is constructed as an unbiased estimation constraint term to form a causal regularized loss function; the unbiased estimation constraint term is anchored to the loss amount based on the reporting rate of the actual loss reduction sub-stage, rather than the total loss reduction amount; S53: Train gradient boosting trees based on causal regularization loss function. For samples in high intervention periods, take the average of multiple random seed predictions as the counterfactual expected compensation.

[0014] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by the present invention, step S6 includes the verification and accounting steps: S62: Placebo test at the implementation point, advance the policy time point by 2-3 years, and supplement it with random permutation of the intervention degree, repeatedly perform difference-in-differences estimation, and test whether the true estimated value is in the tail of the random distribution to eliminate the interference of accidental factors; S63: Execution mechanism test: The vulnerability of the insured amount-weighted object is incorporated as a time-varying covariate into the difference-in-differences model to control for changes in the quality of the insured pool; the unit loss indicator is decomposed into the reporting rate and the average claim payout, distinguishing between the actual loss reduction and the screening effect of the insured pool; the robustness of the treatment effect is verified by re-estimating on a county-year subsample with stable insurance penetration. S64: Construct a dual-account accounting framework for a social benefit account and an insurance industry account. Effects confirmed as genuine loss reduction through mechanism verification will be included in the social benefit account, while the net benefits of underwriting screening, actuarial pricing, operational efficiency improvement, and customer relationship enhancement will be included in the insurance industry account. The two accounts will be listed side by side and addition is prohibited.

[0015] According to the quantitative objective meteorological service benefit evaluation method for the insurance industry provided by the present invention, the steps for outputting the evaluation results include: S65: In the disaster prevention and mitigation process, calculate the difference in reporting rates between covered and uncovered areas, and include it in the social benefit account as the actual loss reduction; S66: In the underwriting screening process, the difference in loss rates is compared and the net benefit of claims savings and premium waivers is calculated. This process is recorded as a gain in the insurance industry account and as a risk transfer with negative externalities in the social benefit account. S67: Measure the increase in premium adequacy after the introduction of weather factors in the actuarial pricing process and include it in the insurance industry account; S68: In the operational efficiency improvement stage, the cost of reducing the average investigation time per case is calculated and supported by forecasts. This is recorded as a real resource saving and is included in both the social benefit account and the insurance industry account. In the customer enhancement stage, the renewal improvement is broken down into net operational efficiency improvement and renewal income. Renewal income is listed as a separate operating indicator and is not directly included in the benefit account. S69: Generate a structured benefit assessment report categorized by dual accounts, business processes, and meteorological service products.

[0016] This invention establishes a spatiotemporally matched policy-level multidimensional panel data foundation, extracting actual service reach data from four business process dimensions: proactive loss prevention, underwriting optimization, actuarial penetration, and resource timeliness. A continuous meteorological service intervention index is objectively synthesized using the entropy weight method, and exogenous components are extracted to eliminate reverse causality. This overcomes the bottleneck of traditional methods being limited by missing or poor-quality disaster loss data, ensuring that benefit assessment is based on observable and traceable process evidence, guaranteeing the objectivity and operability of the assessment from the source. By introducing the exogenous intervention component as a weighting factor into the Copula joint distribution model, the fitting effect of risk scores on extreme losses is corrected. Furthermore, a panel fixed effects model is used to estimate the marginal reduction effect, and a propensity score matched staggered difference-in-differences model is used to estimate the local average treatment effect. Inverse variance weighting is used to calibrate the consistency of causal effects. Building upon this foundation, the obtained unbiased causal constraints are embedded into LightGBM to construct a causal-regularized counterfactual loss prediction model. This model accurately calculates expected losses under conditions without meteorological service intervention, transforming the assessment results from a qualitative assessment of "whether it is helpful or not" to a quantitative assessment of "how much loss is reduced." Combined with a point-in-time placebo test to verify statistical significance, this significantly improves the precision and credibility of the portrayal of insurance meteorological service effectiveness. By introducing a dual-account accounting framework, the total loss reduction and efficiency enhancement value is broken down and independently calculated at each stage. This fundamentally corrects the systematic underestimation caused by traditional methods focusing on direct losses while neglecting indirect value-added benefits, as well as the misleading practice of misrepresenting risk transfer as genuine loss reduction. It achieves comprehensive and verifiable monetization accounting covering the entire value chain from risk pricing to customer value enhancement, and can separately meet the decision-making needs of governments and insurance companies, providing a quantitative fulcrum for the closed-loop business of financial meteorology. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of a quantitative objective meteorological service benefit evaluation method for the insurance industry, as described in this invention. Figure 2 This is a flowchart illustrating the quantitative meteorological service intervention index in an embodiment of the present invention; Figure 3 This is a flowchart of constructing a causal regularization model in an embodiment of the present invention. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Example: Figures 1 to 3 As shown, this embodiment provides a quantitative objective meteorological service benefit evaluation method for the insurance industry, including: S1: Collect meteorological monitoring data, insurance institution business operation logs, and control variables. Use a geographic information system (GIS) to align the policy location with meteorological grid points to form a spatiotemporal matching dataset. Meteorological monitoring data includes, but is not limited to, grid or station observations such as precipitation, wind speed, temperature, and relative humidity. Insurance institution business operation logs are time-series records of system instruction changes, underwriting decisions, pricing factor calls, and claims resource scheduling recorded by insurance companies in their daily operations. Control variables are other external factors affecting insurance payouts, such as regional GDP, building structure type codes, and insurance penetration rates. The GIS uses spatial computation to link the spatial location of the policy target with the meteorological grid.

[0021] The steps for aligning the policy location with meteorological grid points using a geographic information system include: S11: Convert the geographic coordinates of the policyholder to a point vector layer, and perform a spatial intersection operation with the meteorological grid polygon layer to assign a meteorological grid number to each policyholder. The spatial intersection operation can be implemented using the ST_Intersects function of the spatial database, which overlays the longitude and latitude of the policyholder with the meteorological grid polygons to assign a unique identifier to the nearest meteorological grid for each policy record.

[0022] S12: Perform nearest neighbor matching on disaster events and insurance business records based on timestamps to form a standard data structure. Disaster events refer to the start and end times of sudden meteorological disasters such as typhoons, rainstorms, and hail. Insurance business records include the reporting time, loss assessment time, etc. When matching, the time of the disaster is used as the benchmark, and the insurance record with the smallest absolute value of the time difference is searched for and associated. The matching tolerance window can be set to ±72 hours.

[0023] S13: Perform outlier removal and multiple imputation of missing values ​​on the matched dataset to output a spatiotemporal matching dataset. Outlier removal uses the interquartile range method to replace extreme values ​​exceeding the upper and lower boundaries with boundary values; multiple imputation of missing values ​​uses a chain equation algorithm to build an imputation model using other complete variables and generate multiple complete datasets, and finally uses the mean to fill in the missing values ​​to ensure the integrity and robustness of the data.

[0024] S2: Based on the spatiotemporal matching dataset, a meteorological service intervention index is generated for each spatiotemporal unit, and exogenous components are extracted. The meteorological service intervention index (MSID) measures the comprehensive degree to which insurance institutions take proactive measures based on meteorological information within a specific spatiotemporal unit, with a value range of [0,1], where 0 represents no intervention and 1 represents maximum intervention. The steps include: S21: Based on system instruction change records, calculate the trigger ratio of loss prevention inspection work orders or material allocation instructions within a preset time after the early warning is issued, and generate a proactive loss prevention action index, expressed by the formula: , In the formula, As a proactive loss prevention action index, The number of loss prevention and inspection work orders actually generated within a preset time after the early warning is issued in spatiotemporal unit i within time t. This refers to the number of material allocation orders issued during the same period. This represents the total number of warning events for that unit within that time period; if ,but Take 0.

[0025] S22: Based on the statistics of underwriting rule trigger records, the percentage of policies whose underwriting conclusions were changed due to the meteorological risk field is used to generate an underwriting intervention sub-dimension, expressed by the formula: , In the formula, For the underwriting intervention sub-dimension, This refers to the number of policies whose underwriting decisions changed from normal coverage to increased premiums, restrictions, or rejection due to meteorological risk fields (such as rainstorm risk level and typhoon path). This represents the total number of quoted insurance policies within this spatiotemporal unit.

[0026] S23: Based on the pricing factor loading record statistical quotation system's retrieval rate of meteorological risk factors, the actuarial pricing penetration rate is generated, expressed by the formula: , In the formula, To accurately calculate pricing penetration rate, The number of policies for which meteorological risk factors (such as extreme rainfall and wind zone categories) were retrieved during the quotation process. This represents the total number of quoted insurance policies.

[0027] S24: Based on resource pre-deployment records, obtain the timeliness score of claims personnel and rescue vehicles arriving after the warning, expressed by the formula: , In the formula, For time-sensitive purposes, This represents the average time (in minutes) for resources to arrive at the scene after an early warning. and These are the minimum and maximum resource arrival times among all spatiotemporal samples, respectively. After normalization, a higher score indicates more timely arrival.

[0028] S25: The proactive loss prevention action index, underwriting intervention sub-dimension, actuarial pricing penetration rate, and timeliness score are weighted and condensed using the entropy weight method to generate a meteorological service intervention index continuously distributed in the [0,1] interval. The entropy weight method is an objective weighting method that calculates information entropy based on the degree of variation of each indicator; the smaller the entropy value, the greater the weight of the indicator. The specific steps are as follows: First, the... Min-Max standardization is performed to obtain dimensionless values. The information entropy of the k-th indicator is calculated using the following formula: , in, Let be the information entropy of the k-th indicator, and N be the total number of samples, i.e., the number of spatiotemporal units. Let be the proportion of the k-th indicator in the sample (i,t). The indicator values ​​are dimensionless values ​​after Min-Max standardization; indicator weights ,in Let j = 1, 2, 3, 4 represent the information entropy of each indicator, corresponding to AI, UI, PP, and TS respectively. The final meteorological service intervention index is expressed as follows: , In the formula, This is a meteorological service intervention index, with a value range of [0,1]. The weights are obtained by the entropy weight method. These are the values ​​of each sub-index after Min-Max standardization.

[0029] S26: Extracting the exogenous and endogenous components of intervention degree. Since the proactive loss prevention action index, underwriting intervention sub-dimension, and actuarial pricing penetration rate are all determined by the market behavior of insurance institutions, there is a reverse causal risk—high-risk targets are more likely to be subject to loss prevention investigations or underwriting reversals, leading to a correlation between intervention degree and risk itself. Therefore, supply-side variables from the meteorological sector are used as exogenous tools, including warning levels, warning lead time, and meteorological service product supply intensity (IMSL), to extract these components. Regressing the above supply-side variables and control variables, the fitted values ​​are used as the exogenous component MSID. exo The residual as an endogenous component MSID endo In subsequent distribution weighting, propensity score matching, and causal identification, the exogenous component MSID is preferentially used. exo Simultaneously, the multiplicative structure MSID, which combines the exogenous component of the intervention degree with the meteorological supply intensity, can be adopted. exo ×IMSL describes the synergistic effect of "supply-side supply × demand-side adoption".

[0030] S3: Using the exogenous component of the meteorological service intervention index as weight, the marginal distribution estimation and Copula joint distribution construction process of disaster intensity and insurance loss sequence are weighted and corrected to generate a conditional probability loss matrix and assign expected loss risk scores to each spatiotemporal unit.

[0031] The core idea of ​​weighted correction is to amplify the contribution of high-intervention samples during the distribution fitting process, so that the subsequent dependency structure and risk score can better reflect the actual loss characteristics after meteorological service intervention. The steps include: S31: When fitting a generalized extreme value distribution to the meteorological disaster intensity sequence and a gamma distribution or log-normal distribution to the insurance loss sequence, and estimating the distribution parameters using the maximum likelihood method, sample weights constructed from the exogenous component of the meteorological service intervention index are introduced into the likelihood function; the objective function of the weighted maximum likelihood estimation is the weighted log-likelihood function. ,in, The weighted log-likelihood function value. Let N be the vector of distribution parameters to be estimated, and N be the total number of samples. Let be the probability density function of the selected marginal distribution. For the j-th observation, the sample weights are... Obtained by the exogenous intervention degree mapping of the corresponding spatiotemporal unit, for example λ is a magnification factor greater than 0, to enhance the impact of high-intervention samples on the estimation of tail parameters; Let be the exogenous meteorological service intervention index corresponding to sample j; the generalized extreme value distribution is used to capture the extreme value characteristics of disaster intensity, and its distribution function is expressed as: , In the formula, Let x be the cumulative distribution function of the generalized extreme value distribution, and let x be a random variable, such as disaster intensity. For position parameters, For scale parameters, These are shape parameters that control tail behavior.

[0032] S32: Determine the goodness of fit using a weighted KS test and select the optimal marginal distribution that passes the significance test; the weighted KS test statistic... Defined as: , in For the weighted empirical distribution function, Let I be the distribution function to be tested, and let I(⋅) be the indicator function, which is 1 when the condition is met and 0 otherwise.

[0033] S33: When selecting the optimal Copula, the empirical Copula adjusted by the exogenous component of the intervention degree is used as the fitting target to generate the conditional probability loss matrix, expressed by the formula: , In the formula, S is the disaster loss matrix. The level of disaster intensity. According to the level of insured loss, In order to achieve a disaster intensity of Under the conditions, loss level occurs The probability of is calculated using a weighted Copula function and is expressed as: , in The optimal Copula function selected after weighting the intervention degree. and These represent the marginal cumulative probabilities of disaster intensity and loss level, respectively. Loss level The corresponding marginal cumulative probability, This represents the edge cumulative probability corresponding to adjacent lower disaster intensity levels.

[0034] S4: Input the expected loss risk score and meteorological service intervention index into a pre-set panel fixed effects model to estimate the marginal mitigation effect; using regional sudden meteorological disasters as breakpoints, estimate the additional mitigation amount of the intervention group using a propensity score-based staggered difference-in-differences model; perform consistency calibration on the marginal mitigation effect and the additional mitigation amount, and merge them into candidate quantities for monetized objective direct benefits; the formula of the panel fixed effects model is expressed as:

[0035] In the formula, The earned loss ratio or combined ratio, where α is a constant term; The expected loss risk score, for The main effect coefficient, As an indicator of meteorological service intervention, for The main effect coefficient, For interactive items, capture how the degree of intervention plays a mitigating role as the risk level changes; The coefficient of the interaction term represents the marginal loss reduction effect. Its economic meaning is the change in the marginal impact of the expected loss risk score on the payout rate when the intervention level increases by one unit, that is, the degree to which meteorological services reduce payouts in a high-risk environment. Multiplying this effect by the total risk exposure of the sample yields the total marginal loss reduction. It should be noted that this total marginal loss reduction includes both the actual loss reduction and the underwriting pool screening effect, and must be decomposed and distributed after verification by the S6 mechanism. It cannot be directly used as a social loss reduction benefit. For individual fixed effects, For time-fixed effects, This is the kth control variable; for coefficient, For random error term, i is the cross-sectional individual identifier, t is the time identifier, and k is the control variable number.

[0036] Expected loss risk score The loss is calculated by combining the conditional probability loss matrix with the disaster intensity level of the unit, and is expressed as follows: , In the formula, M represents the total number of loss levels. Loss level The representative value (usually taken as the midpoint of the grade range). The disaster intensity level observed in the spatiotemporal unit (i,t).

[0037] Using regional sudden meteorological disasters as breakpoints, a difference-in-differences model based on propensity score is used to estimate the additional loss in the intervention group. The steps include: S41: Individuals whose meteorological service intervention index consistently exceeds the first threshold are assigned to the treatment group, while those below the second threshold are assigned to the control group. The first and second thresholds are not set empirically but determined using a data-driven approach: the 75th and 25th percentiles of the intervention distribution can be used, respectively, or the optimal cut-off point can be determined by maximizing the Youden exponent of the receiver operating characteristic (ROC) curve. In the robustness test, sensitivity analysis is performed on the threshold within its neighborhood to verify that the direction and significance of the treatment effect do not reverse with fine-tuning of the threshold.

[0038] S42: Estimate propensity score using covariates, perform caliper nearest neighbor matching, and generate a balanced panel. Propensity score is defined as the probability of being assigned to a treatment group given the covariates. Regression via logit It is estimated that among them For covariate vectors, The coefficient vector of the covariates; in addition to regional economy and target attributes, the covariates also include expected loss or risk scores to control for differences in target vulnerability between the treatment group and the control group; the matching is performed using caliper nearest neighbor matching, and the caliper width is usually set to 0.05 times the propensity score standard deviation to ensure that the treatment group and the control group are evenly distributed on the covariates.

[0039] S43: Using a single, highly disastrous meteorological event as the time discontinuity indicator variable, a difference-in-differences model is established. When the assessment scope involves multiple regions and multiple time points with overlapping treatments, the Callaway-Sant'Anna estimator or the Sun-Abraham estimator is used to estimate the group-time average treatment effect, and the effects are weighted and aggregated according to treatment group and time point to avoid the negative weighting bias of traditional two-way fixed effects estimators under heterogeneous treatment effects. The difference-in-differences model formula for the single-event case is expressed as follows: , In the formula, For unit loss indicators, The coefficient of variation between the treatment group and the control group is the intrinsic difference coefficient. This is the time effect coefficient. To handle group indicator variables, δ is a time discontinuity indicator variable, and δ is a difference-in-differences estimator that measures the additional loss reduction effect of intervention under disaster impact, reflecting the unit loss reduction of the high intervention group relative to the low intervention group under disaster impact. This is the random error term.

[0040] The marginal diminishing effect and the additional diminishing effect are calibrated for consistency and combined into a single monetized objective direct benefit. This consistency calibration aims to improve the robustness of benefit estimation by integrating the results of panel regression and causal inference methods. If the direction is consistent with δ and the difference between the two does not exceed the preset compatibility interval (e.g., the relative difference is less than 20%), then the inverse variance-weighted average of the two is taken as the objective direct benefit of monetization. , in, To monetize the objective and direct benefits, The total marginal reduction is calculated as follows: Total risk exposure is typically the sum of risk scores for all observations in the sample or the base for compensation. Additional reduction for difference-in-differences estimation, Weights are the inverse variance. These are the standard errors of the two estimators; if they are inconsistent, the difference-in-differences result takes precedence, and the fixed effects estimators are adjusted using a bias correction method before being re-merged. This candidate quantity is still defined as "reduction in compensation" and must be confirmed as a genuine reduction through mechanism verification before it can be included in the social benefits account.

[0041] S5: Based on the marginal reduction effect and the additional reduction in the intervention group, construct a causal regularization model. This model uses the intervention effect derived from causal inference as a priori constraint on the prediction model to enhance the causal consistency and interpretability of counterfactual predictions. The steps include: S51: Spatiotemporal samples with meteorological service intervention indicators below a preset low threshold are selected as the no-intervention sample set. The preset low threshold is determined using a data-driven approach and verified through sensitivity analysis to retain a pure control group that is almost unaffected by meteorological services, ensuring that the basis of counterfactual prediction is not contaminated by intervention. The input features are disaster intensity level, building structure type, target geographic topography code, regional GDP, and insurance penetration rate; the output label is the actual compensation amount. S52: In the loss function of LightGBM, a monotonic derivative penalty term based on the marginal reduction effect is added, and the additional reduction amount is constructed as an unbiased estimation constraint term, forming a causal regularized loss function, expressed by the formula: , in, For causal regularization loss function, The original loss function for LightGBM is set to Tweedie distribution to accommodate zero-payout heavy-tailed features. As a monotonicity penalty, the predicted value is forced to increase monotonically as the intervention level decreases; For unbiased constraints, the model's average prediction error on the control group should be equal to the actual loss reduction based on the reporting rate. ; The hyperparameter is selected through cross-validation; j is the index of the sample that satisfies the intervention degree being greater than 0. Let be the model's predicted value for sample j. The partial derivative of the predicted value with respect to the degree of intervention is forced to be non-positive, meaning that the higher the degree of intervention, the lower the predicted payout. This is the sample size for the control group. For the control group sample set, This represents the model's predicted payout for sample i in the control group. This represents the actual compensation amount for sample i. Here, the anchor point for the unbiased constraint term is the actual loss reduction amount calculated based on the difference in reporting rates between covered and uncovered targets in the disaster prevention and mitigation process. Instead of the DID estimate of total payout reduction, it avoids the spread of biases from underwriting screening, changes in the underwriting pool structure, and other effects to machine learning counterfactual models.

[0042] S53: A gradient boosting tree is trained based on a causal regularized loss function. For samples from periods of high intervention, the average of multiple random seed predictions is used as the counterfactual expected payout. For the high-intervention spatiotemporal unit k, the counterfactual expected payout is... ,in Let S be the LightGBM model trained using different random seeds for the s-th time, where S is the total number of random seeds (e.g., S=10). Let k be the feature vector of sample k; and then obtain the loss avoidance method. This represents the actual reduction in compensation due to the intervention of meteorological services, among which For counterfactual expectation of compensation, This represents the actual observed compensation amount for sample k.

[0043] A causal regularized model was trained using samples from periods with intervention levels below a preset threshold as a counterfactual payout predictor to predict the counterfactual expected payout during periods of high intervention and no service. S6: Perform parallel trend tests, point-in-time placebo tests, and mechanism tests on counterfactual expected payouts. After passing these tests, construct a dual-account accounting framework based on the candidate quantities of objective direct monetization benefits. Decompose and output the evaluation results according to account affiliation and insurance business process. Steps include: S61: Plot the mean loss curves of the treatment group and the control group in each period before the disaster tipping point, and test whether there is a significant difference in the pre-disaster change trend between the two groups; at the same time, construct a regression model that includes time indicator variables and group interaction terms, and perform a joint F test on the interaction term coefficients in each period before the tipping point. If the p value is greater than 0.05, the parallel trend hypothesis is satisfied.

[0044] S62: Placebo Test at Time Points and Random Permutation Test of Intervention Levels. For continuous intervention scenarios, instead of randomly shuffling treatment group labels, the following methods are used: First, a placebo test at time points: the treatment effect is re-estimated 2-3 years earlier than the policy time point. If the spurious treatment effect is not significant, the true treatment effect is supported. Second, a random permutation test of intervention levels: the intervention level index is randomly permuted among samples, and the difference-in-differences estimation is repeatedly performed to test whether the true estimate is at the tail of the random distribution, thus eliminating interference from random factors. Specifically, the true difference-in-differences estimate is denoted as δ, and then the policy time point is advanced more than 500 times, with each iteration yielding a spurious difference-in-differences estimate. p-value of placebo test If the p-value is less than 0.05, it indicates that the actual additional reduction effect is not driven by accidental factors; where B is the total number of advance timing points and b is the index of the random simulation.

[0045] After the verification is passed, based on the candidate quantities of the objective direct benefits of monetization, a dual-account accounting framework is constructed, and the total loss reduction benefits are decomposed according to account ownership and insurance business links, and the evaluation results are output. The steps include: S63: To distinguish between "true impairment" and "cleaning of the insurance pool," three tests are performed: First, the vulnerability of the insured asset, weighted by the insured amount, is incorporated as a time-varying covariate into the difference-in-differences model to control the impact of changes in the quality of the insured pool on the loss ratio. Secondly, the unit loss indicator is broken down into the reporting rate and the average claim payout per case. If the decrease in the claim payout rate is mainly due to the decrease in the reporting rate of the same type of insured in the same region, it is determined to be a real reduction in loss; if it is mainly due to the decrease in the proportion of policies in high-risk areas, it is determined to be a screening effect of the underwriting pool and should not be included in the social benefit account. Third, the treatment effect is re-estimated only on county-year subsamples where the insurance penetration rate is stable (i.e., within a continuous observation period in the target county (such as several years before and after a disaster), the insurance penetration rate fluctuates around its own mean within that window without showing a monotonically increasing or decreasing trend across periods). If the treatment effect is still significantly negative, the credibility of the conclusion on the actual loss reduction is greatly improved.

[0046] S64: Construct a dual-account accounting framework consisting of a social benefit account and an insurance industry account. Effects confirmed as genuine loss reduction through mechanism verification will be recorded in the social benefit account, while the net benefits of underwriting screening, actuarial pricing, operational efficiency improvement, and customer relationship enhancement will be recorded in the insurance industry account. The two accounts will be presented side by side and addition is prohibited.

[0047] The accounting standards for social benefit accounts and insurance industry accounts are as follows: Social benefit account Where ΔD is the actual loss amount, calculated based on the reporting rate; To calculate the indirect spillover effect, the industry spillover effect is obtained by injecting the net benefits of the insurance industry into the "Monetary and Financial Services and Other Financial Services" section of the input-output table. Incremental costs for loss prevention operations (such as surveying, reinforcement mobilization, and material allocation).

[0048] Insurance industry account I = I_underwriting screening + I_actuary pricing + I_operational efficiency improvement + I_net customer growth; It should be noted that: the actual reduction in insured amount ΔD×π is a subset of ΔD, not an independent increment. It also represents the amount less paid by the insurance company, but it cannot be counted repeatedly as an independent addition to the insurance industry account; the two accounts are listed side by side. If it is necessary to issue a total "social + industry" figure, the overlapping part must be explicitly removed; π is the insurance penetration rate.

[0049] S65: In the disaster prevention and mitigation process, the difference in reporting rates between covered and uncovered areas is calculated using the following formula: , in, Due to differences in reporting rates, , and These are the number of reported claims and the number of insurance policies, respectively. To improve the reporting rate of the target of the early warning, The reporting rate for uncovered targets, The number of cases reported that were not covered by the interim report. The number of policies with uncovered assets; ,in The average claim amount is ΔD, which represents the actual loss reduction and is recorded in the social benefit account. The grouping of covered or uncovered cases is based on the proactive loss prevention action trigger records of loss prevention inspection work orders or material allocation instructions within a preset time after the warning is issued.

[0050] S66: In the underwriting screening process, pair and compare the differences in claim rates and calculate the net benefit E of claims savings and premium waivers, expressed by the formula: , in, This refers to the number of policies that were refused coverage or had increased premiums due to weather conditions. The expected payout for such businesses, predicted using counterfactual analysis of matched samples, should be determined without intervention. For the average claim payout in normal underwriting business, This refers to the average premiums forgone. This step is recorded as a gain in the insurance industry account, but as a risk transfer in the social benefit account, and is explicitly marked as "negative externality: the payouts saved by the insurance company are borne by the policyholder or covered by government disaster relief," and is not included in the actual social loss reduction.

[0051] S67: Measuring the increase in premium adequacy after the introduction of weather factors in the actuarial pricing process. The formula is expressed as: , in, Loss and expense ratios for policies priced solely on weather factors. For the new loss rate, For the new expense ratio, For the old loss rate, The old rate applies; if the new rate applies... If the total ratio is closer to 100%, the adequacy is improved; the benefit of this step is reflected in the underwriting losses avoided due to the enhanced risk differentiation, which are included in the insurance industry account.

[0052] S68: Calculate the cost reduction in average case investigation time supported by forecasting in the operational efficiency improvement phase. The formula is as follows: , in The cost savings resulting from the reduction in the average case investigation time and The average investigation time (in hours) is calculated for cases without forecasts and with forecast support, respectively. The hourly labor cost per claims adjuster. The total number of cases is represented by TS, which is the time-to-delivery score. Under high intervention, the time-to-delivery score approaches 1, and the arrival time approaches 1. This step represents genuine resource savings, with real savings in labor costs. It can be recorded in both the social benefits account and the insurance industry account, but the reporting method for its presentation in both accounts must be specified in the report.

[0053] The effect of weather service outreach on renewal rates is evaluated in the customer enhancement process, and the formula is expressed as follows: , Customer groups are determined by whether they have received services such as alert push notifications, which are queried through intervention-related records. To increase renewal volume, To improve the renewal rate for customers who receive weather services, For customers who do not accept weather services, the renewal rate For the number of relevant insurance policies, For average premium, The life value multiplier for customers is usually the discount factor for expected future renewal annuities. To increase premium income, this stage is broken down into operational efficiency (E1) and revenue growth (E2). Specifically: E1 corresponds to cost savings resulting from reduced customer churn due to improved service reach, which can be included in the revenue account; E2 is the net renewal income after deducting corresponding claims costs and service costs. If claims costs and service costs cannot be reliably separated, it is listed separately as an operational indicator and not directly entered into the revenue account to avoid misclassifying revenue as revenue. E1 = Cost savings resulting from reduced customer churn due to improved service reach, included in the insurance industry account; .

[0054] S69: Generate a structured benefit assessment report categorized by dual accounts, business processes, and meteorological service products. Clarify that the social benefit account is for the government and meteorological authorities to demonstrate the rationality of public financial investment; and the insurance industry account is for insurance companies to demonstrate the feasibility of commercial payment.

[0055] In summary, the quantitative and objective meteorological service benefit evaluation method for the insurance industry provided by this invention establishes a spatiotemporally matched policy-level multidimensional panel data foundation. It extracts actual service reach data from four business process dimensions: proactive loss prevention, underwriting optimization, actuarial penetration, and resource timeliness. A continuous meteorological service intervention index is objectively synthesized using the entropy weight method, and exogenous components are extracted. This overcomes the bottleneck of traditional methods being limited by missing or poor-quality disaster loss data, ensuring that benefit evaluation is based on observable and traceable process evidence, thus guaranteeing the objectivity and operability of the evaluation from the source. By introducing the exogenous intervention component as a weighting factor into the Copula joint distribution model, the fitting effect of risk scores on extreme losses is corrected. Furthermore, a panel fixed effects model is used to estimate the marginal loss reduction effect, and a propensity score matched staggered difference-in-differences model is used to estimate the local average treatment effect. Inverse variance weighting is used for causal effect consistency calibration. Building upon this foundation, the obtained unbiased causal constraints are embedded into LightGBM to construct a causal-regularized counterfactual loss prediction model. This model accurately calculates expected losses under conditions without meteorological service intervention, transforming the assessment results from a qualitative assessment of "whether it is helpful or not" to a quantitative assessment of "how much loss is reduced." Combined with a point-in-time placebo test to verify statistical significance, this significantly improves the precision and credibility of the portrayal of insurance meteorological service effectiveness. By introducing a dual-account accounting framework, the total loss reduction and efficiency enhancement value is broken down and independently calculated at each stage. This fundamentally corrects the systematic underestimation caused by traditional methods focusing on direct losses while neglecting indirect value-added benefits, as well as the misleading practice of misrepresenting risk transfer as genuine loss reduction. It achieves comprehensive and verifiable monetization accounting covering the entire value chain from risk pricing to customer value enhancement, and can separately meet the decision-making needs of governments and insurance companies, providing a quantitative fulcrum for the closed-loop business of financial meteorology.

[0056] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A quantitative objective meteorological service benefit evaluation method for the insurance industry, characterized in that, include: S1: Collect meteorological monitoring data, insurance institution business operation logs and control variables, and align the policy location with meteorological grid points through a geographic information system to form a spatiotemporal matching dataset; S2: Based on the spatiotemporal matching dataset, quantify and generate meteorological service intervention indexes for each spatiotemporal unit, and extract their exogenous components; S3: Using the exogenous components as weights, the marginal distribution estimation of disaster intensity and insurance loss sequence and the Copula joint distribution construction process are weighted and corrected to generate a conditional probability loss matrix and assign expected loss risk scores to each spatiotemporal unit. S4: Input the expected loss risk score and meteorological service intervention index into a preset panel fixed effects model to estimate the marginal loss reduction effect; take regional sudden meteorological disasters as the breakpoint and use an alternating difference-in-differences model based on propensity score to estimate the additional loss reduction of the intervention group; perform consistency calibration on the marginal loss reduction effect and the additional loss reduction, and merge them into a candidate quantity of monetizable objective direct benefits; S5: Based on the marginal loss reduction effect and the additional loss reduction of the intervention group, construct a causal regularization model; train the causal regularization model as a counterfactual payout predictor using samples from periods with intervention levels below a preset threshold to predict the counterfactual expected payout in the absence of service during periods of high intervention. S6: Perform a test on the counterfactual expected compensation. If the test is passed, construct a dual-account accounting framework for the social benefit account and the insurance industry account based on the candidate quantity of the monetized objective direct benefits. Decompose the evaluation results according to the account affiliation and insurance business links, and output the evaluation results.

2. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S1, which involves aligning the policy location with the meteorological grid space, includes: S11: Convert the geographic coordinates of the policy target into a point vector layer, perform spatial intersection calculation with the meteorological grid layer, so that each policy target obtains its own meteorological grid number; S12: Perform nearest neighbor matching on disaster events and insurance business records based on timestamps to form a standard data structure; S13: Perform outlier removal and multiple imputation of missing values ​​on the matched dataset, and output the spatiotemporal matching dataset.

3. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S2, the steps for quantifying the meteorological service intervention index, include: S21: Based on the system instruction change record, calculate the trigger ratio of loss prevention inspection work orders or material allocation instructions within a preset time after the early warning is issued, and generate an active loss prevention action index. S22: Based on the statistics of underwriting rule trigger records, the proportion of policies whose underwriting conclusions were changed due to the meteorological risk field is generated as an underwriting intervention sub-dimension; S23: Based on the pricing factor loading record statistical quotation system, retrieve the call rate of meteorological risk factors to generate the actuarial pricing penetration rate; S24: Based on resource pre-deployment records, obtain the timeliness score of claims personnel and rescue vehicles arriving after the warning; S25: The proactive loss prevention action index, underwriting intervention sub-dimension, actuarial pricing penetration rate and timeliness score are weighted and condensed using the entropy weight method to generate a meteorological service intervention degree index that is continuously distributed in the [0,1] interval; S26: Using meteorological supply-side variables as exogenous tools, extract the exogenous components of intervention degree through instrumental variable method or orthogonalization to eliminate reverse causal influence; the exogenous components are preferentially used in weighting, matching and causal identification.

4. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S3, the weighted correction steps include: S31: Fit a generalized extreme value distribution to the meteorological disaster intensity sequence, and a gamma distribution or a log-normal distribution to the insurance loss sequence. When estimating the distribution parameters using the maximum likelihood method, introduce sample weights constructed from the exogenous component of the meteorological service intervention index into the likelihood function. S32: Determine the goodness of fit using a weighted KS test, and select the optimal marginal distribution that passes the significance test; S33: When selecting the optimal Copula, the empirical Copula adjusted by the exogenous component of the intervention degree is used as the fitting target to generate the conditional probability loss matrix.

5. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 4, characterized in that, In step S3, the formula for the conditional probability loss matrix is ​​expressed as follows: ， In the formula, M is the disaster loss matrix. The level of disaster intensity. According to the level of insured loss, In order to achieve a disaster intensity of Under the conditions, loss level occurs The probability of.

6. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, In step S4, the formula for the panel fixed effects model is expressed as follows: ， In the formula, The earned loss ratio or combined ratio, where α is a constant term; The expected loss risk score, for The main effect coefficient, As an indicator of meteorological service intervention, for The main effect coefficient, For interactive items, capture how the degree of intervention plays a mitigating role as the risk level changes; This refers to the interaction term coefficient, i.e., the marginal loss effect; For individual fixed effects, For time-fixed effects, This is the kth control variable; for coefficient, For random error term, i is the cross-sectional individual identifier, t is the time identifier, and k is the control variable number.

7. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S4, which involves estimating the additional impairment in the intervention group using a difference-in-differences model based on propensity score, includes: S41: Individuals whose meteorological service intervention index is higher than the first threshold are treated as the treatment group, and individuals whose index is lower than the second threshold are treated as the control group; the first threshold and the second threshold are determined by data-driven quantile method or Youden index, and sensitivity analysis is performed on the thresholds; S42: Estimate the propensity score using covariates, perform caliper nearest neighbor matching, and generate a balanced panel. S43: Using a single, highly disastrous meteorological event as the time breakpoint, a time breakpoint indicator variable is defined. For cases involving multiple overlapping regions, a difference-in-differences model is established using the Callaway-Sant'Anna or Sun-Abraham estimators, expressed as follows: ， In the formula, For unit loss indicators, The coefficient of variation between the treatment group and the control group is the intrinsic difference coefficient. This is the time effect coefficient. To handle group indicator variables, δ is a time discontinuity indicator variable, and δ is a difference-in-differences estimator that measures the additional damage reduction effect of intervention under disaster impact; This is the random error term.

8. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S5, the steps for constructing the causal regularization model include: S51: Select spatiotemporal samples with meteorological service intervention indicators below the preset low threshold as the non-intervention sample set, with disaster intensity level, building structure type, target geomorphological code, regional GDP and insurance penetration rate as input features, and actual compensation amount as output label; S52: In the loss function of LightGBM, a monotonic derivative penalty term based on the marginal loss reduction effect is added, and the additional loss reduction is constructed as an unbiased estimation constraint term to form a causal regularized loss function; the unbiased estimation constraint term is anchored to the loss reduction amount based on the reporting rate of the actual loss reduction sub-stage, rather than the total loss reduction amount; S53: Train a gradient boosting tree based on the causal regularization loss function. For samples in high intervention periods, take the average of multiple random seed predictions as the counterfactual expected compensation.

9. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, In step S6, the verification and accounting steps include: S61: Plot the mean loss curves of the treatment group and the control group in each period before the disaster tipping point, and test whether there is a significant difference in the pre-disaster change trend between the two groups; S62: Placebo test at the implementation point, advance the policy time point by 2-3 years, and supplement it with random permutation of the intervention degree, repeatedly perform difference-in-differences estimation, and test whether the true estimated value is in the tail of the random distribution to eliminate the interference of accidental factors; S63: Execution mechanism verification: The vulnerability of the insured amount-weighted object is incorporated as a time-varying covariate into the difference-in-differences model to control for changes in the quality of the underwriting pool; the unit loss indicator is decomposed into the reporting rate and average claim payout, distinguishing between the actual loss reduction and the underwriting pool screening effect; the robustness of the treatment effect is verified by re-estimating on a county-year subsample with stable insurance penetration; the execution effect is decomposed using the aforementioned candidate quantity of objective direct monetization benefits as the total reconciliation amount. S64: Based on the effect decomposition results, construct a dual-account accounting framework of social benefit account and insurance industry account. Effects confirmed as real loss reduction through mechanism verification will be included in the social benefit account, and the net benefits of underwriting screening, actuarial pricing, operational efficiency improvement and customer relationship enhancement will be included in the insurance industry account. The two accounts will be listed side by side and addition is prohibited.

10. The quantitative objective meteorological service benefit evaluation method for the insurance industry according to claim 1, characterized in that, Step S6, the steps for outputting the evaluation results, include: S65: In the disaster prevention and mitigation process, calculate the difference in reporting rates between covered and uncovered areas, and include it in the social benefit account as the actual loss reduction; S66: In the underwriting screening process, the difference in loss rates is compared and the net benefit of claims savings and premium waivers is calculated. This process is recorded as a gain in the insurance industry account and as a risk transfer with negative externalities in the social benefit account. S67: Measure the increase in premium adequacy after the introduction of weather factors in the actuarial pricing process and include it in the insurance industry account; S68: In the operational efficiency improvement stage, the cost of reducing the average investigation time per case is calculated and supported by forecasts. This is recorded as a real resource saving and is included in both the social benefit account and the insurance industry account. In the customer enhancement stage, the renewal improvement is broken down into net operational efficiency improvement and renewal income. Renewal income is listed as a separate operating indicator and is not directly included in the benefit account. S69: Generate a structured benefit assessment report categorized by dual accounts, business processes, and meteorological service products.

Images