Processing method and device for expected credit loss of credit assets, equipment and medium

Abstract

The invention relates to a processing method and device for expected credit loss of credit assets, equipment and a medium. The processing method for the expected credit loss of the credit assets comprises the following steps: acquiring multi-modal data of a target user in the aspect of the credit assets; inputting the multi-modal data into a preset expected credit loss prediction model; performing fusion processing and prediction on the multi-modal data by using the expected credit loss prediction model to obtain a prediction result; determining an expected credit loss measurement result based on the prediction result; and performing accounting processing based on the expected credit loss measurement result to obtain a target audit track record corresponding to the target user. The accuracy of risk prediction can be improved, and then the accuracy of credit loss calculation and extraction is improved.

Description

Technical Field

[0001] This application relates to the fields of financial technology and intelligent risk management, and in particular to a method, apparatus, equipment and medium for processing expected credit losses of credit assets. Background Technology

[0002] Currently, when implementing International Financial Reporting Standard 9 (IFRS 9) for the measurement and accounting of expected credit losses (ECL), financial institutions mainly rely on data integration solutions based on traditional ETL (Extract, Transform, Load).

[0003] Traditional ETL data integration solutions mainly use ETL tools such as DataStage and Informatica to periodically extract, transform, and load structured data from various systems for ECL metering.

[0004] However, traditional ETL tools mainly rely on preset rules to perform data transformation operations, such as field mapping, type conversion, and aggregation calculations. They cannot perform deep semantic analysis of the data content, resulting in the neglect of key risk information. Furthermore, since traditional technologies usually only process structured data, risk features are not fully extracted, ultimately leading to relatively low accuracy in credit loss provisioning. Summary of the Invention

[0005] This application provides a method, apparatus, device, and medium for processing expected credit losses of credit assets, aiming to solve the problem that the accuracy of credit loss provision is relatively low due to the traditional use of ETL tools to periodically extract, transform, and load structured data from various systems for ECL measurement.

[0006] In a first aspect, embodiments of this application provide a method for processing expected credit losses on credit assets, the method comprising: Acquire multimodal data of target users regarding credit assets, wherein the multimodal data includes structured data, time-series data, text data, and graph data; The multimodal data is input into a preset expected credit loss prediction model; The multimodal data is fused and predicted using the expected credit loss prediction model to obtain prediction results, which include the probability of default, the loss rate of default, and the default risk value. Based on the prediction results, the expected credit loss measurement results are determined; Accounting processing is performed based on the expected credit loss measurement results to obtain the target audit trajectory record corresponding to the target user.

[0007] The further technical solution is that the expected credit loss prediction model includes a feature extraction layer, a feature fusion layer, and a prediction layer; The process of fusing and predicting the multimodal data using the expected credit loss prediction model includes: The multimodal data is preprocessed to obtain preprocessed multimodal data; The preprocessed multimodal data is input into the feature extraction layer; The feature extraction layer is used to extract features from the preprocessed multimodal data to obtain multimodal features; The multimodal features are then subjected to multimodal encoding to obtain a multimodal feature encoding vector; The multimodal feature encoding vector is input into the feature fusion layer for fusion processing to obtain fused features; The fused features are input into the prediction layer to obtain the prediction result.

[0008] A further technical solution is that the multimodal features include basic features, statistical features, temporal features, textual features, and cross features. The feature extraction layer is used to extract features from the preprocessed multimodal data to obtain multimodal features, including: Extract basic features from the structured data; Statistical and temporal features are extracted from the time-series data; Extract text features from the text data; Extract the cross features from the graph data.

[0009] A further technical solution is that the multimodal feature encoding vector is input into the feature fusion layer for fusion processing to obtain fused features, including: The multimodal feature encoding vector is input into the feature fusion layer. Self-attention is used to calculate the attention weights between features within the same modality, cross-attention is used to calculate the cross-attention weights between different modalities, and multi-head attention is used to obtain feature relationships at different levels. Based on the attention weights between features within the same modality, the cross-attention weights between different modalities, and the feature relationships, the multimodal feature encoding vectors are fused to obtain fused features.

[0010] A further technical solution is that the accounting treatment based on the expected credit loss measurement results includes: The expected credit loss measurement results are matched with accounting entry templates to obtain matching results; Based on the matching results, the journal entry parameters are automatically filled in to obtain the filling results; The filling results are verified using preset accounting rules; If the verification is successful, automatic posting, impairment and financial statement updates will be performed to obtain the target audit trail record corresponding to the target user.

[0011] A further technical solution is that, before performing accounting entry template matching on the expected credit loss measurement results, the method further includes: Credit risk stage identification is performed on the expected credit loss measurement results; If the expected credit loss measurement result is identified as belonging to the low-risk stage, then accounting treatment shall be performed based on the expected credit loss over 12 months. If the expected credit loss measurement result is identified as belonging to the medium-risk stage, then accounting provisions shall be made in accordance with lifetime expected credit loss. If the expected credit loss measurement results are identified as belonging to a high-risk stage, then accounting provisions will be made for lifetime expected credit losses and interest will be suspended.

[0012] A further technical solution is that, after determining the expected credit loss measurement result, the method further includes: Based on preset economic indicators and scenario probability allocation values, determine macroeconomic scenario information; Based on preset adjustment factors, preset adjustment ranges, and preset adjustment rules, forward-looking adjustment parameters are determined, wherein the preset adjustment factors include macroeconomic factors, industry factors, regional factors, product factors, and time factors. Based on the macro-scenario information and the forward-looking adjustment parameters, a preset dynamic update mechanism is used to dynamically adjust the expected credit loss measurement results.

[0013] Secondly, embodiments of this application also provide an apparatus for processing expected credit losses of credit assets, which includes a unit for performing the above-described method.

[0014] Thirdly, embodiments of this application also provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0015] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the above-described method.

[0016] This application provides a method, apparatus, device, and medium for processing expected credit losses of credit assets. The method includes: acquiring multimodal data of a target user regarding credit assets; inputting the multimodal data into a preset expected credit loss prediction model; using the expected credit loss prediction model to fuse and predict the multimodal data to obtain a prediction result; determining an expected credit loss measurement result based on the prediction result; and performing accounting processing based on the expected credit loss measurement result to obtain a target audit trajectory record corresponding to the target user.

[0017] This application's embodiments introduce multimodal data. Since multimodal data includes not only structured data but also time-series data, text data, and graph data, it allows for comprehensive fusion analysis based on diverse data, avoiding misjudgments of risk due to missing information and improving the accuracy of expected credit loss prediction. Furthermore, by using a pre-set expected credit loss prediction model for fusion analysis and prediction, the default risk value can be directly predicted, improving the accuracy of risk prediction and thus improving the accuracy of credit loss provision. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0021] Figure 1 A flowchart illustrating the first embodiment of a method for handling expected credit losses of credit assets provided in this application; Figure 2 A schematic diagram of the deep learning ECL econometric model architecture provided in this application; Figure 3 A schematic diagram of the overall architecture of the system for handling expected credit losses of credit assets provided for this application; Figure 4 The system deployment architecture diagram provided for this application; Figure 5This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0025] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0028] To address the aforementioned issues, this application provides a method for handling expected credit losses on credit assets, which can improve the accuracy of credit loss provision.

[0029] See Figure 1 , Figure 1 The flowchart of a first embodiment of a method for handling expected credit losses of credit assets provided in this application is shown. The method for handling expected credit losses of credit assets includes the following steps: Step 110: Obtain multimodal data on the target user's credit assets.

[0030] The multimodal data includes structured data, time-series data, text data, and graph data.

[0031] For example, structured data can be customer basic information, product information, financial data, etc.; time-series data can be historical repayment records, account change trajectories, etc.; text data can be customer reviews, news and public opinion, policy documents, etc.; graph data can be customer relationship networks, guarantee relationship diagrams, etc.

[0032] In some embodiments, PostgreSQL can be used to store structured business data, InfluxDB to store time-series data, Neo4j to store customer relationship network data, MongoDB to store unstructured data, and Redis to provide high-performance caching services.

[0033] In some embodiments, the data storage process can perform horizontal data sharding by customer ID hash; vertical table splitting by business module; sharding historical data by time dimension; and support dynamic expansion using consistent hashing; thus, intelligent sharding routing and load balancing can be achieved.

[0034] In some embodiments, real-time data can be integrated and processed, wherein Debezium can be used to capture database changes; Kafka can be used for data stream buffering and distribution; Flink can be used for real-time data processing and transformation; real-time data synchronization across systems can be achieved; and distributed transactions can be used to ensure data consistency.

[0035] Step 120: Input the multimodal data into the preset expected credit loss prediction model.

[0036] Step 130: Use the expected credit loss prediction model to fuse and predict the multimodal data to obtain the prediction results.

[0037] The prediction results include the probability of default, the loss rate due to default, and the risk value of default.

[0038] Step 140: Based on the prediction results, determine the expected credit loss measurement results.

[0039] Step 150: Perform accounting processing based on the expected credit loss measurement results to obtain the target audit trajectory record corresponding to the target user.

[0040] This embodiment introduces multimodal data. Since multimodal data includes not only structured data but also time-series data, text data, and graph data, it can perform comprehensive fusion analysis based on diverse data, avoiding risk misjudgment due to missing information and improving the accuracy of expected credit loss prediction. Furthermore, by performing fusion analysis and prediction through a preset expected credit loss prediction model, it can directly predict the default risk value, improve the accuracy of risk prediction, and thus improve the accuracy of credit loss provision.

[0041] See Figure 2 In some possible implementations, the expected credit loss prediction model includes a feature extraction layer, a feature fusion layer, and a prediction layer; Step 130, namely, using the expected credit loss prediction model to fuse and predict the multimodal data, includes: Step 131: Preprocess the multimodal data to obtain preprocessed multimodal data.

[0042] Preprocessing may include the following steps: 1) Data source identification: Comprehensively sort out and identify all relevant data sources, including internal business systems (such as credit management system, customer relationship management system, financial accounting system) and external data sources (such as central bank credit reporting system, third-party big data platform, macroeconomic database, etc.).

[0043] 2) Data extraction: Adopting Change Data Capture (CDC) technology, we monitor and capture data changes in the business system in real time (such as new loans, repayment records, and customer information updates) to ensure the timeliness and continuity of data acquisition and avoid the delays caused by traditional batch processing mode.

[0044] 3) Data cleaning: Perform quality control on the extracted raw data, including removing duplicate records, handling missing values ​​(such as filling, interpolation or labeling), and identifying and correcting outliers to ensure the accuracy and reliability of the data used in subsequent analysis.

[0045] 4) Data standardization: unify the format and specifications of data from different sources, such as unifying the time format (e.g., ISO8601), the amount precision (e.g., retaining two decimal places), and the coding specifications (e.g., customer classification, industry codes), to eliminate data ambiguity caused by inconsistent standards.

[0046] 5) Data association: Establish the association between customers, products, and transactions based on key identifiers (such as customer ID, product number, and transaction serial number).

[0047] In some embodiments, the preprocessing stage can also control the quality of the data, such as performing integrity checks, accuracy verification, consistency checks, timeliness monitoring, and outlier detection.

[0048] Integrity checks can involve monitoring the missing rate of key fields such as customer identity, loan amount, and repayment status, for example, by setting a threshold of 5%. When the missing rate exceeds this threshold, the system automatically triggers an alarm and initiates a data completion or marking process.

[0049] Accuracy verification can be based on business logic rules to validate data and ensure it conforms to real-world scenarios. For example, it can verify that "the loan amount should not be negative" and "the repayment date should not be earlier than the loan disbursement date," thus promptly identifying and correcting logical errors.

[0050] Consistency checks can involve cross-system data comparisons, such as verifying whether the balance of the same loan is consistent between the credit system and the financial system. When inconsistencies are found, the system automatically initiates reconciliation and repair mechanisms to ensure data consistency and synchronization across different systems.

[0051] Timeliness monitoring involves real-time monitoring of the update frequency and transmission latency of various data sources to ensure data arrives on time. Alarms are issued for data streams with latency exceeding preset thresholds to guarantee the real-time requirements of ECL metering.

[0052] Outlier detection can employ unsupervised machine learning algorithms such as Isolation Forest to automatically identify data points that deviate from normal patterns (such as abnormally high transactions or sudden changes in customer debt), preventing extreme values ​​from interfering with model training and prediction results.

[0053] Step 132: Input the preprocessed multimodal data into the feature extraction layer.

[0054] Step 133: Use the feature extraction layer to extract features from the preprocessed multimodal data to obtain multimodal features.

[0055] Step 134: Perform multimodal encoding on the multimodal features to obtain a multimodal feature encoding vector.

[0056] Specifically, multimodal features can be input into a multimodal encoder for multimodal encoding. Specifically, fully connected layers and batch normalization can be used to process numerical features to encode structured data; LSTM or GRU can be used to process time series data to encode sequence data; pre-trained BERT models can be used to extract textual semantic features to encode textual data; and GraphSAGE can be used to process customer relationship network data to encode graph data. Then, the encoded features from different modalities are mapped to a unified dimensional space for feature dimension alignment.

[0057] Step 135: Input the multimodal feature encoding vector into the feature fusion layer for fusion processing to obtain fused features.

[0058] Step 136: Input the fused features into the prediction layer to obtain the prediction result.

[0059] In some possible implementations, the multimodal data includes structured data, time-series data, text data, and graph data, and the multimodal features include basic features, statistical features, temporal features, textual features, and cross features. Step 133, namely, extracting features from the preprocessed multimodal data using the feature extraction layer to obtain multimodal features, includes: 1) Extract basic features from the structured data; The basic characteristics may include customer age, income, debt ratio, product term, interest rate, etc.

[0060] 2) Extract statistical and temporal features from the time-series data; Among them, statistical characteristics may include the mean, variance, and maximum number of overdue days of historical repayments; time characteristics may include seasonality, trend, and periodicity.

[0061] 3) Extract text features from the text data; Textual features can include customer reviews, semantic features of news, etc., which can be extracted using the BERT model.

[0062] 4) Extract the cross features from the graph data.

[0063] Among them, cross features can include combinations of features such as customer level × product type, revenue × debt ratio, etc.

[0064] In some possible implementations, step 135, namely, inputting the multimodal feature encoding vector into the feature fusion layer for fusion processing to obtain fused features, includes: Step 11: Input the multimodal feature encoding vector into the feature fusion layer, use self-attention to calculate the attention weights between features within the same modality, use cross-attention to calculate the cross-attention weights between different modalities, and use multi-head attention to obtain feature relationships at different levels; Step 12: Based on the attention weights between features within the same modality, the cross-attention weights between different modalities, and the feature relationships, the multimodal feature encoding vectors are fused to obtain fused features.

[0065] For steps 11 and 12, to achieve deep interaction and effective fusion of multimodal features, an attention fusion mechanism based on the Transformer architecture can be adopted, which specifically includes the following core components: Self-Attention: Within each modality, the correlation weights between feature vectors are calculated through a self-attention mechanism, enabling the model to focus on more discriminative key features within the same modality and enhancing the expressive power of internal features.

[0066] Cross-Attention: This mechanism establishes interactions between different modalities, allowing features from one modality to "pay attention to" and integrate relevant information from other modalities. For example, it leverages textual semantic information to enhance the representation of structured data, achieving cross-modal semantic alignment and information complementarity.

[0067] Multi-Head Attention: This method employs a multi-head mechanism to perform multiple self-attention or cross-attention computations in parallel, with each "head" independently learning feature relationships in different subspaces. By fusing the outputs of multiple heads, the model can simultaneously capture diverse dependencies, both local and global, explicit and implicit, thereby enhancing the richness and robustness of feature fusion.

[0068] Positional encoding: For sequential data (such as time series and text), positional encoding information is introduced to provide the model with the sequential relationship of elements. This mechanism compensates for the insensitivity of attention mechanisms to position, ensuring that the model can effectively utilize temporal or word order information.

[0069] Residual Connections: Residual connections are introduced after the attention computation at each layer, directly passing the input information to the output. This design effectively alleviates the vanishing gradient problem in deep networks, improving the training stability and convergence speed of the model.

[0070] This fusion mechanism achieves efficient integration of multi-source features, including structured, temporal, textual, and graph features, through multi-level and multi-dimensional attentional interactions, providing high-quality joint representations for subsequent accurate predictions of PD, LGD, and EAD.

[0071] In some possible implementations, step 150, namely the accounting treatment based on the expected credit loss measurement results, includes: Step 151: Match the expected credit loss measurement results with accounting entry templates to obtain matching results.

[0072] Step 152: Automatically fill in the entry parameters based on the matching results to obtain the filled results.

[0073] Step 153: Verify the filling result using preset accounting rules.

[0074] Step 154: If the verification is successful, perform automatic posting, update impairment and financial statements, and obtain the target audit trail record corresponding to the target user.

[0075] For steps 151-154, in some embodiments, an intelligent accounting processing algorithm may be used, which includes accounting rule parsing, journal entry generation processing, and automatic posting processing.

[0076] In this process, accounting rule parsing can involve first building a rule knowledge base and then setting up a rule matching algorithm. Specifically, the construction of the rule knowledge base can be found below: 1) Accounting Standards Analysis: Analysis of relevant clauses of IFRS 9 and Enterprise Accounting Standards. 2) Business rule extraction: Extract accounting processing rules from business processes; 3) Rule structuring: Converting natural language rules into structured representations; 4) Rule Relationship Modeling: Constructing dependency and conflict relationships between rules; 5) Knowledge graph construction: Use Neo4j to build an accounting rules knowledge graph.

[0077] The rule matching algorithm can include semantic vectorization, similarity calculation, rule ranking, confidence assessment, and multi-rule fusion. Semantic vectorization can be performed by using the BERT model to convert rules into semantic vectors. Similarity calculation can be performed by calculating the semantic similarity between the business scenario and the rule. Rule ranking can be performed by ranking the matching rules according to their similarity. Confidence assessment can be performed by evaluating the confidence score of rule matching. Multi-rule fusion can be performed by handling the case where multiple rules are matched at the same time.

[0078] In some embodiments, rule conflicts can be handled during rule matching. For example, logical reasoning can be used to detect conflicts between rules; priorities can be determined based on the source and importance of rules; expert systems can be used to resolve rule conflicts; and the logical consistency of solutions can be verified. Audit logs can also record the conflict resolution process and decision-making basis.

[0079] In some embodiments, the entry generation process may include template matching, automatic parameter filling, and entry verification checks.

[0080] Template matching can include the following: 1) Business Scenario Identification: Identify specific business scenarios based on ECL results; 2) Template library retrieval: Retrieves matching templates from a predefined template library; 3) Fuzzy matching: Fuzzy template matching is performed using edit distance; 4) Template scoring: Scoring based on relevance and historical usage performance; 5) Optimal template selection: Select the accounting entry template with the highest score.

[0081] The automatic parameter filling can include the following: 1) Data mapping: Establishing a mapping relationship between ECL results and journal entry parameters; 2) Calculation logic: Implements complex amount calculation and allocation logic; 3) Account Determination: Accounting accounts are automatically determined based on business rules; 4) Auxiliary accounting: Automatically fills in auxiliary accounting items such as customers and products; 5) Parameter verification: Verify the rationality and completeness of the filled parameters.

[0082] In some possible embodiments, journal entry verification checks may include debit and credit balance checks, account compliance checks, amount reasonableness checks, business logic checks, and approval process checks. Specifically, the debit and credit balance check verifies whether the debit and credit amounts are balanced; the account compliance check verifies whether the use of accounting accounts complies with regulations; the amount reasonableness check verifies whether the amount is within a reasonable range; the business logic check mainly verifies whether the journal entry conforms to business logic; and the approval process check mainly determines whether manual approval is required.

[0083] In some possible embodiments, automated posting processing includes pre-posting checks, batch posting, and post-posting verification.

[0084] Pre-posting checks may include system status checks, permission verification, period checks, data integrity checks, and backup preparation.

[0085] For example, system status checks can check whether the accounting system is in a posting-ready state; permission verification mainly verifies whether the system has posting permissions; period checks can verify whether the posting period is correct and open.

[0086] In this way, data integrity checks can ensure that journal entries are complete and error-free.

[0087] In some embodiments, batch posting can be achieved by grouping entries according to posting rules; by using multi-threading to process different groups of entries in parallel; by using database transactions to ensure the atomicity of posting; by performing error handling, such as ensuring that the failure of a single entry does not affect the posting of other entries; and by monitoring the posting progress and success rate in real time.

[0088] In some embodiments, post-posting verification may include the following process: 1) Balance Verification: Verify the accuracy of the account balance after posting; 2) Report consistency: Check the consistency between the general ledger and the subsidiary ledgers; 3) Audit trail: Records the complete posting audit trail; 4) Anomaly Alarm: An alarm will be issued promptly upon detecting any abnormal situation; 5) Rollback mechanism: Provides a rollback mechanism for failed postings.

[0089] In some embodiments, real-time risk monitoring algorithms can be used to monitor changes in the risk of credit assets in real time, dynamically adjust ECL measurement results, and provide timely warnings of risk anomalies.

[0090] The real-time risk monitoring algorithm can include operations such as real-time data monitoring, anomaly detection, and intelligent alarms.

[0091] Real-time data monitoring can be achieved through data stream processing architecture, risk indicator calculation, and dynamic adjustment of thresholds.

[0092] Specifically, the data stream processing architecture can involve real-time access to business systems, market data, and external data; using Kafka as a buffer and distribution mechanism for the data stream; using Flink for real-time data processing and computation; using RocksDB to manage the state information of the stream processing; and employing fault tolerance mechanisms to implement checkpoints and fault recovery mechanisms.

[0093] Risk indicators can be calculated by updating default probabilities in real time based on the latest data; monitoring the overall risk level of the credit portfolio; monitoring customer, industry, and regional concentration risks; monitoring asset liquidity and maturity mismatch risks; and monitoring changes in market factors such as interest rates and exchange rates.

[0094] Dynamic threshold adjustment can be achieved by calculating dynamic thresholds based on historical data; adjusting thresholds considering seasonal business characteristics; adjusting thresholds based on the macroeconomic environment; using online learning algorithms to adaptively adjust thresholds; or setting upper and lower limits for thresholds based on expert experience.

[0095] In some embodiments, anomaly detection may include statistical anomaly detection, machine learning anomaly detection, and temporal anomaly detection.

[0096] Among these methods, statistical anomaly detection can include using three times the standard deviation to detect numerical anomalies; using quartiles to detect outliers; calculating standardized scores to detect anomalies; using Mahalanobis distance to detect multidimensional anomalies; and using statistical tests such as t-tests and chi-square tests.

[0097] Machine learning anomaly detection can be achieved by using IsolationForest to detect anomalous samples; using One-ClassSVM to identify anomalous patterns; using deep autoencoders to detect reconstruction anomalies; and using LSTM to detect temporal anomalous patterns.

[0098] Thus, the accuracy of detection is improved by integrating multiple algorithms.

[0099] Time series anomaly detection can include using the CUSUM algorithm to detect time series turning points; detecting abnormal trend changes in time series data; detecting abnormal changes in periodic patterns; detecting abnormal deviations from seasonal patterns; and detecting abnormal deviations based on time series prediction.

[0100] In some embodiments, intelligent alerting includes an alert rule engine, intelligent alert filtering, and multi-channel notification.

[0101] The alarm rule engine allows for flexible alarm rule configuration; supports complex alarm rules with multiple conditions; allows setting different alarm priorities; avoids duplicate alarms and alarm storms; and supports an automatic alarm escalation mechanism.

[0102] Intelligent alarm filtering can use machine learning to filter noisy alarms; analyze the correlation between alarms; automatically analyze the root cause of alarms; assess the impact of alarms on business; and provide handling suggestions based on historical experience.

[0103] Multi-channel notifications can include email notifications, SMS notifications, instant messaging notifications (such as DingTalk and WeChat Work), voice notifications, and mobile push notifications.

[0104] In some possible implementations, a forward-looking ECL adjustment algorithm can be used to dynamically adjust the ECL measurement results based on macroeconomic scenario information and forward-looking information, thereby improving forecast accuracy. Specifically, refer to steps 21-23. After step 140, i.e., determining the expected credit loss measurement results, the method further includes: Step 21: Determine macroeconomic scenario information based on preset economic indicators and scenario probability allocation values.

[0105] The preset economic indicators can include macroeconomic indicators (such as GDP growth rate, inflation rate, unemployment rate, and interest rate level), industry indicators (such as industry prosperity index, capacity utilization rate, and profitability), financial market indicators (such as stock index, bond yield, exchange rate, and volatility), policy indicators (such as changes in monetary policy, fiscal policy, and regulatory policy), and external shock indicators (such as natural disasters, geopolitical events, and sudden events such as the epidemic).

[0106] Scenario probability allocation can be done by assigning 50% probability to the most likely baseline scenario, 25% probability to the optimistic scenario of economic improvement, and 25% probability to the pessimistic scenario of economic downturn. Furthermore, the probabilities can be dynamically adjusted based on the latest economic data, and extreme scenarios can be designed for stress testing.

[0107] In some embodiments, scenario generation algorithms can be used to generate scenarios, and then scenario probabilities can be assigned. These scenario generation algorithms can include: generating multiple economic scenarios using Monte Carlo methods; predicting macroeconomic variables using vector autoregression models; simulating economic state transitions using Markov chains; setting extreme scenarios by incorporating expert experience; and constructing reference scenarios based on historical economic cycles.

[0108] Step 22: Based on preset adjustment factors, preset adjustment ranges, and preset adjustment rules, determine forward-looking adjustment parameters, wherein the preset adjustment factors include macroeconomic factors, industry factors, regional factors, product factors, and time factors.

[0109] For example, macroeconomic factors mainly calculate the impact coefficient of macroeconomic variables on the default rate; industry factors mainly analyze the sensitivity of different industries to the economic cycle; regional factors mainly consider the differences in economic development in different regions; product factors mainly analyze the risk characteristics of different product types; and time factors mainly consider the time lag effect of economic impact.

[0110] The preset adjustment range can be determined through historical backtesting, expert judgment, peer benchmarking, regulatory guidance, and model validation.

[0111] The preset adjustment rules can be tiered adjustments, such as adjustments based on customer level or product type; they can use smoothing algorithms to avoid overly drastic adjustments; or they can set upper and lower limits for the adjustment range.

[0112] In addition, consistency checks can be used to ensure the logical consistency of the adjusted results, and approval processes can be set up for major adjustments.

[0113] Step 23: Based on the macro-scenario information and the forward-looking adjustment parameters, dynamically adjust the expected credit loss measurement results using a preset dynamic update mechanism.

[0114] The preset dynamic update mechanism can include periodic updates (such as monthly updates to proactively adjust parameters), threshold triggers (such as triggering updates when key indicators exceed thresholds), event triggers (such as triggering updates when major economic events occur), model drift (such as triggering updates when model performance is detected to decline), or updates as required by regulatory agencies.

[0115] In some embodiments, incremental updates (e.g., updating only the changed parameters), full updates (e.g.), version control (e.g., maintaining a version history of parameters), and rollback mechanisms (e.g., supporting rollback operations for parameter updates) can be adopted.

[0116] In some embodiments, an intelligent impairment provision adjustment algorithm may be used to intelligently adjust the amount of impairment provision based on ECL measurement results and accounting standards requirements, thereby ensuring the accuracy of financial statements.

[0117] In some possible implementations, before performing accounting entry template matching on the expected credit loss measurement results, the method further includes: identifying the credit risk stage of the expected credit loss measurement results.

[0118] If the expected credit loss measurement result is identified as belonging to the low-risk stage, then the accounting treatment is carried out according to the expected credit loss of 12 months. If the expected credit loss measurement result is identified as belonging to the medium-risk stage, then accounting provisions shall be made in accordance with lifetime expected credit loss. If the expected credit loss measurement results are identified as belonging to a high-risk stage, then accounting provisions will be made for lifetime expected credit losses and interest will be suspended.

[0119] Impairment provision calculation includes stage judgment, ECL calculation logic, and provision adjustment.

[0120] Specifically, if the credit risk of a financial asset does not increase significantly after initial recognition, the expected credit loss measurement result is identified as belonging to the medium-risk stage, and accounting provisions are made according to lifetime expected credit loss.

[0121] If a financial asset has significantly increased credit risk but has not experienced credit impairment, and the expected credit loss measurement result is identified as belonging to the medium-risk stage, then the accounting treatment should be based on lifetime expected credit loss.

[0122] For financial assets that have already experienced credit impairment, it indicates that the expected credit loss measurement results belong to a high-risk stage, and accounting treatment is carried out according to lifetime expected credit loss and interest is suspended.

[0123] This includes the ability to monitor the transition of assets between different stages and to record the timing and reasons for these transitions.

[0124] In some embodiments, reserve adjustments may include making impairment provisions for new assets; reversing over-provisioned provisions when credit risk improves; writing off assets deemed unrecoverable; adjusting for exchange rate fluctuations of foreign currency assets; and adjusting interest income from impaired assets.

[0125] In some embodiments, the pre-defined expected credit loss prediction model can employ a multi-task learning architecture. For example, a shared encoder can be used, which constructs a unified deep neural network as the underlying feature extractor, shared by the three tasks: PD, LGD, and EAD. This encoder learns a general, high-order joint feature representation from multimodal input data, improving the model's generalization ability and reducing redundant computation.

[0126] Building upon a shared encoder, task-specific heads can be employed, allowing for the design of independent output layers (i.e., "prediction heads") for each prediction task. The PD head focuses on classification or probability estimation, the LGD head handles bounded continuous value regression, and the EAD head predicts the amount of risk exposure, ensuring that each task can be modeled in a refined manner according to its characteristics.

[0127] In addition, weighted multi-task loss functions, gradient balancing mechanisms, and uncertainty quantification can be employed. The weighted multi-task loss function can be a weighted sum of the losses of each task (such as binary cross-entropy, mean squared error, etc.) according to learnable or preset weights. By dynamically adjusting the weights, the contributions of different tasks during training can be effectively balanced, preventing any single task from dominating the training process.

[0128] Gradient balancing mechanisms can introduce gradient normalization techniques (such as GradNorm) to automatically adjust the scale of gradients for each task, alleviate training imbalance caused by differences in the magnitude of loss or convergence speed between tasks, and improve the stability and efficiency of joint optimization.

[0129] Uncertainty quantification can be achieved by integrating methods such as Bayesian neural networks or Monte Carlo Dropout to generate uncertainty estimates during the prediction process.

[0130] In some embodiments, the pre-defined expected credit loss prediction model can undergo training strategy optimization, hyperparameter optimization, and evaluation and verification optimization during training. The training strategy optimization can be found below: 1) Data Partitioning: The dataset is divided into a training set (70%), a validation set (15%), and a test set (15%) in chronological order using a time-series partitioning method. This approach simulates the time evolution scenario in real-world business operations, preventing future information leaks and ensuring the credibility of the model evaluation results.

[0131] 2) Batch Sampling: A stratified sampling strategy is employed during training to ensure a balanced proportion of samples from each risk level (e.g., low, medium, and high risk) in each training batch. This method effectively mitigates model bias caused by uneven sample distribution and improves the ability to identify minority class samples.

[0132] 3) Learning Rate Scheduling: Cosine Annealing Learning Rate Scheduling is employed to dynamically adjust the learning rate during training. This strategy enables the model to converge quickly in the early stages and then finely search for the optimal solution in the later stages, helping to escape local minima and improve the final performance.

[0133] 4) Regularization techniques: Combining regularization methods such as Dropout and Weight Decay, the model complexity is suppressed and overfitting is prevented. Dropout randomly blocks some neurons during training to enhance the robustness of the model; Weight Decay constrains the parameter size through L2 penalty terms to improve generalization ability.

[0134] 5) Early Stopping Mechanism: During training, the model's performance on the validation set (such as AUC, RMSE, etc.) is continuously monitored. If no performance improvement is observed after 10 consecutive training epochs, training is automatically terminated to prevent the model from overfitting on the training set and to retain the model parameters that best perform on the validation set.

[0135] Hyperparameter optimization can be achieved through grid search, such as performing a grid search on key hyperparameters; or through Bayesian optimization, such as using the TPE algorithm for efficient hyperparameter search.

[0136] Evaluation, validation, and optimization can be carried out through backtesting using historical data; online performance evaluation using A / B testing; and stress testing can be performed by analyzing feature importance using SHAP values ​​and testing model performance under extreme market conditions.

[0137] Based on the above embodiments, see Figure 3 and Figure 4 This application also provides a system for processing expected credit losses of credit assets. The system for processing expected credit losses of credit assets adopts a microservice architecture and mainly includes a data access layer, a data processing layer, an AI model layer, an ECL measurement layer, an accounting processing layer, and a monitoring and management layer.

[0138] The data access layer is mainly used for real-time data stream access (supporting real-time data streams from message queues such as Kafka and Pulsar), batch data synchronization (supporting batch data synchronization methods such as direct database connection and file transfer), and external data interfaces, such as connecting to external data sources like credit reporting agencies and macro data providers.

[0139] The data processing layer is mainly used for data quality control (such as data integrity, accuracy, and consistency checks); feature engineering (such as automated feature extraction, transformation, and selection); and data standardization (such as unifying data formats and coding standards).

[0140] The AI ​​model layer can employ deep learning models or traditional machine learning models, and supports ensemble learning algorithms such as XGBoost and LightGBM.

[0141] The ECL metering layer is primarily used for calculating risk parameters (such as PD, LGD, and EAD) and the ECL metering engine (such as the automatic calculation of 12-month ECL and lifetime ECL) as well as adjusting forward-looking parameters.

[0142] The accounting processing layer is mainly used for intelligent journal entry generation, impairment provision management, and financial statement updates. Intelligent journal entry generation primarily involves the automatic generation of accounting entries based on business rules and AI algorithms; impairment provision management mainly involves dynamically adjusting the amount of impairment provisions; and financial statement updates automatically update relevant financial statement items.

[0143] The monitoring and management layer primarily includes functions for real-time monitoring, anomaly alerts, and compliance checks. Real-time monitoring refers to the real-time monitoring of system operating status and business metrics; anomaly alerts mainly involve intelligent anomaly detection and multi-channel alert notifications; and compliance checks primarily involve automated compliance checks and report generation.

[0144] In some embodiments, a microservice architecture comprises three core components: service decomposition, service governance, and containerized deployment.

[0145] I. Service splitting strategies may include the following: 1) Data Service: Responsible for the access, cleaning, standardization and unified view construction of multi-source heterogeneous data, providing high-quality data support for upper-layer services.

[0146] 2) ModelService: Encapsulates the full lifecycle management functions of AI models, including model training, batch / real-time inference, version control, and performance monitoring.

[0147] 3) Measurement Service (ECLCalculationService): Enables automated measurement of expected credit loss (ECL), performs calculations of risk parameters such as PD, LGD, and EAD, and completes stage judgment and ECL result generation according to IFRS 9 standards.

[0148] 4) Accounting Service: Automatically generates accounting entries based on ECL results, performs impairment provision adjustments, and synchronizes accounting postings with financial statements to ensure the compliance and accuracy of accounting treatment.

[0149] 5) Monitoring Service: Provides real-time monitoring of system operating status, model performance, data quality and business metrics, and supports anomaly detection and multi-level alarm mechanisms.

[0150] II. Service governance mechanisms may include the following: 1) Service registration and discovery: Nacos is used as the service registry center to achieve automatic registration and dynamic discovery of service instances, and supports load balancing and failover.

[0151] 2) Configuration Management: Nacos enables centralized management of configurations, supporting dynamic updates and environment isolation (development, testing, production), with changes taking effect without requiring a service restart.

[0152] 3) Circuit breaking and degradation: Integrate circuit breakers such as Sentinel or Hystrix. When a dependent service experiences delays or failures, the circuit breaking mechanism is automatically triggered, and a degradation response is returned to prevent the failure from spreading and ensure the availability of core links.

[0153] 4) Distributed tracing: Based on Jaeger, distributed tracing is implemented to record the call path, time, and status of requests between microservices, which facilitates performance analysis, fault location, and root cause diagnosis.

[0154] In some embodiments, containerized deployment implementations may include the following: 1.Docker containerization 1) Base image selection: The lightweight AlpineLinux is used as the base image, which significantly reduces the container size and improves deployment efficiency.

[0155] 2) Multi-stage build: Using Docker's multi-stage build technology, complete dependencies are included in the build stage, and the final image only retains the files required at runtime, optimizing security and performance.

[0156] 3) Image layering design: Rationally plan the image layer structure (such as basic environment, dependency library, application code layering) to improve image build speed and cache utilization.

[0157] 4) Security Scanning: Integrates open-source tools such as Trivy to automatically scan images for CVE vulnerabilities and configuration risks during the CI / CD process, ensuring container security.

[0158] 5) Image repository management: Build a private image repository based on Harbor, which supports image version management, access control and vulnerability scanning, and realizes centralized and secure image storage.

[0159] 2. Kubernetes Orchestration Management 1) Cluster planning: Design a highly available Kubernetes cluster architecture, including multiple master nodes and cross-availability zone deployment, to ensure the disaster recovery capabilities of the control plane and worker nodes.

[0160] 2) Resource quota management: Set CPU and memory resource requests and limits for each microservice to prevent resource contention and ensure service quality.

[0161] 3) Automatic scaling: Configure HorizontalPodAutoscaler (HPA) to automatically scale Pods based on CPU, memory usage or custom metrics (such as request volume) to flexibly respond to traffic fluctuations.

[0162] 4) Rolling Update Strategy: Adopt Kubernetes' rolling update mechanism to gradually replace old version Pods, achieve zero-downtime deployment of services, and ensure business continuity.

[0163] 5) Health check mechanism: Configure LivenessProbe and ReadinessProbe so that Kubernetes can automatically detect the container status and restart abnormal instances or remove them from the service endpoint in a timely manner.

[0164] To unify the management of the microservice system's external interfaces and ensure system security, stability, and observability, the system adopts an API gateway as the unified entry point for all external requests. The API gateway undertakes core responsibilities such as service routing, access control, and traffic management, with the following specific functional design: 1) Request routing and forwarding Client requests are dynamically forwarded to the corresponding backend microservices based on predefined rules (such as request path, HTTP method, request header information, etc.). Flexible routing configuration is supported to achieve service decoupling and version management (e.g., / api / v1 / credit → metering service, / api / v1 / accounting → accounting service).

[0165] 2) Load balancing When routing to multiple instances of the target service, various load balancing strategies are provided, including algorithms such as Round Robin, Weighted Round Robin, and Least Connections, to ensure that requests are distributed reasonably among service instances and improve system throughput and resource utilization.

[0166] 3) Rate limiting and flow control To prevent sudden traffic surges from overloading the system, the gateway integrates rate limiting algorithms such as TokenBucket and LeakyBucket to finely control API call frequency. It supports rate limiting based on client IP, user ID, or application key to ensure the stability of core services.

[0167] 4) Authentication and Authorization It integrates standard security mechanisms such as OAuth 2.0 and JWT (JSON Web Token) to perform identity authentication and authorization verification on all API requests. Before forwarding requests, the gateway verifies the validity of the token and extracts user identity and authorization information to ensure that only authorized users can access protected resources.

[0168] 5) Monitoring and Statistics It collects and reports API call data in real time, including key performance indicators such as call volume, response time, error rate, and latency distribution. It supports integration with monitoring systems such as Prometheus and Grafana, providing data support for system operation and maintenance, performance optimization, and capacity planning.

[0169] Through the unified management of the API gateway, the system achieves centralized governance of external interfaces, effectively improving the system's security, maintainability, and scalability, and is a key component for building a highly available microservice architecture.

[0170] Thus, by employing a deep learning ECL measurement model to fuse multimodal data, this application can improve the accuracy of ECL measurement and accounting processing, as well as increase processing speed and automation.

[0171] In addition, in terms of risk management, it can monitor risks in real time, identify potential credit risks and issue early warnings; in terms of financial management, it can realize the automatic accrual and adjustment of impairment provisions, greatly improve the timeliness and accuracy of financial statement preparation, and reduce costs, effectively solving the problems of low accuracy, poor efficiency and insufficient automation of traditional solutions.

[0172] Corresponding to the above-described method for handling expected credit losses of credit assets, this application also provides an apparatus for handling expected credit losses of credit assets. This apparatus includes a unit for executing the above-described method for handling expected credit losses of credit assets, and can be configured in a desktop computer, tablet computer, laptop computer, or other terminal.

[0173] like Figure 5 As shown in the figure, this application provides a computer device including a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114. Memory 113 is used to store computer programs; In one embodiment of this application, when the processor 111 executes the program stored in the memory 113, it implements the method for processing expected credit losses of credit assets provided in any of the foregoing method embodiments, including: Acquire multimodal data of target users regarding credit assets, wherein the multimodal data includes structured data, time-series data, text data, and graph data; The multimodal data is input into a preset expected credit loss prediction model; The multimodal data is fused and predicted using the expected credit loss prediction model to obtain prediction results, which include the probability of default, the loss rate of default, and the default risk value. Based on the prediction results, the expected credit loss measurement results are determined; Accounting processing is performed based on the expected credit loss measurement results to obtain the target audit trajectory record corresponding to the target user.

[0174] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program may be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0175] Therefore, this application embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method for processing expected credit losses of credit assets as provided in any of the foregoing method embodiments, including: Acquire multimodal data of target users regarding credit assets, wherein the multimodal data includes structured data, time-series data, text data, and graph data; The multimodal data is input into a preset expected credit loss prediction model; The multimodal data is fused and predicted using the expected credit loss prediction model to obtain prediction results, which include the probability of default, the loss rate of default, and the default risk value. Based on the prediction results, the expected credit loss measurement results are determined; Accounting processing is performed based on the expected credit loss measurement results to obtain the target audit trajectory record corresponding to the target user.

[0176] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk, or any other physical storage medium capable of storing program code. The computer-readable storage medium can be non-volatile or volatile.

[0177] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0178] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0179] The steps in the methods of this application embodiment can be adjusted, merged, or deleted according to actual needs. The units in the apparatus of this application embodiment can be merged, divided, or deleted according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0180] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0181] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0182] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Since these modifications and variations fall within the scope of the claims and their equivalents, this application also intends to include these modifications and variations.

[0183] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for handling expected credit losses on credit assets, characterized in that, The methods for handling expected credit losses on the aforementioned credit assets include: Acquire multimodal data of target users regarding credit assets, wherein the multimodal data includes structured data, time-series data, text data, and graph data; The multimodal data is input into a preset expected credit loss prediction model; The multimodal data is fused and predicted using the expected credit loss prediction model to obtain prediction results, which include the probability of default, the loss rate of default, and the default risk value. Based on the prediction results, the expected credit loss measurement results are determined; Accounting processing is performed based on the expected credit loss measurement results to obtain the target audit trajectory record corresponding to the target user.

2. The method according to claim 1, characterized in that, The expected credit loss prediction model includes a feature extraction layer, a feature fusion layer, and a prediction layer; The process of fusing and predicting the multimodal data using the expected credit loss prediction model includes: The multimodal data is preprocessed to obtain preprocessed multimodal data; The preprocessed multimodal data is input into the feature extraction layer; The feature extraction layer is used to extract features from the preprocessed multimodal data to obtain multimodal features; The multimodal features are then subjected to multimodal encoding to obtain a multimodal feature encoding vector; The multimodal feature encoding vector is input into the feature fusion layer for fusion processing to obtain fused features; The fused features are input into the prediction layer to obtain the prediction result.

3. The method according to claim 2, characterized in that, The multimodal features include basic features, statistical features, temporal features, textual features, and cross features. The feature extraction layer extracts features from the preprocessed multimodal data to obtain multimodal features, including: Extract basic features from the structured data; Statistical and temporal features are extracted from the time-series data; Extract text features from the text data; Extract the cross features from the graph data.

4. The method according to claim 2, characterized in that, The step of inputting the multimodal feature encoding vector into the feature fusion layer for fusion processing to obtain fused features includes: The multimodal feature encoding vector is input into the feature fusion layer. Self-attention is used to calculate the attention weights between features within the same modality, cross-attention is used to calculate the cross-attention weights between different modalities, and multi-head attention is used to obtain feature relationships at different levels. Based on the attention weights between features within the same modality, the cross-attention weights between different modalities, and the feature relationships, the multimodal feature encoding vectors are fused to obtain fused features.

5. The method according to claim 1, characterized in that, The accounting treatment based on the expected credit loss measurement results includes: The expected credit loss measurement results are matched with accounting entry templates to obtain matching results; Based on the matching results, the journal entry parameters are automatically filled in to obtain the filling results; The filling results are verified using preset accounting rules; If the verification is successful, automatic posting, impairment and financial statement updates will be performed to obtain the target audit trail record corresponding to the target user.

6. The method according to claim 5, characterized in that, Before performing accounting entry template matching on the expected credit loss measurement results, the method further includes: Credit risk stage identification is performed on the expected credit loss measurement results; If the expected credit loss measurement result is identified as belonging to the low-risk stage, then accounting treatment shall be performed based on the expected credit loss over 12 months. If the expected credit loss measurement result is identified as belonging to the medium-risk stage, then accounting provisions shall be made in accordance with lifetime expected credit loss. If the expected credit loss measurement results are identified as belonging to a high-risk stage, then accounting provisions will be made for lifetime expected credit losses and interest will be suspended.

7. The method according to claim 5, characterized in that, After determining the expected credit loss measurement result, the method further includes: Based on preset economic indicators and scenario probability allocation values, determine macroeconomic scenario information; Based on preset adjustment factors, preset adjustment ranges, and preset adjustment rules, forward-looking adjustment parameters are determined, wherein the preset adjustment factors include macroeconomic factors, industry factors, regional factors, product factors, and time factors. Based on the macro-scenario information and the forward-looking adjustment parameters, a preset dynamic update mechanism is used to dynamically adjust the expected credit loss measurement results.

8. An apparatus for processing expected credit losses on credit assets, characterized in that, Includes a unit for performing the method as described in any one of claims 1-7.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, can implement the method as described in any one of claims 1-7.

Images