AccuRate is a software solution for development, validation and automated selection of optimal credit rating and scoring models, using both tried and innovative machine learning and econometric methods.
Once the data is loaded, the end-users can build a credit rating model and receive a credit score and alphanumeric rating grade for all instruments or entities in their portfolio. Modeling process is fully aligned with regulatory requirements. Outputs can be used to transition from standardized to internal ratings-based approach in calculating RWA in Basel III framework.
In order to allow for efficient mining of Big Data, advanced machine learning algorithms are implemented to search for a globally optimal model from an immense space of potential models and their predictive variables.
AccuRate employs logistic regressions as well as advanced machine learning techniques for searching and calibrating optimal credit rating and scoring models. The result of this AI search and calibration is a set of Probability of Default (PD) models that are subject to thorough validation, using in-sample and out-of-sample statistical tests, as well as by performing out-of-time model performance tests. Subsequent automat-ed selection of the optimal model is based on information criteria. Lastly, borrowers and exposures are assigned credit scores and mapped onto alphanumeric rating scale, using various techniques of statistical clustering
• Calibration of PD models, using advanced econometric methods, and employing both quantitative and qualitative data (soft facts);
• Automated selection of optimal PD models among several candidate-models;
• Advanced model search based on hierarchical clustering technique;
• Validation of PD models, using in-sample, out-of-sample and out-of-time statistical tests;
• Assignment of credit scores to individual borrowers in sample, based on estimated PD;
• Mapping borrowers and exposures onto alphanumeric rating scale;
• Generating various predefined reports. AccuRate utilizes advanced methods in model calibration and validation to avoid common problems of standard logit models: multicollinearity, extreme factor values, missing values, and non-linear relationship between log odds and factor itself. AccuRate avoids these problems with careful data transformation and Firth’s bias reducing regression.
AccuRate is ahead of its peers both in terms of complexity and speed of calculations.
• Built-in tool which enables users to create custom factors based on raw data;
• Allows for expert opinion in factor selection process, in order to best capture specific local market conditions;
• Multiple competitive models. Users can choose between successfully validated models.
• Immense flexibility in parameter selection. This includes a variety of stratification methods – by sector, region, period etc. in order to improve model performance;
• Advanced machine learning models applied by AccuRate allow our clients to search large space of possible predictive factor combinations efficiently;
• Highly intuitive user interface, allows users to revert to previous steps, if necessary;
• AccuRate has its own relational database and built-in tool for data import;
• AccuRate performs automated out-of-sample and out-of-time validation of models, which is prerequisite for transition to internal ratings-based approach and for internal ratings audit;
• AccuRate creates rating categories based on calibrated model by applying K-means clustering technique and calculates optimal number of rating categories for any given model;
• AccuRate enables archiving relevant steps, intra-steps, results, calibrated and validated data, and allows for export into Microsoft Excel.
• AccuRate comes with accompanying methodology with introduction to statistical methods applied in modeling.