Collection of Tools for PD Rating Model Development and Validation
Univariate analysis
Area under curve (AUC)
Bivariate analysis
Bootstrap model validation
Categorical risk factor binning
Slice categorical variable
Confusion matrix
Constrained logistic regression
Create partitions (aka nested dummy variables)
Palette of cutoff values that minimize and maximize metrics from the c...
Custom decision tree algorithm
Testing the discriminatory power of PD rating model
Embedded blocks regression
Encode WoE
Ensemble blocks regression
Modelling the Economic Value of Credit Rating System
Model fairness validation
Testing heterogeneity of the PD rating model
Herfindahl-Hirschman Index (HHI)
Testing homogeneity of the PD rating model
Imputation methods for outliers
Imputation methods for special cases
Extract risk factors interaction from decision tree
Indices for K-fold validation
K-fold model cross-validation
Multi-period predictive power test
Slice numeric variable
Near-zero variance
Power of statistical tests for predictive ability testing
Testing the predictive power of PD rating model
Predict method for custom decision tree
Population Stability Index (PSI)
Replace modalities of risk factor with weights of evidence (WoE) value
Risk factor clustering
Extract interactions from random forest
Calibration of the rating scale
Scaling the probabilities
Model segment validation
Synthetic Minority Oversampling Technique (SMOTE)
Staged blocks regression
Customized stepwise regression with p-value and trend check
Customized stepwise regression with p-value and trend check on raw ris...
Stepwise logistic regression based on marginal information value (MIV)
Stepwise logistic regression based on risk profile concept
Stepwise regression based on risk profile concept and raw risk factors
U-shape binning algorithm
Testing for U-shape relation
Weights of evidence (WoE) table
The goal of this package is to cover the most common steps in probability of default (PD) rating model development and validation. The main procedures available are those that refer to univariate, bivariate, multivariate analysis, calibration and validation. Along with accompanied 'monobin' and 'monobinShiny' packages, 'PDtoolkit' provides functions which are suitable for different data transformation and modeling tasks such as: imputations, monotonic binning of numeric risk factors, binning of categorical risk factors, weights of evidence (WoE) and information value (IV) calculations, WoE coding (replacement of risk factors modalities with WoE values), risk factor clustering, area under curve (AUC) calculation and others. Additionally, package provides set of validation functions for testing homogeneity, heterogeneity, discriminatory and predictive power of the model.