IFRS 9 PIT & Lifetime PD Modelling (SAS, English Course)

Learn how observation windows capture borrower characteristics and how performance windows track defaults, aligning predictors with outcomes in IFRS 9 credit risk modelling.

Learn mean and median imputation to replace missing values with the mean or median, showing how the median is more robust to skewed distributions and outliers in credit risk modeling.

Apply regression imputation in IFRS 9 credit risk modeling to predict missing data from observed variables using a regression model, leveraging multivariate relationships while noting potential correlation inflation.

Demonstrate hot-deck imputation in IFRS 9 credit risk modelling by filling missing values from a similar record, retaining realism through careful matching criteria.

Explore fine classing to model detailed relationships and coarse classing for stability. Target monotonic WOE trends and minimum bin populations, and validate with IV, PSI, and simple logistic checks.

Demonstrate a fine-to-coarse classing workflow by starting granular, merging by similar WOE or event rates, validating stability, and preparing bins for monotonic effects.

Explore correlation analysis with Spearman and Hoeffding, detect monotonic and broader dependencies missed by Pearson, and prioritize signals that remain stable across samples.

Combine stepwise heuristics with professional judgement for variable selection in predictive modelling. Evaluate candidate sets using AIC/BIC, c-stat, and calibration on validation to lock a compact, explainable feature set.

Assess the linearity of the log-odds in IFRS 9 credit risk by an empirical-logit check, binning scores, plotting elogit versus mean score, and applying regression or loess.

Learn logistic regression in SAS with PROC LOGISTIC, using class, model, and store statements for reproducible scoring, export ODS OUTPUT, and save predicted PDs for audit-ready report with key diagnostics.

Explore logistic regression for credit risk, model default_12m(event='1'), interpret odds ratios for each driver, and assess discrimination (AUC/KS) and HL test with lift to deliver a transparent baseline model.

Explore how to handle categorical variables in credit risk modelling by computing odds tables and bad rates, and group noisy categories before multivariate modelling to assess level risk.

Assess the linearity of the log-odds with empirical-logit checks by binning the score, plotting elogit against mean score, and applying regression or loess; use transforms or spline effects if curved.

Import and validate the IFRS 9 dataset, verify structure and statistics, and create a stratified 70/30 train-validation split with reproducible defaults for logistic regression modeling.

Explore Gini, lift, and gains metrics to assess credit risk models; learn that Gini is a scaled AUC, and how lift and gains charts convey ranking of borrowers for stakeholders.

Explore Kolmogorov-Smirnov statistic and gains and lift curves through a SAS-based decile analysis of predicted defaults, highlighting K-s values, decile event rates, and model rank ordering in training and validation.

Demonstrates the Hosmer-Lemeshow test for calibration of predicted defaults, with observed vs expected defaults and a chi-square of 10.17 (p=0.2533), plus a Brier score of 0.126 indicating moderate accuracy.

Explore D-statistic and concordance in credit risk modelling, measuring how often predicted risk aligns with observed defaults and the separation between good and bad borrowers.

Advance model evaluation by balancing discrimination, calibration, and business impact under IFRS 9, using metrics like ROC AUC, Gini, and Brier score to ensure accurate ACL and governance readiness.

Calibrate 12 month point-in-time PD using calibration in the large and Platt scaling in SAS, evaluating AIC, likelihood, odds ratios, and the C statistic to improve calibration.

Learn to calibrate macro inputs for pd modeling using arima forecasts of GDP, unemployment, and interest rates, then build baseline, upside, and downside scenarios to drive IFRS 9 ECL.

Explore IFRS 9 staging and SICR assessment, where stage 1 uses 12-month ECL and stages 2–3 use lifetime ECL, with triggers like PD increases, credit score drops, and qualitative events.

Explore the IFRS 9 modelling workflow by setting up the environment, importing and cleaning data, and evaluating staging with 12-month proxy bad rates.

Explore lifetime probability of default modelling with cohort analysis, survival analysis, and transition matrices, and apply it to stage 2 and stage 3 ECL.

Master survival analysis for credit risk and learn forward-looking PD modeling with Kaplan-Meier, Cox proportional hazards model, and transition matrices, preparing for IFRS 9 lifetime and 12-month ECL calculations.

Learn how to prepare a quality LGD data set for IFRS 9 and Basel 3.1, by defining LGD, setting observation and performance windows, and addressing missing values, outliers, and segmentation.

Explore linear regression as a transparent LGD benchmark, showing how LTV, collateral coverage, and GDP growth influence losses, with SAS and Python implementations.

Explore the mixed effects LGD model that combines fixed and random effects to capture unobserved recovery factors across industry, region, and macro conditions, aligned with IFRS Nine and Basel.

Learn to apply logit transformation to LGD, back-transform predictions with inverse logit, and build OLS model on logit space in SAS for IFRS 9 lifetime PD with 70/30 train-test split.

Learn to apply beta regression with SAS GLIMMIX to estimate LGD, bound predictions between 0 and 1, use the logit link, and score new data with PLM.

Present a SAS decision tree regression for loss given default, with a 70/30 train-test split, LTV ratio feature, and pruning, highlighting seniority and LTV as key predictors.

Master exposure at default (EAD) modelling as the final pillar of the ECL framework, combining PD, LGD, and expected drawdowns with CCFs for IFRS 9 and Basel calculations.

Explore how the credit conversion factor converts undrawn commitments into drawn exposure to estimate exposure at default, and apply it under IFRS 9 and Basel 3.1 for forward-looking EAD modeling.

Explore data preparation for EAD modelling, designing robust, regulator-ready datasets with observation and default windows, calculating CCF, and validating data quality for IFRS 9 compliant exposure analysis.

Learn Tobit and truncated regression to model EAD and CCF within natural bounds, implement in SAS and Python, and interpret results for IFRS 9 and Basel 3.1 compliance.

Transform raw exposure data into an IFRS 9 compliant end-to-end EAD modelling case study, from data prep to ECL integration, using OLS, Tobit, beta regression, and two-stage logistic modeling.

Recap the transition from exposure at default to expected credit loss by integrating PD, LGD, and EAD within IFRS 9 and Basel 3.1 frameworks, with forward-looking, scenario-based validation.

Understand how to compute expected credit loss by multiplying PD, LGD, and EAD, discount with the effective interest rate, and generate portfolio-level reports and IFRS 9 disclosures.

Finish section eight to learn how to combine pd, lgd, and ead into the IFRS 9 ecl formula, calculate 12-month and lifetime ecl, apply discounting, and test scenario sensitivity.