Basel IRB Credit Risk Modeling Using Python

Explore the Basel framework for credit, market, and operational risk, its three pillars including minimum capital requirements, and how standardized and internal rating based approaches estimate risk capital.

Explore the loan loss distribution under Basel IRB, distinguishing expected, unexpected, and stress losses, and show how PD, LGD, and EAD drive VaR in Python.

Compute risk-weighted assets (RWA) by applying exposure amounts to weights (0% government bonds, 50% residential mortgages, 100% corporate loans) and maintain the capital adequacy ratio under Basel 3.

Apply the Basel standardized approach under Basel II/III to assign predefined risk weights to exposures using external ratings. Compute the capital charge as exposure amount times risk weight times 8%.

Explore the internal rating based (IRB) approach under Basel to estimate PD, LGD, and EAD using bank data, and compare foundation and advanced IRB with standardized methods.

Under Basel's IRB framework, banks classify banking book exposures into asset classes, retail (QRRE, mortgages) and wholesale (SMEs, large corporates, specialized lending) plus sovereign, bank, and equity exposures.

Basel defines probability of default as the one-year default likelihood, with 90 days past due and no cure; it uses through-the-cycle PD and segmentation for PD modeling with scorecards.

Load the dataset into Python, inspect the 6,000 observations across 10 variables, and identify the default_flag as the one-year Basel IRB pd target with key retail drivers.

Analyze the portfolio's default rate and class imbalance, noting 4,731 non-default and 1,269 defaults for a 21.15% unconditional PD, illustrating real-world imbalances.

Explore exploratory data analysis for Basel IRB credit risk modeling in Python, and identify key risk drivers like credit score, utilization, and DTI.

Understand why variable transformation is essential for PD modeling and how fine classing, coarse classing, WOE, and IV create stable, interpretable inputs for Basel IRB PD models.

Implement weight of evidence and information value in Python to convert raw predictors into modeling-ready features, compute WoE and IV, and select influential variables for credit default risk modeling.

Analyze multicollinearity via VIF for the five IV-selected predictors to keep the Basel IRB PD model stable and interpretable, retaining credit score, utilization rate, months on book, past delinquency, DTI.

Learn to build a Basel IRB compliant probability of default model with logistic regression in Python, covering variable selection, model estimation, interpretation, PD prediction, and decile ranking checks.

Assess the risk ranking performance of the PD model by analyzing decile buckets and observed default rates, ensuring monotonic increase and strong discriminatory power for Basel IRB rank ordering.

Evaluate the four validation pillars of a PD model: discriminatory power, calibration, stability, and Basel IRB compliance, for accurate risk estimates and Basel aligned credit decisions.

Validate the Basel IRB credit risk model by measuring discrimination with AUC, KS, Gini and ROC curve, supported by practical steps using scikit-learn.

Define loss given default (LGD) under Basel and compute it after recoveries from defaulted accounts to estimate potential losses, guide loan pricing, and support capital adequacy.

Learn how downturn LGD uses the recovery rate to assess losses under Basel regulatory capital. Explore factors affecting LGD, including collateral, debt seniority, recovery costs, and the economic environment.

explain the cooling off analysis in LGD modeling, showing how the cooling off chart identifies the recovery workout period and cumulative recoveries before applying chain ladder extrapolation for incomplete data.

Examine the workout method for measuring loss given default (LGD) by tracing default cash flows, incorporating direct and indirect costs and recoveries, and discounting to the default moment.

learn how to estimate loss given default using the chain ladder method, extrapolating missing recoveries with development factors, discounting to present value, and computing LGD from exposure at default.

Explore Basel IRB credit risk modeling by enforcing LGD between 0 and 1, cabbing negative values at 0 and values above 1 at 1, using long-run averages and stress testing.

Segment data by product to capture distinct recovery and loss characteristics for LGD modeling, and apply loan-to-value ratio and time since default segmentation to reflect recovery dynamics.

Prepare LGD data from defaulted accounts in the observation period, and use chain and ladder to extrapolate recoveries for recently defaulted cases per Basel guidance on cure for LGD.

Present a mortgage loan dataset prepared for LGD modeling, detailing transformed LGD, recovery rate, loan purpose indicators, and the default flag used to build the Basel IRB style model.

Explore the Basel IRB LGD data in Python, load the 2542 defaulted exposures, perform data quality checks, and note the bounded, right-skewed LGD distribution guiding beta or Tobit regression.

Assess correlation and multicollinearity to guide LGD model development, exclude mechanically linked and derived variables to avoid target leakage, and confirm ltv as a stable predictor.

Compare beta regression and Torbit regression for LGD estimation, showing how beta directly models LGD as a proportion, while Torbit uses an underlying loss with boundary 0 and 1.

Model loss given default with beta regression, using pre-default inputs like LTV and loan purpose, addressing zeros and ones, and Basel IRB conventions for bounded LGD.

Explore Tobit regression as a Basel IRB LGD challenger model, using censored 0–1 outcomes, with LTV and loan purpose, and clipping predictions for comparison to beta regression.

Validate Basel IRB LGD models by comparing beta regression and Torbit predictions using MAE and RMSE to assess stability and avoid underestimation.

Explore exposure at default (EAD) under Basel guidelines and its conservative computation for term loans and revolving facilities, including credit conversion factors (CCF).

Model the credit conversion factor for exposure at default on off-balance-sheet exposures, using drawn amount, limit, and headroom to estimate EAD, with CCF floored at 0 and capped at 1.

Learn to compute the credit conversion factor and exposure at default using fixed and variable time horizon methods, including the cohort approach and handling zero headroom with a modified ccf.

Learn to segment revolving and non-revolving exposures into homogeneous pools, estimate downturn CCF and downturn EAD by segment averages and regression with explanatory variables.

Explore the dataset for Basel IRB credit risk modeling using Python, detailing exposure at default, ccf, undrawn limit, and related variables in a retail credit card portfolio.

Describe the EAD dataset with basic statistics to assess data quality and economic meaning, highlighting credit score, utilization, undrawn limit, and CCF.

Develop an end-to-end exposure at default workflow using logistic regression to convert CCF to a binary flag, estimate expected CCF, and compute EAD under Basel IRB.

Model continuous credit conversion factors with beta regression in Python, adjusting 0–1 boundaries, and compare results to logistic regression for Basel IRB EAD.

Validate Basel IRB CCF and EAD models through portfolio-level checks and bias analysis. Compare observed and predicted CCF and EAD, assess conservatism, and review segment-level results by utilization rate.