Skill Guide

Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD) modeling

PD/LGD/EAD modeling is the quantitative process of estimating the three core parameters of credit risk (the likelihood of borrower default, the severity of loss given a default, and the monetary exposure at the time of default) to calculate expected loss for regulatory capital and business decisions.

This skill is the engine of risk-based pricing, capital adequacy (under Basel II/III/IV), and portfolio management. It directly determines a bank's profitability and regulatory standing by quantifying credit loss in monetary terms.

1 Careers

1 Categories

8.7 Avg Demand

25% Avg AI Risk

How to Learn Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD) modeling

1. Master the regulatory definitions (Basel IRB approach). 2. Understand the input data structure: loan-level data, borrower financials, collateral details, macroeconomic indicators. 3. Grasp the fundamental difference between through-the-cycle (TTC) and point-in-time (PIT) modeling philosophies.

Focus on model development lifecycle: data cleaning (handling missing values, outlier treatment), segmentation (pooling data by product/risk), and statistical techniques (logistic regression for PD, Beta/Tobit regression for LGD, CCF calculation for EAD). Avoid overfitting by rigorously applying out-of-time validation.

Direct strategy by integrating models into decision engines (e.g., for loan origination or provisioning). Develop challenger models using machine learning (XGBoost, survival analysis). Master model risk management (MRM): designing validation back-testing frameworks and stress testing models against severe macroeconomic scenarios.

Practice Projects

Beginner

Project

Build a Basic PD Scorecard

Scenario

You are a junior credit analyst with a sample dataset of 10,000 personal loan applications with a 12-month performance window and a default flag.

How to Execute

1. Clean the data and perform Exploratory Data Analysis (EDA) to identify strong predictors (e.g., debt-to-income, credit history). 2. Apply Weight of Evidence (WOE) transformation and Information Value (IV) analysis to select variables. 3. Build a logistic regression model in Python or SAS. 4. Score the out-of-time sample and calculate the Gini coefficient and Kolmogorov-Smirnov (KS) statistic to measure discrimination power.

Intermediate

Project

Develop a Workout LGD Model for a Corporate Loan Portfolio

Scenario

A bank's special assets group needs to estimate LGD for its distressed commercial real estate loans. Data includes loan terms, collateral valuations, recovery cash flows, and workout timelines.

How to Execute

1. Define the default point and calculate the discounted present value of all recovery cash flows. 2. Segment loans by collateral type and seniority. 3. Model LGD using a Beta distribution regression or a two-stage model (probability of cure * loss if not cured). 4. Validate model accuracy by comparing predicted LGD versus realized LGD on a hold-out sample using Mean Absolute Error (MAE).

Advanced

Case Study/Exercise

Design a Dynamic EAD Model for Revolving Credit Facilities

Scenario

You are the lead model developer tasked with updating the EAD model for a credit card portfolio to comply with new regulatory expectations that require capturing behavioral responses to economic stress.

How to Execute

1. Analyze historical drawdown patterns on defaulted accounts to calculate the Credit Conversion Factor (CCF). 2. Integrate macroeconomic variables (e.g., unemployment rate) into the CCF model to make it dynamic (CCF_t = f(macro_t)). 3. Develop a stress-testing scenario engine that simulates how EAD increases under a severe recession. 4. Document the model assumptions and present a challenger model using a different methodology (e.g., machine learning) to the Model Risk Committee.

Tools & Frameworks

Statistical & Programming Software

Python (scikit-learn, statsmodels, lifelines)SAS/ETSR

Core tools for model development. Python is dominant for new development and ML integration. SAS remains entrenched in many large banks for legacy models and regulatory compliance. Use for data manipulation, statistical modeling, and back-testing.

Regulatory & Methodological Frameworks

Basel IRB Approach (Foundation & Advanced)IFRS 9 Expected Credit Loss (ECL) FrameworkSR 11-7 (OCC/Fed Model Risk Management Guidance)

Non-negotiable knowledge. The Basel framework defines model components and supervisory floors. IFSR 9 dictates the accounting implementation. SR 11-7 outlines the governance and validation standards all models must meet in the US.

Data Infrastructure & Model Ops

SQL for data extractionMLflow / Airflow for model pipelinesTableau/Power BI for monitoring dashboards

Essential for operationalizing models. SQL is required to query vast credit data warehouses. MLOps tools manage versioning, retraining, and deployment. BI tools are used for ongoing model performance monitoring and drift detection.

Interview Questions

Answer Strategy

The candidate must distinguish between through-the-cycle (TTC) and point-in-time (PIT) calibration. A strong answer will reference Basel's focus on long-run averages over an economic cycle, while IFSR 9 requires lifetime ECL estimates sensitive to forward-looking macroeconomic forecasts. Sample answer: 'Basel PD models are typically TTC, calibrated to a long-run default rate to ensure capital stability across cycles. IFRS 9 models must be PIT, incorporating 12-month and lifetime macroeconomic scenarios to reflect current conditions, leading to more volatile provisions.'

Answer Strategy

Tests analytical problem-solving and understanding of model validation. Sample answer: 'I would first segment the validation results by collateral type and economic period to confirm the underprediction. The issue likely stems from the model's failure to capture the pro-cyclical nature of real estate valuations. The solution is to incorporate a time-varying collateral hair-cut factor or a macroeconomic variable (e.g., commercial property price index) into the LGD regression to improve downturn performance.'