Mortgage Credit Risk Modeling with R

About This Course

This course takes you from R-comfortable analyst to building fully validated, production-grade mortgage credit risk models — the same models used by banks, GSEs, fintech mortgage lenders, and the regulators who examine them. Across the main core (Modules 0-12, including Module 2A on Quarto) and optional extensions (capstones and an ARM appendix), you will construct a complete PD/LGD/EAD/CECL modeling framework for conventional conforming fixed-rate mortgages using real Freddie Mac Single-Family Loan-Level data. You won't just learn about mortgage credit risk modeling — you will build every component yourself in R, step-by-step, with a practitioner guiding you through the exact workflow used in industry and reviewed by independent validation and regulatory examiners. What You'll Learn Set up a professional R environment with renv-locked reproducibility, R Project structure, and Git-based version control for sensitive loan data. Ingest, clean, and join Freddie Mac Single-Family Loan-Level origination and performance files, HMDA, FHFA HPI, and FRED macro data into a unified loan-month panel. Master mortgage mathematics for fixed-rate products — level-payment amortization, prepayment, curtailment, LTV trajectory, and HPI-driven mark-to-market LTV. Build a full Probability of Default scorecard using WoE/IV, logistic regression, monotonic binning, and points-to-double-the-odds scaling. Develop survival-based and competing-risks PD models — Kaplan-Meier, Cox proportional hazards, parametric Weibull, and Fine-Gray for prepayment vs. default. Build Loss Given Default models using beta regression, two-stage default-flag-plus-severity, and LTV-based curves with HPI stress sensitivity. Implement Exposure at Default using amortization schedules and curtailment behavior. Construct lifetime CECL reserves with vintage loss curves, roll-rate models, macro-linked satellite models, scenario blending, and Q-factor framework. Build a modern machine learning challenger — XGBoost with SHAP explainability and adverse action reason code generation. Conduct fair lending analysis using HMDA: denial rate disparities, rate spread analysis, redlining tests, BISG proxies, and disparate impact regression. Produce audit-ready documentation in Quarto: Model Development Documents, Validation Reports, Ongoing Monitoring Reports, and CECL disclosures. An optional appendix extends the framework to adjustable-rate mortgages using Freddie Mac's Non-Standard Dataset, with caps, floors, reset schedules, and payment shock. Who This Course Is For Mortgage credit risk analysts, model developers, and model validators. Bank, credit union, GSE, and fintech mortgage professionals. Data scientists transitioning into mortgage or consumer credit risk. Audit and Model Risk Management staff supporting mortgage portfolios. Career-changers preparing for risk, analytics, or model development roles in mortgage lending. Module 0 covers the R toolchain, RStudio configuration, package management, and Git basics from a practitioner's angle — but you should be comfortable with basic R syntax, data manipulation, and undergraduate-level statistics. This is a practitioner course, not an introduction to programming. Hands-On Projects You Will Build By the end of the main core, you will have created a reproducible RStudio project with Freddie Mac Single-Family Loan-Level data, renv-locked dependencies, and Git history; a working teaching subset of SFLLD built from raw quarterly text files via a documented Arrow/Parquet pipeline; a fully exploratory analysis of origination attributes and performance trajectories with vintage and seasoning views; a validated PD scorecard with KS, Gini, AUROC, calibration, and out-of-time / out-of-sample validation; a survival-based lifetime PD model with competing-risks treatment of prepayment; a two-stage LGD model with HPI stress sensitivity; an EAD model accounting for amortization and curtailment; a complete CECL engine with vintage curves, roll-rate models, scenario blending, and back-testing; an XGBoost challenger model with SHAP explanations and reason codes; a full HMDA-based fair lending analysis with disparate impact testing and remediation; and an audit-ready Model Development Document, Validation Report, and PSI/CSI monitoring dashboard. Optional capstones extend the framework against the full Freddie Mac dataset; an optional ARM appendix module applies the methodology to adjustable-rate mortgages. Every module ends with a tangible, portfolio-ready artifact. Every section ends with a hands-on lab. Every capstone ends with a complete documentation package. Course Structure The main core covers Modules 0 through 12, including Module 2A on Quarto for regulatory reporting. The Foundation phase (Modules 0, 1, 2, 2A) covers the R toolchain, mortgage credit risk fundamentals, data ingestion, and Quarto reporting. Mortgage Mechanics (Module 3) covers fixed-rate amortization, prepayment, LTV trajectory, and HPI mark-to-market. Core Models (Modules 4, 5, 6, 7) cover PD scorecards, survival PD, LGD, and EAD. Application (Modules 8, 9, 10) covers the ML challenger, fair lending, and CECL. Governance (Modules 11, 12) covers validation and ongoing monitoring. Optional extensions include capstones (Module 13) and an ARM appendix (Module 14). Every lecture includes a complete instructor script, branded slide deck, runnable R code, exercises, knowledge checks, and session notes — produced and quality-controlled to a single consistent standard. Why This Course Is Different Most credit risk courses teach theory on toy data. Most R courses don't touch regulation. This course does both — using real Freddie Mac Single-Family Loan-Level data spanning the 2005-2007 crisis vintages through recent originations, with every method implemented in production-quality R code and every output framed for SR 11-7-compliant model documentation. You'll learn not just how to build mortgage credit risk models, but how to defend them in independent validation, document them for regulatory examination, monitor them in production, and remediate them when fair lending or stability issues surface. By the end, you'll be able to independently develop, validate, document, and govern mortgage credit risk models in a real-world environment — across PD, LGD, EAD, and CECL, with the regulatory and fair lending lens that examiners actually apply.