The Model Risk Manager
No Cover

The Model Risk Manager

Inside Model Risk Management at a Bank: Inventory and Tiering, Independent Validation, Backtesting and Monitoring, Overlays, Vendor and AI Models, Governance, the Rules and the Career

Julian R. Sterling

On one Tuesday in June, a bank's model risk committee has to decide whether the model that sets the probability of default of 1,240,000 credit card accounts is still fit for use, when it ranks risk well but expected 5,667 defaults where 6,002 came.

The Model Risk Manager is a practitioner's guide to model risk management in a bank: what counts as a model, the inventory and tiering, development and documentation, independent validation and effective challenge, conceptual soundness and data quality, discrimination and calibration, backtesting, benchmarking and sensitivity analysis, the validation report, findings and remediation, ongoing monitoring, credit, market, stress testing and financial crime models, model limitations and overlays, vendor models, machine learning models and their explainability and bias, governance and model risk appetite, the rules and the career. One fictional US regional bank is followed through one year and one focal day, and every figure comes from one model.

Some of what the arithmetic shows: how 12 validators plan 15,090 hours of work across 166 models; how an AUC of 0.805 and a Gini of 0.610 can sit beside two red grades in a binomial test; how a backtest moves from overprediction to underprediction in four cohorts; how 7 exceptions in 250 days fail a Kupiec test; how an overlay of $19.82 million is sized, approved and tracked; why a population stability index of 0.121 is a warning and not a verdict; how a vendor score loses a quarter of its Gini on the bank's own loans; and how a fraud model that catches 360 more frauds with 2,400 fewer alerts can still fail a fairness tolerance.

Current to October 2026, with each real-world fact dated: the US supervisory guidance of 2011 and its revision of April 2026, the OCC's 2025 clarification for community banks, the UK's SS1/23, the ECB's guide to internal models and the EU AI Act's treatment of credit scoring. Case thresholds are teaching assumptions, not any bank's or supervisor's requirements. Not legal, regulatory, compliance, accounting or investment advice.

The model is not described, it is supplied: a companion workbook rebuilds the case by live formula, and the checklists come as an editable document.

What you will be able to do:
- Decide what counts as a model and keep an inventory tiered by materiality and complexity
- Plan validation work against capacity and scope a full validation, a periodic review or a change review
- Judge conceptual soundness, data quality and implementation
- Measure discrimination with AUC, Gini and KS and calibration with a binomial test by grade
- Backtest, benchmark and stress a model, and write a validation rating and findings
- Backtest a value at risk model with the Kupiec test and review AML scenario thresholds
- Monitor models with a population stability index and size and govern an overlay
- Review vendor models and validate machine learning models for explainability and bias
- Describe the three lines, the model risk committee, the US and UK rules, and plan a career in model risk

Seventeen chapters and four appendices: a guide to the workbook, a model risk toolkit of fifteen checklists, thirty-six interview questions with model answers, and a glossary of sixty terms.

Written for analysts and graduates joining a model validation or model risk management team; for model developers, credit risk modelers and data scientists whose models are validated; for risk managers, internal auditors and compliance staff who rely on models; and for candidates preparing for model risk interviews. Part of The Private Markets Career Guides.

How this book was made: written with the help of artificial intelligence tools, under the author's direction; every figure was checked against its source.

Book Info

Publisher Julian R. Sterling
Release Date 2026-10-21
Page Count 223 Pages
File Size 412.8 KB
Language en-US

No download link available yet. You can .

Related Books

Similar recommendations you might enjoy

Join Telegram