Managing Model Risk: Lessons and experiences from industry and research on the challenges and dangers of analytical models

The authors take one through the field of Model Risk in both a very efficient and concise matter. As such, the book works almost like a checklist, as a part of a Model Risk overview, one can go through the book one chapter at a time and see if one's own organization covers these areas. The focus of the book is not on the mathematical structure or specific models, but for that, there are other books. Instead, it really tries to capture the essence of Model Risk, which is about looking out for misuse of both algorithms of data, and the book is rich in both advice and industry wisdom. For anyone that is serious about using models, and therefore need to understand the risks that come with them, this book is a must!

If you are looking for a guide to model risk you will not find it here. Each section, when read in isolation seems to convey useful reflections on model risk, for example, the section on data leakage rightly points out why this is a problem along with some useful anecdotes across six pages. However, other than throwing a collection of topics at the reader, the authors do little in providing a useful framework to assess model risk in a more general way.In fact, the whole book can be summarized as: "Here is a collection of war stories. Be careful out there!" These little warnings would be useful only if they weren't painfully obvious and this is my main grief with the book. Regression and machine learning models seem to be the main focus but at the same time these are the type of models for which there are already thousands or articles out there.The book avoids the really difficult topics, for example, model documentation is the bane of every model risk manager's existence, and the authors spend a whooping three paragraphs on it, barely one page. This isn't an isolated issue; on the topic of model monitoring (another pivotal topic) the authors merely mention drift over a handful of paragraphs assuming this is sufficient and that it could be generalized to every model (it isn't).In conclusion, don't waste your money on this book as there is nothing of value even for entry-level practitioners.