Measuring Market Risk, 2nd Edition (The Wiley Finance Series) [Hardcover]

I would have returned this book to Amazon.com and asked for my money back if I had not already been reimbursed by my employer for it. I am a risk manager, and so was looking forward to reading this cover to cover. Well, my opinion of this book changed before I got out of the preface!!!! The author states that all of the most important results will be provided as Matlab programs in the attached CD. Why? Because he doesn't think that usual programs like Excel/VBA are strong enough and yet programs like C and others are not user friendly. He then urges the reader to buy Matlab regardless of it being quite expensive.

Ok, my advice is to beware anyone who says that there is one and only one program that you can do something in. What he did was program in what he was comfortable with regardless of the reader's potential skills and resources. There are a lot of programs out there. But his whole book is targeted to using ONE third party vendor program. This was not disclosed in the book's descriptions. This is misrepresentation. If the book is so targeted toward one program, that should be indicated explicitly. The title of this book should be "Measuring Market Risk using Matlab". I suspect the reason this was not done was because it would have limited sales. I really think the author shows fundamental contempt for his readers by not appropriately disclosing the tie to Matlab on the cover.

Why this is particularly important is that this book is so targeted toward the CD. It looks like you have to have Matlab to run his examples. Without Matlab, all you have is code. And this is not some easy to read pseudo-code you can translate into another program. For example, one line taken from his Matlab code is:

[U,S,V]=svd(return_data,0);

What is that? If you guessed svd is standard deviation, you would be wrong. Do you think you are going to translate this spaghetti into something useful? I don't think so. There are a few Excel programs in a subdirectory on the CD, but they are just a scattered collection of bits and pieces of functions. One I opened up that looked promising could not run without the user buying yet another program called "Crystal Ball". This just got worse and worse.

The book is divided into two parts. The first is an overview of risk ideas with minimal derivation or explanation. The second half seems to be a more of a catalog of models and approaches that then points the user toward his CD. For example, there are 5 pages on Principal Components Analysis and Factor Analysis and then points the user to the CD. There are 4 pages on Copula functions, and then it points the user to the CD. If you don't already know how to do PCA and Copula, you are not going to learn it in 4 or 5 pages!!!! So what do we have here? It does not go into meaningful detail to explain the concepts. So it is not a strong risk book. So, this whole package is more like a software program with significant documentation. But wait. Its not a software program either because you don't get Matlab with it. You just get some prewritten functions from Matlab that don't appear useful unless you have Matlab. So, its not a software package, and its not a stand-alone risk book. What is it? Unless you already own or will buy Matlab and are already up to speed on market risk, then DO NOT get this.

Incidentally, I own Dowd's "Beyond Value at Risk". It is getting dated, but I thought that book was excellent. I think the author can do great work. My issue is not with the author's knowledge or skills. The problem is this book.

I fundamentally disagree with the reviewer stating that this book should not be bought. The fact that that reviewer isn't familiar with Matlab is a shame given that he or she works as a risk manager and since Matlab is often a basic pre-requisite for doing good quantitative finance. Serious risk management demands serious numerical software and Matlab is one such tool which allows quick model implementation in the fast paced business world. Excel/VBA are definitely not suitable for good work in this field (just look up the many statistical problems that can be found in Excel's functions, for example, or try to implement some basic matrix operations using VBA). C is not great either given that the Dowd's didactic message would be lost in a sea of imperative coding logic. Not understanding that the 'svd' function is shorthand for singular value decomposition, makes me suspect that that reviewer's quantitative abilities may not at the level needed to read this book. However, I don't want to turn this review into a flame-fest advocating Matlab over all else and ignoring the content of the book itself. I think Matlab has its faults too, but Dowd made a sensible choice in using this pseudocode-like language for the examples of the models he presents. (And if Matlab is too expensive to purchase there are many free clones that work just as well: just search for 'scilab' or 'octave' on the web). But on with the review of the book itself...

As I said in the subject heading, this book is not suitable for beginners. There is not much in the way of justification of the aims of VaR or the field of market risk management, while much time is spent on classroom level theory. For example, chapter one contains a very brief recap of the highlights of the history of portfolio risk measurement, while chapter 2 already attempts to rip VaR apart with the justification of using coherent risk measurements instead of VaR. A beginner is just not going to be able to grasp all that's going on at this early stage without a good number of practical examples. Even *with* the examples, it's often hard for people new to the field to get the kind of intution that only comes after years of practice and working through real problems. Dowd doesn't do much to alleviate that kind of confusion. However, for the practicing, well-read risk manager or quant, the book is a veritable encyclopediac reference of the field. Dowd does for VaR what Fabozzi does for fixed income securities. He covers practically all the major models and their many variations and gives much more information about the mathematical tools needed to make these models tick than do most of the classic references (e.g. Jorion). What's more Dowd does an admirable job of describing important complements to VaR such as stress testing, backtesting, and model risk. Finally, the citations that Dowd includes in the book are useful in and of themselves as they include the main readings in the field.

My biggest complaint with the book is that's it's no more than a survey of the *theory* of VaR. To that end, Dowd's academic focus is present throughout and this focus book does not easily lend itself to practical issues in risk management and analysis. For example, one of the most important practical issues in risk management is the mapping of securities into their building block risk factors, yet Dowd spends a paltry 11 pages discussing this topic. What's more there is no mention of the very real world need to model portfolio VaR in the case of missing market data. Dowd too often assumes a perfect world of complete data which is simply not the case. Moreover Dowd does not discuss many of the real world issues involved in the development and maintenance of living breating enterprise risk systems.

Overall I think this book is an extremely useful addition to every risk manager's bookshelf, but I only gave it 4 stars because I feel there is a fair bit of room for improvement.

It hurts me to see negative Amazon reviews because this is one of the single best references on market risk VaR (i.e., market VaR not credit VaR). I do agree with previous reviewers is three respects: Dowd's book has a specific non-beginner audience, this is a theoretical (academic) rather than practice-oriented text on VaR, and its strength is not really portfolio VaR. Okay, so clearly this is not really for an introduction to VaR nor is it for VaR-in-the-trenches.

But within the (advertised) scope, there simply isn't a better survey of theoretical market risk VaR tools. Dowd goes into much more detail that Jorion (who is a leading author). And, very few will appreciate the analytical rigor and precision of Dowd; I use Dowd to teach VaR, and where other authors make formula mistakes, he doesn't seem to. Other authors, like Culp and even Wilmott, use a VaR that leads new learners into certain math errors, but Dowd's VaR formulations are thoughtful and less prone to deployment errors. Each year I use this book, it continues to impress me more.

Special strenghts are: reviews of non-normal parametric value at risk; succinct and accessible introduction to extreme value theory (EVT); careful review of VaR's weaknesses (e.g., non coherent)--Dowd is expert on the limitations of VaR; and liquidity-adjusted VaR.

The bent is mathematical, for sure, but most of the book is within reach of basic/intermediate calculus; if you have a background, you'll find that some of the first one-third is even probably too much introduction (e.g., volatility, covariances and correlations).

The FRM (financial risk manager) has assigned only Chapter 16 (Model Risk) to FRM candidates. Which is ironic, because this chapter isn't so strong, FRM candidates should be reading the rest of the book. But, for now, three years on, this is still the leading book in a very specific category.