I've enjoyed Ernie's posts in the past; I cannot comment on his success as a definitive book author but his blog commentary with respect to relative value trading is usually a worthwhile read. Having said that, it is indeed extremely relevant to separate the academic exercise from the practical application in the particular case of statistical arbitrage. This tactic has progressed from simple pairs trading in the 1980's to huge baskets or portfolios of long/short stock positions designed and tested to pare down beta risk as much as possible and to capture these co integration moves away from the established mean - hopefully many times over the course of a trading day by means of automation. This stuff can be easy enough for a decent HF analyst or good quantitative soloist to model and test for - if you have a Bloomberg terminal and can find the beta scatter plot functionality you're walking in tall cotton already. But here's the catch, and the reason why IMHO why the separation between academic and practical exercise is so important: the execution of the trade makes it or breaks it. As I type this, there are literally hundreds of firms in Chicago, NYC, Greenwich CT, London, etc. that invest huge sums of money for ECN infrastructure and to pay for access to order flow and exchange order queue priority ( even front running ) in order to conduct statistical arbitrage. Let's say, for example, as an independent trader clearing, let's say, IB, that you've carefully identified and tested a pair of stocks that your rigorous back testing says you should be able to capture 2 cents per trade, let's say, on average, 30 times per day, on these volatility blips away from the established mean. Looks superb on paper. You open up TWS and create the appropriate combination order selection tab. My prediction would be that the leg slippage in the live markets makes the trade that looked so good on paper just disappear. Or worse, you are repeatedly getting hung and over the course of the trading day you have on a directional unhedged stock position on numerous occasions. This is just my opinion, but legitimate and consistent statistical arbitrage success for the independent trader with a $50K retail account would be a very difficult proposition based purely upon execution slippage.
His second book is the one I read, titled "Algorithmic Trading". It wasn't a terrible read, but if it is "way more deep" than his first book, I will go ahead and cross that one off my list. As for him pushing matlab because he gets kickbacks.... doubt it. Matlab is probably the best tool there is for rapid proof of concept type trade studies. It has terrific graphics, math, and other toolboxes that just aren't offered elsewhere. Well worth the money for a license. In my opinion, it is a great first coding language to learn.
Agreed @IAS_LLC, no one would have to push MATLAB to the quantitative finance, academic, biomed, or engineering communities. It has a strong foothold there...and no...I am not interested in a MATLAB vs. R vs. whatever war. I'm merely saying it's no surprise that he uses MATLAB based on his academic and industry background. Mathworks has absolutely no need for his advertising, they have probably half of all colleges and engineering firms paying for subscriptions and minting new users already. You could certainly do everything he does in his books in R though.
I havent found a lot of trading books I like because i tend to be a very technical person and most of the books just seem to scratch the surface. I did however really enjoy these two: Options Volatility and Pricing Professional Automated Trading. Neither of these books claim to be strategy books. The latter gives some good ideas on what an automated trading architecture should look like. Ive been searching for a good modern electronic market making book, but havent landed on one yet.