Customer Reviews

Trading System Analysis: Using Trading Simulations and Generated Data to Test, Evaluate and Predict Trading System Performance

5 star:		(0)
4 star:		(0)
3 star:		(0)
2 star:		(1)
1 star:		(1)

 

 

This is one of those books where you think maybe the author knows the material he's talking about but doesn't have very good skills at clearly conveying the information he wants to present to the reader. It's like a university class where going to the lectures seems pointless so you just try and rely on the text - except this is supposed to be the text. Most of the time Barnes seems to be talking *around* the subject rather than addressing the topics directly, and all the parenthetical comments (literally hundreds of them) became quickly annoying. The book seemed at times to be more a collection of elaborated notes rather than a coherent whole.

I was about 1/2-3/4ths the way thru when it struck me that the book wasn't even really about the analysis of trading systems. While there are a few pages at the beginning on the book's alleged subject (including tables which are printed too small to be readable), most of the material is about methods for generating synthetic price data. After a discussion of the theory behind the intricate details of price movement, the idea is that you start with a year or three of actual (daily) price data, characterize it using one of three different models (random walk, sequential walk, or "price vector"), and then use that characterization to generate unlimited amounts of synthetic data. The reason why one might want to do this is a bit fuzzy, especially considering the vast amounts of actual price data the real markets generate every day. And I'm not sure I could successfully execute the recipe myself given the poor way it's presented; maybe with much hard study his actual method could be puzzled out.

As the other reviewer noted, the generated charts don't look very real (way too many gaps, for one), implying something essential is missing from the model(s). I also kept waiting for him to take some of this synth-data and put it into the top of the process as if it were real data to see if the original characterization would be recovered; this would seem to be a crucial test.

There's a final chapter called "The Complete Trading System", but it's only 17 pages long! This would have been an overview or an intro chapter of a book I might like to have had rather than this one.

There is about enough content in this book for a Stocks and Commodities magazine article, but not a book. The creation and use of simulated data is better covered elsewhere, whether off the internet or perhaps a book such as Chande's Beyond Technical Analysis.

This book is heavily padded out with endless tables of generated data (one example would have sufficed) and endless charts of generated data. The charts are not presented in a very useful way - they are sideways on the page, labeled in small print, and the generated data is of a different length than the actual data, so it is difficult to compare.

However, one thing is clear from the charts, and that is that Barnes' methods for creating simulated data don't pass the test - they don't look like the markets they are supposed to simulate. However, his "price vector" model is an interesting idea, and conceivably provides a starting point for additional research.