I have looked at a number of books on Trading Systems and the best seems to me to be http://www.amazon.com/Beyond-Technical-Analysis-Develop-Implement/dp/0471415677 by Chande. IMO it is better than others because it gives in-depth reasoning and not just optimization. Anyone else have a view? Oh by the way, although the book is expensive, I have discovered that there is a way to have access to it, and in fact almost all trading books (and computer technology and other books as well) which is http://www.safaribooksonline.com/ for a cost of $20 a month for one plan. Not only that, there is a 10-day free trial - you can read a lot in 10 days!

Well most of these books could be renamed, "Throw Spaghetti on the Wall and See if it Sticks". However, David Aronson's book and Perry Kaufman (advanced trading systems) are good authors. As mentioned, the issue with all of these guy is that their method for finding strategies is not scientific. You have to test your hypothesis or use bayesian statistics to get to the next level of enlightement. For every one of these books that you read, pickup a book on econometrics or at least some sort of book on Probability.

Is this legal? I mean do these guys pay the authors something or they provide access to their works without paying them royalties?

I still have not found a good read on Bayesian statistics applied to trading. There is a lot of free material in places like the dekalog blog and the price action blog with examples of standard statistical tests applied to random entries, specific systems, etc. http://dekalogblog.blogspot.gr/2013/05/random-entries.html http://www.priceactionlab.com/Blog/2013/05/significance-of-a-system-for-trading-spy/ Their methodology is standard and based on calculating a test statistic from Monte Carlo simulations and then testing the significance of a trading system based on its distribution.

Bayesian statistics literature is sparse compared to classical stat's books and less so in the way of trading. Some issues with using B.S. are deciding on the prior distribution and slower computation speed. I would suggest studying Bayes' theorem and work some examples to see how it could be used. Otherwise, for modeling purposes google AIC and BIC for an easier implementation.