Some excellent points here. The author of the book doesn't show much backtesting of different input numbers/periods/variables and therefore doesn't give us a good sense of robustness. I would also add that a lot of the tests (though not all) were performed over periods of time when there was a strong bull market and perhaps that is a consideration too. In his defense, some of the strategies were tested and yielded good results in multiple stocks and other securities, increasing the total number of trades tested and indicating some level of extrapolation of the positive expectancies across markets. Despite the concerns we've all noted, the book is filled with actionable ideas for future testing, including some strange-looking rules for entries and exits that I would have never thought of in a million years, and I'm impressed by the effort.
I would imagine that the strategies also tested - that didn't have good results before the variables were changed - would create a book that might be visible from space.
No Kindle. Had a BN Nook at one time. When I want to read a book ... I read a book. Nonetheless if testing was done on anything but tick data I always question the validity of the results.
Automated systems that give lots of signals and win rates much north of 50% (eg 65,70,80,90% win rate systems) with at least a 1:1 win loss ratio don't exist accept in back testing. The market works on a low win% percent but potential for big wins when they do eventually happen. Everyone who has watched the markets for any length of time knows this already.
100% correct. You really have to love probability theory and do the math for this to become clear. Reading about "the math" in the context of some specific domain, such as trading, isn't the way to learn this stuff.
All good systems involve fitting . You are fitting the future based on the past . Tell me how do you make decisions about anything in the future unless its anchored to previous events ? Even crooks do this when they pull the same boilerplate scams over and over. (ie Crypto)
This is why a better term is 'over fitting' I build systems around my market observations, then use back testing to test and fine tune those ideas.