I actually agree with much of what you say. A lot of info is people pulling statistics out of their rear end.
"I, like other traders for the past 15 years of trading systems have been trying to reconcile the bewildering vast number of statistics that are spewed from guruâs books and articles on trading."
Which I wouldn't do. I found in a related field that it's best to develop my own approach. My trading strategy, for example, is something I developed based off my trading philosophy. I like to read, find nice ideas, and some of these just click and become fundamental in my mind.
In sportsbetting I used to test for statistical significance, but I dropped that after I found it's better to judge the methods than the results. This has continued to my trading, and I have no intention of testing my results. I'm certainly not one of those "+ev traders" you talking about. I'm completely aware of probabilities and expectency, but I prefer sound logic to determine my trading, rather than backtesting.
"1. Performance measurements must be consistent with each test.
2. What you apply in testing (optimization) must be verified in production (live trading).
3. Performance must be integrated between all components to be effective (integrated with other parts of trade management)."
I agree with these rules, and I think anyone backtesting or testing over a large sample should following these. If the method can't be exactly replicated in live trading, there's not much point in testing.
"Guru rule 1 from one web site: ââ¦Use a fixed fractional position sizing⦠For example, you might risk 2% of your account equity on each trade (the "2% rule"). â¦â
Guru rule 2 from another web site ââ¦we cannot risk more than 6% of our trading account when we enter multiple positions at the same time â¦â
Guru rule 3 from another web site ââ¦You should not exceed 8% over all losses in month. In other words, the most you can lose in month or total trading account loss is 8%...â
The first rule is understandable. Most people won't be able to judge the quality of the opportunity, so trading a fixed amount is fine. The "2% rule" is a bit blind though, as I think people should understand the underlying mathematics before deciding size. 2% is too high for most anyway, since even 2% leads to large swings, and most aren't profitable anyway so they shouldn't be trading in the first place.
The second two rules, for me these are just stats pulled out of their rear end. Multiple positions should be considered, but 6% is a meaningless figure. Losing 8% in a month is also ridiculous, especially for those that follow the 2% Rule (you could easily lose 8% in a day!).
"Then there is expected value. When I optimize a typical trading system I get 200 profitable settings. When I dump the trade data to my database and SQL it all 200 have a positive expected value? However when I forward test only 8 settings are profitable? So expected value has no predictive quality and is not consistent across testing (same result last 8 years)."
This is where we differ. I don't think your conclusion here is correct. We're dealing with uncertainty, so a strategy that yields profitable results over a sample isn't necessarily +ev, even if our numbers suggest it is. This is why I use the terms "perceived ev" and "true ev". A system that tests well might have perceived +ev, but it may not actually be profitable, and therefore the true ev may be negative.
This is quite important to understand for people who backtest systems, as blindly trading systems that have tested well is not a profitable strategy. If you tested 100 systems, 10 may come back profitable, but that doesn't mean they're actually +ev. Chances are they are not, and this is why there must always be strong underlying logic. The test for a system with solid underlying logic is much stronger than just testing random systems.
"When the forward tested strategy settings live trades are examined, guess what? The expected value from testing has no relationship or correlation to actual results."
That's because the expected value over testing is not equal to the true expected value, and then there's variance to consider. I've had a profitable (sportsbetting) system experience completely different results over two statistically significant samples.
"Expected value doesnât work for me. Expected value like many other guru statistics does not correlate (because it constantly changes as you note) to building better positive trading results for me. It is a simplistic general statistic."
Whether it is useful or not depends on how it is used, but where we differ here is mainly in our interpretation of ev. You're thinking of it as a statistic generated by testing over a sample. To me, that is just perceived ev, since it may very well be incorrect. If the perceived ev is incorrect, then naturally there must be a value that is correct, and I call this the true expected value. This is what is important as it exists for every trade, but unfortunately we never know what it is. This is the joke in my original point.
