Original question: What (kind of backtesting) do you do to understand if your rules or system is likely to have a positive expectancy?
I don’t do any backtesting at all. It doesn’t work as no algo can see what a trained chart eye can see. My “backtesting” is simply my own trading history. History is 2400 trades rich (1700 with basically an identical system) so I was able to drew a lot of conclusion from analyzing my own data. I believe that analyzing ones past trades and getting rid of stuff which simply doesn’t cut it is a major part of refining or honing any trading system.
I do a basic t-test analysis of my trades to learn about the probability that my expectency is real and not just part of a normal distribution with a mean value of 0. Once that probability is at >99.9% I call it good and try to keep it at that level. I also include the initial risk of my wins in my calculation of the expectency. Then I go on and only monitor this calculated expectency and try to keep it positive. Doing it that way helps you to be detached from any monetary value. I even go one step further and cap all real wins at ~3 times my average loss for the sake of the expectancy calculation and try hard to keep that capped system positive. In reality I do have many wins above that capping threshold but for the sake of the calculation I ignore them. Psychology wise this simple technique is almost holy grailish, but don’t tell anyone 😉
As you may know, trading outcomes shouldn’t be normally distributed in a >3R system with a winrate of 30-40%. You typically have a huge peak at -1R, which is basically your maximum risk. Ideally 1R should be between 0.5 and 2.5% of your account balance! Most of your trades will end up being stop loss hits at around that level, hopefully ;-). Then there should actually be a void all the way up to around 2 – 3R where you realize your baseline R-multiple wins. Most traders have a lot of trades around breakeven. This is a signal of a weak trading system and shows that the trader in question likely moves his stops to breakeven due to psychological pressure. Anyway, it is clear from the above that the outcome isn’t normally distributed but rather lognormal (inlcuding negative values) with a peak around 2-3R. I applied some transformations to my data to make it normally distributed before applying the t-test but that didn’t change much hence I skipped it.
As a matter of fact the outcome histogram of your past trades is actually your trading fingerprint. Analyzing it tells a lot about your weaknesses and strengths at a glance. Therefore this will be a major part of the upcoming methodology/coaching service. One can use a histogram of the R-multiples or the “% of account balance” directly. Former distribution can be tricked by trading small positions so I advice you to use the latter in combination with a calculated expectancy and statistical validation (t-test) including the initial risk of winning trades as outline above.
If you are not aware of your outcome histogram you lack one vital tool in your toolbox for sure.