It's 44 trades out of 282. If you define it a bit more strictly as only the 4th trade after a 3 trade win streak, it's 25 trades. Some of the 44 are the 5th trade after a 4 trade streak with an embedded 3-trade streak at the end. That was how I wrote the Excel logic, but I don't think that's right, now that I look at it closer, so I'd say the 25 is the better number for directly testing this. Fortunately, for my hypothesis, the winning percentage for that 25 is actually lower than the 44.
BTW, I always thought the gambler's fallacy was a rationalization for increasing the number of bets because the next one would recoup all losses. I'm aiming to reduce the number of bets, if I can identify a systematic way to ignore signals.
I think your sample is too small still to draw any conclusions honestly. Forward testing will confirm or deny this, but I think unfortunately it is too small though. I was thinking previously if it was all intraday that perhaps the time of day had something to do with it, but it doesn't sound like that is the case either.
Yeah, with 44 I was more comfortable, but the stricter definition is a better test of the systematicity or lack thereof. Hopefully, I will have many 3-trade win streaks in the future to generate a larger sample. I still am interested in the general idea of how to calculate the most likely size of a win streak for a given winning percentage and if that calculation can be used to avoid taking low-percentage trades, even if they are the same setup as the ones which generate the streak. Could just be barking up the wrong tree.
You figured out how I play black jack, in that like you I have noticed a statistical error in actual real world applications that don't support that each hand has a 50% chance of winning or losing if a certain number of hands have similar outcomes.
In any system I have ever worked with the trades are to varying degrees related to what has happenned before. You could come at this another way. The chances of 4 winners in a row if the occurences were ramdom would be win% * win% * win% * win%. So a 50% system would be .0625 or once ever 16 four trade sequences. You then would calculate how many 4 trade sequences you had in your data and see how many 4 streak winners you had. Lets say you had 1600 4 trade sequences. Assuming 50% system wins, you would expect to have 100 four streak wins. If you had a number substantially less then you would have a good indication your trades are related. Like I said I haven't seen any good edges where the sequences were not related, though the correlation can go either way. In your case with your potential observed correlation, I suspect your system characteristics are trend following with a win% less then 50% and a high PayOut Ratio.
Agree on both points. Need forward testing. Saying something is true in hindsight is curve fitting. Another 282 trades might be prudent. Need to take time into account. What were the 3 winning trades followed by 1 losing trade? Mon - Weds followed by Thurs? 9am - Noon followed by 1pm? Time effect might in part or in full explain the losing trade tendency.