Sorry I haven't been keeping up this journal, but I wanted to have something worth sharing. So here is where I am: 1. My manual strategy worked. The success probability is about 70% 2. I have backtested to 2008 3. It is consistently profitable week after week, year after year. It never lost money. So now I'm deathly afraid I'm screwing something up. I have backtested using the following methods: 1. Manual 2. Coded in two separate platforms 2a. Platform A: let it manage the exit via a trailing stop 2b. Platform B: I managed the stop manually (in code) The results were the fucking same within a few $10K. I'm losing it, because if this works, I will have achieved my childhood dream. I don't know what to do next, but I guess I'll allocate about $5K to this strategy in the new year and let it go. If it works successfully in month 1, I'll put in the full $50K towards the strategy. If I lose it all, ah well. Merry Christmas, and happy new year! Here are the backtest results:
There is only one place where there could be lookahead bias, and I have tried to confirm that it doesn't exist. How would you know for sure? Edit: I do not know the number of trades yet over the whole period.
Well, it's just one guess - the curve is eerily smooth for anything that's not a proper high frequency strategy. There is a number of possible issues that could cause this (I'll go through some ideas shortly), but rehashing them here would be a bit too public. PS. Just tried sending you a PM - apparently, you have that turned off
Right. In previous incarnations it looked extremely uneven until a few hours ago where I doubled down on getting the right exit conditions. Everything else remained the same, but once I nailed the exit condition, it became what you see :-/ It is intraday, but not high-frequency. It's designed to be neutral to market direction.
I feel like the next steps are to print out all the trades and manually go over them using a third data provider to make sure everything is lining up and my backtests are executing correctly without lookahead bias. Thoughts?
Everyone here knows how you feel. Use this as an opportunity to remind yourself that if you do see something that looks too good to be true you should assume a bug somewhere until proven otherwise
Well, I was deathly afraid that I was wrong The thing I'm still trying to reconcile is why the other platform (which can't have the same bug) gives similar results.