I tried that but got no encouraging results from backtests, by using small and not so small WFO windows. I'm loosing faith in the idea that parameters should be reoptimized often, but that's just my experience. If an algo is to produce good results, it can't be too sensitive to param adjustment, because if it is, it could also be too senstive to market conditions. This is a general suggested guideline in system design. When I sense that an algo may no longer be working as expected, I prefer to rethink it completely or disable it.
+++ Update +++ Just closed out three positions opened during the last four days. Currently 35.5% up from the beginning (November 14). Comparing the equity curve with the QQQ during the same period it can be seen that the system did a pretty good job in reducing the drawdown and still gaining more than the index, which rose about 22%. A more aggressive system could have gained much more, considering the huge swings, but all in all I'm satisfied for now. In June I'm up more than 3.000$ and the only month that suffered a little loss was May (-176$). Little money, yes, but I'm leaving almost all of it in the account for reinvesting. I've switched to a long-only, multi-position algo which I had already developed but not used so far. Then I did a little modification to it and now it appears to work ok. I discontinued the old algo because this one, all in all, produces a much better backtest curve over 8 years of 1-minute data (third image below). Still working on neural nets, interesting results yet not convincing enough to be used.
do you have any algos that trade actual money? if no, why not start one up with 2500, or 250, or 25.00$? get the beak wet, so to speak.
Thank you. The IB equity curve (upper graph) clearly shows the difference between the two algos, before and after the flat segment at the center. Although the previous one looks smoother, my expectation is that the one I'm using now will produce better results overall.
If you're referring to the lower graph (backtest) yes, it includes slippage and fees. Actually slippage isn't a problem, considering the typical holding time (a few days) and that index futures are very liquid. X axis = number of trades, total period is a little less than 8 years Y axis = cumulative %, by just adding up (not compounding) the trades About 90/100 trades per year on average. This particular algo can add up to 4 position over the first one, hence 5 micro e-mini MNQ contracts.
Compounding: I started with 25k and the balance is currently 41k (also thanks to my son's 8k that he added along the way). If and when I hit 50k, I will double the position size, hence 2 contracts per position.
Automation: this is certainly an automated trading system. However, intervening on it by designing/modifying/enabling/disabling algos is still "trading", although of a different kind. I guess it could be defined as meta trading (no pun intended).