I actually like no1: It has more trades so less chance of overfitting and easier to observe that forward testing matches backtesting. The ratio of winning trades to losing trades is lower, again this gives me comfort there's less overfitting. The backtested drawdown is longer so you'll be able to trade it and take losses with more patience.
I would think all systems are curve fit to some degree. It is the overfitting that gets people into trouble.
Hey, you never know. He may have tested extensively with out-of-sample data. Possibly even removed a few of the largest wins from each strategy to ensure that the results are not attributed to a disproportionately few winning trades. Guy could be a genius. All else being equal, I'm putting everything on #4.
He didn't ask anyone to evaluate the validity of his backtesting. He merely asked that we choose from the proffered results. #4, all in.
Correction: ...to ensure that the results are not attributed to a few disproportionately large winning trades.
I appreciate the responses. I didn't really want to get into the strategies, but several have expressed a opinion that there is "over-fitting or curve fitting". I am providing some more information. Any additional thoughts would be appreciated. Curve-Fitting: I chose equity indexes with good volume that work well. I would like to use ES, but it doesn't work as well (but maybe I should take another look given the responses I have received). So, I "fit" the instruments to the strategy. I found that being flat at the end of the day worked better than holding until the next day, so I closed at the end of the day (profitable both ways). I found that a "stronger" signal gave better results - all signals were profitable, but I did "fit" to the "best" signal for each instrument. Disproportionate Trades: For #2 (the one that I was leaning towards, so did a little further studying before posting here) the most profitable trade was $6,672 - perhaps an anomaly or data error. The next most profitable was $2,977, and the third most profitable was $2,477. The largest losing trade was -$1,553. If I eliminate the top 10% of trades, I get: 420 winning trades, 218 losing trades, max draw-down of $4,100, a max draw-down/breakeven period of 150 days, and profit of $71,000. It seems that some would consider this to be a superior strategy, but I wouldn't know which profitable trades do avoid until after the fact Additional Testing: I did test for six months out of sample with the same basic results... I'm not sure if this information sheds any light on whether the system is "over fit" or not, but I would appreciate any constructive comments.
Are you limited by the amount of intra-day data you have? The Nasdaq emini (NQ) was launched in 1997. I would be very curious to see how these strategies performed on older data.
When I started looking at this, I downloaded daily data from Yahoo. One of the futures - I think it was EMD - seemed to run out of data. I guess I should find some older data and see what happens for another 5 or 10 years...
None of them. Instead, I'd trade mine. Trading QQQ daily bars, $100k beginning bankroll. 43 trades, ALL profitable. Profit for the year = $86,425.54, so far. Max draw down = $11,651. All commissions and fees included. It's not for sale, I haven't written a book about it, and I don't offer classes, online or otherwise. I'd be interested in hearing from others who have something better. NB: I doubt that this will continue. We've had a very good year. I'm testing a follow on strategy using options for possible use next year.