Mechanical Day Trading

I've been backtesting a variety of mechanical trading strategies and chosen one with the best average performace over the last four weeks. The system will trade automatically through programs I've written and tested using TD Ameritrade's Active-X control. The plan is to use the best performing strategy over the last four weeks for the next week of trading. This particular strategy was the best out of over 7,000 permutations of possible entry and exit triggers for the most liquid ETF's.

I'll be trading QLD (Proshares ultra QQQ) at 15 minute intervals triggering on crosses of price over and under SMA 30 (contrary) with appropriate stops and some other filtering.

I'd be very interested in hearing from other traders who have experience or interest in mechanical day trading.

I've done a bunch of back testing of systems. Do you think that 4 weeks of testing is sufficient?

I had been looking for that one set of rules that worked over long periods of time. Without complex heuristics I do not believe a single set of rules will work.

In all my testing (10k+ permutations through 10+ years of historical data) I found that no single strategy works all the time. In fact, over any extended period every system I tested fails miserably. What I found is that for periods of time certain systems perform very well. This is why I'm looking at the best average performers over the last four weeks. These are the ones that are trading well in the current market. When the market shifts, other systems will bubble to the top of the list. Of course this is all theory, I'm just putting my money where my mouth is this week. Fake-trading this way for the last three weeks has had great results.

What you have just described should be a strategy itself (perhaps a meta-strategy) that should work over long periods of time and should be back-testable.

i.e. When do you abandon the current strategy and switch to the next one? I would assume that you do that after your first week of loss...? Then how do you determine which one to switch to? I assume that you find one that's worked for the last 3 to 4 weeks...

So your back-tester could do that as well. It could trade a strategy until the first week of fail and then switch to another one and you could run that across your 10+ years of data and determine if it will work for long periods of time.

Wow, backtesting the strategy rollover is a great idea... thank you. I have to think about how I can retool my engine to do that.

Currently on the weekend I run my analysis which picks the best performing strategy (a weighted average) from the last four weeks. I rollover the strategy regardless of whether the current week's strategy had a profit or loss. If the same strategy wins multiple weeks then I will use it multiple weeks.