Its fine if you are using SPY. The system should translate to any index product at this point. You need to configure your contract specification in AB by assigning big point value, scale and minimum move. That way, your testing will reflect the 0.25 tick increment in the ES. Otherwise just use the SPY. Here's the test I get for the ES. Two things to notice, this system only works during decent volatility, as expected. $75 per trade on the ES is not bad but not good for an intraday ES system. I get $6.25 slippage + $2.40 commissions per side for the ES, so subtract ~$18 from that $75 to get your per trade profit. Max drawdown is about $5100, which isn't great either (about 1/4 the profit for a 12 year period). What else can we deduce from this system report? What are some improvements we can make?
Good question, I was hoping to get his input for this system. Here's the SPY return distribution with histogram. Can someone do all the pertinent analysis? Skew, kurtosis, etc in R or something other than Excel? Excel fudges certain #'s. Mike
I'm new to R but gave this a shot...basically just tried to recreate one of Talon's sheets from BoWo's thread. Let me know if there are other figures we should include.
Hey guys, I'm still around. Sorry for my absence. We have been very busy losing a ton of money doing stupid trades to start the new year off. Kind of funny what a couple weeks away from the desk will do for your brainpower. Actually... I guess not losing a lot of money, but doing a lot of trading and not really getting paid for it. Anyway, I've been reading the posts here but haven't had time to do it carefully. Is the point of this development work to specifically look at the volatility filter? If so, I'd suggest running the system you have with the filter and without so you can compare the two. Also, i've never done a volatility filter on bar ranges squared. Not saying it doesn't make sense to do so, but the reason for squaring returns is obvious, but not so much for ranges. This is important: *** The solution to the overnight gap problem for an intraday indicator is pretty simple actually. Use range not true range. The inter-bar gaps are not captured in H-L so it's no problem. Also on question of adjusting volatility indicators for the normal day's "smile" (higher at opening and close.) I guess you can do that but the fact is that most intraday systems that need a lot of volatility shouldn't be traded in the middle of the day and this indicator would also show you that volatility is lower at those times... so no problem. You could easily write an indicator that adjusts for this, and I could see a use for it... but not sure it applies to what you're doing here. I think this thread is on the right track. What questions can I answer?
Glad you're back. I was starting to think this discussion was waning interest... too bad if it did cause this is where is gets interesting IMO! Attached please find the same system with the volatility filter excluded. You can see this system is worthless without some sort of filter. This is a simple example based on dtrader's realized vola. calculation and some very basic math - so outside of me providing a basic concept, this filter is ET's creation. The main idea here being that a vola filter - even a crude one; that may or may not have the exact properties we are looking for, can do a decent job of making a system much better. Given we obey the basic facts of vola. of course... In this case, the math may not be kosher, but it gets the job done. That said, there are several conditions (and some better math) that can easily be added to this system that make it - dare I say it - tradeable, but, I'll leave that up to the readers to figure it out. I've been spoon feeding it too much as it is. Anyway, the hard questions still exist: assuming we have scracthed the surface of an "edge" here, what can we do to improve it conceptually? And perhaps most perplexing is how do we make good money during low vola? That latter question has been baffling me for years BTW.... Who knows, we may stumble onto something here ... or not. Mike
Thanks for doing this. It looks like the normality test passed, yes? That's usually an encouraging sign. Any improvement ideas? Mike
Actually, based on the test i think we'd have to reject normality here with that small p-value. Didn't have a chance to look at improvements today (trying to get the scaling right for that R chart took an embarassing amount of time today), but will work on some tomorrow afternoon. Quick question if someone has a chance in the meantime: other than beating B&H and costs as Talon stated, do you guys have a rough benchmark for when you consider something to be tradeable (x% per day, a risk adjusted figure, etc.)?
Ahh, glossed over the exponent on the p-value. Thanks for clarifying. Thanks again for doing this analysis. To roughly answer your question: I usually look for a high net profit/drawdown ratio over about 5-10 years for intraday. Usually anything over 5 to 1 gets my attention and over 10-1 is where I may start forward testing live with small positions. The Net/DD ratio assumes fixed position size and zero money management. That said not all systems are meant to be traded stand alone. Conditions exist where an average system can be combined into a portfolio of systems to balance out long/short biased returns. Mike
Hi Did you use RExcel to do that? How have you found it? I thought about trying it but heard it really slows down VBA. Not sure it's worth the effort. Any observations? It might be worth integrating if performance doesn't suffer. If it does, however, may be just better to run the tools separately.