...I recommend that you look into E-Signal as well. They now offer one year of one-minute data. IMO the real issue is your preference for the style of coding in the testing language. My read of most system testers who post here is that they use TradeStation, for a long time the only choice for intraday testing and with lots of algorithms floating around. I prefer E-Signal's Java-based language, which looks and programs like "real code". Be careful what you choose however, because once you build up a library of code in one it would be painful to switch to the other.
hypo...when did ESignal start providing 1 year of data? i only get 6 months right now. did this just happen?
i was reading the katz and mccormick book, and they mentioned trying to stick with compiled (into native code) languages. i don't think either javascript, or easylanguage fits that bill. i am wondering myself if i should look into some c/c++ systems, because some of my tests -- just on the 6 months of 5-min data, are getting to be longer than an hour on esignal. and that isn't even optimization (which esignal doesn't even have). another rant on esignal -- i tried to set a time frame between 6:30 - 1:00 (pst) for 60 minute bars -- well, esignal doesn't quite do that. the last 60 min bar (12:30) includes up to the 1:30 time, even if you specify it to cutoff at 1. pretty slick, eh? don't know if tradestation does that yet.
...my mistake. I got confused because when they first started offering more data I once downloaded nearly a year of NQ #F in a one minute chart. I see from their promo literature that it is currently only six months, but I believe Chuck T. said somewhere it goes to a full year next year. Also, if you can say, I would like to know why you BT so far back. My current experience in one-minute NQ is that increasing volatility makes 3 months almost too long, and I find that I need to check optimizations every week to make sure I haven't missed a volatility shift. Further, I am totally in awe of code so complex that it takes an hour to BT. How many instructions do you typically have? My longest stuff is maybe 400 instructions, but the simplest (and best) is only about 70 instructions and runs fast enough to finely optimize conveniently. How the hell could you trade it? I run my codes in real time with helper panes to backstop me in case my mental codes miss a trade. As always, hope I'm not asking anything proprietary, but I am so struck dumb that I had to ask. Thanks.
interesting you bring this up. i really don't know how far back i should test. right now i'm just dumbfounded at how good this system works (i guess i never knew what trading with an edge meant). my fears say backtest farther than 6 months. but i could be doing my system injustice by backtesting stuff that might not necessarily be representative of what i intend to trade, as you say. nevertheless, i hope to uncover something by backtesting it further -- if i don't, then i won't feel like i wasted my time. my code is pretty simple (to me). now i have it down to 281 lines, and i javadoc like crazy and write very clean code so alot of that is not logic. i think its my indicator i wrote, which i can't get into of course, but it does alot of computations each bar. i am now running my backtests on the 5 min chart and the system's efficiency is just getting better. i can actually run this on many more markets now. a question for you, do you know any way to auto-optimize in esignal? do they have add-on .efs scripts for optimization? this would really help me when i get to the point of optimizing my input params (to think i haven't even done this yet!) god i love this shit merry xmas to me
funky, What platform did you use to come up with this system and do the equity curve test? Is the same platform available for real-time trading?
...thanks for providing that perspective. Re BT'ing for six months, over the last year of testing, I have found that most of the systems I have developed fade away after 3-6 months. In particular, one that I use now which is unbelievably good (and unbelievably simple) only started working 3 or 4 months ago. I watch it like a hawk every week for signs of failure. So now even though more data is available from E-Signal I only use it to see if a potential system worked further back than 60 days. Otherwise IMO you risk missing systems which have started working recently. No doubt many people here think it is idiotic to trade evanescent systems, but I don't know any better. My favorite current example (which I give to you for nothing because that's exactly what it is worth) is "Sell Wednesday Morning". Didn't work 3 months ago (when "Buy Wednesday Morning" worked, haha!), but it works now. I don't trade something this stupid, but it makes my point. Re auto-optimization, they obviously have the BT entry screen set up like TradeStation does, but who knows when they'll implement it. I probably wouldn't use it anyway, as I only optimize in price ranges where I know from experience the optima are likely to lie. Re code efficiency, are you doing the obvious stuff, like setting the time format for only the time of day you trade? Making sure that all variables that can be precalculated outside a bar are by calculating at the begginning of "main" or by using global variables? Applying tests to avoid calculations in a bar at times you aren't interested in (e.g., not running trade entry calcs if you are already in a trade)? Cutting down on the number of loops by combining multiple Boolean tests in one "if" statement? How many parms you got? I try to keep mine to 7 max, and optimize the important ones first so that when I get to the soft ones the performance is already broadly optimal and fine tuning them is less important.
Congrats! it looks like you've made terrific progress. On the auto-optimization. I have found that the "better" solution is to parameterize via volatility or some other factor that measures what you want to optimize for. That way you can often eliminate the need for frequent optimization and in many cases eliminate variables. Simple example for a breakout system would be that in stead of looking to buy e.g. .50 pts above a recent high, in stead buy 25% of ATR above recent high. I have found that the "25%" changes much less frequently than, in this case, the 1/2 point i used as an example.