| stevehopwoodforex.com https://www.stevehopwoodforex.com/phpBB3/ Print view |
|
| Stupid Q re strategy tester inputs https://www.stevehopwoodforex.com/phpBB3/viewtopic.php?t=3233 |
Page 1 of 1 |
| Author: | nonlinear [ Fri Nov 01, 2013 6:51 pm ] |
| Post subject: | Stupid Q re strategy tester inputs |
This is just silly, but here goes. I have been optimizing a few dozen parameters across different correlated pairs to see robust input ranges that appear to work somewhat consistently. When I am looking at the interim optimization results (before the test is fully completed), I only see less than all of the inputs for each pass. What I want to know is there an easy way that I can see all of the inputs for a specific pass before all of the passes are done? I have tried copying and pasting the inputs from the optimization results tab, but that doesn't collect them all. The only way I know of to do this is to set input parameters for the specific pass I am curious about and then look at the inputs in the EA properties -- all after the entire test is done. This is a major PIA because I have to wait for the entire optimization test to complete to check the inputs I might want to test immediately on another instance of Empty4, particularly on tick heavy pairs. There has to be an easier way to just see all of the inputs from one pass before the test is done. |
|
| Author: | SteveHopwood [ Fri Nov 01, 2013 9:47 pm ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
I will make this one single attempt to dissuade you from wasting your time. I will make no further reply here. The reasons why craptesting in crapergy tester is a waste of time are well documented in this forum. Optimisation is great fun, but bears less and less relation to reality the more you do it. My advice is to forget it. Its only value is to curve fitting peddlers of crap commercial EA's specifically written to separate the ignorant and naive from their cash.  |
|
| Author: | garyfritz [ Sat Nov 02, 2013 12:03 am ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
I'm a bigger fan of backtesting than Steve is, but his comments about the craptester in Empty4 are spot on. I don't trust anything it says. Backtesting is very valuable if you have a decent reliable backtester -- but ST in Empty4 ain't it. Also: even with a perfect backtester, optimizing "a few dozen parameters" is just about 100% guaranteed to fail. The more you optimize, the more parameters you tune, the more likely you are custom-fitting a solution to the past price behavior -- and ONLY the **PAST** price behavior. It will fall apart in live forward trading, almost every time. Study any reference on trading system development and they'll agree with me. Simplicity is golden in system development. Simple systems are much more likely to act in the future as they did in your backtest. |
|
| Author: | nonlinear [ Sat Nov 02, 2013 1:40 am ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Appreciate the comments, sincerely. I started back testing to find bugs and ended up trying the optimizer. I am not trying to "optimize" an EA to a specific pair, but I am, however, trying to find input settings ranges that appear to work across multiple correlated pairs, even if those settings are in individual pairs not even close to the "best" parameters for that specific pair, which is always the case it seems. Maybe this still is optimizing? Anyway, my thought was if the EA proves profitable on multiple USD crosses like CHF, EUR, and GBP on forward tests with the same settings, then it has a better chance of long term profitability on my chosen pair going forward. If the Empty4 optimizer is not good for this, then I will go the slow route. I welcome the advice of others who, unlike myself, have been down this road many times before. This process is not easy. Thanks. |
|
| Author: | RedLineFred [ Sat Nov 02, 2013 4:39 am ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Hi Gary, (and others) What is a trader to do? I have an EA that looks promising. I need to prove to myself that its worth risking on my live account. I cant use Empty4. I could run it in demo for a couple of months, but if I need to make changes or correct errors it could be years before its ready. Any advice welcome. thx Fred |
|
| Author: | garyfritz [ Sat Nov 02, 2013 2:04 pm ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Fred, you're absolutely right that forward testing is a poor substitute for proper backtesting, for several reasons: 1. It's slooowww. 95%+ of all system ideas I test don't meet my standards. I don't want to have to spend months testing each one. 2. It might not thoroughly exercise your code. You might have code to handle conditions that don't happen all that often. The Craptester does at least do a decent job of this. 3. The big one: it's too short a test. Many systems perform great for a month or two at a time, but lose money most of the time. Your forward test might hit one of those good windows, and you think you've hit the jackpot. Then you risk real money on it, and it blows up. I don't trade anything unless it performs pretty consistently for at least the last 3-5 years. I want to see how it handled conditions like the 2008 financial crisis. I need to know what to expect when I actually trade it. So what to do? You could use the Craptester, but I don't recommend it. You could use another platform -- Tradestation is pretty good, it's free for forex, and their brokerage is very good. phil_trade uses Forextester, and it looks pretty promising but it's not free. There are a few others. But those all require either learning to program, which some people can't do or don't want to do, or teaming up with someone who can. And people who can program and properly backtest a system are in short supply. Steve is incredibly generous in coding EAs for people, but then you're stuck with Empty4 and the Craptester. nonlinear, the more parameters you select, and the more conditions you add to the system logic, the more likely you are cherry-picking past behavior that won't carry forward. The markets change constantly so you can't expect past behavior to repeat exactly in the future. Your trading logic has to be general enough to handle "about" the same behavior. Testing with the same parameters on multiple pairs is a very good idea, since that makes it much less likely that the tuning is over-fitting to one particular past history. |
|
| Author: | nonlinear [ Sat Nov 02, 2013 3:50 pm ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Gary, Do you know if it is difficult to convert Empty4 to ForexTester format? All these problems with Empty4 just make me want to use Ninjatrader. I would consider tradestation and have an account with them, but I don't like being limited to one broker. |
|
| Author: | phil_trade [ Sat Nov 02, 2013 3:54 pm ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Hi Fred Don't forget that with FT and also Empty4 backtester, you can just ForwardTest your strategy from the past without optimizing it ! That a way to do forward test and avoid curve fitting. Also, good practices for backtesting are 1 - avoid lower timeframe which are very sensitive to data quality 2 - Optimize from a period, let say 2009 - 2012 and then forward test with same parameters on 2013 to see if you're EA well performed on a non seen period If you want to purchase FT (about 150$), you can find on it's WEB Forum, some good developers who can translate for you in Delphi your Mql4 strategy. cheers Philippe |
|
| Author: | RedLineFred [ Sun Nov 03, 2013 12:55 pm ] |
| Post subject: | Re: Stupid Q re strategy tester inputs |
Gary and Philippe, Thanks for the advice. I cant get Tradestation here in Australia - last time I tried they couldn't offer spot FX. I have previously looked at FT but wasn't sure if it was any better than Empty4, and then there is the issue of converting the code. I write my own, and have a fairly good handle of mq4, but have not tried any other language, so it seems my options are somewhat limited. What I am trying is to write my EA to only trade off the open of a one minute bar. My logic still looks at higher time frames for decisions, but does this by counting M1 candles. In this way I am trying to avoid the random nature of the higher timeframe candles, and so improve reliability and repeatability of testing. Any thoughts on this approach? Fred |
|
| All times are UTC | Page 1 of 1 |
|
Powered by phpBB® Forum Software © phpBB Limited |
|