Settings Files - page 3
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
BackTesting GBPUSD
Here is a .set file that has yielded promising results on 2 months of backtesting (Oct-Nov 2006) with by-the-tick data downloaded from GainCapital. Please note that this is a fairly preliminary venture into backtesting, and the results may be improved upon by someone with more experience than I have. Still, I thought I would bring the inputs to the thread's attention for possible feedback. Just rename the .txt file to .set.
Question: can someone provide a detailed description of the meaning of each of the input settings? The old Phoenix manual is rather sketchy on some of the parameters. I'm wondering, for example, if there are trailing stops in modes 2 and 3 (and how to set them).
Boola
Setting
This is just suggestion. Maybe we could keep all BEST settings in one place - like in this attached file.
Fikko, I think you also have settings with TakeProfit=80; StopSell=80 and TrailingStop=35... but I didn't find it. Also I wouldn't know which setting numbers are the latest/best anyway. Would you please check if these are last/best?
MarioHi Mario & Daraknor,
After a few weeks of forward test on single currecy USDJPY, I haven't found good way to do setting.
1. I use 2 months past data, tuning for the best setting (eq. October 1st - December 1st 2006)
2. Applying it for the next following week on Monday (Dec 4) and use it until friday (Dec 8)
3. Then re-calculate the best setting with 2 months data (eq. October 8th - December 8th)
4. Use the new setting for week (11-15 December).
I've done this for a few weeks, but the result is not good. Phoenix can have more than 90% winning in back test, but it couldn't predict future.
The big question is still Historical time frame to be used and when to use it and its lifetime. We need to use setting ahead the market.
One more thing, on slower market, the Take Profit should be adjusted to 30-35 for it to perform. SL should be 60... TP/SL =80/80 wouldn't be good in slow market.
is Phoenix good using real account?
Hello Fikko, Daraknor,
is Phoenix profitable when using demo account only ?
i am testing with demo account at NF broker. the performance of phoenix 5.6.8 is really GOOD.
btw, is it safe if we change the lot with 0.1 and true for micro account. i did, bu t seems not working..
once more..how to change the EA, i just want the EA to execute 1 open trade only..
need ur advice friends..
thanks.
jonisu
Yep, I know the feeling...
Been rather busy (behind the scene this time) the last few weeks doing similar things. Still running 9 different forward tests as we speak but I find it hard to say which strategy works best..
I have done half yearly backtest, 2 monthly backtests etc etc. All with different results. The problem remains that it is hard to tell what works best in the long run. Results of just a few weeks are not telling us a lot I think... On top of that we have the difficulty in choosing a version and stick with this one for a while to see difference in settings (and not just differences caused by version changes...)
Will keep everyone posted,
Vince
Hi Mario & Daraknor,
After a few weeks of forward test on single currecy USDJPY, I haven't found good way to do setting.
1. I use 2 months past data, tuning for the best setting (eq. October 1st - December 1st 2006)
2. Applying it for the next following week on Monday (Dec 4) and use it until friday (Dec 8)
3. Then re-calculate the best setting with 2 months data (eq. October 8th - December 8th)
4. Use the new setting for week (11-15 December).
I've done this for a few weeks, but the result is not good. Phoenix can have more than 90% winning in back test, but it couldn't predict future.
The big question is still Historical time frame to be used and when to use it and its lifetime. We need to use setting ahead the market.
One more thing, on slower market, the Take Profit should be adjusted to 30-35 for it to perform. SL should be 60... TP/SL =80/80 wouldn't be good in slow market.After a few weeks of forward test on single currecy USDJPY, I haven't found good way to do setting.
1. I use 2 months past data, tuning for the best setting (eq. October 1st - December 1st 2006)
2. Applying it for the next following week on Monday (Dec 4) and use it until friday (Dec 8)
3. Then re-calculate the best setting with 2 months data (eq. October 8th - December 8th)
4. Use the new setting for week (11-15 December).
I've done this for a few weeks, but the result is not good. Phoenix can have more than 90% winning in back test, but it couldn't predict future.
The big question is still Historical time frame to be used and when to use it and its lifetime. We need to use setting ahead the market.
One more thing, on slower market, the Take Profit should be adjusted to 30-35 for it to perform. SL should be 60... TP/SL =80/80 wouldn't be good in slow market.Let me cry wolf one more time, Daraknor.
This is what I been saying, your optimization is of no use. If you would of done the testing as seggested you'd be ahead of the game.
Theres a lot of knowledge embedded in the above statments. You all need to open your eyes and look at this a new.
Some of you dont like the way I do things, thats fine. You will not get any strate answers from me, just directions to the answers and more questions that need answering. If your not into doing the testing or work needed to find the answers then what I have to say will not be of any value to you. You can not learn by having some one give you the answers. You out of school now and in the real world.
The CockeyedCowboy
Yep, I know the feeling...
Been rather busy (behind the scene this time) the last few weeks doing similar things. Still running 9 different forward tests as we speak but I find it hard to say which strategy works best..
I have done half yearly backtest, 2 monthly backtests etc etc. All with different results. The problem remains that it is hard to tell what works best in the long run. Results of just a few weeks are not telling us a lot I think... On top of that we have the difficulty in choosing a version and stick with this one for a while to see difference in settings (and not just differences caused by version changes...)
Will keep everyone posted,
VinceVince
Very glad to see that your still at it. The problem is that we are trying to optimize this thing to death. With defferent currencies, timeframs, modes and yes even brokers. If one finds which variables that actually control the out come of this system then the system can be made to have maket dynamic adjustments inbedded in it. I have in the past seggested that several tests be made useing a random number generater. Those test would of showed which variable control performance and which to give little attention to. You and fikko are wise to have done this testing but the work has just stated.
The CockeyeCowboy
Do you not think that any system should be able to work stable in the long run?
When an EA is programmed for market conditions it performs well in those market conditions. When the market leaves those conditions (becomes strongly trending for Phoenix) then it needs to stop trading or only trade in the direction that works (TP with the trend, doing counter-mini-trend on dips). Long term stability is achieved by knowing when to *not* trade.
Question: can someone provide a detailed description of the meaning of each of the input settings?
Very soon this will be my focus. I think I have Mode3 finally fixed in 5.7. I ripped out most of the existing detection code and replaced it with 2 lines that work well. Less flexible, more reliable. If we don't get any bugs today, then I work on settings and documentation.
The old Phoenix manual is rather sketchy on some of the parameters. I'm wondering, for example, if there are trailing stops in modes 2 and 3 (and how to set them).
Trailing Stop is only available in Mode1. Mode2 sets a breakeven SL on Trade2 after Trade1 sort of exists. I haven't debugged it much or seen it in use much. Now that we can achieve similar functionality in Mode3 without changing any code I may make Mode2 something completely different. Mode3 opens 3 trades. When any trade exits, Trade2 and Trade3 have SL=Breakeven. If Trade3 is the last trade (meaning Trade2 has exited) then SL=TP/2. Now that additional detection is made, we could activate Trailing Stops as well. (Maybe as the new Mode2?) If anyone uses Mode2 please let me know.
I've done this for a few weeks, but the result is not good. Phoenix can have more than 90% winning in back test, but it couldn't predict future.
We have seen it done, we just don't know how to do it. Right now the trending market is straining for Phoenix. Another complication was the bugs in Mode3. I am beginning to think we should do all optmization in Mode1 (for reasons explained in the main Phoenix2007 thread). If you optimized Mode3, then you optimized bugs :/ 5.7.0 seems stable for Mode3, I tested 5 times on 3 different data sources for months of data. It looks... different now.
Rather than picking a fixed 2 month time, I'm thinking of setting MaxTrades to 10, setting signal_count to 10, MaxRisk very low and then optimizing that *since market shift*. Every time the market shifts style, treat that as ground zero and reoptimize. Don't include the most recent week of data. Then apply those settings to the most recent week and test for profitability (the top 5 or 10 results). The 6 month rule from Hendrick worked very well because the market had been the same for about 6 months. That is no longer true. I will follow up on this soon.
is Phoenix profitable when using demo account only ?
People are still making money on real accounts with the default settings. We just haven't figured out how to make new better settings yet. My main focus has been fixing bugs. It turns out that many people were optimizing bugs. If the bugs are fixed, I will turn my attention to opimization more.
Vince
Very glad to see that your still at it. The problem is that we are trying to optimize this thing to death. With defferent currencies, timeframs, modes and yes even brokers. If one finds which variables that actually control the out come of this system then the system can be made to have maket dynamic adjustments inbedded in it. I have in the past seggested that several tests be made useing a random number generater. Those test would of showed which variable control performance and which to give little attention to. You and fikko are wise to have done this testing but the work has just stated.
The CockeyeCowboyThose tests with random number generator proved you are full of crap. Say something concrete or stop posting in *every* thread saying the same message of vague doom. No one listens to "Chicken Little."
Let me cry wolf one more time, Daraknor.
Some of you dont like the way I do things, thats fine. You will not get any strate answers from me, just directions to the answers and more questions that need answering. If your not into doing the testing or work needed to find the answers then what I have to say will not be of any value to you. You can not learn by having some one give you the answers. You out of school now and in the real world.
The CockeyedCowboyDaraknor has served all of us with his valuable time. It is not easy to be in his shoes. A standing aplause for him..
I think it is still worth to spend more months with Phoenix.
Yep, I know the feeling...
Been rather busy (behind the scene this time) the last few weeks doing similar things. Still running 9 different forward tests as we speak but I find it hard to say which strategy works best..
I have done half yearly backtest, 2 monthly backtests etc etc. All with different results. The problem remains that it is hard to tell what works best in the long run. Results of just a few weeks are not telling us a lot I think... On top of that we have the difficulty in choosing a version and stick with this one for a while to see difference in settings (and not just differences caused by version changes...)
Will keep everyone posted,
VinceI also have completed forward testing with 2 weeks historical timeframe and also 1 month, both are not good. 2 months time frame is better. 3 months setting will have similar values with 2 months.
I still have to do more forward testing scenarios. No conclusion at this time.