Settings Files - page 7
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
Settings for USDCHF
Daraknor,
I'm not sure I follow all the steps you outlined, but the method appears to be promising. What if you were to backtest on a month of data before the period you used in the testing?
I'm curious to know why you don't use Trailing Stop. Would that not be a more efficient way to secure gains, rather than relying on TP and SL to lock in profits (and losses)?
Actually, perhaps you could clarify something for me. When Trailing Stop is triggered, is it not supposed to switch off both Take Profit and Stop Loss? Or does it only affect Stop Loss?
thanks for all your advice and wisdom
m6m6 I was looking at machines, I priced out a fairly fast dual core AMD64 system for $660-750 USD depending on the processor, using an old graphics card I had laying around. The CPU time is part of the issue of course, but it appears that there is a lot of manual time spent parsing the results and no way to export the data in a usable, time efficient manner. Anything we do with the initial results needs to tested in Phase 2, and the most promising results in Phase 3. I can probably write a script (if nobody else does after a few weeks) converting the "Copy All" page to a series of settings files. These could be imported and used for Phase 2 and 3, but it would still be a good deal of button mashing.
Of course this doesn't sit well with a programming type, so I will try to come up with a "good solution" soon. Including a macro with AutoHotkey will allow some automation of phase2, but the macro would be 100% linked to my exact window layout, desktop size, etc. An uberBot would take some time write, but building it incrementally would be best.
My Athlon 3200 did fairly well, and I plan to get a dual core system like the one I specced out already which should decrease the grinding time for Phase 1. I'm sure a more intelligent settings file can be created for Phase 1, looking over my results and then tweaking the settings file would probably be the biggest advantage we could apply. Some settings were tested with 100 permutations but the same results were achieved by most of the values. Identifying the variables that don't affect trade performance in most situations really is job #1.
I have a programming deadline on an EA I'm writing, so it will be a few days before I take a chunk of time to look at it. There are some obvious edits that can be made if you look at the phase2 data.
Great work Daraknor!
Really thankful to you for sharing.
By the way, just confuse bout the TP and SL for best setting.
Is it 42 or 51 for TP(mode 1) . SL equal to 90..is it right?
I will start to forward test this setting and will post the results later.I included TP as one of the adjustable parameters. SL I set to 90 and didn't tune at all, since I was trying to tune the signal. If you only look at the TP/SL levels, you get NO optimization. The optimizations were all in the actual indicators used to trade with.
Daraknor,
I'm not sure I follow all the steps you outlined, but the method appears to be promising. What if you were to backtest on a month of data before the period you used in the testing?
I'm curious to know why you don't use Trailing Stop. Would that not be a more efficient way to secure gains, rather than relying on TP and SL to lock in profits (and losses)?
Actually, perhaps you could clarify something for me. When Trailing Stop is triggered, is it not supposed to switch off both Take Profit and Stop Loss? Or does it only affect Stop Loss?
thanks for all your advice and wisdomWe're doing a trend reversal system, a Trailing Stop might be ideal theoretically, but I didn't test it. Some changes made in 5.7.2 would make tuning for trailing stop a more viable strategy. (Version unreleased, I wanted to make a few more tweaks to the trailing stop system & break even system first.)
These optimizations were all about the entry signal. I didn't tune SL at all. I tuned TP in chunks of 3 pips as a minor tuned parameter.
Phase 1 optimization should theoretically be configured based on market type.
Phase 2 optimization is ideal when market type in Phase 1 still applies. New optimizations could be performed when the market type changes. Phoenix 7 will automatically do this, alternating between collections of settings when markets change.
Phase 3 optimization is really just a tie breaker. It can be done over months of data. Settings with higher 'general trading performance' outside of the market conditions are (at this time) much more desirable than settings specific to market conditions. When Phoenix 7 comes around in a few months, we would use Phase 3 tick data from other times the current market conditions are in effect.
In other words, right now we're looking for 'generally better' settings. In a few months we will be able to look for 'appropriately better' settings and apply them automatically. After Phoenix 6 is complete, I will be working on finding "what is appropriate" more and more.
I have limited permission to share code between the current EA I'm working on, and Phoenix 6. So I'm actively developing shared code. (The code sharing is generally limited to a template system much larger than Phoenix requires so adding new features to the template will be easy.) I can't really promise Phoenix 7 or features in it, but I know that I'm going to be working on AI based EA systems, and I can probably release some of those features. I plan to make an "EA Breeder AI" in the next 4-9 months, and creating entirely new EA trading strategies would be a matter of CPU time. Right now I'm creating a lot of 'generics' and 'fundamentals' as I go with these features as a long term intent.
Great Work Daraknor
Some guys have all the brains
Exellent work, i will work on understanding the new testing methods. Unfortunatly i am not as smart as you are but i am working on it
Thanks for everything. keep it up.
Doing a Genetic Algorithm walk of the settings file I posted in the optimization thread, here are some early returns. The chopped off section is the default settings for trading hours, when 5.7.2 is released that won't happen anymore.
The results were optimized from Dec2 to Dec30 on USDJPY using the master optimization file set in the optimization thread. I turned on Genetic Algorithm searching to help find the best settings. This is 3000 files tests complete out of 10,496 to test. Possible combinations are 10^18. This *early* random walk found better settings than I have seen before using other methods, including my "one at a time" signal optimization.
All of these settings (and the others that are created) still need testing on a timeframe they weren't optimized for and possibly some 'local maximum' checking as well. By this I mean that we should adjust each of the parameters up one and down one to make sure we don't have a fluke great result. (Curve fitting) These settings are based on 4 weeks of data instead of 6, I wanted the optimization to finish while I was awake. Compared to the 6 week tests of USDJPY I did before, these 4 week tests should have better results.
Once the first pass of Genetic Algo is completed, I will do the Phase 2 and Phase 3 testing. The more promising returns from Phase 3 testing I will do a 'local maximum' search on a few times, to make sure we didn't have a fragile setting. Then we should have some new settings files to use!
I'm chronicling my work so people know how to do this, and can test in parallel with me.
Phoenix 5_7_2 - Current Default Settings
Probablly not too usefull, but anyway, I've put all preffered settings into this Excel file for easy comparing and easy entering them before starting with optimization.
Mario
Probablly not too usefull, but anyway, I've put all preffered settings into this Excel file for easy comparing and easy entering them before starting with optimization. Mario
Now those settings (default) are for a 15min chart?
Thanx,
GK
Now those settings (default) are for a 15min chart?
Sorry for late respond.
Yes, those are default settings for M15 charts. There will be better settings in the future. But you can use these instead looking through the code, when entering numbers before starting with optimization.
Mario
after seeing bertbin's awesome results with the EURJPY pair, i put the original settings thru the "optimizer", and this is what i got out of it.
data taken from metaquotes (1M) and converted to 15M
FXDD demo account
Phoenix 5.7.0
time period for optimizing 08/01/06 - 12/31/06
then i applied the found settings to the time period from 01/01/07 to 02/16/07
the backtesting grew a 10K account by $700 in 6 weeks using 3x0.1 lots. this is great! although only 60% of the trades were profitable the drawdown was very minimal at 4%.
AZBOfin