| Pages:
1
2 |
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
Promising EC Build
For years I've been building and trading the EC with pretty much breakeven performance, nothing stellar I can assure you so the EC has always been a
challenging one for me. Recently I've been working with the new system build process (as Peters been sharing) using the EC.30min data1 and EC.120min
data2, 2am - 1400 CET. There are a few subtle differences the build which I'll share here.
I did the training built using dates for a training period (20090101 - 2015631) and a validation period (20150701 - 20171231). I was attempting to
filter a better quality of OOS systems in the validation period. The hypothesis is to build the 50k systems and get a better average quality of
systems as opposed to using 100% training only and getting a lot of poorer quality OOS results within the validation period that pollute the total
build.
Trading times only 730am - 1300 MOC=1400
I left the period 20171231 - 20190301 completely unseen and with nth to all I did a final filter for the top 150 systems on NP/DD and put them to a
final WF across ALL Dates.
This has delivered some great looking systems, a sample of which can be seen here. I'm sure that better systems can be built given this was a new
build with a new methodology. I hope that makes sense, more to come I'm sure...
Thanks received (1):
+1 cotila1 at 2023-11-24 10:17:03
|
|
|
admin
|
Thread Moved
26-4-2019 at 10:21 PM |
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC?
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by Carl  | Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC? |
Correct Carl. I've moved this to TS to see how it runs over the next few months before putting live, that'll happen as long as the equity curve
remains above it's average or I build something better! 
|
|
|
cotila1
Junior Member
Posts: 78
Registered: 8-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Bruce | | Quote: Originally posted by Carl | Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC? |
Correct Carl. I've moved this to TS to see how it runs over the next few months before putting live, that'll happen as long as the equity curve
remains above it's average or I build something better! |
Well Done Bruce!
Do you mind to share the performance report - equity curve?
thanks
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cotila1 | | Quote: Originally posted by Bruce | | Quote: Originally posted by Carl | Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC? |
Correct Carl. I've moved this to TS to see how it runs over the next few months before putting live, that'll happen as long as the equity curve
remains above it's average or I build something better! |
Well Done Bruce!
Do you mind to share the performance report - equity curve?
thanks |
Thanks. Attached is the TS report of the EC system as GSB generated. There's a bit that could be done to tweak this with entries and exits.
Attachment: Login to view the details
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Bruce | | Quote: Originally posted by cotila1 | | Quote: Originally posted by Bruce | | Quote: Originally posted by Carl | Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC? |
Correct Carl. I've moved this to TS to see how it runs over the next few months before putting live, that'll happen as long as the equity curve
remains above it's average or I build something better! |
Well Done Bruce!
Do you mind to share the performance report - equity curve?
thanks |
Thanks. Attached is the TS report of the EC system as GSB generated. There's a bit that could be done to tweak this with entries and exits.
|
I just finished up a build using the EC 30 and 120 and the same session and trading times using the most recent version of the methodology (+ as the
only operator along with prescribed indicator weights and parameters) and wound up with dozens of great systems as well even after WF and looking at
the last 1+ year as totally unseen data.
Attached is a sample MC performance report of 2 of the systems that I combined together, each trading 1 contract (data was IQFeed continuous
backadjusted).
Very healthy metrics for an intraday system even after factoring in transaction costs.
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi avatartrader / Bruce
You both done some great work.
Have either of you done market validation, and compared with and without data2 60 minute, or varied the session times?
Were the session times found using looking where the range and volume are?
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | | Quote: Originally posted by Bruce | | Quote: Originally posted by cotila1 | | Quote: Originally posted by Bruce | | Quote: Originally posted by Carl | Hi Bruce,
Thanks for posting the details.
I suppose it's your goal to build day trading strategies on EC? |
Correct Carl. I've moved this to TS to see how it runs over the next few months before putting live, that'll happen as long as the equity curve
remains above it's average or I build something better! |
Well Done Bruce!
Do you mind to share the performance report - equity curve?
thanks |
Thanks. Attached is the TS report of the EC system as GSB generated. There's a bit that could be done to tweak this with entries and exits.
|
I just finished up a build using the EC 30 and 120 and the same session and trading times using the most recent version of the methodology (+ as the
only operator along with prescribed indicator weights and parameters) and wound up with dozens of great systems as well even after WF and looking at
the last 1+ year as totally unseen data.
Attached is a sample MC performance report of 2 of the systems that I combined together, each trading 1 contract (data was IQFeed continuous
backadjusted).
Very healthy metrics for an intraday system even after factoring in transaction costs.
|
avatartrader
This is a great result, well done! Certainly warrants further testing. I've been using the + / new weightings methodology with SI / GC and also
getting numerous systems that have great metrics.
Now how to narrow the final selection to just a few to put to live
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Hi avatartrader / Bruce
You both done some great work.
Have either of you done market validation, and compared with and without data2 60 minute, or varied the session times?
Were the session times found using looking where the range and volume are? |
I've been thinking about this and if we had the ability to 'slide' the start time by the number of bars or a 'window of time' as part of the testing,
that would be interesting, also it could be used to remove some of the gap effect between sessions.
IMO, the recent + and weighting changes have had a noticeable impact on the quality of the system builds.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Your idea has merit Bruce, but then we would need to do a stats on the best performing systems, and see what session time was most common.
However you didnt answer my original question.
Great to hear that + works well. I still dont know if we should WF the weights if we build systems with fixed weights.
Time will tell, more research and a lot of cpu time needed
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Hi avatartrader / Bruce
You both done some great work.
Have either of you done market validation, and compared with and without data2 60 minute, or varied the session times?
Were the session times found using looking where the range and volume are? |
No, I haven't got to do any validation yet.
Yes, I first looked at the range and volume to select the session, I did an initial build on one data stream using 29,30,31. Results were okay but not
stellar. Adding the Data2 made a significant improvement to the build results.
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Bruce | | Quote: Originally posted by admin | Hi avatartrader / Bruce
You both done some great work.
Have either of you done market validation, and compared with and without data2 60 minute, or varied the session times?
Were the session times found using looking where the range and volume are? |
No, I haven't got to do any validation yet.
Yes, I first looked at the range and volume to select the session, I did an initial build on one data stream using 29,30,31. Results were okay but not
stellar. Adding the Data2 made a significant improvement to the build results. |
I have not compared a similar study in GSB without data2 yet (e.g. just EC30) - however, I can tell you that based on the initial study that many of
the best performing strategies do use a combination of indicators from both data1 and data2.
For the session times, I confirmed and used them based on a combination of range/volume and prior systems development and experience derived from my
relationships with and first-hand knowledge of the business practices and cycles of large importers/exporters, bankers and institutional money
managers - things that generally can be exploited in spot FX, but can be adapted for futures - in other words, I agreed with Bruce, for perhaps the
same + additional fundamental and/or practical or experiential reasons.
You can vary the sessions a bit either way for testing keeping in mind the business driver behind them, but for the Euro and most other currencies
related to the EU and quoted in or based on USD, testing and working between the EU and US sessions like that is a good choice and has a strong
logical and statistical basis.
I did mean to ask about market validation however - I have gone through the relevant videos and documentation multiple times, and as of my most recent
builds, have been first testing to determine the top 1/3 of indicators, etc before I do any actual builds on a given instrument.
However, in this case, the top indicators and subsequent studies were determined using the 30/120. From your perspective and most recent research,
what do you mean by market validation? Do you mean validating on a single market/time frame strictly and then testing and building from there?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi avatartrader
I mean follow this approach, and each time you make a tweak, Ie session times, or time in that session, redo the steps to see what your stats are
like.
https://trademaid.info/gsbhelp/Provingthemethodology.html
IE see the first picture
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
I did follow that procedure, but began with the EC 30 and 120 and the session times in question as my starting point. The results were so good that I
then proceeded to at least run a real build with the resulting best indicators just due to time constraints and wanting to get in some repetitions of
running through the updated methodology end to end.
Now that I pretty much have it down, I am going to begin doing some more foundational benchmarks to begin working from, especially because I would
like to also begin exploring building some swing systems.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader |
I did follow that procedure, but began with the EC 30 and 120 and the session times in question as my starting point. The results were so good that I
then proceeded to at least run a real build with the resulting best indicators just due to time constraints and wanting to get in some repetitions of
running through the updated methodology end to end.
Now that I pretty much have it down, I am going to begin doing some more foundational benchmarks to begin working from, especially because I would
like to also begin exploring building some swing systems. |
Keep us posted on progress. There are now emini's on some markets that have 20% the leverage. Not sure where I saw this, but that might suit some
people for over night systems which do have more risk
|
|
|
Daniel UK1
Member
Posts: 470
Registered: 4-6-2019
Member Is Offline
|
|
Just wanted to follow up on this thread.. i have been doing some dev work over the past week on EC, and i have gone trough about 10- 15 runs of
building about 20k systems in each run, however despite multiple variations, doing nth 1/1, 1/80, 1/160.. testing all second filters, 15, 30, 60 and
120 min bars, using latest methology settings... using dates 1900-2015 to build on.. i can not get below around 70% degradation in stats... its
impossible.. for me... is there a tweak to EC that i am missing, or is the high degradation not to be concerned of for this market? would be
great to hear any experience about EC and what settings has been used to get better results than myself, i have run out of ideas to try.. What
degration number is reasonable and acceptable for a market?
|
|
|
waldocktrades
Junior Member
Posts: 14
Registered: 12-7-2018
Member Is Offline
Mood: No Mood
|
|
I'm having a similar situation with 30yr T-Bonds. One thing I've discussed with Peter is the notion that we're looking for a consistent algorithm.
Degradation may be poor as a whole but, there may still be algos within the set that perform consistently both in and out of sample. However, when we
see bad degradation numbers on the sample, we delete them and move on.
I've also found that date ranges make a huge difference. I'm uncomfortable with small timeframes because I don't want to constantly walk-forward and
reconfigure portfolios based on current settings. I prefer a long timeline that covers multiple economic cycles. I create my data streams back to the
90's using the old pit trading sessions as a starting point.
Best, Andy.
|
|
|
saycem
Junior Member
Posts: 48
Registered: 13-7-2018
Member Is Offline
Mood: No Mood
|
|
I've also been wondering about this of late. Am I throwing too many babies out with the bath water!? A poor degradation score tells me that on my
current settings I am not producing a high proportion of systems with good oos results. It does not mean that allthe
systems are bad does it? What if there was a few gems in the build that could work really well - could it be possible they perform better than systems
built with better degradation settings. Is it true that some of the best edges can come from some quite illogical settings?
I like the idea of testing degradation as it improves your odds of selecting a good system, but it doesn't guarantee you pick the right one. I still
feel the selection process you use determines whether you pick one of the systems that will work or not. I also wonder if we are all working to find
the best degradation scores aren't we ultimately all heading towards building the same systems with the same settings?
In the end, does it really matter how you get there or just that you pick the most robust system with good performance? If the latter is true then the
method to separate the good from the bad regardless of how many bad there are would be most important wouldn't it?
What if we had a standardised scoring system to determine the robustness of a system individually. Then users can share system scores and the settings
they used in forum for others to replicate/improve on? Could this take some of the heavy lifting off Peter and get people trading GSB systems quicker?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Im going to publish my settings, and supply the config files and macros in the private forum for ES in about a week.
The standardized testing is going to be an excellent idea.
All my focus has been on ES, and im happy with the results.
The soybeans market had high degradation, but the out of sample results were fine. They are also developed > 1 year ago where we didnt have the
robustness features we have now.
As for will we make the same systems? Parameters will tend to vary, and there is still lots of scope to vary from the defaults.
We also will add to indicator library and will be adding pattern filters + more over time.
|
|
|
Daniel UK1
Member
Posts: 470
Registered: 4-6-2019
Member Is Offline
|
|
Many thaks Peter for alwasy helping out and now sharing the ES setting and macros etc.. very helpfull..
As in my original question, its not just about EC, but it concerns also for example GC, s, etc, i find some markets among these to have quite high
degradation when i do m y market validation for the whole build... i test and try for days, and try all king of variations, however as in my example
above EC, i can not, get under 70% degradation, impossible for me.. what i am wondering about is... is this to high to settle for.. would this be
acceptable? what is acceptable for others here to build systems on and pick from ?
I also find that, i see lots of systems in my builds, that when i remove all with VSS below -10 from a build as you recommend and then pick what i
like to WF, i remove in general all the "best" systems, highest PF, highest NP, highest Pearson etc... and i find in general that the systems left is
ok, but not great..
Is this just the "cost" of getting better robustness in the system, or is it just me that is thinking incorrectly here.
Comments on both my questions would be greatly appreciated.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Daniel,
I think high degradation is ok, but not ideal. Bruce is going to revist the market and do more testing with what we have learnt on ES.
Currently I can only comment on ES, as its what I have spend my time for months refining.
Are you using 29,30,31 min bars, but look at the degradation on 30 min only?
What is your secondary filter? GA, closed-close and or closed/close ?
It would be good to compare how the final systems metrics were compared to Bruce's published one.
|
|
|
Daniel UK1
Member
Posts: 470
Registered: 4-6-2019
Member Is Offline
|
|
Many thanks for commenting Peter,
So, how i go about things, first i test all kind of larger variations for a market to see which direction i shall go , and do larger main,changes,
such as bar size, 15, 30 etc, test data 2, SF, larger tweaks, and i then check the stats with the market validation macro to see degradation between
IS and OOS, when i am ok with my larger tweaks, and have gotten down the degradation score to something acceptable, i then reduce the indicators to
what is among the top systems have used, i then after this, make very small tweaks, one at a time to see if i can improve my degradation scores and
stats in general, when i am happy i start to build systems..
When i reach this stage, i reduce all systems with anything below -10 in VSS, and require VS to validate on all.. from these systems left i pick the
top 250, based on PF, pearso, NP and NP/DD.. i then WF all, and pick from there..
This is how i go about it, and believe its within the methology?
... however i am wondering about how i shall think when it comes to degradation in my validation for the market such as EC or GC since i cant follow
the above due to the fact that i cant get below 70% degradation for the market validation and most of the systems that is great or even good, has
individual VSS numbers below -30 .. what is acceptable for individual systems as VSS and where shall i be ok and where shall i accept that market
validation is not working to build on or to risky?
Thanks for sharing
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Daniel,
let me see your setup via teamviewer early tomorrow your time. I had similar issues today with another user that got resolved by doing some mild
tweaks.
|
|
|
waldocktrades
Junior Member
Posts: 14
Registered: 12-7-2018
Member Is Offline
Mood: No Mood
|
|
I've noticed that the most consistent verification scores and most stable out of sample performance occur when the system selected closely matches the
performance characteristics of the population initially used to determine degradation.
Logically, this makes sense. However, it also revisits, "throwing the baby out with the bathwater." Systems selected near the mean exchange robustness
for sexy performance metrics.
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
Just been revisiting EC again given comments above and the huge advances with GSB and Peters methodology since I did my original post on this thread.
There are now so many choices with weight mode, sign mode, entry levels, etc. there's a lot more to work through ( I hope the AI will help with that).
Just wanted to show some initial results from Opt Price EC30, 5 and 10 tick rand #1, verify with 29, 31, 5 and 10 tick rand #2, WF just with 30min.
Still a lot more to do. I think there's a possible issue generating enough OOS trades with the GSB metrics, GA with the price series data given EC
prices are 1.xxx, need Peters input on that one.
|
|
|
| Pages:
1
2 |
![[*]](images/davis/default_icon.gif){kind=icon}
