| Pages:
1
..
79
80
81
82
83
..
98 |
goodoboy
Junior Member
Posts: 38
Registered: 27-7-2021
Member Is Offline
|
|
Quote: Originally posted by admin  | @goodoboy Sorry i missed your last post.
1) The trial version of GSB is working with NT, but missing some functionality.
The beta version of NT has much more functionally but not ready yet. Its the number one focus of lead programmer once some critical bugs are fixed for
some other users.
2) GSB works with tick date etc, and I suspect in theory tick data is much better.
In practice tick data is really problematic. Many data providers give only short tick data history. Thats often not usable as the data is too short
and or the sample size too small to make working systems
Trading on a different data provider to your historical data also is very problematic.
The GSB users who I consider experts in this field, most have given up. Not a GSB issue but data feed issues.
Please read this
https://trademaid.info/gsbhelp/RenkoPointKasebars.html
|
Hello Peter,
Great response.
Do you recommend I just use GSB with Tradestation for now until GSB is fully ready for Ninjatrader 8?
I will stay away from tick data and keep life simple, only minute bars with GSB.
Thank you
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Much better for us both to use TS if you are familiar with TS. TS/MC will always be leading in features as we test things on TS/MC, then if we like
the feature later add to NT
|
|
|
Chituan
Junior Member
Posts: 14
Registered: 21-11-2022
Member Is Offline
|
|
Hi Peter, I use TradeStation data window with @CL to download data and compare it to the data in GSB CL.1.Minute.0900-1430_ExchangeTime and see big
difference in the prices. Do you know why ? I attached the screenshot.
Thanks
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Chituan | Hi Peter, I use TradeStation data window with @CL to download data and compare it to the data in GSB CL.1.Minute.0900-1430_ExchangeTime and see big
difference in the prices. Do you know why ? I attached the screenshot.
Thanks |
Hi Chituan,
Continuous contract prices are always adjusted after every contract rollover.
More information here:
https://blog.quantinsti.com/continuous-futures-contract/
Thanks received (2):
+1 BlackBox at 2022-12-25 18:13:54 +1 admin at 2022-12-25 17:09:39
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@Chituan
I should say there are issues with GSB and CL. Its possible all issues fixed in the next build, but not sure yet
One of the issues is CL price <0 and the lowest low GSB assumed was possible in price was zero.
This will cause missmatch.
|
|
|
Chituan
Junior Member
Posts: 14
Registered: 21-11-2022
Member Is Offline
|
|
Hi Carl,
Would it cause issue ? say I am using the default GSB data CL.1.minute as above, then GSB generates trading systems and in these systems, there is an
indicator based on some sort of prices difference from time 1 to time 2 . Then we move to tradestation using @CL, trying to check if tradestation
backtest match with GSB, but continuous futures prices got adjusted based on return (i.e log of prices difference) . I would think this will
invalidate the systems learned by GSB .
Please correct me if I am wrong here.
Thanks
|
|
|
goodoboy
Junior Member
Posts: 38
Registered: 27-7-2021
Member Is Offline
|
|
| Quote: Originally posted by admin | | Much better for us both to use TS if you are familiar with TS. TS/MC will always be leading in features as we test things on TS/MC, then if we like
the feature later add to NT |
Thank you Peter for responding
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Your welcome. there is a thank button below post you can click.
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Chituan |
Hi Carl,
Would it cause issue ? say I am using the default GSB data CL.1.minute as above, then GSB generates trading systems and in these systems, there is an
indicator based on some sort of prices difference from time 1 to time 2 . Then we move to tradestation using @CL, trying to check if tradestation
backtest match with GSB, but continuous futures prices got adjusted based on return (i.e log of prices difference) . I would think this will
invalidate the systems learned by GSB .
Please correct me if I am wrong here.
Thanks |
Hi Chituan,
GSB mainly uses normalized indicators, so the signals generated are the same for both data sets.
However the trade result can vary somewhat.
Overall I found the differences to be so small, they can be ignored.
You can test this yourself:
1. build on the old price data set dated 2020
2. verify a strategy on the recent price data set (GSB left window)
Thanks received (1):
+1 admin at 2022-12-27 00:07:28
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
Latest Systems/Diag and Statistics
I admit to being new to GSB (142 intensive hours) and watching dozens of videos, but certainly not new to trading (60 years).
In a couple of videos, Peter has shown the Latest/diag and Statistics tabs to be populated with very useful information. What do I have to do to
populate those tabs?
What does the verify process do and how is the verify data different from the trade data?
Thanks in advance for the help.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@ChuckNZ
not sure what video you refer to, but this is done via macros, and much easier done by gsb automation
The gold videos on my web site show how this was done in automation, but the nq settings are best to start with
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...
Also verify info. You could mean we verify a system on differnt markets / time frames. (I now have much less faith in this) of you could mean veryify
the method via statistics. Hard for me to reply without more specifics
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
Verification vs. Validation?
I would appreciate if someone can help me to understand the difference between "verification" and "validation". When I set "validation %" in the
optimization section, what data is it using? If it's just using the same data that it's using for the training and testing, why not just make the
testing period bigger?
This is all a very "gray" area for me. Thanks in advance for helping me to understand.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Chuck
its ok to do that, but I no longer user verification / validation like what you propose
You have to have some out of sample data, else you are going to curve fit and find live trading tanks
If you follow the nq methodology, it has multiple years out of sample, with another year ++ still not seen by gsb
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...
When you understand current methodology of testing concepts on 10,000++ systems, you see looking at one system is legacy method.
You can verify a system on other markets or bar intervals. After much research the experienced GSB user had little faith in this working.
The objective proof is make 10,000 systems ->300, take the top 30% by whatever verification method you chose, and see how it compares say to the
bottom 30%
(say in sample 2007 to say 2018in sample) and oos 2017-Feb-28
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
Thanks Peter. I think I finally "got it". To be fair, however, I was working my way through a bunch of old videos lingering around the place.
It's the content of the old videos that prompted my question.
I applied your "latest" technology to six instruments and the number of favourites were typically (300,250,50,30) or similar. In other words, a
diminishing number of strategies as the process worked away.
When I ran 60,000 strategies on NQ, however, I ended up with (300,300,300,300). Is this an indication that I must have done something wrong or were
the strategies simply "too good".
Thanks
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I saw someone else with 300,300,300,300 recently. I dont remeber who, or if it was you
Proof of 2 good is ther eis nearly 2 years oos that gsb has not seen when you add into ts.
i stoped doing videos as they take forever to make, and can be out of date in no time. Hence I do the documatation and thats easy to change
Normally 20,000 systems is fine
you can also look at the stats F I think was the entire out of sample period.
|
|
|
SwedenTrader
Junior Member
Posts: 63
Registered: 3-8-2020
Location: Stockholm, Sweden
Member Is Offline
|
|
I also got 300 in all A,B,C,D recently on NQ and did not find anything strange in settings.Happened several times. However when looking at the
families in D, only about 4 of them had more than a couple of members. So picked only them for going forward with WFO.
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
Performance filters not working for me
For the last ten days or so of intensive testing, something has puzzled me. I set the Minimum Pearson to 0.9 for training and testing periods. I
also set Minimum Profit Factor to 1.8 for those two periods. But, I see hundreds of strategies in the grid that don't even come close to those
minimum requirements. Pearson's, for instance. Some are even negative. Same thing for Profit Factor.
I must have a setting wrong, or I would be seeing others mentioning this.
|
|
|
Daniel UK1
Member
Posts: 470
Registered: 4-6-2019
Member Is Offline
|
|
| Quote: Originally posted by ChuckNZ | For the last ten days or so of intensive testing, something has puzzled me. I set the Minimum Pearson to 0.9 for training and testing periods. I
also set Minimum Profit Factor to 1.8 for those two periods. But, I see hundreds of strategies in the grid that don't even come close to those
minimum requirements. Pearson's, for instance. Some are even negative. Same thing for Profit Factor.
I must have a setting wrong, or I would be seeing others mentioning this. |
I do believe you are doing yourself more harm than good, by setting the filter that high also for the period after build.. you only then end up with
good systems.. what good will that do during your research ? I would focus on understanding whats bad, and try to minimize the bad.. for this you
need to see and understand the bad ... not just filter it away so you dont have to see it...
Just my 2 cents... focus on what is causing the "bad" so you can remove that.. and you end up with something good in the end 
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by ChuckNZ | For the last ten days or so of intensive testing, something has puzzled me. I set the Minimum Pearson to 0.9 for training and testing periods. I
also set Minimum Profit Factor to 1.8 for those two periods. But, I see hundreds of strategies in the grid that don't even come close to those
minimum requirements. Pearson's, for instance. Some are even negative. Same thing for Profit Factor.
I must have a setting wrong, or I would be seeing others mentioning this. |
Are you using the Nth method or IS/OOS/VAL?
Because when you use the Nth method, during the build process GSB is only using the training filter to filter strategies .
Thanks received (1):
+1 admin at 2023-01-05 17:25:38
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
Can anyone enlighten me about what the highlighted area on this chart is telling me? I'm aware that many of you think that verifying with random
ticks is "old school", so just ignore those lines.
Comments about the walkforward graph are welcome along with all comments.
[img] [/img]
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@chuck
this means the wf stability of parameter is 60% for rolling, and 52% for coarse (reasonable)
the faster a system gets stable wf parameters, the better it is. If its a curve fit more likely to jump all over the place and have a low score
I have zero faith in random noise as validation.
You are asking questions which imply your doing your own learning (great if your an expert in trading and GSB), and not going down the path where
years of collective research is clearly laid out in the recent nq methodology. It means you waste lots of your own time with poor outcomes.
the bolleneger bands means the equity curves are close (good)
In all fairness, awesome your able to try new things at a mature age
|
|
|
SwedenTrader
Junior Member
Posts: 63
Registered: 3-8-2020
Location: Stockholm, Sweden
Member Is Offline
|
|
If doing ES with the new NQ methodology, should we change only the data/symbol or also something else in the standard settings for NQ ? Not including
stop level.
Also regarding other markets like energy and metals, have you tried the latest NQ metgodoly there also or are these markets still recommended being
created by the latest methodoly described specific for them ?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@SwedenTrader
Good question. I think change just the symbol. Ideal stop might vary between markets but its not critical and err on the side of big.
I need to publish the NG settings. Im trying to be in holiday mode, but working more hours than normal.
I have done all the NG work, but not tried to build systems.
If you wait till I publish, it will save you massive amounts of work.
A lot of CL work done, but too early to tell results. You must have the next build as significant GSB bug with CL.
GSB wrongly assumed lowest price possible was zero. This is fixed in the next build that I will release.
the macros will be the same for NG, but the inital gsb setup will not be the same
session hours, secondary filter, entry type etc different.
I think GSB works well on NG, which you can also see in the NG systems being sold
Thanks received (1):
+1 SwedenTrader at 2023-01-05 18:33:57
|
|
|
ChuckNZ
Junior Member
Posts: 50
Registered: 22-12-2022
Member Is Offline
|
|
For the record, I followed the directions in excruciating detail in the "recent nq methodology" document. I found the directions and the concept to
be very worthwhile. The fact that I like to see how random data may or may not affect the action shouldn't really offend anyone.
| Quote: Originally posted by admin | @chuck
this means the wf stability of parameter is 60% for rolling, and 52% for coarse (reasonable)
the faster a system gets stable wf parameters, the better it is. If its a curve fit more likely to jump all over the place and have a low score
I have zero faith in random noise as validation.
You are asking questions which imply your doing your own learning (great if your an expert in trading and GSB), and not going down the path where
years of collective research is clearly laid out in the recent nq methodology. It means you waste lots of your own time with poor outcomes.
the bolleneger bands means the equity curves are close (good)
In all fairness, awesome your able to try new things at a mature age |
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
No offense taken. Great to hear you followed the methodology. GSB users often have very varied opinions on some things, and often we can now
objectively confirm concepts that are assumptions traders have made.
Its great to have you on board and asking questions.
|
|
|
| Pages:
1
..
79
80
81
82
83
..
98 |
![[*]](images/davis/default_icon.gif){kind=icon}
