| Pages:
1
..
34
35
36
37
38
..
47 |
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
No, that cant be done as the files gsb writes for ewfo are massive.
You should not have to wf many systems to ewfo
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by portfolioquanttrader2020  | Hi peter
I do not understand how to choose the best of the 300, which are also 300
2) Build 50,000 systems, and allow GSB to choose the top 300, and the best of these 300 systems into Favorites B
Favorites B systems are all the systems that have good out of sample results.
All data before Feb 28 2016 is in sample, except that the indicator selection has only seen 50% of this data. |
You are correct what you say.
FAV b contains the best systems oos
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...
see points 3 4 5 6
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
June 5 2021
https://trademaid.info/gsbhelp/BuildingSP500swingsystems.htm...
https://trademaid.info/forum/viewthread.php?tid=262
At top of file repository data and macros for ES/YM/NQ systems and swing trading ES long only systems
I did not get as far as building the final systems due to distractions on more critical issues.
Thanks received (6):
+1 avatartrader at 2021-06-07 10:23:25 +1 Bruce at 2021-06-05 17:13:45 +1 bartek at 2021-06-05 10:47:00 +1 bizgozcd at 2021-06-05 07:47:42 +1 Daniel UK1 at 2021-06-05 04:12:37 +1 Carl at 2021-06-04 21:01:44
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Next few days I will publish more on long es swing systems.
Here is my system. OOS 20210226
Below the older benchmark system
Original file in repository did not have the data and macros used (by mistake)
fixed now.
Thanks received (3):
+1 Daniel UK1 at 2021-06-09 14:51:10 +1 bizgozcd at 2021-06-09 09:35:40 +1 Carl at 2021-06-09 07:32:50
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Next few days I will publish more on long es swing systems.
Here is my system. OOS 20210226
Below the older benchmark system
Original file in repository did not have the data and macros used (by mistake)
fixed now.
|
Hi Peter,
Thanks for sharing your findings!
The session used is ES.D 30 min 0830-1500 Monday, Tuesday, Wednesday, Thursday, Friday?
So positions over the weekend are possible too?
|
|
|
bartek
Junior Member
Posts: 40
Registered: 26-3-2020
Member Is Offline
|
|
@Peter
How to set SF CloseLessPrevCloseDBPV as non normalized?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Good question. See this.
stock index typically work best with non normalized CloselessPreveClosedBPV.
Note however our more recent default is normalized Closetohighlow3
to use Closetohighlow3 you would use GA
|
|
|
bartek
Junior Member
Posts: 40
Registered: 26-3-2020
Member Is Offline
|
|
| Quote: Originally posted by admin |
Good question. See this.
stock index typically work best with non normalized CloselessPreveClosedBPV.
Note however our more recent default is normalized Closetohighlow3
to use Closetohighlow3 you would use GA |
I used this as on screenshot but got in the script :
vSF = GSB_CloseLessPrevCloseDBpv2(iSFbpv) of Data(iSFData);
vnSF = GSB_Norm4(vSF, 0, 100) of Data(iSFData);
sfResult = vnSF * iSFWeight;
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Carl | | Quote: Originally posted by admin | Next few days I will publish more on long es swing systems.
Here is my system. OOS 20210226
Below the older benchmark system
Original file in repository did not have the data and macros used (by mistake)
fixed now.
|
Hi Peter,
Thanks for sharing your findings!
The session used is ES.D 30 min 0830-1500 Monday, Tuesday, Wednesday, Thursday, Friday?
So positions over the weekend are possible too?
|
correct on all points.
shown here is entry Monday,tue,wed and force exit friday 1500 if not flat
shown here is entry Monday,tue,wed thur and force exit friday 1500 if not flat
shown here is system as is.
zero slippage used here
As these systems can be traded on micros, the risk should be acceptable
not related, but the overnightES system under systems for sale doesn't have the weekend risk.
If it enters after Friday, its Sunday night where the market is always open till exit Monday at day session close.
see https://trademaid.info/gsbhelp/OvernightES.html
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Swing systems docs are finished.
Includes the out of sample results from feb 27 2021 to today.
https://trademaid.info/gsbhelp/BuildingSP500swingsystems.htm...
Thanks received (6):
+1 avatartrader at 2021-06-11 18:43:00 +1 bizgozcd at 2021-06-11 15:29:32 +1 Daniel UK1 at 2021-06-11 11:34:47 +1 Bruce at 2021-06-10 21:54:33 +1 bartek at 2021-06-10 13:40:03 +1 Carl at 2021-06-10 10:57:27
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I used the same settings, but long and short.
Here is wf of all members in the family
Note there is only 15 systems in favouritesB (compared to 100 to 200-fitness dependent for the long only version)
Its possible there is a lot of improvement in the setup used
  
Thanks received (2):
+1 Piet at 2021-06-11 00:12:54 +1 Systemholic at 2021-06-10 22:05:46
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I manually optimize this in ts with ewfo. (good policy I make first system that's very different to anything else I've done)
Observartion. Very easy to wf and this system like the long only system too has inidcators where weight is not needed (redundant)
basically on 2 indicators the real logic was
oversimplied old logic
v1=inidcator1
result=v1^weight1*v2..
new logic
v1=sign(inidcator1)
result=v1*v2..
Trading this live and long now.
note sign returns only 1,0 or -1
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
more updates here on long short es systems
https://trademaid.info/gsbhelp/BuildingSP500LongShortsystems...
|
|
|
REMO755
Member
Posts: 181
Registered: 11-4-2021
Member Is Offline
|
|
| Quote: Originally posted by admin | New methodology video
Info and video at the start of this thread is obsolete now
the two pass method here using green/green applies to all markets.
https://trademaid.info/gsbhelp/Crudeoilfastrevisedmethodolog...
Simulated data, 29,30,31 minute bars.
Simulated data, 26,30,34 minute bars.
Simulated data, 29,30,31 minute bars. ES, IDX, SPX
Until more recently I was not aware that GSB could work on more markets than I originally found. There were 3 keys to this. The first one was to get
the correct session times.
The second was the finer details of the settings, and the 3rd was pure CPU grunt. (Except natural gas)
One of my favorite jokes is a joke about knowledge extrapolation.
Smart people assume that knowledge in one field, amounts to knowledge in another.
GSB users including myself have also done some dangerous knowledge extrapolation.
A doctor, a lawyer, a little boy and a priest were out for a Sunday afternoon flight on a small private plane. Suddenly, the plane developed engine
trouble. In spite of the best efforts of the pilot, the plane started to go down. Finally, the pilot grabbed a parachute, yelled to the passengers
that they had better jump, and then he bailed out. Unfortunately, there were only three parachutes remaining. The doctor grabbed one and said "I'm a
doctor, I save lives, so I must live," and jumped out. The lawyer then said, "I'm a lawyer and lawyers are the smartest people in the world. I deserve
to live." He also grabbed a parachute and jumped. The priest looked at the little boy and said, "My son, I've lived a long and full life. You are
young and have your whole life ahead of you. Take the last parachute and live in peace." The little boy handed the parachute back to the priest and
said, "Not to worry, Father. The 'smartest man in the world' just took off with my back pack
source: http://www.jokes4us.com/morekiddiejokes/alawyeronaplanejoke....
|
It's very good, I laughed this morning !! 
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
This has small but significant updates.
https://trademaid.info/gsbhelp/BuildingSP500LongShortsystems...
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
As per the most recent updates to the method on Gold and Indexes, I just wanted to check - is the best practice now for finding the best indicator to
use 3 indicators again in the builds for the indicator testing versus 2 as per the CL methodology? Or is this something that could still be variable
per market? Or is either acceptable for the purposes of generating a potentially greater variance of systems?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | | As per the most recent updates to the method on Gold and Indexes, I just wanted to check - is the best practice now for finding the best indicator to
use 3 indicators again in the builds for the indicator testing versus 2 as per the CL methodology? Or is this something that could still be variable
per market? Or is either acceptable for the purposes of generating a potentially greater variance of systems? |
Good question. I suspect 3 is best on all markets.
Its certainly faster as there are too few combinations with 2 inidcators, especially if you are using entry aic (any indicator cross) which has
weights of zero
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I hope to release compare3 entry type in less 2 weeks.
Its basically compare2 with weights and entry level set to zero.
see here https://trademaid.info/gsbhelp/BuildingSP500swingsystems.htm...
and scroll down to June 17 2021 update.
I suspect compare3 entry type works better than other entry types in some circumstances. This will be in the next build of GSB
Compare3 is compare2 with weights and entrylevel all set to zero.
So we go long when
result=inidcator1*inidcator2*indicator3
if result Crosses result[1] then buy...
Indicators (Which are normalized) as far as we are concerned are wither below zero, zero or > zero.
|
|
|
avatartrader
Junior Member
Posts: 60
Registered: 1-10-2018
Member Is Offline
Mood: No Mood
|
|
I had a question about the use of the "Filters" in light of the settings supplied in the new macros for index systems: in the latest version of the
methodology, I am assuming the new best practice based on the research is to enable the CloseLessPrevCloseD in the "Filters" section now, and then set
the SF to Genetic Algorithm?
I am looking at the EL code of generated systems and am wondering what the relationship is, if any, between the "Filter" and "Tertiary Filters"? For
example, in an NQ system I am looking at, the TF filter decision is set to CloseLess... even though there are no TF's enabled in the indicator list.
So, if I am understanding it correctly, the way the TF works is:
1. The number of TF's is evaluated, and if greater than or equal to 1, a TF is evaluated for use based on what is enabled or forced in the "Filters"
section (CloseLess... is the non-normalized that was previously set via SF)
2. If IndicatorLevelCompare is enabled, then the indicators that are enabled for TF (if any) are evaluated as part of this - or is this not correct
and they are evaluated regardless if the number is >= 1 for "# of Tertiary Filters" and the indicator is enabled for evaluation in the TF in the
indicator list?
|
|
|
bartek
Junior Member
Posts: 40
Registered: 26-3-2020
Member Is Offline
|
|
@Peter
Don't you think interesting correlation for research for finding the best fitness criteria? Am interesting how far it will go for more systems...
Example of VBasePlusMinusMedium for ES.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by avatartrader | I had a question about the use of the "Filters" in light of the settings supplied in the new macros for index systems: in the latest version of the
methodology, I am assuming the new best practice based on the research is to enable the CloseLessPrevCloseD in the "Filters" section now, and then set
the SF to Genetic Algorithm?
I am looking at the EL code of generated systems and am wondering what the relationship is, if any, between the "Filter" and "Tertiary Filters"? For
example, in an NQ system I am looking at, the TF filter decision is set to CloseLess... even though there are no TF's enabled in the indicator list.
So, if I am understanding it correctly, the way the TF works is:
1. The number of TF's is evaluated, and if greater than or equal to 1, a TF is evaluated for use based on what is enabled or forced in the "Filters"
section (CloseLess... is the non-normalized that was previously set via SF)
2. If IndicatorLevelCompare is enabled, then the indicators that are enabled for TF (if any) are evaluated as part of this - or is this not correct
and they are evaluated regardless if the number is >= 1 for "# of Tertiary Filters" and the indicator is enabled for evaluation in the TF in the
indicator list? |
Filters are not normalized (forced to range of -100 to 100)
TF are normalzied. This can give very different results.
Best practise depends on the market.
for example es/ym/nq we are used to use SF closelessPrevClosedbpv not normalized
Now we use sf SF Closetohighlow3 with filter closelessPrevClosedbpv
#ofTertiaryfilters must be set to >0 for filters to work.
2) If I understand this correctly, you are correct.
Very deep and important questions.
Whats in filters will expand greatly in time. ie time masks, pattern filters etc
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by bartek | @Peter
Don't you think interesting correlation for research for finding the best fitness criteria? Am interesting how far it will go for more systems...
Example of VBasePlusMinusMedium for ES.
|
Hi Bartek
to be objective its best to use GBS automation with 4 identical tests and see the ave bcd stats and favb stats. I have no interest in things like
pass-score myself(PS)
ewfo can also batch test numerous fitness types as well. VBasePlusMinusMedium is rarely the best, but I thought it was a good idea. (Fitness relative
to current volatility)
|
|
|
REMO755
Member
Posts: 181
Registered: 11-4-2021
Member Is Offline
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by avatartrader | I had a question about the use of the "Filters" in light of the settings supplied in the new macros for index systems: in the latest version of the
methodology, I am assuming the new best practice based on the research is to enable the CloseLessPrevCloseD in the "Filters" section now, and then set
the SF to Genetic Algorithm?
I am looking at the EL code of generated systems and am wondering what the relationship is, if any, between the "Filter" and "Tertiary Filters"? For
example, in an NQ system I am looking at, the TF filter decision is set to CloseLess... even though there are no TF's enabled in the indicator list.
So, if I am understanding it correctly, the way the TF works is:
1. The number of TF's is evaluated, and if greater than or equal to 1, a TF is evaluated for use based on what is enabled or forced in the "Filters"
section (CloseLess... is the non-normalized that was previously set via SF)
2. If IndicatorLevelCompare is enabled, then the indicators that are enabled for TF (if any) are evaluated as part of this - or is this not correct
and they are evaluated regardless if the number is >= 1 for "# of Tertiary Filters" and the indicator is enabled for evaluation in the TF in the
indicator list? |
Filters are not normalized (forced to range of -100 to 100)
TF are normalzied. This can give very different results.
Best practise depends on the market.
for example es/ym/nq we are used to use SF closelessPrevClosedbpv not normalized
Now we use sf SF Closetohighlow3 with filter closelessPrevClosedbpv
#ofTertiaryfilters must be set to >0 for filters to work.
2) If I understand this correctly, you are correct.
Very deep and important questions.
Whats in filters will expand greatly in time. ie time masks, pattern filters etc |
Hello,
1) When # of Tertiary = 1 and non-normalized S.F becomes normalized T.F, correct Peter?
2) If CloseLessPrevCloseD (enabled) and tertiary #of = 0, is it the same as flagLevelCompare (enabled)?
3) S.F input mode, sometimes Compare1 other times Compare2, where is this handled, is it automatic?
4) Could you explain what Compare1 or Compare2 does?
|
|
|
REMO755
Member
Posts: 181
Registered: 11-4-2021
Member Is Offline
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by avatartrader | I had a question about the use of the "Filters" in light of the settings supplied in the new macros for index systems: in the latest version of the
methodology, I am assuming the new best practice based on the research is to enable the CloseLessPrevCloseD in the "Filters" section now, and then set
the SF to Genetic Algorithm?
I am looking at the EL code of generated systems and am wondering what the relationship is, if any, between the "Filter" and "Tertiary Filters"? For
example, in an NQ system I am looking at, the TF filter decision is set to CloseLess... even though there are no TF's enabled in the indicator list.
So, if I am understanding it correctly, the way the TF works is:
1. The number of TF's is evaluated, and if greater than or equal to 1, a TF is evaluated for use based on what is enabled or forced in the "Filters"
section (CloseLess... is the non-normalized that was previously set via SF)
2. If IndicatorLevelCompare is enabled, then the indicators that are enabled for TF (if any) are evaluated as part of this - or is this not correct
and they are evaluated regardless if the number is >= 1 for "# of Tertiary Filters" and the indicator is enabled for evaluation in the TF in the
indicator list? |
Filters are not normalized (forced to range of -100 to 100)
TF are normalzied. This can give very different results.
Best practise depends on the market.
for example es/ym/nq we are used to use SF closelessPrevClosedbpv not normalized
Now we use sf SF Closetohighlow3 with filter closelessPrevClosedbpv
#ofTertiaryfilters must be set to >0 for filters to work.
2) If I understand this correctly, you are correct.
Very deep and important questions.
Whats in filters will expand greatly in time. ie time masks, pattern filters etc |
Hello,
1) When # of Tertiary = 1 and non-normalized S.F becomes normalized T.F, correct Peter?
2) If CloseLessPrevCloseD (enabled) and tertiary #of = 0, is it the same as flagLevelCompare (enabled)?
3) S.F input mode, sometimes Compare1 other times Compare2, where is this handled, is it automatic?
4) Could you explain what Compare1 or Compare2 does?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@remo755
Your questions are very good and show you have understanding of how GSB is working
1) im not clear.
you need this setup. see picture
2) Correct 
3) This is chosen genetically.
4) compare1
if result >0 then trade// Never seen this work well in practise.
compare2
result=indicat*weight*....
if result CrossOver zero.
compare3 {most recent build}
result=indicat{*weight}*.... //weights removed
if result CrossOver zero.
|
|
|
| Pages:
1
..
34
35
36
37
38
..
47 |
![[*]](images/davis/default_icon.gif){kind=icon}
