| Pages:
1
..
44
45
46
47 |
thowoc213
Junior Member
Posts: 39
Registered: 22-8-2021
Location: Poland
Member Is Offline
|
|
A question
Hi Peter,
I do agree that there is a lot of things going on.
Perhaps too much to keep up with your pace.
I’ve been away for a quite a while and I’m both impressed and terrified.
We have a great tool and great methodology. The trouble is the constant evolution that is not supported in user friendly way.
May I suggest a very simple structure for everything which has been released and for the future changes and updates?
I think it would do the trick. It wouldn’t create much additional job for you.
Would the other GSB users support my proposal? I’m keen to know.
Thomas
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
@thowoc213
your points are very valid & I totally agree.
The work involved however is what I spend about 30% of my time on. Its massively time consuming, however that depends on
how much you want to improve things. You hit that point of diminishing returns, but also get break through s from time to time
Here is what happens
we find something that works on a market. current example CL. Often thats a GSB user figuring out a new way to use GSB on a specific market
The GSB community is fantastic as there is collectively a great deal of knowledge. The user base input improves GSB for the benefit of all
we then improve it. Often new features in GSB are made to get these improvements.
We then publish settings
Then we revisit another market, with what we have learnt on the last market + new GSB features, and apply to another market.
IE whats been done on CL, highly likely as a concept will apply to natgas
I am days away from major CL update
The current Tuesday Long Short setting were published about 3 weeks ago, but there is significant improvements
Whats most recent, with no improvements + going well is the NQ/ES long only swing systems
The same setup also I think works on day session only data to. Proof of that is one of the oldest gsb systems
https://trademaid.info/gsbhelp/GSB_AG_swingAG101ES.html
Example of new features released recently is, extensive day of week combinations in GSB automation.
New forms of atr indicators (wilder atr) (in GSB but hidden short term)
atr filter (nearly finished)
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Here is the CL work,
the docs could be more polished, and users are welcome to add comments, clarify and document this further.
Enjoy. Its been about 2 to 3 weeks work
Attachment: Login to view the details
Thanks received (5):
+1 cotila1 at 2024-10-25 10:46:03 +1 TradingPrice at 2024-06-20 20:27:27 +1 JozefSusko at 2024-06-19 07:54:31 +1 Ketil at 2024-06-19 04:50:58 +1 Carl at 2024-06-19 02:00:17
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
This was my first system from my results. It will be for sale soon
the WF is as good as WF ever get
Thanks received (2):
+1 Bruce at 2024-06-19 15:50:32 +1 Carl at 2024-06-19 01:59:55
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
There is a mistake in the data file uploaded in the last settings
the file cl.1.minute.2007_20240515-tick-800-1430exchangeTimeB last date is 20230515 not 20240515
Next release I will call it cl.1.minute.2007_20240515-tick-800-1430exchangeTimeC
|
|
|
TradingPrice
Junior Member
Posts: 21
Registered: 4-8-2022
Member Is Offline
|
|
@admin
Amazing work on that Excel sheet, thanks
I'd like to suggest if we all could work on a methodology to discover new market, for example:
1- find base session to work with + CloseLessPrevCloseDBpv as SF
2- Entry types
3- Secondary Filters
4- Filters
5- Sessions
6- Trading times
7- Day of week
8- Bar intervals
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Good subject Tradingprice
Remember also that for stock indices CloseLessPrevCloseDBpv as SF needs to be not normalized,
while for most other markets it needs to be normalized.
I have some ideas, next time we chat will discuss
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
more results on CL to come in a week or so.
I found one mild improvement
Also tried the concept of, turn on many things at once
honing in using automation, i can get 200+ systems that end up in favourites B using only 2500 systems
using turn on almost the lot, out of 200,000 systems, I am getting zero systems in Favourite B
But the question which relates to the comments above is, are there clues in 200k systems as to what will work in a market you have no idea on what
works?
I observed a few systems with day of week choosen genetically had wednesday only as true. Didn't get much time to investigate more.
This highlights why im damming on the concept of turn everything on and expect to find systems, rather than get a benchmark and change one setting at
a time.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Updated CL testing. Big picture summary updated and secondary filter info
Update to best bar interval
Likely I will stop here.
Hope you enjoy build systems. They should be fast to make them with this info
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
so much is happening
Here is some info on filters vs secondary filter
lets say a secondary filter using atr is used
high atr we would go long, low atr we go short. You would expect that to be the wrong way to use atr. (fine for rsi etc)
but atr as a filter
high atr we can trade long and short, low atr no trade
related to this
we normally (but not always) use cross over of the primary filters
There are filters however that have cross.
In practice the cross of a filter and primary indicator will almost never occur at the same time.
However a filter with cross over and compare 1 on indicators, works, and works brilliantly on the market im testing it on. So far its the best of all
combinations
compare 1 is if indicator>0 then buy and if indicator < 0 then sell
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Below are tips for somewhat competent ts / mc users, and not for novices.
The last month, I have spent much of my time building systems, and pulling each module apart to see whats redundant / weak. I can say I have learnt a
great deal.
For experimenters, you can optimize any system that uses a variant of accdist. GSB has many of these, and not all are released to end users.
Im interested in feedback if anyone wants to put this code in an existing system that uses some sort of accumdist aleady.
In the one market I tested this in, one was ahead a long way
here is the code
inputs:typeAcc(1);
vars:v1(0);// dont define this twice
Switch(typeAcc)
Begin
Case 1:
v1 = GSB_AccumDist(Ticks) of Data1;
Case 2:
v1 = GSB_AccumDistClosev2(Ticks) of Data1;
Case 3:
v1 = GSB_AccumDistCloseCUpDnv2 of Data1;
Case 4:
v1 = GSB_AccumDistCloseCUplessDn of Data1;
Case 5:
v1 = GSB_AccumDistCloseOUpDnv2 of Data1;
Case 6:
v1 = GSB_AccumDistCloseOUplessDn of Data1;
Case 7:
v1 = GSB_AccumDistCloseUpDnO of Data1;
Case 8:
v1 = GSB_AccumDistCUplessDnv2 of Data1;
Case 9:
v1 = GSB_AccumDistDR(Ticks) of Data1;
Case 10:
v1 = GSB_AccumDistDR0v2(Ticks) of Data1;
Case 11:
v1 = GSB_AccumDistMomv2(Ticks) of Data1;
Case 12:
v1 = GSB_AccumDistMultCloseOUpDnv2 of Data1;
Case 13:
v1 = GSB_AccumDistNoVolumev2 of Data1;
Case 14:
v1 = GSB_AccumDistOUpDnv2 of Data1;
Case 15:
v1 = GSB_AccumDistOUplessDn of Data1;
Case 16:
v1 = GSB_AccumDistUpDnv2 of Data1;
Case 17:
v1 = GSB_AccumDistOMomv2(Ticks) of Data1;
Case 18:
v1 = GSB_AccumDistMomCloseOv2(Ticks) of Data1;
Case 19:
v1 = GSB_AccumDistOUplessDnv2 of Data1;
end;
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
test 1 is the gsb equivalent of ts accumdist(ticks) and test 20 is ts accumdist(ticks)
so we can see one test that massivly better, and another 2 thats greatly better.
This system used was a 3 inidcator system, with the other two indicators removed
for aic (any indicator cross systems) you can just comment the line, and you will see the results of jsut the one inidcator.
//v2=inidcatorx; {the // is a comment which removes the line}
//v3=inidcatory;
for entry type cross you can change this line
result = Sign(Sign(vn1) + Sign(vn2) + Sign(vn3))* Absvalue(result);
to
result = Sign(Sign(vn1) + Sign(vn2) + Sign(vn3));
and then comment out v2=... and v3=..
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
tips for expert traders.
GSB is not a black box, so dont treat it like it is.
Here is an example
for ES/NQ AIC (any indicator cross) is the best entry types.
this means any of 3(default) indicators cross, you get a trade.
What happens if you use 5 inidcators? So many indicators are true, bad indicators get less significant - but are still active and you get higher
chance of systems with inidcators that are bad and give negitve, or erratic postive results
thats why I will test every inidicator, one at a time too see the equity curve
ie
v1 = GSB_S2R2LessClosev2 of Data(i1Data);
v2 = GSB_Dist(Ticks) of Data(i2Data);
v3 = GSB_LessCloseS1R1v2 of Data(i3Data);
change to
v1 = GSB_S2R2LessClosev2 of Data(i1Data);// and note the np, dd , pf here etc
//v2 = GSB_Dist(Ticks) of Data(i2Data);
//v3 = GSB_LessCloseS1R1v2 of Data(i3Data);
repeat for v2, and v3
if you have one of the many stop methods as well as fixed stop, what happens if you remove it from the code?
if it makes little difference, then remove the logic.
Trading systems should be as simple as possible
unrelated
Im re-doing nq & ES day trading. Any contribution of workers appreciated - and I will send you results of the outcome. Please give me you share key so
I can use your workers
here is an example of work so far. Test of 126 secondary filters. I had also done bar interval and session times, and trading times
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Ive had an extremely fruitful week of research
Ive tested 195 different secondary filters on es nq, long , short and long and short
What ive learnt is
We have many secondary filters, different for long, different for short and different for long & short
Day of week bias's seem to be consistent over numerous secondary filters (a good sign)
Massive improvements in results can be found by using DOW filters.
New release of gsb automation has totally programmable dow patterns
ie you can do
tue LngSht Wed LS Fri Lng
then
Mon Sht, tue Sht Wed LS Fri Lng
So what does this look like at the end of the day
lets say nq 30 min system suing highlowless3 secondary filter and filter closelessClosedBPV
this gave 102 systems in my favorites B
With the correct DOW LS mask, we got 170 ave systems in Fav B
However with the best secondary filter I found so far, and best DOW LS mask I got 268 systems in favorite B
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
had a very productive week as far as research goes, but too early to share more results. Not quite finished the first batch of NQ short only systems,
and hope to have them released well under 2 weeks time.
Ive learnt a lot manually optimizing NQ systems.
IN EWFO im liking fitness vbase(m)^1 with drawdown ^1
vbase factors in daily range which puts a bit less weight on highly volatile things.
Its still just an idea, but hope to have a new product for TS which is really unique. Will know more in 1 or 2 weeks.
Thanks received (3):
+1 Ketil at 2024-09-28 04:47:34 +1 erlendsolberg at 2024-09-27 09:10:04 +1 JozefSusko at 2024-09-27 09:05:21
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSBshort2nq series are finished, but not tidied up for sale. IE documented, auto detect for micros etc, paypal link etc
However im keen to roll them out, especially as October statistically tends to be a bearish month and the market is over bought. Many indexs are above
or at all time new highs
This also is good for Bounce, which loves the violent retracements of a bear market
https://trademaid.info/gsbhelp/BounceES-NQ-DAX-RUS-DOW.html
enclose are reports of GSBsys2short series (7 systems in the family)
The systems are on special for this month at $600 for the entire family
As there is no paypal link, you can just buy another system for $600, and send me an email that you want the GSBNQshort2 series.
These systems are quite different to what will eventually come out when the short only day trading nq systems are done.
Attachment: Login to view the details
Attachment: Login to view the details
Thanks received (1):
+1 cotila1 at 2024-10-25 10:49:09
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Updates for the week.
After I finish the short only systems (significantly different to the 7 i released for sale this last week) i,m going to work on Long & short day
trading NQ systems.
There has been a mild update to pa pro where systems you dont like under the market performance tab can be de-selected. I have not got to test this
year as took some day of this week
The new product I was hoping to have work has started, but will test the proof of concept on monday. Its more complex than expected
Thanks received (1):
+1 cotila1 at 2024-10-25 10:47:16
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Ive made a complex but simple to use strategy, you can add onto any TS/MC chart your are building for GSB and non GSB systems.
My practice to to test every module in GSB code, to make sure it makes a difference, and optimizes with acceptable optimization curves.
lets say you want to make a change to your TS/Mc code.
IE test one indicator at a time.
i would put // in front of the lines to remove
then do a screen shot of the chart before changes, then after, then compare them
with gsb_performancemetrics, it writes the performance metrics in the print log, including pearsons correlation to a straight line of the equity
curve.
v1 = GSB_CounterTrend(Close, i1length, i1numDevs) of Data(i1Data);
//v2 = GSB_CounterTrend2(Close, i2length, i2numDevs) of Data(i2Data);
//v3 = GSB_MedianOsc(Close, i3length) of Data(i3Data);
Now all you need to do is put the gsb_performancemetrics strategy on the chart, and compare the lines in the print log.
This will be pushed with the next build of GSB
GSB_EquityMetrics.NQ 15 tot net pft:137125 tot trades:527 tot ave: 260.20 tot pf: 1.82 tot dd:7820 Pearson:0.936 long net pft:137125 long trades:527
long ave trade: 260.20 long pf: 1.82 long dd:7820 short net pft:0 short trades:0 short ave trade: 0.00 short pf: 0.00 short dd:0
Thanks received (4):
+1 TwntySQ at 2024-11-08 03:34:33 +1 Ketil at 2024-10-26 14:57:23 +1 Bruce at 2024-10-25 19:40:26 +1 cotila1 at 2024-10-25 10:48:05
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
What an amazing week trading!
Here is the high level of whats going on.
Im doing a lot of learning, and shortly back to system building. This research in time is going to make its way back into GSB.
My latest thoughs are going to be experimented with in the next set of GSBshortnq systems.
I will start on them in a weeks time roughly.
Thanks received (1):
+1 TwntySQ at 2024-11-08 03:34:22
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I have built 2500 systems, and picked the top 300 systems ranked by in sample data only
200 inidcators were used (all the not released ones) which was deliberate in that I did not do the two pass method to get the top 10 indicators.
I did not use the best systems ranked by out of sample put into favourites B,c,d. So I expect setup to be worse than what im using to build systems.
These are for nq short only day trade systems
Then did the family command to reduce it to 225 systems
The systems were verifed against emd, ym rty es, nq 28 29 31 32 minute bars, nq30 with 4* 5 tick random noise and 4*10 tick random noise
so 4 lots of data to verify against
what I want to know is what in sample metrics were the best predictor of sample np/dd
I have done the work, but likely others can do it much better and faster than me. I couldnt get chatgpt to succeed.
The results were a surprise to me. I used only abs of correlation.
chat gpt said
Compute correlation coefficients between each column A to S and column AB to understand linear relationships.
Use regression or feature importance analysis to identify which metrics contribute most to AB.
Rank the columns A to S based on their influence on AB.
Attachment: Login to view the details
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Peter,
This is your file including the correlation matrix
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Carl.
thank you. You should be a data analyst.
this looks like my results are correct.
do you know how to "Use regression or feature importance analysis to identify which metrics contribute most to AB."?
i want to do one or two more test of this to confirm results again to get some more data sets.
Likely done monday
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
I must have a Python script somewhere to do the regression/feature analysis, but I have to first find it. Probably on an old desktop somewhere... I
let you know when I find it.
Thanks received (1):
+1 admin at 2024-11-22 05:05:36
|
|
|
Frank
Junior Member
Posts: 2
Registered: 29-4-2024
Member Is Offline
|
|
Peter, Carl, interesting for linear correlations could also be the Pearson Correlation in Excel and I hope that I will find out how to attach files
soon  . Feature importance helps to improve machine learning predictions, both
could benefit from using more lines of data. Happy to contribute as well!
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Frank
to attach files, make a new post, do preview post and then attach files.
Here are the results of my tests. - different set of systems. Very successful proof of concept.
Setup is a good nq short only settings I have been working on for some time.
Too make it degrade a bit more than my best settings, I did a one pass build with all 200 indicators (numerous indicators not released, and some of
them might work poorly), not the recommend 2 pass method using the top 10 indicators.
I verified with 15 20 40 60 minute nq data. This data was by far the best data to verify systems with compared to the others I used.
Even the worst of these systems likely are good, but we get a 12 to 19% improvement in fitness (np*at)
I suspect if there were more poor systems, the improvement would be better.
|
|
|
| Pages:
1
..
44
45
46
47 |
![[*]](images/davis/default_icon.gif){kind=icon}
