| Pages:
1
..
16
17
18
19
20
..
98 |
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by boothy  | Hi,
I need a bit of help to make sure I have everything set up properly for using closeDBpv, as I want to be able to verify on other markets.
when I build systems using ES.30 data1 and SPX.30 data2 using the original closeD I get OOS degradation in the mid 20% range, but whenever I use
closeDBpv the degradation blows out to 50 - 60% range.
would I be correct in saying that if I have everything set up correctly and point values are normalised, shouldn't the OOS degradation be similar?
attached are screenshots of my contracts list and price data file that references the new $SPXBPV contract. Have I got this correct?
Thanks.
|
What you've publishes is great, but you have not shown what secondary filter is set to. bpv or GA with bpv used in GA
Also the parameters with sf entry level and sf indic weight
|
|
|
boothy
Junior Member
Posts: 54
Registered: 21-5-2018
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by boothy | Hi,
I need a bit of help to make sure I have everything set up properly for using closeDBpv, as I want to be able to verify on other markets.
when I build systems using ES.30 data1 and SPX.30 data2 using the original closeD I get OOS degradation in the mid 20% range, but whenever I use
closeDBpv the degradation blows out to 50 - 60% range.
would I be correct in saying that if I have everything set up correctly and point values are normalised, shouldn't the OOS degradation be similar?
attached are screenshots of my contracts list and price data file that references the new $SPXBPV contract. Have I got this correct?
Thanks.
|
What you've publishes is great, but you have not shown what secondary filter is set to. bpv or GA with bpv used in GA
Also the parameters with sf entry level and sf indic weight
|
Secondary Filter - Enabled
Secondary Filter Indicator - CloseLessPrevCloseDBpv
Parameters SF Entry Level - (0:100:1)
Parameters SF Indic Weight (1:1:1)
I have nerver touched the parameters, always left as standard.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Not touching the defaults is generally good, but not in this case.
Weight should be -1 to 1 step 2. This means it will try trend and counter trend. - though I never noticed counter trend systems.
SF entry level will be around $800 mark. You could use 0 to 3000 step 10 if your on a wide range of markets,
or $200 to $1200 if you want to be tighter. Not critical if you go to wide. To narrow (as in your current settings) is bad and the likely cause if
your issues.
If you go back to SF closeD, you will need to be 0 to 100 step 1 (roughly)
Ideally gsb should have different variable for each type of Secondary filter
|
|
|
boothy
Junior Member
Posts: 54
Registered: 21-5-2018
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Not touching the defaults is generally good, but not in this case.
Weight should be -1 to 1 step 2. This means it will try trend and counter trend. - though I never noticed counter trend systems.
SF entry level will be around $800 mark. You could use 0 to 3000 step 10 if your on a wide range of markets,
or $200 to $1200 if you want to be tighter. Not critical if you go to wide. To narrow (as in your current settings) is bad and the likely cause if
your issues.
If you go back to SF closeD, you will need to be 0 to 100 step 1 (roughly)
Ideally gsb should have different variable for each type of Secondary filter |
ok, thanks for clearing that up,
is it correct what I have done with contracts list and price files? I wasn't sure If I had to create new price file wich linked to new BPV contract?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by boothy | | Quote: Originally posted by admin | Not touching the defaults is generally good, but not in this case.
Weight should be -1 to 1 step 2. This means it will try trend and counter trend. - though I never noticed counter trend systems.
SF entry level will be around $800 mark. You could use 0 to 3000 step 10 if your on a wide range of markets,
or $200 to $1200 if you want to be tighter. Not critical if you go to wide. To narrow (as in your current settings) is bad and the likely cause if
your issues.
If you go back to SF closeD, you will need to be 0 to 100 step 1 (roughly)
Ideally gsb should have different variable for each type of Secondary filter |
ok, thanks for clearing that up,
is it correct what I have done with contracts list and price files? I wasn't sure If I had to create new price file wich linked to new BPV contract?
|
You dont have to do that, but I like what you have done.
GSB will always use GSB contract list bpv, no longer TS bpv
|
|
|
boothy
Junior Member
Posts: 54
Registered: 21-5-2018
Member Is Offline
Mood: No Mood
|
|
Thanks for your help 
|
|
|
ProbTrader
Junior Member
Posts: 16
Registered: 3-7-2018
Member Is Offline
Mood: No Mood
|
|
Hi Peter
Product development suggestion...
I think it would be great to be able to see the underlying asset, overlaying the equity graph somehow. Such feature exists on other platforms and it
helps greatly in the process of understanding how the system has behaved historically, with respect to different market conditions. Most of us
probably know exactly how say SP500 has moved over time. But for many other assets, I find myself having to have both GSB open (equity curve) as well
as another system (market direction). It would be neat to be able to see both these curves in the same graph.
Many thanks for all the other improvements as of lately. Amazing how quickly and well GSB has developed. Exciting to be part of the process.
|
|
|
ProbTrader
Junior Member
Posts: 16
Registered: 3-7-2018
Member Is Offline
Mood: No Mood
|
|
Underlying asset graph...
Hi Peter
Product development suggestion...
I think it would be great to be able to see the underlying asset, overlaying the equity graph somehow. Such feature exists on other platforms and it
helps greatly in the process of understanding how the system has behaved historically, with respect to different market conditions. Most of us
probably know exactly how say SP500 has moved over time. But for many other assets, I find myself having to have both GSB open (equity curve) as well
as another system (market direction). It would be neat to be able to see both these curves in the same graph.
Many thanks for all the other improvements as of lately. Amazing how quickly and well GSB has developed. Exciting to be part of the process.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by ProbTrader | Hi Peter
Product development suggestion...
I think it would be great to be able to see the underlying asset, overlaying the equity graph somehow. Such feature exists on other platforms and it
helps greatly in the process of understanding how the system has behaved historically, with respect to different market conditions. Most of us
probably know exactly how say SP500 has moved over time. But for many other assets, I find myself having to have both GSB open (equity curve) as well
as another system (market direction). It would be neat to be able to see both these curves in the same graph.
Many thanks for all the other improvements as of lately. Amazing how quickly and well GSB has developed. Exciting to be part of the process.
|
Thank you again for your kind words.
Happy to do that. It was in the back of my mind yesterday doing video on CL when I was noting that 2017 a harder year for CL systems. Maybe have daily
atr too
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
If you're building a system to daily bars (390min), what associated data streams would you suggest to validate?
Ie. as you do for 30min, you use 29, 31 to optimise (discover) and then use 25,26,27,28,32,33,34,35 to validate.
How would this look for a system built for 390min?
I hope that makes sense...
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by TradingRails |
If you're building a system to daily bars (390min), what associated data streams would you suggest to validate?
Ie. as you do for 30min, you use 29, 31 to optimise (discover) and then use 25,26,27,28,32,33,34,35 to validate.
How would this look for a system built for 390min?
I hope that makes sense... |
I think thats not going to validate so well.
Lets say your on ES where you can shift the session time range roughly 815 to 1515 instead of 830 to 1500
It means you can go say 831 to 1501 or 829 to 1459 etc. But this will not vary the amount of bars in a day, which is what stresses the oscillators
out.
(and stressing them is the point)
My un-tested theory is using 29,30,31 min bars works better than random noise on just 30 min bars. The reason is the noise doesnt stress the
oscillators as much as changing the osc lengths (done indirectly as you change the number of bars in a day with 29,30,31)
Much down the track I will add random noise as a stress test option on GSB.
I dont expect it to be as good as multi time frame / market
|
|
|
JasonT
Junior Member
Posts: 61
Registered: 6-6-2018
Member Is Offline
Mood: No Mood
|
|
Turning Off Performance Filters
Hi Peter,
if we wanted to turn off all performance filters and see all the systems that GSB builds, how do we do that? For example do we use the settings as
per the attached picture?
Thanks,
J
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by JasonT | Hi Peter,
if we wanted to turn off all performance filters and see all the systems that GSB builds, how do we do that? For example do we use the settings as
per the attached picture?
Thanks,
J
|
That's a good thing to do, but you need filters in training.
Your settings were correct. Need them in full period, latest days, test, test& validation.
|
|
|
saycem
Junior Member
Posts: 48
Registered: 13-7-2018
Member Is Offline
Mood: No Mood
|
|
Switching Nth from no trade to trade is giving me this error message. When it completes stats are the same for both.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by saycem | Switching Nth from no trade to trade is giving me this error message. When it completes stats are the same for both.
|
postt the message in the exception folder that occurs at the same time
|
|
|
saycem
Junior Member
Posts: 48
Registered: 13-7-2018
Member Is Offline
Mood: No Mood
|
|
It had something to do with Nth period. I'm not quite clear how to use this. It was set on 90 from building swing strats (read from a previous post)
I have switched it back to 1 as I am building day trading systems now and seems to have fixed the issue.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by saycem | It had something to do with Nth period. I'm not quite clear how to use this. It was set on 90 from building swing strats (read from a previous post)
I have switched it back to 1 as I am building day trading systems now and seems to have fixed the issue. |
thanks for update. If you want it fixed post the exception message from exception folder and do a support ticket.
|
|
|
JasonT
Junior Member
Posts: 61
Registered: 6-6-2018
Member Is Offline
Mood: No Mood
|
|
Unexpected behaviour on short term tests
Hi Peter,
I'm testing an idea about the extent to which back testing over a longer period of time versus shorter periods of time gives you systems that are more
stable/profitable into the future.
To do this I've used ES/SPX 30 min chart price data with date ranges like:
-1 Jan 2018 to 28 Nov 2018 (the subject of this post)
-from 1 Jan 2017 to 28 Nov 18
-from 1 Jan 2016 to 28 Nov 18
...
-from 1 Jan 2000 to 28 Nov 18. (you get the idea).
Each one of these has its own price data file downloaded rather than using the same data file and applying the 'Trading Period > Dates' filter. The
reason for that is if I use a price data file that goes back to, say, 1 Jan 2000, and apply a starting date of 1 Jan 2018 using the 'Trading Period >
Dates' filter I get no trades at all (version 1.0.49.12). This is another unexpected behaviour which is the reason I am using separate price data
files for different time periods. But that's not the problem I want to discuss.
When I run GSB using 1 Jan 2018 to 28 Nov 2018 price data, GSB seems to ignore the first 169 days of data and trading from 19 June 2018. This is
common across all the systems it is creating not just because a particular system hasn't had any valid trades set up for the fist 169 days.
This affects the in-sample training/ out-of-sample test proportions because if you say allocate 40% test and 60% training you get hardly any test,
which means you are not able to perform the back test you expected.
You can see how in attachment "A1" I've selected 40% training and 60% test. In "A2" I've selected 70% training and 30% test. Just to make the point
really clearly, in "A3" I've selected 90% training and 10% test. This shows that the number of trades that appears in the 'training' section compared
to the 'test' section do not represent the proportions selected. You can see in "A1 Trades" that this is because GSB isn't producing any trades until
19 June 2018, 169 days after the start date that it is using the calculate the 40%/60%, 70%/30% and 90%/10% training to test ratios. I've attached a
screen shot of the 2018 data file that I'm using so that you can see that the data starts from 1 Jan 2018 and not 19 June 2018.
In diagram "A4" I've use the same 90%/10% ratio but used the 1 Jan 2000 to 28 Nov 2018 data file. You can see that it looks a lot more like 90%/10%.
However if you look closely you can see how it also misses the approx first half of the first year.
So is GSB supposed to work this way? Is there a setting that can be applied to GSB to ensure it considers the first 169 days (or so) in test?
Thanks,
Jason.
     
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
What causes the additional Nth setting and Commission/slippage fields to disappear as can be seen missing from the attached image?
Version 50.25, Manager. Thx.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by TradingRails |
What causes the additional Nth setting and Commission/slippage fields to disappear as can be seen missing from the attached image?
Version 50.25, Manager. Thx. |
A mistake by programmer. It appears if your in gsb admin mode
I will publish fixed version shortly in beta section
|
|
|
coccigelus
Junior Member
Posts: 73
Registered: 11-7-2018
Member Is Offline
Mood: No Mood
|
|
Hello,
I am new here so please excuse me if some questions have been already answered somewhere in this forum :
1)GSB's default setting use 3 indicators: I would be interested to know if this specific number is the optimal number based on your research and if
that is the case what your research suggest if that number is decreased or increased.
2)Assuming I am going to launch a GSB's analysis with the following setup: OPT price data: ES-NQ-RTY-YM- with data2 SPY-NDX-RUT-INDU set up at
29-30-31 and verify price data set up on the same but with 25 -35 minute;
• Based on what I’ve read the best way to use the secondary filter (with this setup) is to set up closelessprevcloseDBPV because we use different
markets. Would be ok to use the default parameter section with the only change in the SF entry level set up at 0 - 2000 -10 or You recommend other
changes such as the weight or entry parameters (0-6000-100)? Entry parameters is a section not very clear and I would appreciate If You can expand a
bit especially in term of variation of interval needed when the indicators increase or decrease.
• Trading period Nth day: If I would want to trade for two consecutive day and keep the third one as OOS I would need to set up Nth day to 2? Let’s
say use a ratio of 66.7% in sample and 33.3% OOS.
• Workspace section: I am really blind regarding the meaning of worker and so on. Can You point out a section in the forum where I can learn how to
(properly set up) and use them? (I do not have any idea about the concept of worker)
• Am I right to think that the Standalone version uses way less memory than the manager version with workers and to some extent is more reliable
in case server resources are limited?
Please keep up with the videos! Very useful for those not yet fully knowledgeable!
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by coccigelus | Hello,
I am new here so please excuse me if some questions have been already answered somewhere in this forum :
1)GSB's default setting use 3 indicators: I would be interested to know if this specific number is the optimal number based on your research and if
that is the case what your research suggest if that number is decreased or increased.
2)Assuming I am going to launch a GSB's analysis with the following setup: OPT price data: ES-NQ-RTY-YM- with data2 SPY-NDX-RUT-INDU set up at
29-30-31 and verify price data set up on the same but with 25 -35 minute;
• Based on what I’ve read the best way to use the secondary filter (with this setup) is to set up closelessprevcloseDBPV because we use different
markets. Would be ok to use the default parameter section with the only change in the SF entry level set up at 0 - 2000 -10 or You recommend other
changes such as the weight or entry parameters (0-6000-100)? Entry parameters is a section not very clear and I would appreciate If You can expand a
bit especially in term of variation of interval needed when the indicators increase or decrease.
• Trading period Nth day: If I would want to trade for two consecutive day and keep the third one as OOS I would need to set up Nth day to 2? Let’s
say use a ratio of 66.7% in sample and 33.3% OOS.
• Workspace section: I am really blind regarding the meaning of worker and so on. Can You point out a section in the forum where I can learn how to
(properly set up) and use them? (I do not have any idea about the concept of worker)
• Am I right to think that the Standalone version uses way less memory than the manager version with workers and to some extent is more reliable
in case server resources are limited?
Please keep up with the videos! Very useful for those not yet fully knowledgeable!
|
1) This is a good question. I dont think its critical. The generation of GSB earlier used 7.
Now we have market validation, we can test the out of sample results for each amount of indicators.
I like to test with nth=1 and leave a number of years out of sample.
Im using 3 but other users use 2, others 5.
I think you need to do at least 30,000 systems. Even then identical tests might vary 5% in output fitness. (rough guess)
2) Are you trying to build systesm for es only? I would just use 20,30,31 and verify on all other bars 25_35. (exculded 29,30,31)
Your cpu and ram are going to limit you a bit. Lots of data streams will use much more ram, esp when you add in the verifaciton for all the other
indices. My last tests showed ES was best with no secondary data, but verifaction helped a lot.
However CL was greatly helped but ng,ho,rb as data2
Stick to defaults but you must have closed or closedbpv for secondary filter.
Ive been testing entrylevel on es the last few days. It helps a lot to keep at at zero. However after you filter out systems that dont pass
verification 7/7 entry level made much less difference.
Nth day. I still like 1:1 ratio with a few years OOS. But remember 2017 will be a flat year.
Just run one or two workers and a manager. It should work as is. But you might have ram for only one worker. Likely you will crash on 2 workers.
Depends a lot what your doing. 28.7 has some memory improvements.
ddr3 ram, is really cheap, so it would be good to buy some. (make sure you get the right type and have ram slots free)
standalone might use a bit less ram than manager and worker, but manager will get some free gsb cloud usage. You could even run manager with no worker
and depend on the cloud. Free cloud is not fast and depends on how many others are using it.
Thanks for the comments on the video. A comment on youtube and or a like would be good. There is no marketing for GSB as im just too busy to do it.
Buy sales pay for the programmers to make the product better.
|
|
|
JasonT
Junior Member
Posts: 61
Registered: 6-6-2018
Member Is Offline
Mood: No Mood
|
|
100% Training, Nth Day and Performance Filters
Hi Peter,
As I am following along in your videos, I had a question I wanted to clarify.
I want to use the Trading Periods > Nth day to create in sample and out of sample data, rather than say 40% training in sample and 60% test out of
sample, based on your comments that Nth day is a better approach.
So to achieve this I have applied the settings as per the attached diagram. Can you please confirm that I've got the settings correct?
Also can you please confirm that in this case, whether or not the Performance Filter - Full Period and Training need to be the same given that the
training period is 100% which is the full period of time. What happens if these settings are different?
Many thanks,
Jason.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by JasonT | Hi Peter,
As I am following along in your videos, I had a question I wanted to clarify.
I want to use the Trading Periods > Nth day to create in sample and out of sample data, rather than say 40% training in sample and 60% test out of
sample, based on your comments that Nth day is a better approach.
So to achieve this I have applied the settings as per the attached diagram. Can you please confirm that I've got the settings correct?
Also can you please confirm that in this case, whether or not the Performance Filter - Full Period and Training need to be the same given that the
training period is 100% which is the full period of time. What happens if these settings are different?
Many thanks,
Jason.
|
These are goog questions.
Im using the same settings for full period and training, but GSB should use the tighter of the two. ie if full period pf was 2.0 and training was 1.7,
1.7 would be used.
Nth is correct but I also like to have a few years using dates. Remember however the weekness of dates is you have to differentiate bad trading
conditions from bad years. so 2017 for es and cl are going to struggle to make a profit.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
This tool works great. win updates cause me lots of issues, mainly with unplanned reboots.
Do we need to have the tool running all the time, or is it a one off event?
|
|
|
| Pages:
1
..
16
17
18
19
20
..
98 |
![[*]](images/davis/default_icon.gif){kind=icon}
