| Pages:
1
..
20
21
22
23
24
..
47 |
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Quote: Originally posted by admin  | | Quote: Originally posted by Carl | | Quote: Originally posted by admin | There is a critical TS 9.5 bug on @GC. Im using exchange time 700 to 1330.
Before march 2020, we get bars before 700am
On local time the issue is fixed.
However systems made in GSB on exchange time work well. Local time work like crap in GSB. (there should be zero difference)
GSB systems will not match ts if they are made on exchange time in TS.
If systems made in gsb on local time, match with ts is perfect.
I want to compare this to MC / iqfeed
Can someone send me gc.1minute bars
time 700 to 1330 exchange time, exported in exchange time and a file in local time
dates from nov 1 2006 till today
Im interested if ts 10 has the same issue, but prefer to try mc first
Later note.
ts 10 has identical issue
This is the date at which the bug shows.
|
Hi Peter,
I have the same issue on other ticker symbols (TS 9.5 update 28).
Using session 0700-1500 on Exchange Time setting: but chart shows 0630-1500 between November and March. This issue returns every year in the chart.
The chart data between April and October is correct.
Update
On one of my laptops I have opened TS for the last time in July this year.
Opening TS offline now shows the correct price data between 2006 and July 2020 (TS 9.5 update 28).
Is it a bug in TS software or in the data?
|
THanks so much for this. Things like this could kill traders profitability, and we have no idea about it. Im going to document this is gsb docs.
Its a bug in ts session time logic. Solution is to use local time. What symbols have you seen it in? I sue local time normally, but developed cl on
exchange time. I need to check my data on this. |
Hi Peter,
I did a few tests (TS 9.5 update 28 build 3344) and saw this issue on all tickers I tested: CL, RB, ES, NG, HO, TY, EC.
Session time 0700-1400
Chart shows 0600-1300 between Nov and March (this is not correct)
Chart shows 0700-1400 between April and October (this is correct)
I am going to contact Tradestation today. This issue needs to be solved quickly, because it is likely to appear again in the beginning of November and
is going to affect our trading.
After you mentioned the issue here on the GSB forum, it is easy to spot on a chart, The double dashed lines show where the times are off.
Update Oct 1
I am pretty sure this issue is related to the summer time / winter time switch.
Thanks received (1):
+1 admin at 2020-10-01 00:00:20
|
|
|
NickW
Junior Member
Posts: 29
Registered: 12-2-2020
Member Is Offline
|
|
Peter,
I have had some major data issues with TradeStation backadjusted contracts for the past two weeks. There is a lot of data missing, like ES missing
all of March this year, etc. It is so bad that I switch to using IQFeed data. I do not know how this company is still in business. Cannot even call
them anymore.
Nick
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by NickW | Peter,
I have had some major data issues with TradeStation backadjusted contracts for the past two weeks. There is a lot of data missing, like ES missing
all of March this year, etc. It is so bad that I switch to using IQFeed data. I do not know how this company is still in business. Cannot even call
them anymore.
Nick
|
I would close ts, delete your cache. Open a dummy workspace that has @es.d from the first possible date. Then once thats loaded open your live
workspace.
TS is not kovid friendly, but we understand why.
My TS data is ok, apart from this one issue ive complained about
|
|
|
NickW
Junior Member
Posts: 29
Registered: 12-2-2020
Member Is Offline
|
|
When I had the data issue (looks like it's fixed now), i tried everything you suggested, in addition to creating a brand new server, installed TS, and
refreshed data without any luck. I even pulled the data into MC from TS and same issues, on original server and on new server.
So i am eventually going to create a program to check all the data when i restart the environment on Sunday. I have lost too much money on phantom
trades due to bad data in the platform at the time.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by NickW | When I had the data issue (looks like it's fixed now), i tried everything you suggested, in addition to creating a brand new server, installed TS, and
refreshed data without any luck. I even pulled the data into MC from TS and same issues, on original server and on new server.
So i am eventually going to create a program to check all the data when i restart the environment on Sunday. I have lost too much money on phantom
trades due to bad data in the platform at the time. |
thats a good idea,
basically make ts code to look for gap >3 days likely would do it.
Great if you publish it.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
I have an idea thats a tiny but significant tweak to how we build systems.
With the two pass method.
choose indicators, every 80 days in sample, 80 days out of sample pre 1.1.2015 {exact date I may vary. New market make early, well known market like
cl a little later}
Then typically 3 periods out of sample after this.
Whats wrong with this?
We train on the lucrative and volatile 2007 year (not great)
We dont get training in later years, we dont get out of sample in early years
My proposal is basically nth periods expanded, but we will just use dates
pretend we have 4 buttons to click. (exact amount user definable)
button
dates beginning to day 2020.1.1.
1) true/false. 1st lot of 80 recurring periods. Use for indicator IS
2) true/false. 2nd lot of 80 recurring periods. Use for indicator OOS
3) true/false. 3rd lot of 80 recurring periods. Validation period1
4) true/false. 4th lot of 80 recurring periods. Validation period1
additional validation period 20201.1. on-wards. (I want to keep lucrative 3.2020 out of the stats)
Comments welcome
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | I have an idea thats a tiny but significant tweak to how we build systems.
With the two pass method.
choose indicators, every 80 days in sample, 80 days out of sample pre 1.1.2015 {exact date I may vary. New market make early, well known market like
cl a little later}
Then typically 3 periods out of sample after this.
Whats wrong with this?
We train on the lucrative and volatile 2007 year (not great)
We dont get training in later years, we dont get out of sample in early years
My proposal is basically nth periods expanded, but we will just use dates
pretend we have 4 buttons to click. (exact amount user definable)
button
dates beginning to day 2020.1.1.
1) true/false. 1st lot of 80 recurring periods. Use for indicator IS
2) true/false. 2nd lot of 80 recurring periods. Use for indicator OOS
3) true/false. 3rd lot of 80 recurring periods. Validation period1
4) true/false. 4th lot of 80 recurring periods. Validation period1
additional validation period 20201.1. on-wards. (I want to keep lucrative 3.2020 out of the stats)
Comments welcome
|
Hi Peter,
So if I understand your suggestion correctly, the IS period covers the whole date range, but only 25% of the days when the other three blocks are set
to "true"?
Same for OOS and VAL1 and VAL2?
If so, I think it's a very good idea!
|
|
|
OUrocketman
Junior Member
Posts: 18
Registered: 10-5-2018
Member Is Offline
Mood: No Mood
|
|
Peter,
Trying to verify I understand what you’re suggesting. So, for example, take ES.
We could use the Indicator IS period with dates from 1/1/2000 to 9/30/2007. Build systems with nth 80 no trade, auto nth to trade. Select the
non-red indicators at this point.
Run the non-red indicators from step above from 10/1/2007 to 6/30/2015. Build systems with nth 80 no trade, auto nth to trade. Select the green
indicators.
Run stats across validation period 1, period 2, and OOS data. We could also examine effect of walk forward as follows:
Walk forward to beginning of validation Period #1 only
Re optimize at end of validation period #1 and update parameters into validation period #2
Don’t walk forward at all.
Am I understanding what you’re suggesting properly?
|
|
|
cico
Junior Member
Posts: 26
Registered: 4-1-2020
Member Is Offline
|
|
Dears, based on your opinion and experience, what is the best time to run "indicator selection process"? At the begining, when I do have all the
Operators, Entry Modes, Normalization Modes, etc. enabled, or at the end of the process, when I do know exactly, which operator is best for the
market, which entry mode is best and run indicator selection for this specific setup only? Thanks for your ideas.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cico | | Dears, based on your opinion and experience, what is the best time to run "indicator selection process"? At the begining, when I do have all the
Operators, Entry Modes, Normalization Modes, etc. enabled, or at the end of the process, when I do know exactly, which operator is best for the
market, which entry mode is best and run indicator selection for this specific setup only? Thanks for your ideas. |
I do it at the beginning. The newest CL video is good to look at.
Only run * operator. Normalization mode CS is slightly better, so I only use that.
Whats important is entry mode. Either compare2 or a cross entry type is clearly the best. But it varies per market.
ES use cross, CL use compare. You ask good and important questions.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by OUrocketman | Peter,
Trying to verify I understand what you’re suggesting. So, for example, take ES.
We could use the Indicator IS period with dates from 1/1/2000 to 9/30/2007. Build systems with nth 80 no trade, auto nth to trade. Select the
non-red indicators at this point.
Run the non-red indicators from step above from 10/1/2007 to 6/30/2015. Build systems with nth 80 no trade, auto nth to trade. Select the green
indicators.
Run stats across validation period 1, period 2, and OOS data. We could also examine effect of walk forward as follows:
Walk forward to beginning of validation Period #1 only
Re optimize at end of validation period #1 and update parameters into validation period #2
Don’t walk forward at all.
Am I understanding what you’re suggesting properly? |
I am thinking of replacing nth with this.
Its nth can be say 80 trades In sample, 80 oos over say the first 70% of years.
this can can 80 day IS, 80 days oos1, {used for indicator selection} 80 days oos2 80 day oos3 re-occurring.
We would do indicator selection on first 80 is) and second 80 as indicator oos
then build systems on these 80iis1, 80 iis2, and use 80iis3 & 80 iis4 as validation
and then maybe use 2020 as another validation period
Thanks received (1):
+1 OUrocketman at 2020-10-05 21:18:24
|
|
|
cico
Junior Member
Posts: 26
Registered: 4-1-2020
Member Is Offline
|
|
| Quote: Originally posted by admin | | Quote: Originally posted by cico | | Dears, based on your opinion and experience, what is the best time to run "indicator selection process"? At the begining, when I do have all the
Operators, Entry Modes, Normalization Modes, etc. enabled, or at the end of the process, when I do know exactly, which operator is best for the
market, which entry mode is best and run indicator selection for this specific setup only? Thanks for your ideas. |
I do it at the beginning. The newest CL video is good to look at.
Only run * operator. Normalization mode CS is slightly better, so I only use that.
Whats important is entry mode. Either compare2 or a cross entry type is clearly the best. But it varies per market.
ES use cross, CL use compare. You ask good and important questions. |
Peter, thanks for the answer. I saw the CL video, you are doing indicator pass with 1 operator, 1 entry mode and 1 normalization mode. I'm more
thinking to try to do indicator pass with full set of operators, full set of entry modes and full set of normalization modes (to diminish potential
overfitting). Or do you think it is not an ideal thinking? :-) Thanks.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cico | | Quote: Originally posted by admin | | Quote: Originally posted by cico | | Dears, based on your opinion and experience, what is the best time to run "indicator selection process"? At the begining, when I do have all the
Operators, Entry Modes, Normalization Modes, etc. enabled, or at the end of the process, when I do know exactly, which operator is best for the
market, which entry mode is best and run indicator selection for this specific setup only? Thanks for your ideas. |
I do it at the beginning. The newest CL video is good to look at.
Only run * operator. Normalization mode CS is slightly better, so I only use that.
Whats important is entry mode. Either compare2 or a cross entry type is clearly the best. But it varies per market.
ES use cross, CL use compare. You ask good and important questions. |
Peter, thanks for the answer. I saw the CL video, you are doing indicator pass with 1 operator, 1 entry mode and 1 normalization mode. I'm more
thinking to try to do indicator pass with full set of operators, full set of entry modes and full set of normalization modes (to diminish potential
overfitting). Or do you think it is not an ideal thinking? :-) Thanks. |
I think its defiantly not a good idea. Your welcome to test the identical setup, and look at the OOS results. However to be precise, all tests have to
be done 4 times as there is a lot of variation of identical tests. I expect OOS to be well down with what you propose.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Here is the update of the week. A lot of babies being born at the same time, and so I am the bottle neck.
1) GSB automation now has the ability to shuffle through session times. So now im testing gold from 00 to 730 in 30 min session time increments.
I will release this build late next week. Likely the last update before official release. Its a fair bit of work for me to document and to the video
on in.
2) A new GSB will come out needed for automation. It has walk forward fix for ewfo, and might have 2 new indicators. Median and closesup&down
Why I might not is they have not turned up as used indicators in the testing I did on gold.
3) I am working with Murray Ruggiero a bit and some of his ideas may make there way into GSB.
Much time was spend looking at ZonetraderES by Murray. Whats significant is the loosing periods trades to GSBsys1ES is about -0.3. This is amazing
considering its day trading the same market
and it has a trend following and counter trend system in it. Whats that got to do with GSB? Right now nothing which is whats so exciting.
ZT has no oscillators apart from ATR, and is mainly focused on pivot points.
IF countertrend / retracement architecture makes its way into GSB, that has lots of great implications. I may with Murray also sell my variation of
Zonetrader2 as well. It has simplified logic, more symmetrical code
0.43 correlation to the original ZT, but most interestingly - trades a fair bit more often and makes more $. Whats stopping me is just investing a bit
more time. Took at least 2 days work for me in the last 2 weeks and I would like to do more work on parameter choice.
4) Alertmon monitoring is 99% ready for release, but much more work might go into this. Cost of project blew out 10 fold to what expected, but even if
was just for me, the cost savings in execution errors was worth it.
Today without monitoring, a freak set of circumstance would have made me miss out on a 2k trade. I'm only intending to release this on the virtual
computer server Im offering
5) VM service. This has had some very time consuming, lengthy issues to resolve, but the backup server has been going well for a few months. Primary
final server I hope to have running in 2 to 5 weeks.
Im exciting about the final product
6) My CL system thats in my CL video https://www.youtube.com/watch?v=QYg5eH2Q1_s (at 3min) is now trading at Striker securities (and their broker network). I released it instead of
the newer CL system in the private forum as it has a longer track record, though its not made in the very newest methodology.
https://www.youtube.com/watch?v=QYg5eH2Q1_s
Note that ZT2 trades more often.
Note the CNTb4 entries. These are the counter trend ones, and its trading the exact opposite of what GSB would do.
  
Thanks received (8):
+1 LucaRicatti at 2020-10-22 17:37:43 +1 cotila1 at 2020-10-16 17:02:48 +1 Bruce at 2020-10-12 21:26:41 +1 OUrocketman at 2020-10-10 13:56:42 +1 bizgozcd at 2020-10-10 06:10:58 +1 saycem at 2020-10-10 03:50:49 +1 Carl at 2020-10-09 21:51:14 +1 NickW at 2020-10-09 20:18:38
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
There are updates to the methodology in the pipeline, but will take some time to finalize.
Automation and macros can now delver the final systems to be trades in favorites.
The count of how many systems pass and make it into favorites is also being added into the export stats file.
Happy to say this is because GSB users think of some good ideas that i Didnt 
|
|
|
cico
Junior Member
Posts: 26
Registered: 4-1-2020
Member Is Offline
|
|
Dears, does it make sense from your experience to build ES systems with different Secondary Filters (even not the best ones)? An example from CL is
showing that it could be interesting from systems correlation point of view.. Let me know your thoughts, please. Many thanks
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Cico,
Welcome to the GSB forum.
Results on ES are best when secondary filter is non-normalized cloeslessprevclosed, but results can also be reasonable good (and low correlated) when
secondary filter is normalized cloeslessprevclosed.
Thanks received (1):
+1 cico at 2020-10-13 12:20:15
|
|
|
cico
Junior Member
Posts: 26
Registered: 4-1-2020
Member Is Offline
|
|
Hi Carl, thank you very much for welcoming me and for your answer. By normalized and non-normalized you do mean Secondary Filter
CloseLessPrevCloseDBpv and CloseLessPrevCloseD secondary filters? Thanks for clarification and have a nice rest of the day.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by cico | | Hi Carl, thank you very much for welcoming me and for your answer. By normalized and non-normalized you do mean Secondary Filter
CloseLessPrevCloseDBpv and CloseLessPrevCloseD secondary filters? Thanks for clarification and have a nice rest of the day. |
IF SF, is on GA, then the 46 to 106 sf can be used - all normalized
if its on Any other option, its not normalized.
Unless your a GSB guru, on ES I would only use closelessPrevCloseDBPV (not normalized)
Auto selects whats recommend for the contract.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Bruces method (which we are experimenting with) ultimately puts all the systems from the top 250 favA into FavB.
With GSB automation, we want to track how many systems made it into favB
This can now be exported in the export stats file
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Bruces method (which we are experimenting with) ultimately puts all the systems from the top 250 favA into FavB.
With GSB automation, we want to track how many systems made it into favB
This can now be exported in the export stats file |
The Key to this is to capture all OOS dates as one period and then filter to FavB. Creat Family, then walk forward that cohort and then review.
This has delivered some very good GC, US, YM, ES, NQ and SI systems which I'm all trading live.
The attached equity chart is from a wee dev fund I use to test systems before I put them into production, hence its a little ropie to start with
however you can see over the past six weeks the impact I'm currently experiencing trading Live with a few of these newer systems.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Bruce,
is that a sim or live ts account?
Is that account gc,us,ym,es,nq in the one account?
Interesting that it seems equity took of in recent times, and not march 2020.
Were all the systems live at feb 2020, or staggered over time as you made them?
|
|
|
Bruce
Member
Posts: 115
Registered: 22-7-2018
Location: Auckland - New Zealand
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by admin | Hi Bruce,
is that a sim or live ts account?
Is that account gc,us,ym,es,nq in the one account?
Interesting that it seems equity took of in recent times, and not march 2020.
Were all the systems live at feb 2020, or staggered over time as you made them? |
This is a LIVE account NOT SIM trading es, nq, gc. US and YM has just been added for the live test.
prior equity is irrelevant as it's not the same systems or same build process and I've been trading other systems from other build platforms. These
are all GSB builds as described.
Thanks received (1):
+1 admin at 2020-10-14 18:48:00
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Bruce
I think your come up with some innovative tweaks.
Some of what you have done could be applied to the existing methodology, but regardless there is likely a month or much more work for me.
Super high level, we can go straight from systems build by automation to live trading.
Fav A contains 300 systems ranked only by out of sample performance. (20170630)
Bruces macro put the top systems from favA into Favb. I then put the dates back to start dates, and did wf
Note we had 13 members in family1, 11 in family2
SO we want systems to have big families = robust regardless of parameters.
I want wf results (brown curve) to be good. ie not tank in the last few years.
If your new to GSB, stick with whats already documented. Gold and methodology tweaks will take some time.
Big picture is system building once we have the correct setup is going to be very low in human time.
Getting the correct setup takes lots of time - but once we have found it, the info is shared to GSB users.
On that topic, if you have free cloud power, please send me your share keys. This is very cpu intensive
Those that contribute cpu power, get in-depth results of test.(Like we did with CL) But this will take me some time.
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
| Quote: Originally posted by Carl | | Quote: Originally posted by admin | | Quote: Originally posted by Carl | | Quote: Originally posted by admin | There is a critical TS 9.5 bug on @GC. Im using exchange time 700 to 1330.
Before march 2020, we get bars before 700am
On local time the issue is fixed.
However systems made in GSB on exchange time work well. Local time work like crap in GSB. (there should be zero difference)
GSB systems will not match ts if they are made on exchange time in TS.
If systems made in gsb on local time, match with ts is perfect.
I want to compare this to MC / iqfeed
Can someone send me gc.1minute bars
time 700 to 1330 exchange time, exported in exchange time and a file in local time
dates from nov 1 2006 till today
Im interested if ts 10 has the same issue, but prefer to try mc first
Later note.
ts 10 has identical issue
This is the date at which the bug shows.
|
Hi Peter,
I have the same issue on other ticker symbols (TS 9.5 update 28).
Using session 0700-1500 on Exchange Time setting: but chart shows 0630-1500 between November and March. This issue returns every year in the chart.
The chart data between April and October is correct.
Update
On one of my laptops I have opened TS for the last time in July this year.
Opening TS offline now shows the correct price data between 2006 and July 2020 (TS 9.5 update 28).
Is it a bug in TS software or in the data?
|
THanks so much for this. Things like this could kill traders profitability, and we have no idea about it. Im going to document this is gsb docs.
Its a bug in ts session time logic. Solution is to use local time. What symbols have you seen it in? I sue local time normally, but developed cl on
exchange time. I need to check my data on this. |
Hi Peter,
I did a few tests (TS 9.5 update 28 build 3344) and saw this issue on all tickers I tested: CL, RB, ES, NG, HO, TY, EC.
Session time 0700-1400
Chart shows 0600-1300 between Nov and March (this is not correct)
Chart shows 0700-1400 between April and October (this is correct)
I am going to contact Tradestation today. This issue needs to be solved quickly, because it is likely to appear again in the beginning of November and
is going to affect our trading.
After you mentioned the issue here on the GSB forum, it is easy to spot on a chart, The double dashed lines show where the times are off.
Update Oct 1
I am pretty sure this issue is related to the summer time / winter time switch. |
Update October 19 2020
This bug seems to be solved today!
|
|
|
| Pages:
1
..
20
21
22
23
24
..
47 |
![[*]](images/davis/default_icon.gif){kind=icon}
