GSB Forums

Update to GSB methodology. A must read, the backpacker and the Art of war by Sun Tzu

 Pages:  1  ..  8    10    12

admin - 30-6-2021 at 07:06 PM

Starting, June 28, 2021 CME Trading Halt between 3:15 and 3:30 p.m CST (Removed)
This doesn't affect any of my systems using @es.d, but might affect some users using @es
Put this file in C:\Program Files (x86)\TradeStation 9.5\Templates\Sessions

Attachment: Login to view the details es24.png - 188kB

This should NOT be used for systems for sale OvernightES and SwingES

Siem - 14-7-2021 at 05:01 AM

Hi Peter,

I’m looking at the swing trading ES long only systems.
With the beta features enabled the included “ES-LongOnly_SwingTrade”-macro turns on 80 main indicators.
Is this an optimised selection of indicators - or would it make sense to enable most of the other indicators too?

admin - 14-7-2021 at 05:09 AM

Quote: Originally posted by Siem  
Hi Peter,

I’m looking at the swing trading ES long only systems.
With the beta features enabled the included “ES-LongOnly_SwingTrade”-macro turns on 80 main indicators.
Is this an optimised selection of indicators - or would it make sense to enable most of the other indicators too?


I think this is the best of them, and the others wont work as well.
Its been a month or 2 since I have done this.
I dont think its critical and im happy with the 80.
Over time I have migrated the best indicators into being availble to all users.
This of course is a work in progress and will be for a long time as new inidicators are added and tested on various markets.

REMO755 - 22-7-2021 at 12:59 PM

Hello,

HOW TO AVOID OVER-OPTIMIZING.

Everything perfect in GSB, W.F ok, so in TS this happened.

It's very easy to over-optimize.

Ideas to avoid over-optimizing?

Am I doing WF out of sample?
I am doing something wrong ?





1.JPG - 121kB

2.JPG - 193kB








3.JPG - 51kB

admin - 22-7-2021 at 05:05 PM

@remo
Your wf looks good.
Check overall, what were your stats in b,c,d,e etc on the entire set your building on?
Cherry picking a system with a nice curve, is deceptively dangerous. Thats why GSB has the foundation
of the top 300 systems need to perform ok.
You can also do a rolling wf. The curves should not fall apart with rolling wf, but normally oos is higher in GSB systems with anchored.
You could look at the chart and see, why the high dd. ie stop too tight.


I assume your big dd happened in future data not seen in the wf?
If your system is bad, likely it will fail out of sample immediately.
If its good, and can fail at any time. This is the biggest risk in trading in my view.
If this is enrty aic, what I also do (in ts) is use only one inidcator at a time
Then check the results.
If one indicator gives bad results, remove it.
This is not so simple if cross or compare2 is used

admin - 23-7-2021 at 05:16 PM

Just a comment.
Sometime you cannot make money on a market. IE Nat gas now, S&P500 over 2005 - 2006.
Other times markets are the best they have ever been in history.
Well long only swing systems fit that category.
Of course that can change at any stage, and you could be fully loaded going long, and get limit move(s) down...
I had this when Trump was elected, so my entire net worth was moving greatly that day. It ended well but might not have.
In those days I was doing numerous emini contracts, and now you can do micros which have 1/10 the kick.
I may also launch gsb swing4 nq for sale, depending on correlation to the other systems.
Existing systems sold are all very lowly correlated on loosing days. (typically near zero correlation)
Another comment. I feel the biggest risk in trading is system failure. Therefor my opinion is its better to add systems than add contracts on existing systems.
Have a balance of intraday vs long term systems, session times and markets.
The systems for sale happen to be a great example of this.
Intraday on various markets, different session times, secondary filters, some GSB systems and some human designed systems, some long only, some balanced
some on 24 hour data, most on day session (even if they trade over multiple days)
https://trademaid.info/gsbhelp/Systemsforsale.html
Stops are typically $400/$4000, profit targets $300/$3000 to occasionally $600/$6000 (micro/emini)



Open TS report with internet explorer.
Disclaimer. TS report is hypothetical. Past returns are not indicative of futures returns



swing-fills.png - 233kB swing-charts.png - 862kB

Attachment: Login to view the details

admin - 26-7-2021 at 12:41 AM

reports of SWING 4 ES & NQ + correlation of all systems is here
https://trademaid.info/forum/viewthread.php?tid=270#pid7226
give me a couple of days & I will have it listed for sale
There will be a discount with system4 es/nq bundled, and for purchasers who have system4 es already

Siem - 6-8-2021 at 04:33 AM

Here's a little overview I made of the Exit Modes used in 1.0.62.88 (GSB_DecisionExit8).
It helped me get a better understanding of it and I hope it will help others here too.


Exit Modes*

ind = indicators
sf = secondary filter
rev = reverse of position
flat = not a change of position
code = put inside the formula-field in a "Custom 1-Line"-study in TS in order to see when the exit modes get triggered.
*Some modes are only available in the test-mode.

0. None

1. IFltRvAndSFFltRv
(ind = flat or ind = rev) and (sf = flat or sf = rev)

2. IRvAndSFFltRv
ind = rev and (sf = flat or sf = rev)

3. IFltRvAndSFRv
(ind = flat or ind = rev) and sf = rev

4. IRvOrSFRv
ind = rev or sf = rev

5. IFltRvOrSFFltRv
ind = flat or ind = rev or sf = flat or sf = rev

6. IRv
ind = rev

7. IFltRv
ind = flat or ind = rev

8. SFRv
sf = rev

9. CloseD
while long: close < “yesterday’s close”
code: close < GSB_CloseD(1)
while short: close > “yesterday’s close”
code: close > GSB_CloseD(1)

10. OpenD
while long: close < “today’s open”
code: close < GSB_OpenD(0)
while short: close > “today’s open”
code: close > GSB_OpenD(0)

11. LowestCloseDOpenD
while long : close < lowest of “yesterday’s close” and “today’s open”
code: close < minlist(GSB_CloseD(1),GSB_OpenD(0))
while short: close > highest of “yesterday’s close” and “today’s open”
code: close > maxlist(GSB_CloseD(1),GSB_OpenD(0))

12. HighestCloseDOpenD
while long : close < highest of “yesterday’s close” and “today’s open”
code: close < maxlist(GSB_CloseD(1),GSB_OpenD(0))
while short: close > lowest of “yesterday’s close” and “today’s open”
code: close > minlist(GSB_CloseD(1),GSB_OpenD(0))

13. LowDHighDToday
while long: close < “low of today”
code: close < GSB_LowD(0)
while short: close > “high of today”
code: close > GSB_HighD(0)

14. LowDHighDYesterdy
while long: close < “low of yesterday”
code: close < GSB_LowD(1)
while short: close > “high of yesterday”
code: close > GSB_HighD(1)

15. LowDHighDLst2Dys
while long: close < lowest of “low of yesterday” and “low of today”
code: close < minlist(GSB_LowD(1),GSB_LowD(0))
while short: close > highest of “high of yesterday” and “high of today”
code: close > maxlist(GSB_HighD(1),GSB_HighD(0))

16. Flat
ind = flat

17. SFFltRv
sf = flat or sf = rev

x. IRvAndSFRv (not in use)
ind = rev and sf = rev

admin - 6-8-2021 at 04:43 AM

@Seim, Thanks for your work. This is still under development and may change in the next 2 builds
I still think the exits are not likely to help with day trading, but useful for longer term trades.
NT code is improved in the next build (not release yet) but is still not correct.

Daniel UK1 - 7-8-2021 at 02:12 AM

Imho, i believe quite strongly that a large part of our edge, is our native use of no customised exits.
It removes a large variable for error in your process and outcome, which, most likely is the biggest reason to your failure OOS.

However, using exits, as those that exist and mentioned above, could very likely give you better results when building your systems.

As a wise man said, curve fitting is the kryptonite of every systematic trader.

:)

admin - 7-8-2021 at 02:26 AM

Good comments. Going though each exit (on say overnight es systems) in automation is going to show clearly what works.
Note however that some exits will do NOTHING, so you could confuse that result as being a littler better (or worse) than your default exit.
The reason is identical tests will vary a bit, and this can lead to wrong conclusions.
Its good to optimize the exit mode (o to 18??(i think)) on a system and note what exit modes make no different, mild improvement and big degradation.
If its a big degradation, likely no point running in in automation,

cico - 15-8-2021 at 01:31 PM

Dears, I would like to ask you, how do you see picking systems for trading out of the new ES/NQ/YM methodology.

- if there are more families with the average fitness above Avg.F, do you look into all those families to pick the system (i.e. one system from each family with the average fitness above) or do you look just into the family with the highest number of members and the stability of results?
- when choosing the system from the family, what is for you more important? WF OOS performance compared to Opt.(Avg.) or Rolling/Anchored Stability/Equity Bollinger parameters or something else?
- have you ever tried in live trading to put more systems into one chart with the same symbol (or put more systems on NQ for example into one system with different entry setups)? I am thinking about that if it would be better than to have like 10 separate charts with the same symbol and time template (one for each system)?

Many thanks upfront for your shared experience.

Daniel UK1 - 16-8-2021 at 02:09 PM

Quote: Originally posted by cico  
Dears, I would like to ask you, how do you see picking systems for trading out of the new ES/NQ/YM methodology.

- if there are more families with the average fitness above Avg.F, do you look into all those families to pick the system (i.e. one system from each family with the average fitness above) or do you look just into the family with the highest number of members and the stability of results?
- when choosing the system from the family, what is for you more important? WF OOS performance compared to Opt.(Avg.) or Rolling/Anchored Stability/Equity Bollinger parameters or something else?
- have you ever tried in live trading to put more systems into one chart with the same symbol (or put more systems on NQ for example into one system with different entry setups)? I am thinking about that if it would be better than to have like 10 separate charts with the same symbol and time template (one for each system)?

Many thanks upfront for your shared experience.



Cico, imho, you "simply", after you run your final macro and want to pick systems to trade, you head over to B, change to dates so you only see OOS, then mark all systems and note down average fitness, then you check each family down to lets say 5, then you look for families that outperform average fitness for oos, the do wf on the families you like ad evaluate further and make your pick..

cico - 16-8-2021 at 11:33 PM

Hi Daniel, many thanks for your reply. I'm following this methodology, I was more asking based on what you make your final pick of the system to trade? If it is WF OOS performance compared to Opt.(Avg.) or Rolling/Anchored Stability/Equity Bollinger parameters or something else? Many thanks.

admin - 17-8-2021 at 12:57 AM

i like if possible the whole family wf to go well, ideally good anchor and or rolling stability, linear equity curves etc.
if wf stability is low, but equity curves are really parallel then thats ok.
I dont want the oos wf curve to tank.
Sometimes I do a rolling wf. I dont use the final parameters but a rolling wf is more stresstest than anchored.
gsb systems work better with anchored parameters normally

Daniel UK1 - 18-8-2021 at 02:04 PM

Quote: Originally posted by cico  
Hi Daniel, many thanks for your reply. I'm following this methodology, I was more asking based on what you make your final pick of the system to trade? If it is WF OOS performance compared to Opt.(Avg.) or Rolling/Anchored Stability/Equity Bollinger parameters or something else? Many thanks.


After i chosen the family based on stability wf OOS, and fitness and by viewing other metrics, i do not want to pick a single system based on its individual result , i simply pick the family head. Meaning the whole family needs to be good enough or its not used at all.


admin - 24-8-2021 at 12:24 AM

Here is a comment I have to make from time to tome. I will make it less often as Im going to lock GSB down a bit more. Its a bit like you cant put your car into park or reverse when driving 100 kmh (60 mph for the Americans :) )
Operators should normally be * only. + works almost as well, but all the weights have to change to totally different values
ie result =50*w1+50w2+50w3
result =50w1*50w2*50w3
if result > entrylevel^4 then buy
if you use all negitve operands it will be the same as + (likely) as ga mutiplies it all by 1 or -1
if you use + and - then your basically saying I want some bullish indicators and at the same instant some bearish inidcaotrs.
it works but not as well and is much slower.


center scaling and hl scaling are 99% the same. CS is a little better. if you use both, you double the ram usage of every worker, and it runs slower as cache doesnt work as well.
the scaling3 mode does not match and results will change if the start date of data ever changes.
(I need to pull the feature.) SO with these 4 modes on, you have tripled the ram usage and the systems with cd3 and hl3 scaling are useless when it comes to live trading.
Don't remember if its the docs or videos, but I have been damming of the practice of turning on all options.
Stick to the defaults, change one thing at a time and NEVER turn everything on. This used 22gb of ram & I killed them.

again entry modes. Use only one at a time.
Many here were experimental and I need to remove lots of them.
compare2, aic ncc & cross are the main ones.
Cross single and cross double may be a little better, but are a pain when it comes to WF time. Many more combinations and parameter stability will be less






toomany.png - 135kB

Siem - 24-8-2021 at 02:46 AM

Quote: Originally posted by admin  

again entry modes. Use only one at a time.
Many here were experimental and I need to remove lots of them.
compare2, aic ncc & cross are the main ones.
Cross single and cross double may be a little better, but are a pain when it comes to WF time. Many more combinations and parameter stability will be less

Hi Peter,

Thanks for the insights, it's very helpful.
In the GC optimization settings both the 'Cross' and 'CrossDualLevels' entry modes are enabled by default.
Should I pick one of these then when generating systems or leave them both enabled?

admin - 24-8-2021 at 06:21 PM

@siem, this is not deliberate. There are numerous markets in GSB default settings and its very easy to make mistakes.
use cross only, but this is not a critical situation.

admin - 24-8-2021 at 09:09 PM

just a comment
under systems for sale there is free gsbsys1es1.3 that has $40 more per trade than the earlier version.
Free for GSB purchasers
These filters likely can be made to other ES systems and with parameter changes to ym nq rty etc
What I dont know is how they work on existing systems with different secondary filter. (We rarely use closelessClosedBPV on ES/nq,ym rty etc_
https://trademaid.info/forum/viewthread.php?tid=270

admin - 27-8-2021 at 05:09 PM

At the end of the month, (or maybe a day or two later as the statement takes its time) Im going to share what I think is by far the hottest of all markets, with my real time results.
While im 100% running on this market, its going to require collective work from a number of GSB users to make this solution available for all.
The issues what Im doing is leverage is extreme, and account size has to be very large, liquidity aweful. The solution I want GSB users to work on will be ok for any account size.
Last 2 months have been hard trading, but a great week & I im on new highs again.
Also some time next week, in the more philosophical and current event thread
https://trademaid.info/forum/viewthread.php?tid=273 Life down under in AU im going to share what I think is one of the most empowering statements I have ever heard. Whats powerful is
it applies to every person on the planet, and its applicable + practical to implement

admin - 29-8-2021 at 07:50 PM

Im gong to start talking about bitcoin. Over the years some markets go balistic from a system point of view, and this is one of them.
The long only systems have been similar, though how long can the market keep going up??? Long term expect return to the average.

Bruce & I are trading bitcoin futures. While I want to open the door to bitcoin trading, im strongly not going to recommend the big bit coin futures contract.
Its the most leveraged contract Ive ever traded, above my comfort level of risk.
The margin to go short at Interactive brokers is $300,000. (thats not a typo) TS is about $65000 from memory.
Read between the lines. IB is a risk adverse company and it doesnt want you to trade BTC. Their commissions are also exorbitant. (20 x TS rates from memory)
IB being risk adverse is good, as it means they a not likely to go broke. (Unlike man financial & pfg)
The liquidity is awful on futures.
Trade 1 &6 I got stopped. $4575 loss. $4000 & $575 slippage both times.
I intended to open an account with binance. As I have complex financial structures, its been a nightmare and I have almost given up.
Whats good on Binance is its liquid, any size leverage can be used. Multihcharts can trade direct to Binance.
An easy option is trade trade EFT gbtc, but the session time is wrong (830 to 1500) and we need 630 to 130 pm central USA time.
GBTC systems I think are harder than BTC.
Many other issues. The micro contract has short data compared to the big futures contract, and systems that are made on big BTC are significantly different to micro btc and significantly different to binance data. So its a complex mess. Basiclly you need to build the systems on the data you trade.
If your going down this path, open a binance account. Response time to look at your documents for a company is 1 month! Personal account might be much faster.
There are notes earlier in this thread how to do so, and tips to get discounts.

admin - 29-8-2021 at 08:00 PM

Micro bitcoin is also bust on TS. Regardless of the Profit/loss, ts will report 0$
So thats the bad news.
The good news my OOS hypothetical's for the last month show August 2021 to be the second the best month ever in BTC history.
My real time results are well down from hypothetical's, but most of that is due to significant code changes from day to day.
So this is my first 30 days BTC trading.
Also note this is 3 BTC systems trading a collective total of 1 contract max.

more to come, and big thank you to Bruce who did much of this work.


btc-realtime.png - 178kB

Bruce - 30-8-2021 at 02:56 AM

Quote: Originally posted by admin  
Micro bitcoin is also bust on TS. Regardless of the Profit/loss, ts will report 0$
So thats the bad news.
The good news my OOS hypothetical's for the last month show August 2021 to be the second the best month ever in BTC history.
My real time results are well down from hypothetical's, but most of that is due to significant code changes from day to day.
So this is my first 30 days BTC trading.
Also note this is 3 BTC systems trading a collective total of 1 contract max.

more to come, and big thank you to Bruce who did much of this work.




Thanks Peter, however without GSB as the 'discovery tool' and providing the robustness with testing I doubt I would've had the confidence to trade a single contract with @90k of margin going short BTC. It's certainly a wild ride however after a lot of builds, testing and refinements we appear to be managing it.
I just want to add that the stops necessary are wide $$$ and I've had a series of 3 losses in a row so the drawdown(s) can be (and will be) deep. That being said, like Peter, its been a rewarding couple of months. Long may it last! :)

admin - 30-8-2021 at 04:49 AM

todays Margins at ib are 225k for long and 275k for shorts, hence binace looks much better

Siem - 30-8-2021 at 09:52 AM

I have mostly used 2 indicators in the green(Identify Indicators)-step and 3 in the build-step (for GC).
In general, is it important the amount of indicators match in the green-step and the build-step? Would the result for 3-indicator systems become better when there were 3 indicators in the green-step instead?

admin - 30-8-2021 at 05:04 PM

Quote: Originally posted by Siem  
I have mostly used 2 indicators in the green(Identify Indicators)-step and 3 in the build-step (for GC).
In general, is it important the amount of indicators match in the green-step and the build-step? Would the result for 3-indicator systems become better when there were 3 indicators in the green-step instead?

Good question. I simply dont have this in my head now, but the default was 2 indicators for the first pass and 3 for the build.
Ive tested both and suspect it might vary from one market to another.
Danger of using 3 for the first step is more poor indicators might slip through

admin - 2-9-2021 at 09:43 PM

Sept 3 2021
I have started on the documentation on Bitcoin, but its progressing very slowly.
see https://trademaid.info/gsbhelp/BuildingBitcoinsystems.html
New Version of GSB Resource manager that has soft kill of workers, and is 64 bit is released by auto update today

admin - 3-9-2021 at 12:21 AM

New RM may have reset rego on users who were on 32 bit version. Ive stopped the update for now, just want to see how many people this affects

admin - 10-9-2021 at 05:28 PM

I have made excellent progress on GBTC EFT swing trading
After a week or so work, I was getting say 31 systems in favorites B
Now im peaking 168 systems.
Just be going through the methodology, finding what helps and what doesnt.
Very simple to do, but very cpu intensive.
Tried bar intervals, different profit targets, entry types, secondary filter (SF) entry types, fitness etc. Every market is unique with something that gives an edge.

Hope to be building systems next week
A little related, here is results for 45 days trading the big btc futures contract.
Ive tried trading the mini contract and thats going poorly in execution issues on ts.

btc-realtime2.png - 380kB

admin - 15-9-2021 at 02:08 AM

I have my first btc swing system. This has not been manually optimized yet. Thats slow as some of the indicators used were cpu intensive

This is how it came out of GSB, but I increased the profit target from 3000 to 5950
The number of trades on the final system is about 100. This is dangerously low,
but the original system before walk forward was done had 247 trades.


Attachment: Login to view the details

sys2.png - 316kB

portfolioquanttrader2020 - 15-9-2021 at 03:55 AM

Peter is for etf or for future?
the performance image from July 28 is with the etf?

admin - 15-9-2021 at 03:58 AM

this is gbtc eft contract. zero slippage was used as early years price was very low, so brokerage and slippage at todays prices would cause results to degrade a lot in early years

Siem - 16-9-2021 at 01:54 AM

Quote: Originally posted by admin  
I have my first btc swing system. This has not been manually optimized yet. Thats slow as some of the indicators used were cpu intensive


Hi Peter,

Could you explain a bit why this specific system could be a potentially good system worth of manually optimizing?
It looks complicated to me: the average trade size and the profit factor are better than in the original test, but the WF OOS equity curve is below the original curve, the anchored stability score is smaller than 40 and the WF OOS NP/DD looks intimidating at me.
How much better do you hope / expect the results get after manually optimizing?

admin - 16-9-2021 at 02:00 AM

Shall do tommrow.
This is likely my finished version.

Attachment: Login to view the details


admin - 16-9-2021 at 02:02 AM

In most cases the oos will be below the orginal curve. Very rare to see otherwise and thats to be expected.
The final curve is done with the benefit of hind site.
Manual opt also shows you if there is bad logic. You might see one indicator doesnt help, or the parameters are irrelevant etc. all a red flag

admin - 20-9-2021 at 01:52 AM

This is the updated Swing2 on gbtc. It took a great deal of time as the indicators used are really CPU intensive. The next system much faster
Swing2 final release has slight different profit target logic for long and short. With such a low number of trades there will be a bit more out of sample degradtion than typical is what im expecting.
I've also heard from one of the Unger Acady guys that nearly all there bitcoin profits were on intra day, and longer term systems didnt go so well.
Regardless its good to diversify, and whats nice about gbtc is you can do any account size you like.
Here is the new report

Attachment: Login to view the details


admin - 21-9-2021 at 10:02 PM

Here is the next system. Zero slippage used in mht file mainly due to fix cost per share would be really high per transaction in the early days when gbtc was very low value.
50k capital was used.
This system too was manually optimized.
I got a comment, that people couldnt open mht files.
basically only Internet explorer can open mht files successfully. (no idea on macs)
ive converted the file using portfolio analyst also.
If you unzip it and open txt_1.htm you can use any browser.
there is 2 cents per side slippage used in the PA report (txt_1.zip)

Attachment: Login to view the details

Attachment: Login to view the details


admin - 22-9-2021 at 01:40 AM

This is summary of system3, with zero slippage

graph3.png - 115kBsummaryswing3.png - 155kB

admin - 23-9-2021 at 09:39 PM

I have sort of fixed the issue where only internet explorer works to export files from ts.
PA pro can now make html that all browsers can open.
Esp mac users, please test this
https://trademaid.info/gsbhelp/reports/Swing3/index.html

admin - 23-9-2021 at 11:24 PM

Bitcoin GBTC documentation done. Settings in next build 62.07 but likely are in current build, but 62.07 has minor tweaks to the settings.
My current PL on the very leveraged illiquid BTC contract is $38,000 for < 2 months trading.

https://trademaid.info/gsbhelp/BuildingBitcoinsystems.html

bartek - 22-10-2021 at 04:56 PM

What is your advice for the method to create uncorrelated systems for one market? Using different fitness criteria, different set of indicators? Do you have any?

admin - 22-10-2021 at 06:23 PM

Quote: Originally posted by bartek  
What is your advice for the method to create uncorrelated systems for one market? Using different fitness criteria, different set of indicators? Do you have any?

thats hard to answer

A few markets can have different session times, secondary filters, but not most cant.
Different indicators will help a lot. With the swing systems I sold, the exits played a big role too. However with day trading you have much less options for exits.

SO when you choose your top systems, as little over lap in indicators is the best place to start. Fitness can be a mild factor to, as this affects the parameters and also how active the system is too.

admin - 17-12-2021 at 11:13 PM

This is a standard nq system I have been trading. I just happened to look at it recently after some great wins. I should have put a bit more effort into it. IE When I look now it was clear one of the 3 indicators was weak.
Likely with more work put into WF, and pattern filters like what I did with gsbsys21es, I could improve the system a lot more.
See the last part of this page for the effect of pattern filters
https://trademaid.info/gsbhelp/GSBsys21ES.html



Here is the original system, and then with the weak indicator removed.
On any-indicator crossed its good to see the results of each indicator, one at a time.
Simple comment out the other 2 indicators

flt.png - 190kBflt2.png - 154kBsf-flt.png - 86kB

Siem - 24-2-2022 at 02:53 PM

In GSB there are Training (IS), Test (OOS) and Validation (OOS) percentages or periods and also Nth day Trade (IS), NoTrade (OOS) and All (IS+OOS).
I understand how the Nth day works, but I find it confusing that a date can be both IS and OOS for instance NoTrade and Training or Trade and Test and it seems that I'm missing a part of the puzzle there.

The modern way of system-building uses the Nth day method - what do you use the Training, Test and Validation percentages / periods settings for - is it to define Performance Filters periods?

admin - 24-2-2022 at 07:45 PM

Hi Siem
good question.
Only time I would ever use % is / oos is the first time I look at an unknown market. It would give you an instant visual clue if GSB is going to work with your current settings
So im summary I do not use it.
Typically you do inidcator testing with from say start of where contract is liquid to say 2017 (varies per market)
This is using nth no trade, but when it appears in gsb manager gui, its nth all. so indicator testing is 50% IS of pre 2017
then we do pass 2 building systems with the reduced indcatior set
it varies per macro / market but then we have a number of oos periods from 2017 ... to 2021 (still leave some oos again)
Exact dates can vary, but thats the gist of it

Any Luck on YM?

OUrocketman - 8-3-2022 at 07:16 PM

I've been making my way through the index markets since we have our latest methodology and have been met with great success on ES and NQ. Dow on the other hand seems much more difficult in 2021 and flat out dreadful in 2022. Is this what others are seeing on their systems or is there something I'm missing? I've been looking at DOW 30M chart, focusing mainly on CloseToHighLow3, but also have a group of other candidate filters including CloseLessPrevHighD, FractalDimsension, PercentR, Adx, CloseLessLowestLow, MoonSine (way skeptical of this one), Skew, FastK, CloseToHighLow1, AdxR, CloseToHighLow2, CloseOverPrevHighD, and S2R2. Thinking of moving to other time frames and potentially look at intermarket and multi time frame setups also. Before I did that, I thought I'd get a sanity check with the community in case I was trying to reinvent the wheel as they say.

Any input is welcome.

Thanks!

admin - 8-3-2022 at 09:32 PM

@OUrocketman, a user is looking into this. Might be a day or two. I have only one ym system given to me which is not brilliant

Its still very rare for 30 minute to not be the best.
We are testing starting session times earlier, but trading times later in the day.
This is very CPU intensive to do lots of tests, and im working on CL now

admin - 14-3-2022 at 03:26 AM

Im close to having CL market redone. If you can donate some workers, that will help. Im testing many fitness's to see whats best.
email me your share keys. You will also get my cl settings, though the fitness i used as default was np/dd

admin - 14-3-2022 at 03:47 AM

just realized, you need 64.64 or 64.65 to run my jobs, Just released it via gsb resource manager

bartek - 14-3-2022 at 03:51 AM

Quote: Originally posted by admin  
Im close to having CL market redone. If you can donate some workers, that will help. Im testing many fitness's to see whats best.
email me your share keys. You will also get my cl settings, though the fitness i used as default was np/dd


What are results so far? Promising?

admin - 14-3-2022 at 03:55 AM

the stats are promising, but i have not yet tried to build systems.
this is my benchmark, b c d e f are one year data each. e is not in the excel spread sheet. Getting an update for it done shortly


cl-bmark.png - 46kB

admin - 28-3-2022 at 12:48 AM

Update on CL.
I was close to system building, but found the last year out of sample results were not great. When I looked into why it was interesting.
The secondary filter has a strong bias to short systems. The net profit and profit factor are strongly on the short side..... and going short in the last year on CL is not a great idea (with the wisdom of hind-site)
Im now using a neutral rather than short bias secondary filter. Thanks also for those who contributed to my cloud. You will also see what exactly im working on by looking at the workers.

portfolioquanttrader2020 - 2-4-2022 at 07:24 AM

Hi peter
Do you propose any changes?

portfolioquanttrader2020 - 3-4-2022 at 01:59 AM

I leave you an article to improve crude oil methodology intraday

Attachment: Login to view the details


admin - 3-4-2022 at 04:49 PM

thanks for the article. Update on CL
My systems had a strong bias to short. One use who must have put in a lot of effort, made some tiny but massively significant tweaks. (just the session time and time of day)
and this gave a long bias. Im testing around these new areas now.
Its very cpu intensive as im not only testing session times, but also testing numerous times of day to start trading.
My first run on all of this is finished later today

Carl - 4-4-2022 at 05:28 AM

Hi Peter,

Here are some of the results of my analysis on price data to determine the best periods of the day.


CL 15 min bars regular session

Trend 2006-2020 average.JPG - 40kB Trend 2006-2020 per year.JPG - 60kB Volatility 2006-2020 per year.JPG - 51kB Volume 2006-2020 per year.JPG - 45kB

portfolioquanttrader2020 - 4-4-2022 at 03:04 PM

Hi Carlos
How do you do that analysis?

admin - 4-4-2022 at 05:31 PM

Thats basically graphing whats in market validation1, what session time to use.
https://trademaid.info/gsbhelp/Videos.html
In many cases this is the correct time to use.
There are significant exceptions that also can be very dependent on what secondary filter is used. IE Gold
Cl is also one where this doesnt apply.
On Cl there are two related but different issues.
The start of session time and the start of when you trade in that session.
Im testing this in the last week, and waiting on update to the GSB automation spread sheet to analysis the data

Carl - 5-4-2022 at 04:14 AM

Quote: Originally posted by portfolioquanttrader2020  
Hi Carlos
How do you do that analysis?


Python script, but can also be done in Excel.

admin - 7-4-2022 at 07:29 PM

This is the new automation spread sheet
Shown on CL is best session times, and trading start times
ie you can have a session time of 2am to 1330, but only start trading after 8am.
Whats interesting is the long short ratio also varies greatly with session time / start time.
long short ration is Long NP*AT/Short NP*AT but soon will only be log (Long NP*AT/Short NP*AT)
LSR IS in build 64.76 and above
This table is times / session times is not yet complete, and more workers from users will help.
You can see the lsr / long bias shows 330 am session time was a strong long bias, while other times were short bias.
Seeing as we are in a bull market for CL, i didnt want a system thats biased to be short. The ideal is neutral or a bit long

I will also give the results shown here (when im done) do anyone who gives me workers to use.
All I need is your share key

new-automation2.png - 119kB

admin - 29-4-2022 at 05:20 PM

Just a comment,
NQ seems to be the hotest market now, and one of the easyest to build. Deafult settings are working well.
Here is one of the systems I sell, but similar results have been made by even some failry new GSB users
Enclosed some screen shots.

nq-systems.png - 908kB

admin - 2-5-2022 at 11:11 PM

this is worth a read as it applies to all systems
https://trademaid.info/forum/post.php?action=edit&fid=2&tid=...

admin - 10-5-2022 at 01:42 AM

Bruce has played around with the idea of testing early session times for indicators to use, but we enter the trade later on.
His logic was he doesnt like gaps between days. I didnt agree with this so much, as gaps tend to continue the next day.
Whats great about GSB is we have all these little disagreements, but the data determines who is right. (Bruce!)
these tests are not conclusive as 4 tests were not done on all tests. IE session 800 trading at 930 only has 2 tests.
For the most promising tests, 4 tests needs to be done.
I will give the final results of this excel to those who contributed the workers. A big thank you to you all
Due to time pressures, I paused CL development work but hope to take that up again in the next week or two.
One of the reasons for focus on nq is, its really easy to build systems, easier than cl, and nq is just a hot market that trends well at this instant in time



nq-tests5.png - 97kB

bartek - 10-5-2022 at 04:19 PM

@peter what is the timeframe for these results? 30min?

admin - 10-5-2022 at 05:29 PM

@bartek, correct 30 minuts

bud67 - 13-5-2022 at 02:01 PM

Quote: Originally posted by admin  

I will give the final results of this excel to those who contributed the workers. A big thank you to you all




How can I contribute ?

admin - 13-5-2022 at 04:38 PM

Hi Bud67. Email me a unique share key in your workers. Under appsettings
I may not be doing any work till monday

admin - 13-5-2022 at 05:00 PM

I built nq with start date of year 2000 (actually 2001) compared to start dates 2007 01 01
Results are inconclusive, so might as well use 2007. This will be faster to build, and give out additional of sample 2001 to 2007. Not 2007 to 2007 are poor years regardless

startdatesNQ.png - 63kB

bud67 - 14-5-2022 at 07:41 AM

Quote: Originally posted by admin  
Hi Bud67. Email me a unique share key in your workers. Under appsettings
I may not be doing any work till monday

Which appsettings do I have to edit ?

GSBManager
GSBWorker Cloud
GSB Worker

admin - 15-5-2022 at 04:49 PM

Thanks for asking
@bud67
can you do gsb worker, but then I will get your cloud whenever RM is open.
If you want to revoke me using it, remove the share key, save app settings, kill all workers, then restart the workers

admin - 15-5-2022 at 05:27 PM

on NQ I tested secondary filter compare type. This had always been hidden in GSB until a year or so ago.
What we see is a mild increase in out of sample fitness with compare2, but amount of systems in favorites bcd was better on compare1
There will still be variation between identical tests, so I would say the results are inconclusive.
Different secondary filter compare may also give a bit more variation in systems, which is a good thing

compare.png - 87kB

bud67 - 16-5-2022 at 04:19 AM

Quote: Originally posted by admin  
Thanks for asking
@bud67
can you do gsb worker, but then I will get your cloud whenever RM is open.
If you want to revoke me using it, remove the share key, save app settings, kill all workers, then restart the workers

Done in GSBWorker.

admin - 10-6-2022 at 01:13 AM

Just an update
I paused working on crude oil to go to nasdaq, as its such a hot market (+$7000 trades in the last month or so)
In summary I have found 2 or 3 things that have improved results.
Ive also done lots of tedious boring checking, tweaking, trying all sorts of things - many not rocket science.
ie 15 min, 15 30 min, 30 60 minute... the possibilities are endless
trying subtle changes to indicators etc.
I hit the point of diminishing returns for effort, but much of my tedious work confirmed the new things I have found are valid.
I hope to publish this in about 2 weeks (roughly)

GSB NT has made significant progress. There were issues with one version of NT, and still have some indicators that dont match.
Apologies that the task has blown out so much

The back end of GSB SQL server has improved WF, and the next build of GSB has improved WF handling. This wont change WF results apart from driving the workers a bit better


admin - 13-6-2022 at 07:29 PM

With fear and trembling I have posted a significant update.
One section is really simple and important. The other on manual optimization is nice to have, but not critical. Its complex, outside the scope of GSB support and I fear users who know nothing trying to grasp a concept thats very complex, somewhat subjective, that likely needs a lot of tuition to grasp. Feedback welcome.

Part of my methodology in life and trading comes from the great book Atomic habits by James clear. Worth reading by every person on the planet. Big improvements come through the cumulative effect of small improvements.

admin - 13-6-2022 at 08:05 PM

Opps, forgot the url
https://trademaid.info/gsbhelp/Advancedoptions.html

OroDan - 14-6-2022 at 11:14 AM

Thanks Peter, very interesting.
I managed to improve a system, -DD, + NP.
I'll try other systems over the weekend.
The weights can also be optimized, are there any contraindications? Thank you

admin - 14-6-2022 at 07:07 PM

Orodan
with entry type AIC/ NCC, weights must be kept at zero
others entry types use
0.25 min
2 max
step 0.25

and for wf weights
others use 0 min
2 max
step 0.25
You really need gsb automation to work out best fitness for each setup.
IM going to publish updates to NQ systems in the next month or so.
If all the hard work has been done for you, its very fast to get good systems.
If its a new market, its a great deal of work

admin - 16-6-2022 at 02:22 AM

The first part of NQ work is finished, but I am still to document it.
Those who contributed workers will get the full details, including all settings times, time zones bar intervals, entry types etc on monday.
For the rest I will show whats close to the ideal setup.
My final system is here. https://trademaid.info/gsbhelp/GSBsysFLT4.html
Pyramiding gave a decent improvement to results.


Im now doing a completely different NQ setup,
so same deal applies. Send me your worker share key and I will give you the results.



admin - 24-6-2022 at 06:46 PM

I have started to update the building NQ ES systems page with the updates from my research in the last month.
Those who contributed workers to this project, will have received an email about this last week.
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...
Next week should be complete. Files used will be included with GSB one or two builds away.

admin - 13-7-2022 at 01:18 AM

Just an update
I spend a entire week trying to improve GSBsys21ES. It was painfully slow work as the code is very CPU intensive compared to other systems.
Ive found one non very original filter that makes a very decent improvement. This will make its way into GSB in time.

Ive also found a second very simple but effective filter that in time I will test in many other GSB systems, and add into GSB if it works on other systems.
This week im working very few hours, as have a bug, so need to rest as much as possible.
GSBsys21esV3 is finished, but not documented. There will be a paid upgrade to the new version as soon as time permits.

Carl - 13-7-2022 at 02:33 AM

Get well soon, Peter!

autotrader2023 - 13-7-2022 at 07:29 AM

Sorry to hear your not feeling well, hope you recover soon.

portfolioquanttrader2020 - 13-7-2022 at 12:23 PM

Peter get well soon

admin - 14-7-2022 at 12:10 AM

Thanks all. Suspect i have Covid. Was happy to say was feeling 100% better in just over 1 day from when I didnt feel well, but then had to do some physical work till 10:30-pm when should have been resting. Nothing could be done about it, had to be done.
So next day wasnt better again. Simply have a muddy head and feel like not slept well. No temperature, no sore throat or any other aches and pains. I could not have been better prepared to get covid - as far as things needed to get better soon.
If anyone is interested in what I did, happy to post that. There is simply so much you can do before and after you get Covid to look after your health.


GSBsys21v3 is out.
https://trademaid.info/gsbhelp/GSBsys21ES.html
Also real time results of GSBSYS21ESV1 published.

GSB1.XES RESULTS updated too. And the winner of the 3 GSBSYS1.X SYSTEMS BY A LONG WAY IS??
https://trademaid.info/gsbhelp/GSBsysES15.html

admin - 22-7-2022 at 12:41 AM

After many months, the NQ / ES documentation is updated.
The documentation is not finished yet, but the GSB files have all been made and will be pushed out in the next build.
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...

If any one wants the entire GSB folder zipped up, with the settings, then let me know.
Users who contributed to the NQ project and got the pcloud url with lots more extra info should not need these files.

admin - 25-7-2022 at 09:54 PM

This documentation is new greatly updated and finished, though it needs some tiny tweaks.
https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...

Hope to push all the files used in the next build of GSB

admin - 7-8-2022 at 06:40 PM

I am revisiting building gold systems with the new macros used in the recent NQ update
see https://trademaid.info/gsbhelp/BuildingNasdaqSP500orDowsyste...
The new macros incorporate some great ideas added by Nick and Bruce
From what I can see there is substantial room for improved results purely by testing the session start times, and (later in day) start of trading times.
This will not be a lot of work for me to redo this market.
I will publish full results to those who supply cloud workers for this project.
Please email me your share key, else have workers running on previous share keys given.

admin - 12-8-2022 at 01:36 AM

Just an update on Gold.
Ive run all my tests using 80 or so indicators I used on gold last time I tested it.
Im now doing the same tests on 180 indicators, and will then do the same tests on only the indicators that were used in the 180 indicator test.
It would seem logical that if you use all indicators, take the ones used or commonly used (and perhaps repeat the process a few more times) that you will get the best indicators.
My limited testing seems to imply this is not correct, or perhaps that the loop to reduce indicators should be done only 1 or 2 times.
Those who contribute the workers will get the results of this research when Im done. Those who dont contribute will get a high level default settings of whats been learnt in this process.

Daniel UK1 - 12-8-2022 at 03:22 AM

Quote: Originally posted by admin  
Just an update on Gold.
Ive run all my tests using 80 or so indicators I used on gold last time I tested it.
Im now doing the same tests on 180 indicators, and will then do the same tests on only the indicators that were used in the 180 indicator test.
It would seem logical that if you use all indicators, take the ones used or commonly used (and perhaps repeat the process a few more times) that you will get the best indicators.
My limited testing seems to imply this is not correct, or perhaps that the loop to reduce indicators should be done only 1 or 2 times.
Those who contribute the workers will get the results of this research when Im done. Those who dont contribute will get a high level default settings of whats been learnt in this process.



Hi Peter, i am not sure i follow.

Is this a test for finding best process for narrowing down your indicators to top 8-10 etc ?, OR is this is test of process for finding some amount of indicators to start of with before you narrow down ?

I am not following :)

I know that you like to narrow down to top N for each test you do, but i cant relate to your mentioning of 80 or 180 etc..

Confused

admin - 12-8-2022 at 06:01 PM

Good question and I see why you asked it.
Its for the selection of indicators before you look for the top 10
Related is if the user is in beta mode where there are mode indicators than the standard GSB purchased version, or if your in trial mode where there is less.

Part of my logic was around the fact that yesterday I did for example 180 indicators (some not available even to beta users) doing a total of 48 tests with different session times / start times with all tests (indentical) done two times
and found gsb used only 105 indicators.
45 of these were used only once.
SO it seems logical for me to do the next test with 105 indicators.
On nq I also removed the indicators that were used only 1 time, and found the indicators used only 1 time will still very useful and helped results.
Why some indicators are avialble only to me is there are 3 criteria for me to release them
1) The indicator must be shown to work on some markets
2) I must have the TS code
3) it must match gsb to ts

Daniel UK1 - 13-8-2022 at 02:01 AM

Quote: Originally posted by admin  
Good question and I see why you asked it.
Its for the selection of indicators before you look for the top 10
Related is if the user is in beta mode where there are mode indicators than the standard GSB purchased version, or if your in trial mode where there is less.

Part of my logic was around the fact that yesterday I did for example 180 indicators (some not available even to beta users) doing a total of 48 tests with different session times / start times with all tests (indentical) done two times
and found gsb used only 105 indicators.
45 of these were used only once.
SO it seems logical for me to do the next test with 105 indicators.
On nq I also removed the indicators that were used only 1 time, and found the indicators used only 1 time will still very useful and helped results.
Why some indicators are avialble only to me is there are 3 criteria for me to release them
1) The indicator must be shown to work on some markets
2) I must have the TS code
3) it must match gsb to ts


Good it, so its a test of the process.. ok make sense now.

On a sidenote, i have never reduced available indicators before a top N ind narrow down, for any other reason that to remove for example up/dn volume based ones (since only TS has this).... but apart from that i use all.

BTW, speaking about indicators, do you have an ETA for VWAP and the other one i sent you code for ?

admin - 14-8-2022 at 06:45 PM

Hi Daniel,
you are like just about all GSB users in that you start with the full range, else the range of indicators chosen in default settings.
Normally a lot of work has gone into the best indicators to be used in default settings. This was the case with recent nq settings release and the private NQ info given to those who contributed workers.
What Im saying is there is significant room for improvement reducing the initial set of indicators . This has been done on the default NQ settings published in the last month
On some markets there is substantial improvement when up down volume is used. With difficulty this can be done if you combine ts historical data with iqfeed live data.
I think iqfeed gives up down volume for a short length of time. This work around is fraught with issues. Its bad practice to build systems on one data set, and trade using another.
Guess its less bad to use OHLC data from the same provider, but volume from the two providers.
Had a trial user get barely any systems in fav D on the new nq settings. Cause was not using volume. So volume is really important for SOME markets. NQ especially.
A cleaner fix is to use TS account live data in MC. I know that has its issues too, as some users in some countries cant get a ts account.

vwap will not happen until the GSB NT is released. This is the top priority and I will give the programmer no distractions unless an issue is critical. Thanks for your patience in this.

admin - 14-8-2022 at 07:13 PM

@Daniel,
I would like to test the effect on NQ not having up / down volume.
Am I correct that you can use volume, just not indicators that use up/down volume?

Daniel UK1 - 15-8-2022 at 02:21 AM

Quote: Originally posted by admin  
Hi Daniel,
you are like just about all GSB users in that you start with the full range, else the range of indicators chosen in default settings.
Normally a lot of work has gone into the best indicators to be used in default settings. This was the case with recent nq settings release and the private NQ info given to those who contributed workers.
What Im saying is there is significant room for improvement reducing the initial set of indicators . This has been done on the default NQ settings published in the last month
On some markets there is substantial improvement when up down volume is used. With difficulty this can be done if you combine ts historical data with iqfeed live data.
I think iqfeed gives up down volume for a short length of time. This work around is fraught with issues. Its bad practice to build systems on one data set, and trade using another.
Guess its less bad to use OHLC data from the same provider, but volume from the two providers.
Had a trial user get barely any systems in fav D on the new nq settings. Cause was not using volume. So volume is really important for SOME markets. NQ especially.
A cleaner fix is to use TS account live data in MC. I know that has its issues too, as some users in some countries cant get a ts account.

vwap will not happen until the GSB NT is released. This is the top priority and I will give the programmer no distractions unless an issue is critical. Thanks for your patience in this.


Hi Peter, Yes i am sure you are correct, and that it surely is better.
I wonder what you mean with "better", as in resulting in a an end opt setting that you build live systems with, that is superior to a process involved where you would use all indicators ?

In regards to up/down volume, its a very complicated area, mainly as you say in terms of execution. I have decided to keep it simple and not use vol indicators. But I am not against dev on TS data, and then trade live in IQ, since you would get up/dwn, but ongoing stats would be complicated.

In my process, i never do top 10 narrow down, until at the very end of my process when i actually select my indicators to trade with. Before that they are fixed to a small number. If i would change indicators during my test process when i am at the same time are changing test values, i would see this as a multivariate test, and that would complicate it for me to analyse.

imho

Daniel UK1 - 15-8-2022 at 02:24 AM

Quote: Originally posted by admin  
@Daniel,
I would like to test the effect on NQ not having up / down volume.
Am I correct that you can use volume, just not indicators that use up/down volume?


Yes, if i understand you correct, I have sometimes ended up with vol indicators, where i only feed up vlm, i.e IQ data historc.


Daniel UK1 - 15-8-2022 at 07:15 AM

Quote: Originally posted by admin  
@Daniel,
I would like to test the effect on NQ not having up / down volume.
Am I correct that you can use volume, just not indicators that use up/down volume?


Re-reading your message, i think i misunderstood your message, yes i can use volume indicators, but i sometimes remove the up down ones to avoid the issues.

admin - 15-8-2022 at 06:47 PM

@daniel, Ideally it would be good to see the 'cost' of doing this. I know volume is critical for nq, but not sure on up down volume

admin - 29-8-2022 at 01:06 AM

I have had a number of users complain about lack of trading activity, even though the markets moved a lot.
Happy to say no one was complaining about this Friday. I had very high slippage but got just under $10,000 per NQ contract.
This is why I have been encouraging users to start on the NQ market. Its very GSB friendly, well documented and we have been in just amazing times. Its been realistic to get $7000 wins but 10k is awesome
Biggest daily win on this system was $11800 in March 2020

Shown here is a print log of the dates when the secondary filter and tertiary filter are both true. You can see a month from 7/19 till 8 / 26 where the filters didnt allow trades


nq-systems.jpg - 362kB

timemeantnothing - 30-8-2022 at 08:11 AM

Friday was definitely a great day. I'm in sys1ES and we ate well that day too. Which NQ system are you referring to? Also, would you ever consider making a system that trades off a smaller time frame to capture more occurrences?


Quote: Originally posted by admin  
I have had a number of users complain about lack of trading activity, even though the markets moved a lot.
Happy to say no one was complaining about this Friday. I had very high slippage but got just under $10,000 per NQ contract.
This is why I have been encouraging users to start on the NQ market. Its very GSB friendly, well documented and we have been in just amazing times. Its been realistic to get $7000 wins but 10k is awesome
Biggest daily win on this system was $11800 in March 2020

Shown here is a print log of the dates when the secondary filter and tertiary filter are both true. You can see a month from 7/19 till 8 / 26 where the filters didnt allow trades

 Pages:  1  ..  8    10    12