admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSB architecture and lifting the average trade & profit factor
One of the things we are all waiting for is GSB to give higher profits per trade.
As is GSB works great for getting robust entry techniques that pass walk forward tests. The gsbsys1es showed how average trade could be lifted from
around $22.15 to $229.93 , Profit factor 1.21 to 2.9. see GSB documentation_1011.docx and GSB Tutorial_number2a_gsb1_es30.xlsx
This was done by adding one filter, and different parameters.
The solution to improve the performance metrics is to use secondary filters.
ie current gsb pre build 19.96 format is
if ... then buy
In build 19.96 it can use a secondary filter.
if .. and oscilator4(len4)*w4>entrylevel2 then buy
There is a very similar concept to improve metrics that is well proven in my previous legacy GSB.
The concept is to increase the number of oscillators from 3 up to 5.
It also worth testing whats going to be better. 4 oscillators and a secondary filter, or 5 oscillators and no secondary filter.
This is now going to be possible because we are becoming much less dependent on TS to WF the system.
With 3 oscillators, if we have 3 lengths, 2 weights and a threshold, and we have 10 parameters per variable, that gives is 7^10 282 million
combinations.
This is at the limit if what TS can optimize.
If we go to 5 oscillators, we hit 3.4 billion combinations. That's still going to take a lot of CPU time to optimize.
GSB is going to be better suited to large optimization sizes than TS. The good news is sometimes we have less variables. Ie AccumDist doesn't have a
length.
I am hoping this is coming not to far away in a future GSB build.
Note also the current WF methodology needs improvement, and its going to be a couple of weeks away. Current WF is not a true Genetic algorithm.
|
|
|
crazyhedgehog
Junior Member
Posts: 14
Registered: 9-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Peter, thank you for sharing your thoughts and the roadmap. The one question I have re filters is just what role they are playing the the system.
Should they filter out some smaller trades or trades during certain trend or volatility periods when this particular strategy doesn't perform well? Or
should they be one of the main signals driving the strategy? The reason I am asking is that in gsbsys1es if you comment out the GSB signal and just
keep the filter, it performs almost as well. Which made me wonder how to make this approach of first generating the main GSB signal and then adding
filters work in a way that the filter doesn't actually become the system which could result in less robustness. What do you think?
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Im not clear on what you say with gsb1es. The closeD filter may be profitable by itself, but i doubt you will get pf of 2.9 as is.
Filters after all the conditions are true tend to filter out the trades that are not profitable.
ie if the close is well below the ClosedD(1) then going long is a bad idea on ES.
I have many systems with varied results that ll use the CloseD filter. So there is much more to making the system than one filter.
If you build another system with closed, the results also will vary a fair bit.
One thing that portfolio analyst also showed me is that systems may appear correlated.
However they correlate more on winning days than loosing days. This is a good think as false moves trigger less systems, and good moves trigger many
systems
The GSB system you make should be proven by a linear equity curve showing it consistently makes profits, and passing a WF test. They way the rate of
them working out of sample is so good.
Hope my comment helps. At the end of the process, a wf test will sort much of this out. If weighting of zero on the indicators was the best result, it
means they are not effective.
If weighting isnt zero, then WF thinks they are useful.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
GSB 20+ now has secondary filter and up to 5 indicators. To lift metrics further we still need to use some sort of additional filter. ie close>closed,
time of day etc.
Its also going to help to have more indicators. They will come fairly soon. As GSB has been selling well to Alpha /Beta testers, its enabled me to put
one more programmer on the job.
Thanks to all of you who support this work and have a vision for where its going.
Enclosed system with average trade $60.69 as an example.
Peter
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Might be a thought to include something where the user can select the maximum number of oscillators to use?
Carl
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Its there on the left side. See picture. Sorry but I haven't done the docs yet on version 20.1
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Great!
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Just built a system on 20.4
average trade $172.29. pf 2.76. I enclose report.
20.4 a bit buggy to release
This has closeD(1)-close filter
 
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Note the x axis on GSB is date, and on TS its time. Hence big difference in curves.
Here is the WF results. This system took me well under an hour to make. Minutes really.
TS report is not using the improved WF parameters.
pdf is using portfolio analyst with slippage and commission
Attachment: Login to view the details
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Here is another one.
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Peter, thanks for your mail. Am in the train right now. Impressive results! Will take a look later today.
Thanks Carl
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Results GSB 20.4
May 20 2017
These steps describe my first test run with GSB version 20.4.
Step 1. Export the data streams from TS to txt files
Remark: I removed the price data after December 31 2015 from all the data streams to be able to see the REAL out of sample performance from January 1
2016 until May 19 2017.
Step 2. Run GSB (i.e. pop 300, gen 5000)
Step 3. Choose interesting looking strategies (pearson, average trade, PF)
Step 4. Perform WF on these favorite strategies
Step 5. Choose a strategy with a good WF/OOS equity
Step 6. Copy the strategy code to Tradestation
Step 7. Compose the corresponding data streams in a TS workspace
Step 8. Run the strategy in Tradestation with the GSB WF parameter values
Step 9. Subtract costs and slippage
Step 10. Try different levels of stoploss
Is this example I didn't optimize any of the parameter value GSB generated. I just used the values after performing the WF.
Please see attached Word document with more screenshots.
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Results GSB 20.4 ( Word document )
Word document with screenshots.
Very nice results after the first test run with GSB 20.4 !
Attachment: Login to view the details
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Hi Carl,
That's a great result. Esp considering day trading the USA indices has been very poor in the Trump era. If anyone else wants to post similar examples,
please give 2 years out of sample to get a better feel of OOS
I suspect you did not use the WF final parameters all the way thought this, but used the original GSB parameters.
If youve restarted GSB, with the current build you wont get the parameters back. There was a feature to re-WF any system, but it was removed when we
went to secondary filters and up to 5 indicators.
The feature will return to GSB in a few weeks.
Picture here summarizes Carl's experience. 2016 on-wards was never seen by GSB
|
|
|
Carl
Member
Posts: 342
Registered: 10-5-2017
Member Is Offline
Mood: No Mood
|
|
Hi Peter,
For this example I used the Tradestation script from the Walk forward tab.
Was this the right choice?
Carl
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Yes, that's what I would use for the final system. The profit was a fair bit higher than the original system, but im not sure if you chose it or not.
Slippage and commission could have reduced it so it looked similar to the original curve.
|
|
|
admin
Super Administrator
Posts: 5060
Registered: 7-4-2017
Member Is Offline
Mood: No Mood
|
|
Here is another system that I'm very happy with.
http://www.trademaid.info/forum/viewthread.php?tid=15#pid104
Good profit per trade, glowing pass of WF test. (EWFO and GSB wf)
|
|
|
![[*]](images/davis/default_icon.gif){kind=icon}
