GSB Forums - Powered by XMB 1.9.11

Walk forward

jptann - 2-8-2017 at 10:44 PM

Peter, one of the most interesting discussions is around how you do the walk forward calculations. After the walk forward, is your GSB recommending the last parameter set used in run 10 or do you use the same parameter set that you started the WF with?

I don't think there is any right answer here. I try to use Monte Carlo and sensitivity analysis on the results to identify the full parameter ranges. Is the GSB capable to provide this?

It is no big deal because there are other applications out there that do a fine job of this.

By the way, I usually run 5000 Monte Carlo runs at a 95% confidence interval and what I am really looking for is a parameter space where the "risk of ruin" is as close to 0 as I can get. Using $10K per contract, it is usually difficult to get below 5% as the risk of ruin but using $30K per contract, this can drop below 1%.

I guess are you thinking of a module that would give us this capability?

Thanks in advance

admin - 3-8-2017 at 04:23 AM

We could write a big book on this.
We use the last parameters. These are the ones that work best over the entire series.
I also like to see stability in parameters. If the parameter set is stable, it means its locked on the best settings. The earlier you see this the better.
EWFO has stress testing. The build with this on is yet released though. I dont think its an essential to stress test.
New EWFO also has parameter stability measurement for each fitness. This is new as of today.
Portfolio analysis does do Monte Carlo. GSB will have direct output to PA with in a few months.

jptann - 4-8-2017 at 12:07 PM

thanks, I will have to start getting into EWFO. I use MSA to do this now. Just a bunch of Excel steps but I like your idea of stability in EWFO.

parrdo101 - 7-12-2017 at 10:43 AM

Ok, the last parameters, with eye on previous stability - got it. How do you insert them (and, they are not named to match them with anything - anywhere). Insert into the Trade Station script? But which to where? No names to associate. See snippet, where I see 13 unnamed parameters for each line, but I'm noticing the Trade Station script that is associated to this walk-forward has 21 inputs (2 of which are dates). Pardom my dumb-in-the-extraordinary, but I at the moment don't even apparently know what the parameters in the snippet are, it looks like.

I was trying to relate this whole area to the Trade Station Optimization Report (SOR), where one can click the start-node grey area far to the left of the parameter set, and the inputs pop into the TS (Trade Station) chart (but not the .ELD script unless you manually reset it). Once they're in the chart, it's the same forward as if they're in the script and active, and if you like what you see, you can permanently embed your script with these popped in values from the SOR. I'm so far not seeing anything like being able to "pop these last parameters into the GSB script for TS"...? Am I overlooking something?

admin - 7-12-2017 at 03:47 PM

Great result in your wf stability. I will explain each one. About to jump on a plane, so am intermittently online in the next 24 hours.
the inputs will all appear on order, like what is below.
As optimize data steams is off as default, they will be omitted, unless this is set to true

Inputs:
i1Data(1), //omitted as its data stream
i1length(10),
i2Data(1),//omitted as its data stream
i2length(21),
i3Data(1), //omitted as its data stream
i3aa(0.35),
i1Weight(1),
i2Weight(1),
i3Weight(1),
entryParams(20),
iSFData(1),//omitted as its data stream
iSFWeight(1),
sfEntryLevel(4),
beginDate(20071012),//omitted as its date
endDate(20171206);//omitted as its date

parrdo101 - 12-12-2017 at 07:16 AM

"I will explain each one"

I eagerly await this; still no idea really what to do in this area - I feel some compel to put them somewhere, still very hazy where they're coming from. The whole tie-in, what to do with them, THESE PARAMETERS SHOWN IN THE WALK FORWARD TAB.

admin - 12-12-2017 at 03:20 PM

Its good for you to have high level understanding as in my post above.
But if youve done a WF, you dont need to do anything but change the enddate input
The code will not trade after this enddate

parrdo101 - 13-12-2017 at 04:58 PM

The circled in the snippet - equivalent to your "Current" status under your GSB Walk Forward, after a GSB wf has completed, I presume? (The snipp is the TS wfo - obvious.)

Given I did the optimizing, (and that I've used the TS WFO a bit), I could walk those current's back into the strategy inputs and affix them there in a rush of the wind.

I simply cannot figure out, to which, to walk yours - missing having watched you pick your inputs to optimize? (This seems to be it.) Given these presumed parallels of the 2 wf platforms, (your GSB plat being a CASB w/integrated wf'g, to be more exact), where do I "watch you pick your inputs to optimize" - or something, I don't know? It occurs to me that I then could walk these inserted "Current's" forward in a finale validation (I've avoided certain forward data in my TS chart just for that purpose), to see the general power of the WF optimizer in this GSB. (Precludes my limited ability to correctly evaluate.)
:blush:

I observe with awe some of the things you know for granted.

admin - 13-12-2017 at 06:30 PM

im not clear on much of what you write.
First of all the wf you show has only 29 trades, and is 2017 only.
This sample is so small, you are wasting your time. I feel 100 trades is the extreme minimiun for a valid sample, and even then expect greater out of sample degradation as the same size is so small.
Also, youve used rolling, not anchored. This is a really big no no for small sample sizes.
Your fitness is np, not np* average trade. This isnt wrong but I would start by using the same fitness as gsb.
Now the bigger picture.
why are you doing a ts wf when you can to a much faster gsb wf. - There are at times valid reasons for doing this, but im not sure why you are doing it.
valid reasons are you have added additional code to a gsb system which cant be wf in gsb
Making the assumption you are only using * or */ operators,
you want to optimize the indicator length(s) * all 3 (or maybe 5 in your case)
and the secondary filter parameters.
if SF is closeDminusClose then you have one parameter. If its genetic then you have normally one length and one other parameter. Hope this reply helps.

parrdo101 - 14-12-2017 at 07:51 AM

My snippet of the TS WFO was simply to show you the inputs I had optimized - very hasty to pin area of interest. Because I personally picked the optimization, I can walk those "Current's" back into the TS chart strategy and observe my reserved OOS from there, play a bit.

This was in no wise a legit TS WFO run.

If your system is "mimicking" TS WFO as to the "Current's," in your system you seem to be referring to "parameters" in the header where maybe in the TS WFO they are calling them "inputs" in the top tab header, right?

THE WHOLE THING HERE I'm asking about, given much of the foregoing, is how to walk the "Current's" out of your system back into the inputs they belong to - I absolutely can't because - if you parallel your steps with the TS WFO steps - I didn't see, can't see which inputs you're optimizing. Don't know how to match your currents to their input name - if paralleling TS steps.

I have no intent of using the TS WFO having seen yours, which I agree is much better. Please try understand my confusion in that I have learned "all I know" as to any WFO (theirs) on their system, and now walking into yours, I am tending to think the steps are similar...if so, I'm therefore trying to figure out how to get your "Current's" back to the inputs ("parameters"? - usage?) they match. No way I can unless (think TS version of steps) I did the optimize, saw what I did, saw which inputs I optimized so I can walk them out of TS's "Current's," or their "Recommended parameters for next X days" - you have to place these back in manually in TS - I'm trying to do it with yours etc.... Can't see to which inputs they belong.

Wean me from TS - light at last! - and running with yours. Just imagine me still able for now to only think in their terms and system somewhat.

admin - 14-12-2017 at 02:47 PM

Being on holiday, this is a bit harder for me to answer due to lack of decent screen capture program on my laptop
Go to the WF tab, and look at the last line of parameters.
These parameters are the ones used in the WF ts code.
The ones to optimize are in red below.
You can also get a feel for the range to use by looking at the range GSB found for each parameter.
(Good to add a bit of margin on values GSB used)
Still not aware why you even need to do any of this.
Inputs:
i1Data(1), //omitted as its data stream
i1length(10),
i2Data(1),//omitted as its data stream
i2length(21),
i3Data(1), //omitted as its data stream
i3aa(0.35),
i1Weight(1),
i2Weight(1),
i3Weight(1),
entryParams(20),
iSFData(1),//omitted as its data stream
iSFWeight(1),
sfEntryLevel(4),
beginDate(20071012),//omitted as its date
endDate(20171206);//omitted as its date

parrdo101 - 15-12-2017 at 03:14 PM

In the TS WFO they say like, "Suggested inputs to be used for the next X days." That I've understood, if one takes the suggestion you put them in as the inputs 1 by 1 going forward.

Your "Current" parameters at the end of a WF optimize in the GSB: You mean this the same way as TS does?

admin - 15-12-2017 at 03:39 PM

Quote: Originally posted by parrdo101

In the TS WFO they say like, "Suggested inputs to be used for the next X days." That I've understood, if one takes the suggestion you put them in as the inputs 1 by 1 going forward.

Your "Current" parameters at the end of a WF optimize in the GSB: You mean this the same way as TS does?

GSB & TS are similar in the way they do WF in many ways.
My EWFO is just a mega upgrade to ts WF, but sort of works the same. The GSB WF is also the same in many ways. (too hard to explain the finer detail without writing a book)
If what you mean you WF one of the 10 runs into another 10, than thats not what any one intended and will work poorly.
(Im not clear exactly what you mean by your words.)
" if one takes the suggestion you put them in as the inputs 1 by 1 going forward."
Technically is ok to re wf after a period of time, and this is why Wouter who wrote ts WF wrote use parameters for x days.
However if your hit parameters stability for a good number of runs. (at least 3) its likely to be a pointless exercise the produced no change in parameters.

Current parameters are the final parameters that were found to be best. This is similar to TS.
Hope this helps but I have not understood exactly what you have asked.