OmniTrader Professional Forum OmniTrader Professional Forum
forums calendars search
today this week
 
register logon control panel Forum Rules
You are currently browsing as a guest.
You should logon to access more features
A Self-Moderated Community - ALL MEMBERS, PLEASE READ!
Vote for Members who contribute the most to your trading, and help us moderate content within the Forums.


  Current location        Thread information  
OmniTrader Professional Forum
OmniTrader Pro General Discussion
OT2019 PR2A Global Optimization
Last Activity 12/9/2018 7:38 PM
10 replies, 406 viewings

Jump to page : 1
Now viewing page 1 [25 messages per page]
 
back reply
Printer friendly version

^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : OT2019 PR2A Global Optimization
Posted : 11/30/2018 4:19 AM
Post #30193

PR2A added support for global optimization to be used in conjunction with dynamic optimization.
"Dynamic Optimization allows you to incrementally optimize your strategy over the entire data set, so you can see exactly how frequent re-optimization would affect strategy performance."

I want to have a look at (say) the last 4 years of data and see what would have happened if I had each month optimized using the last 6 months of data and traded that for the next month, using the standard dynamic and 2 global dynamic methods.

Q1. If I have a dataset of 960 periods, and I want to use a floating backtest of 120 periods with a 20 day forward test, dynamically updating on monthly boundaries, how do I set that up in the Todo List Test Settings?

I tried Floating Backtest 120 periods, forward test 20 periods and optimize on monthly boundaries and ticked Dynamic optimization but that only looks at the last 140 days. How do I "incrementally optimize over the entire dataset" of 960 days as per the explanation and diagram in Omnitrader Help for "Dynamic Optimization" as shown:





Q2. One of the Optimized systems provided by Nirvana is the MJP Long Term Momentum system. In the systems block it uses RSM-C optimized with Performance Metrics based on Next Pivot Point.

Is it possible to use Next Pivot Point AND optimization to simulate how the optimized strategy would have traded if it was reoptimized at each new optimization boundary or will this give spurious results?




[Edited by John W on 11/30/2018 5:44 AM]

Attached file : Dynamic Opt.png (63KB - 351 downloads)

^ Top
Jim Dean

Sage
200010001002525
Posts: 3173

Joined: 3/13/2006
Location: L'ville, GA

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 7:35 AM
Post #30194 - In reply to #30193

1. Use 960-120 = 840 as the forward test periods. The monthly boundaries input does what you were trying to do with the 20 input. This is how DO always has worked.

2. NPP exits (aka Market Reversals) are never realistic. They use a crystal ball to determine when to exit. However, using NPP as a metric during the backtesting to determine the next walkforward Param setting is ok.

[Edited by Jim Dean on 11/30/2018 7:39 AM]

^ Top
Barry Cohen

Icon
100050010010010010025
Posts: 1931

Joined: 1/1/1900

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 8:22 AM
Post #30195 - In reply to #30193

What Jim said, thanks Jim!
^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 2:01 PM
Post #30196 - In reply to #30194

Thanks Jim, does point 2 mean dynamic plus NPP with DO switched on will not give spurious results?
^ Top
Jim Dean

Sage
200010001002525
Posts: 3173

Joined: 3/13/2006
Location: L'ville, GA

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 2:07 PM
Post #30197 - In reply to #30196

I’m not sure what you mean by “spurious”. Illegitimate or fake?

Well, in the strictest sense I suppose that all backtesting is illegitimate if viewed as an accurate predictor of future income. But usually, with adequate sample size, a series of apples to apples backtests, with controlled alternative variables, can provide legitimate help with choosing a method to use for actual trading.

Hmm. Rereading that, it appears that I’ve been watching too many Big Bang Theory reruns. Sounds a bit like I was channeling Sheldon. ;~)

[Edited by Jim Dean on 11/30/2018 2:09 PM]

^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 2:43 PM
Post #30198 - In reply to #30197

Ha! Unlike TV characters I’ve found you do know what you are talking about.
NPP appears to be a form of optimisation but the systems block is not outlined in red when it is used. DO is supposed to deal with optimization and eliminate/ reduce the crystal ball component.
I don’t know how NPP is constructed.
For example if it’s based on ATR then looking at the last 120 days and freezing that setting for the next 20 days using DO would apear a valid approach?

[Edited by John W on 11/30/2018 2:58 PM]

^ Top
Jim Dean

Sage
200010001002525
Posts: 3173

Joined: 3/13/2006
Location: L'ville, GA

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 3:02 PM
Post #30199 - In reply to #30198

There are several things needing checking, etc to get the optimization to be operative. When it’s on, the block should be red (IIRC).

NPP means next pivot point. That is, the Parm’s that are to be optimized in the block are tuned by evaluating the selected metric (APR, etc) using a “faux-exit” determined by the NPP.

Pivot points (ie local extreme high/low prices) are identified once a certain amount of price change (&/or #bars) has occurred after a given H or L (and before it). Thus the “ideal” faux-exit Pivot is not known until several bars later. Therefore, NPP (aka Market Reversal) H/L prices never are representative of actual trading.

But for dynamic optimization runs, the idea is to figure out the “best” parm value for the upcoming month (sic), based on recent prior months. The optimization is turned off during actual trading, as well as in the statistics-gathering from port sim.

I’m not totally sure what happens to trades that are active at the “borderline” between a given walk forward step’s backtest/tuning range and the forward test beginning - I’d hope that they are simply chopped off rather than including bars up to the next NPP that would have extended into the FT zone. Maybe that’s what you are getting at.

Maybe Barry can clarify that - though it probably will require a programmer to dig into the code to figure it out. That’s the only “hmm I wonder” aspect to it, afaict.

[Edited by Jim Dean on 11/30/2018 3:05 PM]

^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 3:43 PM
Post #30200 - In reply to #30199

Thanks Jim, you raise a good question on DO generally "what happens to trades that are active at the “borderline” between a given walk forward step’s backtest/tuning range and the forward test beginning".

There is another question that come to mind from this discussion.

It appears that NPP is an optimization feature.
If NPP is an optimization feature then does DO as it is working now deal with NPP as it does the other optimized strategies.

If not, should it? For instance if NPP is constructed from an algorithm that can be frozen in the forward test period based on recent backtest performance, and also used for trading the RHE, then that could be a valuable feature.



[Edited by John W on 11/30/2018 8:16 PM]

^ Top
Jim Dean

Sage
200010001002525
Posts: 3173

Joined: 3/13/2006
Location: L'ville, GA

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 11:10 PM
Post #30201 - In reply to #30200

Hi John

As you know, DO has been around several years now, and this NPP block-optimization option has been around for well over a decade IIRC. So, nothing new there. These questions relate to how it’s been working for a long while now - nothing to do with Global Optimization.

When DO came out it was a big deal, and my presumption is that the kinks have been worked out by now. As usual, documentation on the inter workings of it is scanty. So if Barry has time to check on it and clarify things, that would be nice. But, as with any other complex software, in the absence of oddball output indicating things might be amiss, we just need to trust that the designers implemented it correctly and sensibly.

The original question was whether the combo of NPP and DO creates “spurious” results. If we assume the program has been well designed, then I believe that my original response is probably correct.

^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 11/30/2018 11:20 PM
Post #30202 - In reply to #30201

That's great Jim, thank you very much for your advice.
As you say the kinks should have been worked out.
I'm happy with that and will use NPP as a metric during the backtesting to determine the next walkforward Param setting.
It's exciting to see this big advance in useability of dynamic optimization! WELL DONE NIRVANA!
^ Top
John W

Regular
2525
Posts: 71

Joined: 6/18/2011
Location: Sydney, NSW, Australia

User Profile
 
Subject : RE: OT2019 PR2A Global Optimization
Posted : 12/9/2018 7:38 PM
Post #30209 - In reply to #30202

I've attached a spreadsheet that shows trades that appear or vanish when using global optimization, depending on the computer run date. Please see attached for more details, snags and settings used.
Attached file : Test Opmzd Strts 120-240 GO Mode @ 20181207.xls (1541KB - 32 downloads)

Jump to page : 1
Now viewing page 1 [25 messages per page]
back reply

Legend    Action      Notification  
Administrator
Forum Moderator
Registered User
Unregistered User
E-Mail this thread to a friend
Toggle e-mail notification


Nirvana Systems
For any problems or issues please contact our Webmaster at webmaster@nirvsys.com.