Search | User Listing Vidsite | Calendars | Quotes
Home Page ->  Optimizing OmniVest -> Equity Curve Analyzer -> View ThreadLogon (or Register, or Join TradeTight)

You are logged in as a limited-access Guest.To join TradeTight, first read the info in the Organization & Content room, then click the link above. 

James 1:5 (NIV) ... If any of you lacks wisdom, he should ask God, who gives generously to all without finding fault, and it will be given to him.


ECA Usage - Warmups & Stitching
Jump to page : 1
Now viewing page 1 [50 msgs/pg]
Jump to room :
JimDean
Posted 7/11/2014 1:19 PM (#5898)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
Steve Mayo and Mark Holstius have done an amazing job ... visualizing, developing, tirelessly testing, and then volunteering that work to Nirvana ... and having to sort of start all over again in a new environment. I highly applaud their innovation, dedication, and sacrifice.

As we gradually learn more about how ECA works (just one recent presentation on it to go on, plus some posted comments), I'm sure that folks will develop effective guidelines, both what to do, and what to avoid. I don't have any "inside track" on the process ... I've just been listening (and asking some questions) as the process continues. So, my comments in this thread are based on applying general stuff that I am familiar with, to the partially-described, partially-understood, and as yet incomplete ECA tool. So, please don't take anything here to be definitive.

From what I've heard, the "Stitching" process is Nirvana-proprietary ... again, I have no inside track. I can "guess" what it MIGHT entail, from the general diagrams and descriptions that have been provided. So, insofar as my guesses are correct, possibly these discussions might help out.

I'll start by attempting to explain how the calc is being done (half facts, half guesswork) ... then move on to suggestions for how best it might be utilized. This thread will probably have a lot of posts, eventually ...

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JimDean
Posted 7/11/2014 1:35 PM (#5899 - in reply to #5898)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
As I understand it, "Stitching" was invented by Ed, to solve the enormous problem of the need to evaluate huge numbers of equity curve permutations, repeatedly every month (or week?) for potentially every OVPro customer, without bogging down the servers.

Automated Automation!

Part of the extremely clever solution is that ECA has taken a "giant step" beyond the original Portfolio Wizard concept, in that once a "template" for the definition-process of an optimized Portfolio or Account has been established by a user, then that will be *automatically* re-run every month (or more frequently, if server load permits), thus providing a "ever-fresh set" of tools as the market changes, which match the goals of the user.

This is a fantastic enhancement, since it means that *we* don't need to do anything extra, but we get the benefits of having done the busywork ourselves, each month. This is yet another demonstration of OVest as being all about automated cutting edge stuff.

Smooth Servers ...

Undoubtedly one of the greatest barriers and most significant planning and programming challenges in all this is to keep the Server CPU cycles manageable. Doubtless many of you have experienced the "terrible twirlies" ... we're all spoiled, expecting computers to respond instantaneously no matter how complex the task ... and OVest certainly has massive computational tasks going on behind the deceptively simple interface. So, it's wonderful when Nirvana does stuff to either shorten up the "twirly-symbol" time, or postpone it to a better (user-chosen) time, or so totally automate it that it can be hands-off for the user (offloaded to a background thread). They've still got a ways to go on that, but during ECA development, they're obviously trying to prevent the problem from the get-go. Hurrah!

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JimDean
Posted 7/11/2014 1:48 PM (#5900 - in reply to #5899)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
Stitching ... known facts, plus my guesswork here ...

It's a process of focusing in on a limited portion ("snippet") of each strategy's equity curve, to find out what it's been acting like in the recent past in comparison to a whole bunch of other strategies (user defines what a "bunch" is, whether canned or custom), then applying a "measurement" to each of those snippets (user defines what the "measurement" is, using supplied or custom), to "rank" them.

The measurements are OmniScript formulae, but NOT like the formulae used for OmniScans or Color Charts or Setups or TradePlans, etc. Those other OScript formulae are required to have a "relational operatior" such as > or <= etc, so that the result of the formula is "true" or "false". In contrast, the OScript formulae used by ECA for measurements (I'm not sure what the official term will be when ECA is released) does NOT include a relation operator ... they are supposed to evaluate to a NUMBER, not to boolean T/F results.

The closest parallel to this in N-land would be the inputs used by some of the ARM4 tools ... but that's NClub-proprietary ARM4 info so I won't discuss it further here. Suffice to say that ECA includes some cutting edge expert-system concepts, and that those tools are "old hat" for the Nirvana developers. So, although OVest is very new to the investing world, it's comprised of well-proven synergistic tools, developed over decades of effort.

Anyways ... the measurements, applied to each strategy-snippet, generate "snippet scores" (I'm just making up scads of terms here ;~) ... and the scores are used to rank the strat's for each new iterative walk-forward step.

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JimDean
Posted 7/11/2014 2:19 PM (#5901 - in reply to #5900)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
The theory/premise of this process seems to be that strategies whose INDIVIDUAL equity snippets rank the highest, are reasonably likely to represent good components for the next-month's COMBINED portfolio or account. But, the "interactive" effects of the snippets are not being evaluated ... if you've heard the term "ISS" discussed, it suggests a different approach that checks the interactions. (This thread is NOT about ISS, btw).

So, why does ECA evaluate the strat-snippets individually, in its initial measurement-ranking? Simple answer ... calculation speed. This is CRITICAL to the viable success of any OVest tool or procedure, especially if it's something that people will be using regularly. By side-stepping the extra work to check all the permutations of interactive behavior, ECA became a practical solution ... otherwise, as has been suggested, Nirvana might need to buy a few Cray's. Hey, that's for later, once they have a scores of thousands of subscribers.

Let's look at this briefly and then move on. Imagine your starting point has 200 pretty-good strategies, that you've done some work to cull out in general, and that (maybe) you've tuned with appropriate Lists &/or Conditions. 200 may sound like a lot but really it's a pretty small number, since every combo with a different List or Condition is a different strategy. In fact, coming up with that starting list is where "TradeTight" kinda "out of the box" thinking might be very helpful ... but that's for another day.

If you're trying to find the top ten INTERACTIVE strategies (and ten is not very many) for the upcoming month, then the procedure might go like this:
1. Evaluate equity curves for 200 strategy snippets, ranked by the measurement you specify
2. Lock in the top-ranked one, then RE-evaluate that one in combo with the remaining 199, ranking by measurement.
3. Repeat step 2 eight more times, incrementally locking in your ultimate list of ten.
... simple math: 200+199+198+197+196+195+194+193+192+191 = 1955 total snippet evaluations required.

ECA developers have (with wise pragmatic caution) chosen to take a different path than this. Instead of doing nearly 2000 calculations for that first-step of the ECA process, they just do the initial 200, and peel off the top ten ranked strat's, and drive on. This saves about 90% of the CPU time ... my guess is that without doing so, ECA would never get off the ground at all, as a USER tool ... it would be restricted only to Nirvana staff to create things. My vote is always for User based flexibility, as long as it doesn't waste the user's time. This approach is a good way to go, I think.

Will the abbreviated method come up with the "best" solution? Maybe, maybe not ... but we also need to ask, would the exhaustive approach be the "best" solution? Keep in mind that this is a *statistical guess* which is *trying to project the future*. Anyone who's been trading a while knows that projections into the future are only APPROXIMATE.

Aristotle said: "It is the mark of an educated man, and a proof of his culture, that when addressing a problem, he looks for only so much precision as its NATURE permits, and its SOLUTION requires." That's been my engineering-mantra for decades, and I believe it's very healthy for Technical Analysis of Stocks.

The NATURE of this problem is choosing between potentially millions of permutations, using *past* data, to attempt an educated guess as to what will work well in the not too distant future. The NATURE of this problem is that we cannot reasonably expect the solution to be "precise", and that there is NO way to know what the "best" solution might be, until after the time has passed.

So, how much precision does the SOLUTION require? That's easy ... almost any kind of organized culling of the millions of possibilities will tend to yield better results than a haphazard, manual process. Our "requirement" is measured by things like CAR and MDD and CALMAR, etc ... and if we are "shooting for" a value of "50" ... but we know that the precision of the tool is not perfect, then we might choose to keep fiddling with things, till the value we SEE coming out of the tool is "100" ... at that point, we might safely assume the Solution is acceptable.

That's what the ECA process enables us to do, without fighting with the thousands of terrible twirlies along the way.

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JimDean
Posted 7/11/2014 2:34 PM (#5902 - in reply to #5898)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
ECA does not stop with the individual evaluation ... there have been some hints that there are SOME synergistic evaluations being done, but only within the top-ranked individual strategies. Maybe we will know more about this, later. But here is a possible scenario (just guesswork).

Let's say that we start with the (fairly small) set of 200, and we are shooting for the best ten. If some arbitrary subset of the individually-ranked snippets are peeled off ... say, 2x the desired final number = 20 in this case, then it would be a lot more practical from a CPU-time perspective, to evaluate what the "best interactive set of ten" are, out of those 20 ... rather than the best ten out of 200 that we talked about earlier. This would require:
200 individual runs, peel off the best twenty and lock in one; then check that one against the other 19 to lock in the second one, and so on: 200+19+18+17+16+15+14+13+12+11 = 335 runs.

This adds 67% CPU time to the original 200 ... but that's a FAR cry from the 1955/200 = 2470% increase needed by a pure "check all permutations" approach.

Ed said that there was "some" post evaluation done ... thus we see a "true Equity curve" after the dust settles, based on interactive vs stitched results. I *like it* that the beta-display which Ed showed us plots both the stitched curve and the interactive curve. That gives us an important clue to whether "Aristotle's rule" has been met ... maybe we should try a different starting set of 200, or maybe a different target number of strategies, etc.

I hope that they are doing something like that ... seems to me that it's the best of both worlds, without incurring high CPU time penalties for doubtfully-useful future-prediction "precision".

OK ... again let me emphasize that this explanation has been a combo of my understanding of the facts, plus guesswork applied to the hints. Hopefully it's not too far off.

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JimDean
Posted 7/11/2014 2:52 PM (#5903 - in reply to #5902)
Subject: ECA Usage - Warmups & Stitching



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
Moving on to the "stitching" process, now that the foundation of the multiple-strat combination concepts have been discussed.

There are several REASONS why using stitching-together the equity curves is important:

1. Every month, a new combo of strat's will potentially be used, so the basis-trading-dataset-baskets from the individual strategies are continually changing

2. The equity values at any point in each strat's curve differ ... that is, the "simulated dollars available for trading" is not a locked in value.

3. The available dollars for trading is an INPUT to the various strategies, to evaluate their ongoing equity curves - that is, the STARTING equity is a function of the ENDING equity of the prior period ... but the strat's used for that period likely will differ with every new period.

4. To save CPU time, it's essential to avoid the repetitive reprocessing of huge amounts (15 years) of historical data for every iteration discussed earlier ... stitching allows that to be cut dramatically (200 days vs 15*250=3750 days.


So ... stitching is *essential* to a viable tool, both to save CPU time and to properly evaluate each new month's snippet.

How is it done? Probably proprietary, but here is a possible not-very-complex explanation:

1. Do the Individual+Limited-Combo Eval's for stitch-segment #1, resulting in an interactive optimized equity curve to be used for the upcoming month
2. Apply that combo of strats to the actual market data for that month, and record the historical simulated performance of the strat-combo, plus its RESULTING FINAL EQUITY
3. Use the result of step 2 as the starting point for evaluating the next stitch-segment #2, to find a new best-combo of strat's for the next month, as was done in step 1
4. repeat steps 2 & 3 for all past history, and STORE THE RESULTS so that, moving forward, that past history does not need to be re-evaluated repeatedly.

As far as I can tell, that's all that really needs to be done. The initial processing takes a long time, but the incremental processing done each month is only dealing with the input lookback period (typically 200 days).

OK ... I think that covers the underlying concepts adequately ... at least insofar as I can guess how they work. From here, I'd like to move on to discuss how to WORK WITHIN that paradigm ... with some hopefully-useful words of caution, and suggestions for success.

(gotta go now though ... I'll continue this thread soon)

Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
JohnWolfraad
Posted 7/11/2014 6:57 PM (#5904 - in reply to #5898)
Subject: ECA Usage - Warmups & Stitching



Veteran

Posts: 208
100100
Location:
Australia: , Church Point, NSW
On behalf of many of your ardent admirers I want to THANK YOU for explaining complex concepts so simply.

It's an amazing gift that you share so willingly with all, much appreciated!

John :cool:

Edited by JohnWolfraad 7/11/2014 6:58 PM


Thread moved by JimDean on 8/4/2014 8:47 AM from Markets & Methods > OmniVest, Money Mgmt & Risk Control > ECA Usage - Warmups & Stitching

Top of the page Bottom of the page
Jump to page : 1
Now viewing page 1 [50 msgs/pg]
( E-mail a Link | Printer Version | Search Room )

Owner of site: Jim Dean -- Forum content is confidential, and may not be distributed without written permission.