PureBytes Links
Trading Reference Links
|
Over-optimize robustness ?!
That's an interesting concept, but I think not possible due to mutual
exclusivity but I suspect at least someone out there won't agree.
--- In amibroker@xxxxxxxxxxxxxxx, "CedarCreekTrading" <kernish@xxxx>
wrote:
> Dave,
>
> Quite honestly, I haven't heard any definitions of robust at this
forum. Lot's of folks are willing to say: "This is not robust...",
but I have asked for specifics and definitions and nobody is
supplying the answers. The system and how it's traded is not the
issue. This is exactly why I decided to present the CMO5 on ETF's.
The concept isn't new and the data is OOS. Without any tweaking or
modification, the approach continues to produce a reasonable equity
line.
>
> Would you like for me to comment on the deviation of the equity
from a chosen linear regression? Fred mentioned it would be nice to
see an equity curve (and there is one on each chart that I sent) and
that should count for one of Mark's nine robust measures.
>
> Since, I'm not bright enough to even guess at nine measures of
robustness (and so far, Mark hasn't shared), I must rely on simple
bench marks: how does it perform over 8,000 issues, how does it
perform in different markets (futures, equities, ETF's), how does it
perform in different time periods, max eod dd's, max neg. excursions,
avg. win/avg. loss, expectancy, blah, blah, blah.
>
> Did I miss something or did someone actually define robust? If so,
I hope it was better than the optimizing thread. Testing is
necessary, optimizing is a spiraling death trap (unless you name has
DT as initials...then you understand the trap and play the game on
the edges of the black hole).
>
> The CMO5 is a dandy indicator and the rules that I apply to trigger
the trades are simplistic. Certainly, anyone with decent software
can better the results (well, maybe). The trick is to modify the
approach ... without over-optimizing (tough trick).
>
> If over the next six months, this system, with it's defined
triggers, trades the DIA, QQQ, and SPY with decent profits...is it
robust? How many years must it continue to produce steady results
before it can be called robust? If we look back ten years from now,
and it has performed as it has in the last four years, is that
robust? Must it also show the same kind of results in silver, bonds
and soybeans? If it performs on EOD, data ... must it also knock em
dead in one minute time frames?
>
> Here's a different view: If a system performs OOS for a period (x)
of time and continues to perform into the future, without
adjustments...then I hereby declare that it is robust. Why?
Because, just like technical analysis: every input, that can be
considered, is discounted by an issue's price. In technical
analysis, we discount all the information that flows in, out and
around a stock and only consider the price. So, why not just judge a
issue's robustness by strictly pegging it to OOS results? Period.
If these ETF's continue to produce steady results for the next five
years, I expect someone to say: "Well, maybe these ETF's with the
CMO5 performed well for the past eight years, but by my definition,
this is not robust".
>
> I'm still waiting for the naysayers to define robust and to post a
system that can be monitored for results. The CMO5 is the default
indicator that I use to demonstrate the character of momentum
oscillators. Many of us have come to realize that the indicator is
not the most important piece of the puzzle. The grail is not a
silver or gold chalice.
>
> Robust and optimizing are difficult terms to define. Any
definition can be argued. Robust is like beauty. My wife is
beautiful (my subjective opinion). You might find her unattractive.
>
> The real silly part is that people want to argue whether something
is robust or not. The rubber meets the road when you trade. Try
flipping around sizeable accounts with something that is not
robust... money flies away.
>
> Sorry that I haven't been able to answer your question. Why don't
you direct you question to Mark. He has a nine point plan to
evaluate what is or is not robust. Is there a possibility that one
might over-optimize robustness?
>
> Take care,
>
> Steve
>
> ----- Original Message -----
> From: Dave Merrill
> To: amibroker@xxxxxxxxxxxxxxx
> Sent: Friday, October 31, 2003 12:33 PM
> Subject: RE: [amibroker] Robustivity
>
>
> thanks for the specifics, glad to have 'em, even though that's
not what I was trying to ask (:-)
>
> what I'm most interested in isn't the system itself, but what you
did to convince yourself that it's robust. the way it was posted
implied that you disagreed with other ideas of robust-ness/ivity
floating around here, so I'm trying to find out what your approach is.
>
> dave
>
> If I am using the CMO5 with triggers of 34/-34, I would go back
and start a test to evaluate this system and triggers. The starting
period would be whatever date you pick (1990, '97, 2000, etc.).
>
> Next, I run the test over 315 trading days (this period gives
me results for approximately one year..it takes "x" amount of periods
to load the TRIX(21), which I use as a trend identifier. My approach
produces about 10 to 15 round turn trades a year... in each stock.
>
> I then rank all issues by one criteria: percent return per day
(while the money is in the market). If you only consider the percent
per day contributions, I think you will find that all other "book
learned" ratios come out just fine. Numbers lie. Would you rather
trade a $100 stock that returns $20 or a $20 stock that returns $10?
Percent per goes a long way to normalizing the comparisons.
>
> I pick the 20 best percent per day stocks and trade them for
the next quarter. At the end of the quarter, I reevaluate the
percentage per day contributions and reshuffle the issues in play, if
necessary.
>
> Symtems don't go bad, stocks and commodities go bad. Going bad
is best defined by a change in the pattern of supply and demand. The
cream rises to the top of the list.
>
> Is this optimizing? Could be, by some definitions. If all the
odds are even money, who would you prefer to bet on: Chicago or
Kansas City? KC is undefeated and Chicago couldn't beat the local
high school. My money is on KC.
>
> The stock betting setup is not handicapped (like almost all
games). This is basically a even money play (with subtractions for
commission and slippage...juice/vigorish). If you have 9,000 issues
to play, why won't someone want to bet on the strongest performance?
>
> I know that the explanation might be over simplified...but, the
people who know me, in and out of this forum, know that this is the
way I do it. I'm not crusading for anything. This works. I've
presented this simplistic approach publicly to large groups and in a
number of internet seminars. It continues to crank out extraordinary
profits.
>
> Please let me know if the paragraphs help to explain the
ranking.
>
> Take care,
>
> Steve
>
>
> ----- Original Message -----
> From: Dave Merrill
> To: amibroker@xxxxxxxxxxxxxxx
> Sent: Friday, October 31, 2003 9:29 AM
> Subject: RE: [amibroker] Robustivity
>
>
> steve, thanks for your response.
>
> from your msg subject and the way you presented this system,
I thought you were offering it as an example of one you had
objectively evaluated and determined to be robust. I was interested
in how you thought "robustivity" should be evaluated, since you
seemed to be contrasting your approach to walkforward optimization
and the various other system measures people were talking about.
>
> what I'm hearing in your response below isn't what I would
describe as a specific method for distinguishing accidentally
gorgeous backtest results from robustness. you do mention testing
also at faster time frames, which isn't a technique that's been
mentioned recently. but mostly, the robustness label here seems to
come from your integration of various aspects of your long experience
with it, like your visual sense of how it behaves. am I missing
something?
>
> another question: you mention issue selection, the idea of
looking for stocks you think will trade well with a particular
indicator, rather than the other way around. how do you do that? by
measuring raw past growth trading that indicator? other measures?
>
> thanks again,
>
> dave
> just for my understanding, in what sense is this
system "robust"?
>
> Well, first, this was presented to the public in the late
90's, at a series of seminars that I conducted for Equis. Same
indicator, same triggers, same everything. This robust "thing" is a
tough one to define. I'll try to explain what's important to me,
but, it's very subjective and just one person's opinion.
>
> is it because results are similar with different similar
periods and thresholds?
>
> If you take this CMO5 indicator and step down in time (5,
10, 60 minutes), you need to widen the triggers to obtain decent
results. Other than that, it trades through time-zones with very
good results.
>
> that seems unlikely, since there isn't very far to go from
5 to hit 1 and 0, which I'd guess are significantly different. what
sort of testing led you to decide on this period and threshold, and
this system for that matter?
>
> If you're referring to the CMO5...I first started testing
it six years ago. I've tested and eyeballed every version of CMO
(x). I've created a few indicators that combines different periods
of the CMO. For my money, for my style, this judge of momentum
trades more things, more accurately than any other indicator I am
aware of. As I have begged many times: give me something
better...I'll use it instead of this.
>
> is it robust because it works well on many stocks, indexes
and funds over a long period of time?
>
> Yes, it works well on many stocks and indexes. I don't
trade funds, but, some fund managers, DTG members, use versions of
the CMO to aid their timing.
>
> because of the concepts behind the indicator itself?
>
> I process visually. The math is beyond me. My bottom line
has always been the same: give me an indicator that is smooth, yet
sensitive to intermediate and major market turns. After gawking
hundreds of charts, everyday, for the last six years, I'm amazed at
how this indicator quantifies momentum. I like versions of the
Stochastic RSI and the Standard Error Oscillator, but dollar for
dollar, the CMO does it for me.
>
> something else?
>
> I think there's a few other things to mention. First of
all, the ETF's that I showed were chosen because they represent a
broad range of stocks and are popular trading instruments. Do I
suggest trading these issues with this system? No way. The CMO5
trades a lot of other issues with better results than the ETF's. I
always allow the issues "to pick themselves". Trade the issues that
return the greatest percentages in a stable system.
>
> In it's stripped down version, as presented, the CMO5 is an
indicator that can return steady profits (see equity lines) in it's
rawest unoptimized form. Is that robust?
>
> Robustness and optimizing/over-optimizing are fascinating
and misunderstood subjects. Over the years, I've constantly
simplified my approaches. I can improve on the results of the three
ETF's by simply "tweaking" the trigger levels. But, will it walk
forward better than the default triggers of 34/-34? At least what I
presented was out of sample.
>
> If an approach does a good job of identifying movement of
supply and demand, the approach should not be expected to work on all
issues. To say a system needs to work on all issues is total
crap. To say that a system sucks because it doesn't work on XYZ is
another large pile. Build simple things and concentrate on issue
selection.
>
> Optimization leads to dark and spooky places. Ranking
leads you down the yellow brick road.
>
> Take care,
>
> Steve
>
> steve, thanks for sharing this (again).
>
>
> just for my understanding, in what sense is this
system "robust"?
>
> is it because results are similar with different similar
periods and thresholds? that seems unlikely, since there isn't very
far to go from 5 to hit 1 and 0, which I'd guess are significantly
different. what sort of testing led you to decide on this period and
threshold, and this system for that matter?
>
> is it robust because it works well on many stocks,
indexes and funds over a long period of time?
>
> because of the concepts behind the indicator itself?
>
> something else?
>
>
> I'm not disputing the system's value, which I haven't
tested yet. I'm trying to understand what kind of process you go
through to settle on a system and settings.
>
> thanks,
>
> dave
>
> 1. This exact system was presented over a year ago at
this forum
> 2. The charts are OOS (since, it's been posted
publicly forever)
> 3. Rules are simple: Buy the opening of the next day
when the CMO5 closes below -34 and sell when it triggers above 34.
>
> Works on most issues (raw). Works better if:
>
> a. You take trades only with the trend
> b. You protect yourself from large drawdowns (stop)
> c. You conjure a profit target (limit)
> d. You put in a time stop
>
> This is the guts of an indicator and a logical
systematic approach. Whistles and bells are optional (but, in my
opinion necessary). Again, if you start with a pig, the prom dress
doesn't make it look any better. Don't hang ornaments on a twisted
Christmas tree.
>
>
> Send BUG REPORTS to bugs@xxxx
> Send SUGGESTIONS to suggest@xxxx
> -----------------------------------------
> Post AmiQuote-related messages ONLY to:
amiquote@xxxxxxxxxxxxxxx
> (Web page: http://groups.yahoo.com/group/amiquote/messages/)
> --------------------------------------------
> Check group FAQ at:
http://groups.yahoo.com/group/amibroker/files/groupfaq.html
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of
Service.
>
>
>
> Send BUG REPORTS to bugs@xxxx
> Send SUGGESTIONS to suggest@xxxx
> -----------------------------------------
> Post AmiQuote-related messages ONLY to:
amiquote@xxxxxxxxxxxxxxx
> (Web page: http://groups.yahoo.com/group/amiquote/messages/)
> --------------------------------------------
> Check group FAQ at:
http://groups.yahoo.com/group/amibroker/files/groupfaq.html
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of
Service.
>
>
> Yahoo! Groups Sponsor
>
>
>
> Send BUG REPORTS to bugs@xxxx
> Send SUGGESTIONS to suggest@xxxx
> -----------------------------------------
> Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
> (Web page: http://groups.yahoo.com/group/amiquote/messages/)
> --------------------------------------------
> Check group FAQ at:
http://groups.yahoo.com/group/amibroker/files/groupfaq.html
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of
Service.
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Ink Cartridges or Refill Kits for your HP, Epson, Canon or Lexmark
Printer at MyInks.com. Free s/h on orders $50 or more to the US & Canada.
http://www.c1tracking.com/l.asp?cid=5511
http://us.click.yahoo.com/mOAaAA/3exGAA/qnsNAA/GHeqlB/TM
---------------------------------------------------------------------~->
Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
|