|
PureBytes Links
Trading Reference Links
|
Jayson,
Here's the article. Unfortunately TASC wants to charge for
everything they ever printed these days.
http://store.traders.com/v15261aryour.html
It begins ...
Are Your Profits Robust? by Steve Notis
Your trading system probably has a set of parameters you think are
the most profitable. But is it the most consistent? Here's a
technique to find the most robust set that will give you regular
profits.
For those who use trading systems, the most important concern should
not be to find parameters that yield the greatest profit; rather, it
should be to find the most robust parameters. Anyone can tweak a
system until it shows startling results, but unless those parameters
are robust, they will not hold up in real-world trading.
Several testing methods can be used to increase a system's
robustness. The first, and simplest, is to test over long periods.
This assures that the test includes a variety of market conditions.
The second method is the walk-forward, or blind, data test. This
typically consists of optimizing trading parameters over a long
period, but stopping short of the present. Finally, the best
parameters are tested over the remaining, nonoptimized, data. This is
also referred to as walk-forward, blind, virgin, out-of-sample, and
it has even been referred to as real-time testing.
If the optimized parameters work well over the blind data, then it
can be argued that the parameters are robust and likely to continue
to work for some time into the future. Since the final test is run
without the benefit of hindsight or optimization, it's as close to
real trading as you can simulate with historical data. However, don't
assume that the most profitable parameters are the best parameters -
meaning those that will be the most profitable in the future.
The rest costs $3.95 if you don't have old issues at home already.
It's worth a read.
--- In amibroker@xxxxxxxxxxxxxxx, "Fred <fctonetti@xxxx>"
<fctonetti@xxxx> wrote:
> Jayson,
>
> That's exactly the point I was trying to make ...
>
> In the example you used and my comments of that example the
parameter
> value of 100 may be the best value using increments are 25, but it
is
> invisibly a much tighter fit then 91 is. The reason for this is
that
> next to 100 is a performance canyon at 102 where the DD's skyrocket
> but surrounding 91 are 10 points on either side i.e. 81 through 101
> where similar results are achieved.
>
> This kind of situation where performance canyons occur next to some
> parameter value are common in system development and testing making
> the most robust values in the middle of the range where similar
MAR's
> are as opposed to maybe where the highest CAR or MAR occurs. The
> idea behind system development and optimization of course is to
give
> the system the highest chance of success going forward as opposed
to
> having bragging rights looking backward. There was a good article
> about this in TASC a few years ago which I'll try and find the
> reference for if you are interested.
>
> I think Herman did a prestation on this and included a usable Excel
> spreadsheet in the files area and although I personally don't think
> it goes far enough because it only looks at CAR, I do think it's in
> the correct direction.
>
> Fred
>
> --- In amibroker@xxxxxxxxxxxxxxx, "Jayson" <jcasavant@xxxx> wrote:
> > Again Fred, it seems to me that what you are accomplishing is a
> fit. Going
> > forward will this extremely fine tune hold true? As a strong
> proponent of
> > system trading you are probably in a better position to answer
that
> question
> > but to me robustness is a measure of how widely the the stock (s)
> may vary
> > from the optimized sweet spot and still remain consistently
> profitable. The
> > one constant in the stock market is that it is not constant. For
a
> system to
> > perform as expected going forward it seems to me that the system
> must be
> > flexible. Optimizing to such tight tolerances makes that goal
> challenging.
> >
> > Jayson
> > -----Original Message-----
> > From: Fred <fctonetti@xxxx> [mailto:fctonetti@x...]
> > Sent: Thursday, February 06, 2003 12:44 PM
> > To: amibroker@xxxxxxxxxxxxxxx
> > Subject: Stock Selection was: [amibroker] Re: NDX / QQQ - Can it
be
> traded ?
> >
> >
> > Jayson,
> >
> > Using your example ...
> >
> > Wouldn't you consider it more appropriate to optimize in steps of
> one
> > and plot the results looking at the CAR & MDD's or some
combination
> > of those such as MAR = CAR / MDD and see where the most stable or
> > robust area of MAR is for the parameters in play.
> >
> > For example lets assume that you first optimize by 25 and find the
> > best MAR is at 100.
> >
> > And lets assume that you optimize again by 1 and find out the
range
> > between 81 and 101 is all pretty much the same but that at 74 the
> CAR
> > goes way down and at 102 the DD's go way up. Are you really
going
> to
> > use 100 as the final value ? or the midpoint between 81 and 101
i.e
> > what I would consider to be the more robust setting.
> >
> > Fred
> >
> > --- In amibroker@xxxxxxxxxxxxxxx, "Jayson" <jcasavant@xxxx> wrote:
> > > Ken,
> > >
> > > Example: lets say one of your optimizations is the length of a
> > given MA. If
> > > your range is set from 20-200 with steps of 1 then the
> optimization
> > will be
> > > lengthy and what you will have accomplished is a tight curve
fit
> of
> > the Ma
> > > that historically worked on this basket over this time frame.
The
> > result may
> > > very well not work as well in RT . On the other hand lets say
you
> > are
> > > optimizing the same MA, 1-200 but instead of increments of 1 you
> > use steps
> > > of 25. This may offer a better representation since it does not
> > necessarily
> > > curve to fit 1 or a few very good trades but instead may
capture a
> > greater
> > > percentage of good or bad trades.....
> > >
> > > Jayson
> > > -----Original Message-----
> > > From: Ken Close [mailto:closeks@x...]
> > > Sent: Thursday, February 06, 2003 11:55 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: RE: Stock Selection was: [amibroker] Re: NDX / QQQ -
Can
> > it be
> > > traded ?
> > >
> > >
> > > Well, I almost did not respond (although thanks for writing).
I
> am
> > not
> > > concerned with the price factors etc.
> > >
> > >
> > >
> > > This is the thing I am concerned with. I hear your other
emphasis
> > and am
> > > taking care of that. The following is the area my previous post
> > referred
> > > to.
> > >
> > >
> > >
> > > "The challenge is to find a good range of settings with out
> risking
> > a curve
> > > fit, other wise the system will not likely trade in the same
> manner
> > as it
> > > tests."
> > >
> > >
> > >
> > > Any further comments on this one aspect of your reply???
> > >
> > >
> > >
> > > Anyone else???
> > >
> > >
> > >
> > > Ken
> > >
> > >
> > >
> > >
> > >
> > > -----Original Message-----
> > > From: Jayson [mailto:jcasavant@x...]
> > > Sent: Thursday, February 06, 2003 11:41 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: RE: Stock Selection was: [amibroker] Re: NDX / QQQ -
Can
> > it be
> > > traded ?
> > >
> > >
> > >
> > > Apples and Oranges Ken. Perhaps we can make a Nectarine! If you
> > choose a
> > > basket of stocks based on a given criteria then testing that
> basket
> > over 10
> > > years may well hold no significance. Lets use a simple filter.
> > >
> > >
> > >
> > > C>10;
> > >
> > > ma(v,50)>100000;
> > >
> > >
> > >
> > > Lets consider that this is the criteria for our initial
> > scan/selection.
> > > Where were the conditions 1-2-5-10 years ago? If the conditions
> > were not met
> > > just 1 year ago would we have selected the stock for use in our
> > system? If
> > > not then what is the value of testing its results 2-5-10 years
> ago?
> > With
> > > this approach you would need to test that the conditions were in
> > fact met X
> > > years ago then test forward using your selection criteria to
exit
> > when
> > > necessary. This insures that you would not be testing stocks
that
> > in RT you
> > > would have no intention of trading.
> > >
> > >
> > >
> > > You may wish to be careful with your Optimization. The
challenge
> is
> > to find
> > > a good range of settings with out risking a curve fit, other
wise
> > the system
> > > will not likely trade in the same manner as it tests.
> > >
> > >
> > >
> > > Jayson
> > >
> > >
> > >
> > > -----Original Message-----
> > > From: Ken Close [mailto:closeks@x...]
> > > Sent: Thursday, February 06, 2003 11:10 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: RE: Stock Selection was: [amibroker] Re: NDX / QQQ -
Can
> > it be
> > > traded ?
> > >
> > > Jayson: your advice is very good as well as comprehensive..I am
> > still left
> > > unsettled as to my approach.
> > >
> > >
> > >
> > > Lets say my indicator/system is relatively short term in nature
> > (lookbacks
> > > are somewhat short). But, I want to employ some of the
principles
> > that Fred
> > > talked about namely to be sure I optimize over a long period so
> > various
> > > market conditions are captured). [I know the argument that the
> > nature of
> > > the market changed on 3/1/2000 and therefore optimizations
should
> > be done
> > > after that point---I tend to want to be more conservative and
> > capture
> > > conditions on both sides of the peak]. Also, a system that
> > optimizes to
> > > good performance across that period seems a more conservative
> > and "robust"
> > > system. Witness Freds model of results in his "spectacular"
> system-
> > the
> > > equity curve kept going up and was smooth as the system passed
> over
> > > 3/1/2000,
> > >
> > >
> > >
> > > As I said previously, the optimization process is a lengthy
one,
> so
> > my
> > > approach is to do all of what you said below in other scans (I
> pull
> > out of
> > > TC2000 many of the conditions you talk about below). This is my
> > selection
> > > watchlist. Then, I do not want to / can not take the time to
run
> > > optimizations on 300 or 800 stocks so I want a screening
approach
> > that has a
> > > good probability of finding stocks that "resonate" with my
system.
> > Then, I
> > > will optimize these stocks.
> > >
> > >
> > >
> > > I still have to try DTs approach and probably alter it to
conform
> a
> > little
> > > more closely with my system. Another approach is that if I have
> > for example
> > > some trigger levels in my model...while these might optimize
out
> to
> > extremes
> > > for individual members (think about RSI crossing 30 and 70),
then
> my
> > > screening formula might set a standard trigger of crossing 50
> > either way.
> > > Stocks that do "well" against the 50/50 cross should do better
> when
> > > optimized to exact levels. Stocks that do poorly against the
> 50/50
> > cross
> > > will still do poorly when optimized. Don't you think this is a
> > correct way
> > > of looking at it?
> > >
> > >
> > >
> > > Thanks for the help. Any further comments on the above point?
> > >
> > >
> > >
> > > Ken
> > >
> > >
> > >
> > > -----Original Message-----
> > > From: Jayson [mailto:jcasavant@x...]
> > > Sent: Thursday, February 06, 2003 10:24 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: Stock Selection was: [amibroker] Re: NDX / QQQ - Can
it
> be
> > traded ?
> > >
> > >
> > >
> > > Ken,
> > >
> > > I agree with Dimitris. To select a basket of stocks you must
first
> > select an
> > > indicator or method/style to trade them with. If your
System/style
> > is trend
> > > following then selecting a basket that trends strongly (up or
> > down) is
> > > critical. The indicators that you read that tend to behave for
> > trending
> > > conditions tend to fail miserably in consolidations. The
reverse
> is
> > true for
> > > Rolling stocks. You can use simple methods to begin the process
> > such as 20
> > > and 50 ma or slope of either. Many use a floor of price such as
> > c>10 or 15
> > > or whatever. Volume is certainly a major consideration since
> without
> > > liquidity the trade may be too challenging. Volatility is also
> > something to
> > > consider and to compare to your trading style. Do you (can you)
> > stomach the
> > > wild rides of stock like BEAS whose current ATR is >7% of the
> > current price
> > > (today moved nearly 8% on the open!!) or are you more
comfortable
> > trading in
> > > the 3-5% range? Some like to add fundamental criteria. Though
not
> > easily
> > > accomplished within AB there are data suppliers (QP and TC2000
to
> > name a
> > > few) that make such an initial screen very easy. Add your own
> > criteria to
> > > pare down your initial universe to a more workable number. A
lot
> of
> > my work
> > > revolves around Sector Rotation. Determining where the money is
> > flowing in
> > > or out can help you to further refine your list. I think Ara is
> > also working
> > > in this area and may wish to add some of his thoughts.
> > >
> > >
> > >
> > > IMO this type of selection process offers insight into what is
> most
> > likely
> > > to work in the short term. Testing a basket selection using this
> > type of
> > > approach over a long period (years) may not yield the desired
> > results
> > > because the above criteria is constantly in motion. However if
you
> > find that
> > > your system or approach tends to work well under those
conditions
> > then
> > > choosing stocks that meet a given criteria can yield impressive
> > results.
> > > Consider this a method of optimizing. Instead of optimizing the
> > settings in
> > > a group of indicators so that the results test well on a given
set
> > of
> > > stocks, over a given time frame you are instead selecting a
given
> > universe
> > > that tends to do well with a given set if indicators.
> > >
> > >
> > >
> > > Regards,
> > >
> > >
> > >
> > > Jayson
> > >
> > > -----Original Message-----
> > > From: DIMITRIS TSOKAKIS <TSOKAKIS@xxxx> [mailto:TSOKAKIS@x...]
> > > Sent: Thursday, February 06, 2003 5:50 AM
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Subject: [amibroker] Re: NDX / QQQ - Can it be traded ?
> > >
> > > Ken,
> > > I will try to describe an objective method I often use to select
> > > stocks.
> > > First of all it depends on the indicator/system you use [since I
> > have
> > > not yet any ...holy grail available]
> > > Let us suppose you want to use the smoothed MeanRSI as a general
> > > indicator and you want to find "good" applications.
> > > The DEMA(MeanRSI,45) is a quite smooth and descriptive
oscillator
> > for
> > > ^NDX market.
> > > If you could buy at troughs and sell at peaks [with some zig
> period
> > > around 20] you would have one of the best
> > > performances for this Market indicator. Does it suites to ANY
> stock?
> > > Certainly not.
> > > Run the
> > >
> > > f=Foreign("~SUMRSI","C")/Foreign("~COUNT","V");F=DEMA
(F,45);z=20;
> > > Buy=TroughBars(f,z)==0;Sell=PeakBars(f,z)
==0;Short=Sell;Cover=Buy;
> > >
> > > from, say, May1, 2000 till now and see the top gainers [NTAP,
> JNPR,
> > > CIEN etc] and the top loosers [FHCC, PDCO, ESRX etc]
> > > Since Peak/Trough system is, as you know, unrealistic, try to
> > replace
> > > it with a closest approximation.
> > > For smoothed oscillators a 2-level system is really interesting.
> > > Try something like
> > >
> > > f=Foreign("~SUMRSI","C")/Foreign("~COUNT","V");
> > > X=45;//Optimize("X",45,35,50,5);
> > > F=DEMA(F,X);
> > > x1=Optimize("X1",42,38,44,1);//buy level
> > > x2=Optimize("X2",56,55,60,1);//sell level
> > > b1=Cross(f,x1);b2=Cross(x1,f);//no need to be fanatic with the
> type
> > > of cross, the system knows better
> > > s1=Cross(f,x2);s2=Cross(x2,f);
> > > nb=Optimize("nb",1,1,2,1);
> > > ns=Optimize("ns",1,1,2,1);
> > > nSH=Optimize("nSH",1,1,2,1);
> > > nCO=Optimize("nCO",2,1,2,1);
> > > Buy=IIf(nb==1,b1,b2);Sell=IIf(ns==1,s1,s2);
> > > Buy=ExRem(Buy,Sell);Sell=ExRem(Sell,Buy);
> > > Short=IIf(nSH==1,s1,s2);Cover=IIf(nCO==1,b1,b2);
> > > Short=ExRem(Short,Cover);Cover=ExRem(Cover,Short
> > >
> > > The good performance of JNPR in the theoretical Peak/Trough
> project
> > > [around +1000% with 4/4/0 success]
> > > leaves room for good optimization results
> > > [sometimes better than the "ideal" model: there are MANY
> > combinations
> > > better than +1000% for JNPR]
> > > The bad performance of ESRX means that this stock need some
other
> > > treatment.
> > > The same optimization confirms the ESRX conclusion : It is
better
> > not
> > > to trade this stock, since the most profitable combination
> > > is the 0 trades !
> > > Of course, this method is not always accurate, but it gives good
> > > advices for the first selection.
> > > Besides that, I have met many times JNPR and RFMD in the 4-digit
> > > profitable stocks, my experience is not great, I come back to
> basics
> > > [CSCO 70% and BEAS 30%] quickly, I feel more safe there, but,
it
> is
> > a
> > > matter of taste [crude oil and copper futures are also
> interesting,
> > > but let us talk for stocks in this group]
> > > Does it help ?
> > > Dimitris
> > > --- In amibroker@xxxxxxxxxxxxxxx, "Ken Close" <closeks@xxxx>
> wrote:
> > > > DT: can you name some additional symbols, besides CSCO and
BEAS,
> > > with
> > > > which you have seen similar success. Thanks for sharing your
> > > > experience.
> > > >
> > > > Ken
> > > >
> > > > -----Original Message-----
> > > > From: DIMITRIS TSOKAKIS <TSOKAKIS@xxxx> [mailto:TSOKAKIS@x...]
> > > >
> > > > Sent: Tuesday, February 04, 2003 3:38 AM
> > > > To: amibroker@xxxxxxxxxxxxxxx
> > > > Subject: [amibroker] Re: NDX / QQQ - Can it be traded ?
> > > >
> > > > Herman,
> > > > An annual system % return 100%-120% is reasonable for many QQQ
> > > > systems from Jan2000 till now.
> > > > [Suppose always buy, sell, short, cover at +1Open, 0.5%
> commission
> > > > and disabled stops]
> > > > I would be surprised indeed to see a double return.
> > > > Above 200%-250% we may find some other popular stocks, but
not
> > > QQQ,
> > > > AFAIK.
> > > > I come to believe there are some "functional" limitations for
> QQQ
> > > > curve to exceed the annual 150%.
> > > > This conclusion is after MANY tests for various trading
systems,
> > > > optimised or not.
> > > > This is a reason I prefer CSCO or BEAS for example, their
curves
> > > are
> > > > more "profitable" and more flexible.
> > > > Of course I [almost] always speak for medium speed systems
[not
> > > more
> > > > than 6 trades per year]
> > > > It is more than 8 months I did not trade a single QQQ share, I
> > > would
> > > > be glad to come back to my old favorite, but for a better than
> > 180%
> > > > annual return.
> > > > This is my experience, I hope it hepls somehow...
> > > > Dimitris
> > > > --- In amibroker@xxxxxxxxxxxxxxx, "Herman van den Bergen"
> > > > <psytek@xxxx> wrote:
> > > > > Well Fred, the real values to use when estimating whether a
> > > trading
> > > > system
> > > > > is practical have never been answered on this list. You are
a
> > > > trader, I am
> > > > > still mostly a tinkerer, so I respect your opinion. As a
rule
> I
> > > > discard
> > > > > systems that do not survive my "acid test" of 0.5%. This
> allows
> > > my
> > > > to fumble
> > > > > placing the trade, allows for some over-optimization,
> slippage,
> > > > slow data,
> > > > > and even for some commission.
> > > > >
> > > > > There ought to be a formula based on parameters like volume,
> > > > volatility and
> > > > > price, to gives us a working estimate. Places that have
lots
> of
> > > > trading
> > > > > histories could crunch that out in seconds. Would be
> interesting
> > > to
> > > > have a
> > > > > poll on this.
> > > > >
> > > > > Herman.
> > > > >
> > > > > -----Original Message-----
> > > > > From: Fred <fctonetti@xxxx> [mailto:fctonetti@x...]
> > > > > Sent: Monday, February 03, 2003 4:03 PM
> > > > > To: amibroker@xxxxxxxxxxxxxxx
> > > > > Subject: [amibroker] Re: NDX / QQQ - Can it be traded ?
> > > > >
> > > > >
> > > > > Herman,
> > > > >
> > > > > Let's use today as an example and assume you are going to
> > trade
> > > > QQQ
> > > > > after you make a decision on where NDX closes ...
> > > > >
> > > > > At 4:00 PM QQQ was as at 24.50
> > > > >
> > > > > Between there and 4:15 QQQ got as high as 24.54 and as
low
> as
> > > > 24.48
> > > > > or 0.16% over and 0.08% under as EXTREMES.
> > > > >
> > > > > Thomas,
> > > > >
> > > > > I'm not quite ready to toss it on the scrap heap yet. I
> just
> > > > started
> > > > > playing with it.
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Yahoo! Groups Sponsor
> > > > > ADVERTISEMENT
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > 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.
> > > >
> > > >
> > > > 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/
> > >
> > >
> > >
> > > 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.
> > >
> > >
> > >
> > > 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.
> > >
> > >
> > >
> > >
> > >
> > > 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.
> > >
> > >
> > >
> > >
> > > 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
> > > ADVERTISEMENT
> > >
> > >
> > >
> > >
> > > 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
> > ADVERTISEMENT
> >
> >
> >
> >
> > 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.
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/
|