|
PureBytes Links
Trading Reference Links
|
I have an understanding of out of sample testing;
that was an issue.
However, doesn't the design range depend on
what type of system, environment and universe one is considering? I doubt
that the 90's have a lot to do with the present, at least for some time
yet. Is this an a priori assessment? Yes.
The recent discussions of an example CMO system
that, when tested on the NDX gave consistent from the 90's and also since 2000
was an example. Using the NDX was certainly biased towards techs.
The only survivor of the system over both periods<FONT face=Arial
size=2> from my calculations and recollection was ALTR.
There may be no universal solution only local
solutions.
Cheers,
Richard
<BLOCKQUOTE
>
----- Original Message -----
<DIV
>From:
wavemechanic
To: <A title=amibroker@xxxxxxxxxx
href="">amibroker@xxxxxxxxxxxxxxx
Sent: Thursday, June 20, 2002 12:41
PM
Subject: Re: [amibroker] Trends, random
series
<BLOCKQUOTE
>
----- Original Message -----
<DIV
>From:
Richard
Alford
To: <A title=amibroker@xxxxxxxxxxxx
href="">amibroker@xxxxxxxxxxxxxxx
Sent: Thursday, June 20, 2002 1:26
PM
Subject: Re: [amibroker] Trends,random
series
Why wouldn't one simply design on now - 6 months, and
confirm on the previous 6 months? The suggestion of designing to
today, then testing for the next 6 months doesn't appear to be any
different - if one is absolutely certain they did not use information
from the preceding 6 months... Of course, selecting systems that
perform well on the last 6 out of sample data does use that data - tis a
puzzlement. Certainly only trying a new system every 6 months would
constrain losses.
Using a portion of existingdata for
system development and the balance for out of data testing is the standard
procedure, thereby having the ultimate Scrambler. However, one ofthe
pitfalls in evaluations of this nature is being able to ensure that the
period used for development includes a variety of market conditions, which
usually requires a fair amount of data. Otherwise, one probably only
has a system that is good for bull/bear/flat markets, etc.
wrt Al's point - I will let him manage that one. I
am marketing bullet proof boots, btw.
Cheers,
Richard
<BLOCKQUOTE
>
----- Original Message -----
<DIV
>From:
Ken Close
To: <A
title=amibroker@xxxxxxxxxxxxxxx
href="">amibroker@xxxxxxxxxxxxxxx
Sent: Thursday, June 20, 2002 11:35
AM
Subject: RE: [amibroker] Trends,
random series
<SPAN
class=170473116-20062002>William:
<SPAN
class=170473116-20062002>
I think the
point Al was making with his idea of using the Simulator is that you save
TIME. If you optimize (bad word--deduce a system) based on the last
two years up until yesterday, and then you wish to test "out-of-sample"
for 6 months, you will have to sit and collect the next six months worth
of data before you can draw a conclusion about the worth of the system you
have developed. The idea behind the simulator as I understand it is
to generate the next six months worth of data NOW and if results are
satisfactory, THEN begin trading tomorrow.
<SPAN
class=170473116-20062002>
The purpose
is to save time.
<SPAN
class=170473116-20062002>
However,
you shoot yourself in the foot if the conclusion is that the simulated
data is worthless as an out-of-sample test.
<SPAN
class=170473116-20062002>
I think I
have Al's purpose stated correctly.
<SPAN
class=170473116-20062002>
Ken
<FONT face=Tahoma
size=2>-----Original Message-----From: wpeters_1
[mailto:wpeters_1@xxxx]Sent: Thursday, June 20, 2002
1:17 PMTo: amibroker@xxxxxxxxxxxxxxxSubject: RE:
[amibroker] Trends, random series
<FONT face=Arial color=#0000ff
size=2>Thank you Al for that..i'm not sure that you
saw this follow up of mine:
<FONT face=Arial color=#0000ff
size=2>
<FONT
face=Arial><SPAN
class=875075116-20062002>"You then have to go on and ask what would
be the value of studying biases introduced in a series of randomised data
by mathematicians which does not occur in the real world market
place.
<FONT
face=Arial color=#0000ff size=2>
<FONT
face=Arial>Surely what would be valuable
is to study and root out biases already in the market due to human
behaviour or whatever. And get to learn about human behaviour and the
effect on the market place.<SPAN
class=875075116-20062002>"
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>Also-
<FONT
face=Arial>I
still don't see the value in this approach over using real data
history. If at the core of your argument you are saying there is some
possibility of this system reflecting the future, what might happen
in the future etc, then so could the data from the
past..
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial>For
me it always comes back to 'where is the value' over straight ticker
history, which we know is as real as its going to get. Humanistic patterns
have been more or less the same over many years at a base level and
they are represented in history of the chart for us to study.
Any tinkering with those patterns and you get some artificial result which
will never happen in the real world market....and again what is the worth
of studying that data over the
real?
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>Appreciate your
courage
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>William
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial><SPAN
class=875075116-20062002>
<FONT
face=Arial color=#0000ff size=2>
<FONT face=Tahoma
size=2>-----Original Message-----From: Avcinci
[mailto:avcinci@xxxx]Sent: Wednesday 19 June, 2002
07:12 PMTo: amibroker@xxxxxxxxxxxxxxxSubject:Re:
[amibroker] Trends, random series
William (and
Richard),
Let me try to explain
Leo’s scrambler. It was developed by Tushar Chande from his book Beyond
Technical Analysis, 2nd ed. Chande randomly rearranges the data of a
ticker to create new sequences. He does this as follows: he observes the
relationship between the O, H, L, and C of the 2nd bar by
using the C of the 1st bar as reference. So, he writes the
relationship as:
DeltaO = O –
ref(C,-1);
DeltaH = H –
ref(C,-1);
DeltaL = L –
ref(L,-1);
DeltaC = C –
ref(C,-1);
He samples with
replacement with these formulas and creates patterns that bear the
market’s signature as defined by relative price relationships. The next
step is to use a random number generator to scramble the bars. Whenyou
have a new sequence, you need a starting point, which is usually the
prior C. The new bar is derived from the prior C as follows (where the
Syn prefix stands for the new synthetic values):
Syn-C = ref(C,-1) +
deltaC;
Syn-H = ref(C,-1) +
deltaH;
Syn-L = ref(C,-1)
+deltaL;
Syn-O=ref(C,-1) +
deltaO;
He calculates the
interbar relationships as defined above for a ticker of your choice. He
uses a random number generator to pick a number from 1 to x (x
determined by the no. of bars you want to scramble). That number isthe
next bar of the sequence. Suppose on the 10th pick, you pick
bar 5. Then the original bar 5 becomes bar 10 of the new sequence. The
bars may repeat. You can generate as long a sequence as desired. You use
the synthetic values determined by the equations above to establishthe
next bar’s appearance. Thus, what you wind up doing is encapsulating the
market behavior in the original bar 10 and reproducing it in another
sequence to create new synthetic data. You can generate a variety of
chart patterns of any length using data scrambling. According to Chande,
you can generate 100 years of data and test your system against a
variety of market conditions. Since these are the types of patternsyou
are likely to see in the future, this is the most rigorous out-of-sample
testing you can achieve.
The above commentary was
partly plagiarized from Chande’s book, and it is merely a synopsis. To
get a clearer understanding of the methodology with spreadsheet
examples, you should read his chapter devoted to scrambling.
I have attached 2 gifs
showing what scrambled data look like on 2 different tickers. Note that
the patterns are not any more unusual than those of normal chart
patterns, and you can see definite trends developing. In fact, if you
look at recent real charts of the market indices, you will see many
instances of V-tops and V-bottoms, which is what William was concerned
about earlier today. Hope this helps. Many thanks to Leo Timmermansfor
programming Chande’s scrambler for Amibroker. I asked him recently if he
could program it in such a way to create synthetic data for an entire
watch list at one time. That might be tricky, but he said he would look
into it.
Al V.
<BLOCKQUOTE
>
----- Original Message -----
<DIV
>From:
<A title=wpeters_1@xxxx
href="">wpeters_1
To: <A
title=amibroker@xxxxxxxxxxxxxxx
href="">amibroker@xxxxxx
Sent: Wednesday, June 19, 2002
1:22 PM
Subject: RE: [amibroker] Trends,
random series, etc : was MetaStock and AmiBroker
<FONT face=Arial color=#0000ff
size=2>AL,
<FONT face=Arial color=#0000ff
size=2>
<FONT face=Arial color=#0000ff
size=2>With randomised data isn't there a chance that the so-called
'trend' can reverse on the next tick 'more-so' than data based on
human intervention. And that includes mechnical systems (ie. if enough
traders started using a certain indicator).
<FONT face=Arial color=#0000ff
size=2>The obvious predictability in human behaviour is what can give
you the little advantage and this is not in totally randomised
data.
<FONT face=Arial color=#0000ff
size=2>
<FONT face=Arial color=#0000ff
size=2>Surely if so your 'randomised' data 'trend' does not have an
equal chance of reversing on each and every tick then its not
randomised.
<FONT face=Arial color=#0000ff
size=2>
<FONT face=Arial color=#0000ff
size=2>Very interesting topic.
<FONT face=Arial color=#0000ff
size=2>William
<FONT face=Tahoma
size=2>-----Original Message-----From: Al Venosa
[mailto:avcinci@xxxx]Sent: Wednesday 19 June,2002
10:54 AMTo: amibroker@xxxxxxxxxxxxxxxSubject:
Re: [amibroker] Trends, random series, etc : was MetaStock and
AmiBroker
Richard,
I'm glad you took the liberty of changing the subject line. I
don't think you are being a pessimistic, grumpy old man (sorry,
don't know how old you are). Let me throw in another 2 cents into
the discussion of trends. You said in order to have a trend, there
must exist some information in the immediate past that would cause
the trend to persist for a long enough time to profit from the move.
I cannot agree more. That is absolutely true. A bad earnings report
comes out on INTC, and the market immediately reacts by selling.
This is further fueled by some nitwit analyst who changes his buy
recommendation to hold or sell (of course, always after the fact!).
Then, all the other individual "investors" climb aboard and do more
selling, driving the prices still further down. Then, the company
foretells that over the next several quarters there will be more of
the same in regards to diminished sales, inducing further pessimism
that drives the prices still further. All of this information
results in a downward trend. This does not nor cannot happen with
random events like coin tosses. So, technically speaking, you are
perfectly correct in your assertions .
Now, the trend follower comes in. He couldn't care less whatis
driving the prices down. In fact, he purposefully never pays
attention to news events and announcements that may drive prices in
one direction or another. He doesn't care about the cause of the
trend. All he does is climb aboard the resulting trend, using
whatever entry he has fashioned to enable him to take advantageof
this 'breakout' and whatever exit that enables him to exit whenthe
trend falters. Who was taking the other side of the market when
Neeson was loading up on Nikkei futures, driving the Barings Bank
into bankruptcy? The trend followers who recognized what was
happening and capitalized. What I'm saying is, the trend follower
buys or sells all breakouts (however you want to define the term
'breakout') and hopes that the resulting price behavior continues
long into the future. With the scrambler, these 'trends' can indeed
take place even with randomized data, as you have already pointed
out. What I'm suggesting is that your system of entries and exits
doesn't have a clue what caused the breakout to occur because it's
nothing more than a mathematical algorithm that trades based on
certain signals programmed by the user when certain price behavior
occurs. I contend you can test your system on such price behavior
just to learn how well the system reacts to price breakouts andto
see if, indeed, it is effective in finding trends in the data,
albeit randomized data. By the way, these comments are not limited
to trend following systems. I chose that trading approach merely as
an example to make my point.
Richard, I'm not trying to be argumentative on this topic. I'm
just pointing out why I think there might be value in testing a
fully optimized trading model on randomized data. Thanks for the
provocative discussion.
Al V.
>From: "Richard Alford"
>Reply-To: amibroker@xxxxxxxxxxxxxxx
>To:
>Subject: [amibroker] Trends, random series, etc : was
MetaStock and AmiBroker
>Date: Wed, 19 Jun 2002 08:35:15 -0500
>
>I agree on the oxymoronic nature of so called random
series. The problem results from using very particular distributions
for the random numbers which allows the mathematicians to write
books and teach classes. The Gaussian functions are one of the most
well-behaved functions around.
>
>wrt trends in scrambler and other "random" time
series - simply stated if they are truly random there is absolutely
no information about the past in the next event. Your coin toss
observation is a great example: given the 1/1000 case of 10 heads in
a row the odds of another head are precisely 50% for the next toss.
>
>In order to discuss a "trend" there must be some
information that persists. One can generally, perhaps always, see
trends in a random series - in hindsight - however it contains no
predictive information. There are systems that seek to identify
trending vs. trading range behavior (I think Ehlers addresses this
issue - can't say with how much success), however, they are
inaccurate at the changes. A moving average on the coin toss series
will identify trends - but the lag in the moving average that we all
know of and try to remove is the "hindsight" problem.
>
>Trend scale is still another issue. The apparent
fractal behavior of the stock market suggests that one can find
trends, and/or search for trends, over any time scale desired. (The
inflationary or perhaps evolving nature of economies have placed a
long term upward bias on the trends that can confuse the issue.)
Once again a moving average of any length on a random series will
display periods of upward and downward "trends".
>
>Sorry to be such a pessimistic old grump.
>
>Cordially,
>
>Richard
>
> ----- Original Message -----
> From: Avcinci
> To: amibroker@xxxxxxxxxxxxxxx
> Sent: Tuesday, June 18, 2002 8:19 PM
> Subject: Re: [amibroker] Re: MetaStock and AmiBroker
>
>
> Richard,
>
> >>distribution of the trends is predictable
for a random series >> This sounds oxymoronic to me, i.e.,
predictable distribution of random trends. But, I'll take your word
for it. :-))
>
> Regarding the scale when speaking of trends, I was
referring to weeks to months, perhaps even a year or two at most
(stocks, not futures), but not decades. This brings us back to
discussing the Scrambler. You said you don't hold much if any value
in it, but can you elaborate just a little more on what you mean in
terms of the scale I am referring to, because in a period of 130
trading days, I noticed some pretty decent trends develop that
lasted long enough to be of value in system testing? Thanks,
Richard. By the way, I agree that this discussion, interesting as it
is, has absolutely nothing to do with Metastock!!
>
> AV
>
> ----- Original Message -----
> From: Richard Alford
> To: amibroker@xxxxxxxxxxxxxxx
> Sent: Tuesday, June 18, 2002 3:55 PM
> Subject: Re: [amibroker] Re: MetaStock and AmiBroker
>
>
> wrt rsi/stochastics/cmo/etc... my point was that
they attempt to identify a phenomenon that has been observed asan
indication of overbought/oversold. There is a reasonable beliefthat
the arithmetic created to identify that behavior can be a useful
indictor of that condition. The comment offered for Elliot waveand
Fibinacci behavior also points to a large community that believes
that there is a hidden structure to the market and human behavior -
I am not a devotee of that belief, but I also have my doubts about
Madam Cleo....
>
> wrt trends in random numbers: there is, of course, a
trend between any two non=equal numbers. Given a suitable sample,
the distribution of the trends is predictable for a random series -
probably also random and if the normal Gaussian (pardon the pun)
distribution is used, as the derivative (or instantaneous
slope/trend) is a skewed Gaussian. Interesting to the statisticians
but not very useful to the average investor.
>
> wrt: trending 30% - that really has to be a matter
of scale. If you chose a suitably long moving average you can
probably even consider today as part of a long time upward trend -
not particularly useful - although even my meager portfolio looks
better today than it did in 1970 - not saying much....
>
> Interesting discussion - don't have the faintest
idea what it has to do with MetaStock :)
>
> Cheers,
>
> Richard
>
>
> ----- Original Message -----
> From: Al Venosa
> To: amibroker@xxxxxxxxxxxxxxx
> Sent: Tuesday, June 18, 2002 2:28 PM
> Subject: Re: [amibroker] Re: MetaStock and AmiBroker
>
>
> Thanks, Richard. I always enjoy your mathematical
insights. I hope you didn't get the impression I was suggesting
'optimizing' on random bars. I was merely suggesting 'testing' your
already optimized system on the scrambled data, seeing if it can
detect those occasional occurrences of non-random behavior of random
numbers, that's all. Certainly if a trend-following system can
detect trends in random price bars, it ought to be able to detect
them in real, non-random price bars, too, don't you think? Markets
trend only about 30% of the time, as I am told, so wouldn't you
consider non-trending markets (i.e., sideways markets) somewhat
representative of random price behavior?
>
> I also get the impression you are not an advocate of
overbought/oversold oscillators, right?
>
> Al Venosa
> avcinci@xxxx
> >From: "Richard Alford"
> >Reply-To: amibroker@xxxxxxxxxxxxxxx
> >To:
> >Subject: Re: [amibroker] Re: MetaStock and
AmiBroker
> >Date: Tue, 18 Jun 2002 12:55:33 -0500
> >
> >The non-random appearance of random numbersis a
very well know phenomenon - probably accounts for the continued
enthusiasm in slot machines in Vegas...
> >
> >I stand by my comments. In particular, the
various oversold/bought indicators rely on the observation thatthe
close tends to be higher in the l-h range in the overbought
condition - the persistence of such a trend is interpreted as
buy-sell behavior.
> >
> >I personally see no use in random data. There is
no information in random data and no expectation of the past
affecting the future - the only reason for technical analysis (other
than too much time on one's hands) is to glean information about the
near future. There are, of course, degenerate uses for random
numbers in testing, but to test, build, or, heaven forbid -
optimize, using random time series is pointless at best.
> >
> >wrt coin tosses: the coin is a Markov process of
length 0 - no memory. If the market is the same, we may as wellhead
for Vegas and get free drinks as we go broke. (OK - the market MAY
be a fair game, but I doubt it.)
> >
> >Just some thoughts...
> >
> >Richard
> > ----- Original Message -----
> > From: Al Venosa
> > To: amibroker@xxxxxxxxxxxxxxx
> > Sent: Tuesday, June 18, 2002 9:57 AM
> > Subject: Re: [amibroker] Re: MetaStock and
AmiBroker
> >
> >
> > Well, Richard, to be honest, I haven't yetdone
any back (er, I mean forward) testing of the random time bars
created by Scrambler. However, I have looked at the results of
randomly scrambling the bars and watching them form on the screen,
and you can see definite trends develop that can last several
months, much like in real life. So, at least theoretically, it seems
to me that, if you are testing a short-term or intermediate-term
trend-following system, it ought to be able to pick up on those
short, seemingly non-random trends (even though they were developed
from a random number generator). Remember, if you flip an honest
coin 1000 times, there is a low but finite probability that it will
come up heads 10 or 12 or 15 times in a row.
> >
> > AV
> >
> >
> > >From: "Richard Alford"
> >
> > >Reply-To: amibroker@xxxxxxxxxxxxxxx
> > >To:
> > >Subject: Re: [amibroker] Re: MetaStockand
AmiBroker
> > >Date: Tue, 18 Jun 2002 09:48:35 -0500
> > >
> > >I understood that the underlying premise of
technical analysis was that there IS information in the price-volume
behavior. If you test against random data the results had better
result in random behavior - perhaps mean behavior is more accurate.
> > >
> > >Perhaps I am delusional, however, I would
appreciate insight into the value of random time bars?
> > >
> > >Cordially,
> > >Richard
> > > ----- Original Message -----
> > > From: Al Venosa
> > > To: amibroker@xxxxxxxxxxxxxxx
> > > Sent: Tuesday, June 18, 2002 9:23 AM
> > > Subject: Re: [amibroker] Re: MetaStock and
AmiBroker
> > >
> > >
> > > Sorry, Dimitris, I thought you knew about
it. Go to post 19331 and download it, following Leo Timmerman's
instructions. What he has done is created a VBscript tool basedon
Tuchar Chande's price bar scrambler that scrambles all the barsof a
given ticker from the previous x bars (bars and ticker set by the
user) and then re-orders those bars randomly into the future, thus
creating a ticker that you can use in forward-testing. If you set
the no. of bars to, say, 130, you can generate 6 months worth of new
data for each ticker you do this on. Very, very cool. It differs
from William's Simulator in that it actually generates new data
rather than repeating existing data from the last x bars. Try it,
you'll like it.
> > >
> > > Al Venosa
> > >
> > > >From: "dtsokakis"
> > >
> > > >Reply-To: amibroker@xxxxxx
> > > >To: amibroker@xxxxxxxxxxxxxxx
> > > >Subject: [amibroker] Re: MetaStock and
AmiBroker
> > > >Date: Tue, 18 Jun 2002 13:49:30 -0000
> > > >
> > > >What is Leo's scrambler ??
> > > >DT
> > > >--- In amibroker@xxxx, "Al Venosa"
wrote:
> > > > >
> > > >
> > >
> > >
> >
>------------------------------------------------------------------------------
> > > Chat with friends online, try MSN
Messenger: Click Here
> > >
> > > Your use of Yahoo! Groups is subject to
the Yahoo! Terms of Service.
> > >
> >
> >
>
>------------------------------------------------------------------------------
> > Get your FREE download of MSN Explorer at
http://explorer.msn.com.
> >
> > Your use of Yahoo! Groups is subject to the
Yahoo! Terms of Service.
> >
>
>
>--------------------------------------------------------------------------
> Join the world's largest e-mail service with MSN
Hotmail. Click Here
>
>
> Your use of Yahoo! Groups is subject to the Yahoo!
Terms of Service.
>
>
>
> Your use of Yahoo! Groups is subject to the Yahoo!
Terms of Service.
>
>
> Yahoo! Groups Sponsor
> ADVERTISEMENT
>
>
>
>
> Your use of Yahoo! Groups is subject to the Yahoo!
Terms of Service.
>
Send and receive Hotmail on your mobile device: <A
href="">Click HereYour
use of Yahoo! Groups is subject to the <A
href="">Yahoo! Terms of
Service. Your use of Yahoo!
Groups is subject to the <A
href="">Yahoo! Terms of
Service. Your use of Yahoo! Groups is
subject to the Yahoo!Terms
of Service. Your use of Yahoo! Groups is
subject to the Yahoo! Terms of
Service. Your use of Yahoo! Groups is subject to
the Yahoo! Terms of
Service. Your use of Yahoo! Groups is
subject to the Yahoo! Terms of
Service. Your
use of Yahoo! Groups is subject to the <A
href="">Yahoo! Terms of Service.
|