NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/misc/invest/17056

Internet Message Format | 1993-01-25 | 8.3 KB

Path: sparky!uunet!gatech!destroyer!fmsrl7!lynx.unm.edu!centauri.unm.edu!scc26mf
From: [e-mail] (Mark Fleharty)
Newsgroups: misc.invest
Subject: Patter Recognition
Date: 26 Jan 1993 00:18:25 GMT
Organization: University of New Mexico, Albuquerque
Lines: 218
Distribution: world
Message-ID: <[e-mail]>
NNTP-Posting-Host: pioneer.unm.edu


Here is an article that a friend of mine wrote.
I did not post this article to promote the buisness of the company he
works for 'Hanseatic', I posted it here to increase the flow of information
and to promote education.  I have permision from the author and ofcorce
All disclaimers apply.

__________________________________________________________

PATTERN RECOGNITION:  A NEW DIRECTION IN TECHNICAL ANALYSIS


INTRODUCTION

Over the past 10 years, trading system designers at Hanseatic
have forged a new computer-
based pattern recognition approach to technical
analysis of markets.  But before
proceeding to a description of the
kinds of patterns we have developed, it should be helpful
to briefly set forth our philosophy about
technical analysis and price behavior in markets.


A CRITICAL VIEW OF TECHNICAL ANALYSIS

Over the years, the basic premise that has
guided our market research is that price
movement is sometimes random, but at other
times manifests trends that are quite non-random
so we believe that price movement does
derive from some underlying price
structure that presupposes an inherent
order to markets.  For us, this has
precluded any
efforts dedicated to methodologies such
as wave theory and Gann techniques, to say
nothing of astrology.  Nor do we think there
is any systematic cyclic component in price
data.  We also do not believe there is any
evidence of an existing method that is very good
at market forecasting; we are much more
interested in following markets well than in
predicting price levels.  Much of our work is
concerned with identifying better than normal
risk-reward market relationships in one
time dimension (e.g. weekly), and then recognizing
low risk entry points in an adjacent time
dimension (e.g. daily).

One final philosophical point has to do with computers.
Computers give us the advantage
of being able to analyze and organize
tremendous quantities of data and explore the
relationships among different categories of
data.  That is the good news.  The bad news is
that without a strong sense of philosophy
about how markets behave, and what system
objectives one is aiming to achieve, would-be
system designers are left largely with
unproductive chases through the vast
labyrinth of technical analysis techniques detailed in
literally hundreds of books, periodicals and software.
Unfortunately, in markets, even
poorly conceived systems work occasionally.


THE HANSEATIC APPROACH

Price is the only input to Hanseatic models.
Sentiment measures and other exogenous
variables may be useful in the U.S. markets
where objective measures can be formulated,
but they are simply not practical in the
analysis of cash currencies and other international
debt and equity markets that are the focus
of much of the work at Hanseatic.  In order to
accommodate a range of investment objectives
and horizons, we apply our models in four
time dimensions: monthly, weekly, daily and hourly.

The platform upon which we have developed
our pattern recognition approach to market
analysis combines an exponential moving
average of price with two momentum-derived
oscillators.  The primary oscillator is
relatively more reactive to price movements; the
secondary oscillator is less reactive but more stable.
An important aspect of this platform
construction is that it is not optimized
in any way.  The degree of smoothing in the monthly
Dollar - D-Mark, the weekly S&P500, the
daily Nikkei-225 and hourly-IBM, for instance, is
exactly the same.  Similarly, the
construction of the oscillators is the uniform across time
dimensions and markets.

The patterns are defined by either specific
relationships among time, price and the
oscillator components of Hanseatic's
models or a sequence of these relationships.  In either
case, the objective is to identify a
specific reward-risk relationship which has above-average
probabilities of generating profit in
either the short or longer term.  To demonstrate the
viability of each pattern, we have used a
so-called "12-stock q^t" which consists of four years
of daily data on the following stocks:

        AMGEN                MERCK
        COMPAQ COMPUTER        PHILLIP MORRIS
        APPLE COMPUTER            U.S. SURGICAL
        MICROSOFT                FANNIE MAE
        AMR                    DISNEY
        INTEL                 DIGITAL COMPUTER

The patterns are in two general categories.
The first category are transition patterns which
basically try to identify a change in trend
from one direction to anry of patterns
relate to the identification of entry/re-entry points into
trends.  These patterns inherently have
higher than normal probabilities because of the
persistence of trends in many markets
and indi3 months, for example,
there may be several daily entry and
re-entry patterns.  These patterns have high batting
averages, are of relatively short duration,
and have returns that are good but necessarily
constrained by the length of the trade.

The second type of trade pattern is designed
to recognize and enter a trend that has the
probability of continuing for some length of
time, and then remain in the trade until it is
recognized as having ended.  These patterns
also have good batting averages, but last much
longer on average and offer the potential
of very large returns.

A word of caution about probabilities of
achieving a positive return, or more simply,
batting averages.  Most high success-rate
systems are poor performers.  There is even
evidence to suggest that there is an inverse
correlation between high batting averages and
return.  So for those patterns that do have high
batting averages, our objective is to have
the profit factor greater than 2.0 before
transaction costs.  (Profit factor is a performance
measure defined as the ratio of the total net
profits from profitable trades divided by the
total net losses from losing trades:  a
break-even system has a profit factor of 1.)

An important aspect of our work is integrating
the parameters of adjoining time
dimensions to improve the risk-reward relationship
associated with a particular pattern.
For instance, favorable trend or momentum c
haracteristics in the weekly time dimension
will improve the performance of a daily trend
pattern.  Used in the opposite way, weekly
trend and momentum measures can be used as
filters to enhance the performance of a
particular pattern.  Lower time dimensions
parameters are generally used to improve the
entry and exit points for next-higher time
dimension patterns.

CONCLUSION

The observation and analysis of price patterns
is the quintessential art form of technical
analysis.  We, however, have never seen
evidence that any of the traditional patterns hold
up to any sort of vigorous statistical
investigation.  Hanseatic has constructed algorithms
which describe several transition and
trend patterns, each of which has been successfully
computer-tested over a large data sample
with a minimum of optimization.








Ed Meihaus is Vice President and Chief Investment
Officer for Hanseatic Group.  The
Hanseatic Group is a New Mexico-based trading
advisor/funds manager with fifteen years
of experience in diversification-based
management of currencies and both financial and
commodity futures.  It has subsidiaries which
are registered with the CFTC as a commodity
trading advisor and the SEC as an investment
advisor.  A third subsidiary participates in
securities lending.  Hanseatic advises 75
international banks on currency and interest rates
and transmits the advice on a 24 hour basis.
Hanseatic manages almost US$ 200 million in
trading accounts for institutions and acts
as a trading advisor to an offshore fund managed
by Cragnotti & Partners, an equity investor in Hanseatic.

------------------------------------------------------

I had to put the above because Ed Meihaus wanted that to be there
other wise I would not have been able to post this article.

If you have any questions.  Feel free to E-Mail me or post
a message to misc.invest.
                Mark Fleharty,
                [e-mail]
'Risk not thy whole wad.'