 What is Pattern Recognition?
The following article was published in
Business Week. The article explores the result by academics at the
MIT, that chart patterns are are reliable indicators of the future
path of stock prices.
This
Alchemy May Yield Real Gold
Technical analysts--the
Street's soothsayers--get a vote of confidence from MIT rocket
scientists
Technical analysts don't
get no respect. Academic economists consider their game--predicting
stock prices by studying charts--on a par with examining the
entrails of a freshly slaughtered goat. Conventional economic wisdom
says that stock prices already reflect all relevant information, so
past movements are no clue to future ones. Even if stock prices are
a little predictable, say most economists, you won't get anywhere by
poring over charts for such technical-analyst arcana as ''rising
bottoms,'' ''double tops,'' and ''head-and-shoulders formations.''
Burton G. Malkiel, author of A Random Walk Down Wall
Street, writes that ''under scientific scrutiny, chart reading
must share a pedestal with alchemy.''
But technical
analysts--also known as chartists--may yet get the last
laugh. A paper written by three authors from the
Massachusetts Institute of Technology, and recently published by the
prestigious National Bureau of Economic Research, concludes that
''certain technical patterns do provide incremental
information, especially for Nasdaq stocks.'' In language
that's bold for academics, the paper goes on to say that ''while
this does not necessarily imply that technical analysis can be used
to generate 'excess' trading profits, it does raise the possibility
that technical analysis can add value to the investment
process.''
Technical analysis might actually add value?
Predictably, technical analysts are overjoyed by the partial
endorsement. ''I'm, like, flabbergasted,'' says Ralph J. Acampora,
the director of technical research at Prudential Securities Inc. The
chartists are particularly pleased that Andrew W. Lo, director of
MIT’s Laboratory for Financial Engineering, who taught many of the
finance rocket scientists now working on Wall Street, led the
project. Says Acampora: ''This gives the field an awful lot of
credibility.''
The paper, by Lo, graduate student Harry
Mamaysky, and finance professor Jiang Wang (http://www.nber.org/papers/w7613),
is tough slogging unless you happen to be on intimate terms with
kernel regression estimators and the Kolmogorov-Smirnov test. But
its basic strategy is simple. First, the authors wrote a computer
program that automated the process of finding 10 of the chartists'
favourite patterns. Then they turned the program loose on daily
stock returns for hundreds of companies on the New York Stock
Exchange, American Stock Exchange, and Nasdaq from 1962 to 1996. Out
of more than 800,000 observations of raw data, the program turned up
about 2,500 head-and-shoulders patterns (three
peaks, the middle one being the highest), about 2,100 triangle tops
(a pattern with progressively lower peaks and rising bottoms), and
so on.
STARTLING RESULTS. The next step was
to see how stocks performed in the period after a pattern was
completed. If technical analysis were hogwash, the authors should
have spotted no regularities at all--just-random ups and
downs.
The results were startling. The most bullish signal,
the inverse head-and-shoulders pattern, produced an average 4%
increase in the price of a stock on the third day after the
pattern's completion. The most bearish signal, broadening bottoms,
produced an average 6.2% decrease (charts). The authors also looked
at other statistical measures such as standard deviation and skew
and found that they also were significantly different from those of
a randomly chosen day in the market.
The reason for waiting
until the third day after a pattern's completion to peek at the
stock performance was that in practice, it sometimes takes a day or
so to recognize that a pattern has formed and to act on it. Plus,
the authors figured it would be more impressive to sceptics if a
stock was still moving as long as three days after a pattern's
completion. While the paper reports only the one-day change, Lo says
the authors got similarly impressive results when they looked at
five- and 10-day returns.
INTUITIVE INPUT.
So, is this bankable? The authors don't offer an opinion. The
problem is that there's no agreed-upon definition of what it means
to ''beat the market,'' because of the trade-off between risks and
rewards. If you earn an above-average return, but take on
above-average risks in doing so, doubters will argue that your
risk-adjusted return was no better than average. For instance, if a
chartist-trading strategy were highly correlated with movements in
the overall market, it would expose an investor to extra market
risk. Rather than get into a protracted debate about what
constitutes beating the market, the MIT group decided to focus on
what they could clearly prove. The paper demonstrates that technical
patterns do contain genuine information about what is going to
happen next in the market.
To be sure, real-life technical
analysis isn't as pure as the MIT research. Flesh-and-blood
chartists operate heavily on intuition. They often don't agree on
whether a particular pattern exists or even what it signals if it
does exist. They'll frequently walk both sides of the street,
predicting that stocks will fall in the short term but rise in the
long term, or vice versa, without precisely defining long or short.
''You scratch three technical analysts, you'll get four opinions,''
says Mike Epstein, the Boston-based director of quantitative trading
for NDB Capital Markets.
There is, for example, no consensus
among technical analysts about where today's gyrating stock market
is headed in the coming days. In the middle of the Nasdaq Composite
Index's roller-coaster ride on Apr. 4, Acampora said, ''I don't
think it's over for the Nasdaq necessarily,'' but added that ''if
they can blow out some more of these stocks, you'll get to the
capitulation phase,'' in which stocks touch bottom and rebound.
Epstein said, ''I'm looking for a trading rally very shortly, maybe
even right here,'' but warned that he might well change his mind the
next minute.
PROFITABLE PATTERNS. Despite
the theatrics surrounding technical analysis, Lo thinks it works
because it provides insight into some of the key forces that
determine prices, such as the warring forces of fear and greed.
Since markets are made up of real people, not economic automatons,
psychological concepts used by technical analysts such as
''resistance levels,'' ''support levels,'' ''overbought,''
''oversold,'' and ''momentum'' can have real meaning. Lo says that
technical analysts ''have been doing informally and intuitively what
academic economists do''--namely, measuring the forces of supply and
demand.
Ultimately, Lo hopes to use computers to detect new
patterns that are better indicators of the stock market than the
ones handed down from generation to generation of technical
analysts. He acknowledges that any pattern will eventually be
arbitraged away, but believes that ''the arbitrage itself can create
new ones....There might be two heads and three shoulders--some
pretty bizarre Quasimodo figures.'' As Lo sees it, the trick is to
develop tools that find exploitable patterns before others
do.
Lo says he and his co-authors have been approached about
commercialising their work, but they aren't looking to go into
technical analysis full-time. ''We actually like being academics a
lot,'' he says. On the other hand, Lo isn't completely immune to the
lure of money. He said he's still curious about whether technical
analysis really can beat the market after adjusting for risk--a
question that his latest paper doesn't attempt to answer. If he and
his colleagues discover that it can, he says, ‘‘we might never
publish that paper.'' Now, there's a professor with a good head on
his shoulders.
By Peter Coy
in New York
 Market Pattern Forecast
We trade the Market
Pattern Forecast alongside, in conjunction with technical analysis or
what we call our quantitative models. Short term cycles and patterns are never to be
traded on a "standalone" basis by themselves, ever.
Just like quantitative models should never be used by themselves.
The more tools that you have in your arsenal, the better.
Has
anyone ever seen an indicator that flashes a buy or a sell signal, and
then bought or sold, only to see the market keep going in the same
direction, without turning? It is our hypothesis that what is
occurring at these points, even with the best
mathematical/statistical models in the world- is that a pattern
formation is in play, being traded, and it is over-riding even the best of
the the best technical indicator sets.
Our computers analyze the
formations, or
patterns from the past several days: You cannot see them with the naked
eye, because, as mentioned in the above article on pattern
recognition, you cannot see where one begins and ends. Some traders
(and computers) see or detect and act on them earlier than others. The program that we have
written does that for us. It gives us the highest probability
pattern; turning points, angles of
ascent/descent along with the relational amplitude of the Y axis which should
occur over the next 6 to 8 hours. We publish our trading tool in the
form of a 5 minute bar chart which is
posted to the site before the open each day.
The
Future:
Can anyone 'predict" the future? Is it random chaos? We don't think so, because we have
several large firms and multi-million dollar traders that subscribe to
our service- many rarely have one losing week per year. Why
have they been subscribers to our service for many, many years?
The software program works on any stock,
commodity or index. We use it in-house on certain equities that
we trade.
In the
past few years the "research" section on many brokerage websites publish opinions on what should
happen to certain stocks. They read: "Our pattern recognition
program is issuing a buysignal, or a sellsignal on XYZ stock". There are two
camps, the fundamental analysts, and the "technical" or quantitative
analysts. The technical / quantitative analysts "chartists"
believe that everything fundamental is contained within the charts. We
are firmly rooted in the technical/chartist camp and believe that
everything fundamental is contained within the most
recent formations. Our measurements and analysis of the current
pattern that
has been developing over the past several hours determines
for us what we should expect to occur over the next 6-8 hours.
This field is the future,
and it sits on the cutting edge of methodologies employed for
extracting profits from the world's bourses. Quantitative
models- technical analysis systems analize price action and then
indicators will issue signals. Throw in sets of data with a
head and shoulders formation. The models issue the same signals
that traders will take, over and over again; buying or
selling when it issues those signals; when those highs and lows
are taking place. This human "patterned
behaviour" affects the direction of markets, and are predictable behaviours which present us with very high probability trading opportunities.
The quantitative model detects the actual accumulation or
distribution that is occurring at each of the turning points defined
within each Market Pattern Forecast: Both models: the scientific
math/stat quantitative model AND our projected formation of the next 6 to 8 hours- line up like
ducks in a row!
So
we trade our
work, and share it with a very small but growing number of professional
traders. We offer for
sale, one copy only- of the marketpatternforecast software: price: US$
15 Million - This includes a copy of the quantitative model software
that is running in more than one Central Bank. I'll retire and go
away somewhere peaceful -Any brokerage firm should love to buy this
company, but we do not advertise, so I guess they just don't know about
us. If you happen to be a business broker, and know of a firm
that would like to aquire us, there's a hefty finder's fee in it
for you. I'm just a quantitative analyst, not a salesman.
Subscribers
can see our projections for the indices/currencies & oil each day for only US$ 99 a month.
The Quantitative Model:
We
are very lucky to have a good math/stat model. We apply it to multiple time frames: tick charts through monthly
charts and everything in-between. We could not be successful traders
without it. It identifies for us whether the market is moving
into a low or a high on the cycle turning points depicted each day on our maps;
it helps us to determine when the cycle inverts, as is with all cycles;
you cannot always tell if it is a high or a low. It is required
that users of our service apply a good set of technical analysis
indicators, ie; a good math/stat quantitative model which will allow
you to do the same: confirm whether or not the cycle point is a high or a low, with
a higher degree of probability. See the tutorial section on
inversions: here. Occasionally I post some charts of the quant model that we use in the "Today's Performance" section. Interested
parties may obtain license: lease cost is US$100K/yr.
From some old notebooks from the early years, below are some types of patterns that occur in all markets; I ran out of
space, or there would be more; the computer at last check has
identified only 8.2 million of them. Our service provides you with a tool that is superior to any timing tool
available in the finanacial markets: The Market Pattern Forecast.
Click here to go to the subscription area.
Basic Patterns
|