A
Marriage in Artificial Intelligence
A
common assumption is that people can make better decisions with more
information. The logical extension is that people will make
even
better decisions with even more information. Hence, you find
content-drenched websites such as TheStreet.com, cnbc.com,
cbsmarketwatch.com and so many more. Add to these all the subscription
based investing magazines and newsletters that are available and can
there be any doubt about the thirst for limitless information?
But,
do people really make better choices with more information, or can a
task become too complex and, by its very complexity, undermine the
decision making process?
Among economists, it is an axiom that
choice is good and more choice is better. Giving buyers more choice
means more -- and more intense -- competition, which lowers prices,
raises quality and fosters innovation. In the end, workers are more
productive, consumers are better off and the economy is bigger and more
efficient.
It's a lovely theory, and one that is particularly
attractive to those who use it to justify replacing government
services -- Medicare, Social Security, public housing, public schools
-- with market-based solutions.
"Unfortunately,
it turns out not to be true. Yes, up to a point, choice does enhance
efficiency and consumer welfare. But at some point, there get to be so
many options about what to buy or what career to go into or which
mutual fund to invest in that many people make worse decisions than
they would if they had fewer choices……………..."
Washington Post, September 10, 2004
People
can only process a finite amount of information at any moment in time.
Once exceeded, these limitations give rise to information overload.
Numerous studies have shown that as a decision maker is initially given
information, decision quality improves. However, once the information
level reaches a certain point (saturation), the decision making quality
actually begins to deteriorate. At some point, people become
overloaded with information and make worse decisions. Their cognitive
capabilities become strained and overload sets in.
Admittedly,
one would believe that your investment decisions should depend upon as
large a body of expertise and information as possible. But so much
information truly can’t all be processed effectively and that births
the problems of information overload and results in inferior decisions.
Widespread
acceptance of this fact is responsible for the huge swell of interest
in the use of Artificial Intelligence in the investment arena.
Artificial
Intelligence (AI) is the use of computer algorithms, models and systems
to emulate human perception, cognition, and reasoning. In the broadest
sense, it is the range of technologies that allow computer systems to
perform complex functions mirroring the workings of the human mind.
This technology is not plagued by the limitations and restraints
previously mentioned.
On
the contrary, Artificial Intelligence thrives on input. Since computer
programs are not heir to the human frailties of bias, variability,
neglect of important data and emotion, the accuracy of their
conclusions can be far superior to that of humans.
“If
ever there were a field in which machine intelligence seemed destined
to replace human brainpower, the stock market would have to be it.
Investing is the ultimate numbers game, after all, and when it comes to
crunching numbers, silicon beats gray matter every time.”
The
February 13, 2004 edition of CNN Business 2.0
If further proof were required for the efficacy of the field of
Artificial Intelligence, the same article goes on to say:
“Andre
Archambault, for example, manages Standard & Poor's Neural Fair
Value 20, an AI-enhanced model portfolio………………… Since adding AI in
2000, Archambault's portfolio has increased in value by 55 percent,
while the S&P 500 has declined 26 percent.”
Today,
artificial intelligence based decision systems are being used by Credit
Suisse, Standard & Poors, Morgan Stanley,
Fidelity Management & Research and others too numerous to
mention.
Some
forms of Artificial Intelligence are designed around expert systems. An
expert system is a computer program that uses deductive logic to
simulate the decision-making process of humans. The program contains a
knowledge base consisting of if-then-else rules and facts, a user
interface, a database/spreadsheet interface and an inference engine
that makes logic-based decisions. These systems are built by
interviewing large numbers of experts about a particular subject,
building huge database of statistics and then distilling all this
knowledge for the benefit of non-expert users.
For expert knowledge to be effectively incorporated into a program, it
must be expressible in one of three forms:
1. Facts.
2. Procedures or
3. Rules.
For
example, General
Motors trades on the New York Stock Exchange is a Fact. How to go about building a
simple moving average is a
Procedure. Sell
when the GM trades below its 200 day
moving average is a Rule.
However, there are limitations;
expert systems are dynamic but not adaptable. They are exclusively
rule-driven. These rules appear at different levels of a decision tree
and, at any given level of the tree, there are certainty factors that
are associated with each rule or probability of outcome. As information
is presented to the expert system, rules are activated at various
levels of the decision tree.
As the trading facts for the
particular period are input into the system and the respective rules
are activated, the probabilities at the various levels of the tree are
combined. This creates a confidence or certainty factor decision for
that period in question. As such, you'll get different probability
decisions with different facts. An expert system adjusts, but it
doesn't learn. That is to say, it can't adapt.
The other form of
Artificial Intelligence is a neural network; neural networks begin as a
complete void of knowledge. They know nothing until you demonstrate
something to them. A neural network can find cause and effect
relationships or patterns. It adapts when you show it new information.
It actually learns and knows to de-emphasize older cause and effect
relationships that are no longer relevant and to emphasize the new
ones. Neural networks process and operate with feedback.
Neural
networks mimic the human brain's most powerful ability, they can
recognize patterns. Neural networks can be viewed as heuristic
procedures best applied where:
1. One can specify particular influences on a phenomenon whose outcome
is known with certainty.
2. The
relationship can not be described.
3. The
relationship is not necessarily linear or
4. There are no
known models.
Neural
network systems are most effectively applied to pattern recognition
tasks, such as classification. Classification involves the assignment
of input to predefined groups or classes based on patterns that exists
in the input data. The neural net can make a decision by drawing on its
inventory of previously learned patterns and accessing the most
relevant to the current environment. This is where neural nets are
invaluable. What neural nets lack is anchor knowledge, that is,
knowledge that isn’t subject to changes in the data environment.
The
most exciting development in artificial intelligence is the marriage of
neural networks and expert systems technologies. This hybrid technology
allows the creation of intelligent systems where the whole is greater
than the sum of its parts and exploits the complementary strengths of
each. Neural networks and expert systems represent two major aspects of
human intelligence and, therefore, are appropriate for integration.
Neural networks represent the visual, pattern-recognition types of
intelligence, while expert systems represent the logical, reasoning
processes. Together, these technologies allow applications to be
developed that are more powerful than when either technique is used
individually.
A Neural Network Expert System (NNES) is arguably
one of the best simulated approximations of a human brain. Neural
Networks have the uncanny ability to discover subtle relationships in
data if they exist, and can function with incomplete data. The tree
structure of an Expert System allows the knowledge base creator to
build links between relevant neurons. The results are adaptable,
rule-based decision systems incorporating all the advantages of both
modalities.
To build a NNES, knowledge from the expert system is
used to set the initial conditions, the training data and input
variables for the neural network, which evolves from there. The
interaction of the systems can be described in the four following steps:
1. Based on its rules, the expert system extracts instances of objects
(patterns) and their associated attributes (input variables), from a
database.
2. The data is
intermediately stored in a
depository (database or file), where preprocessing of the data is done,
if necessary.
3. The neural net
reads the data from the
depository (instances of objects and associated attributes) and
commences training and
4. The resultant
NNES produces an output at its designated task.
This
is the model in its simplest form; in essence, it combines an expert
system that learns with a neural network that learns to be an expert
system. The appeal of such a NNES is that it provides an effective
means of trade/investment selection and portfolio management through
the simulation of a fully integrated expert thought process.
Academics,
financial mathematicians, professional technical and fundamental
analysts, and investors the world over have all come to realize that
successful investing requires a comprehensive approach which can
exploit all the available information about the financial markets. This
information includes such things as inter-market relationships and
influences, trends and momentums on market activities, business
fundamentals, and mass psychology.
However, such a complete
approach may only be possible through the use of comprehensive
intelligent systems that are able to implement and integrate financial
market predictions, technical trading strategies, fundamental investing
methodologies, behavioral finance and quantitative portfolio management.
Investing
for a consistent annual return measured in multiples of the performance
of the S&P 500 average is a lofty goal. To accomplish this,
one’s
research and inquiry must include:
1. Financial Mathematics and Statistics
2. Technical
Analysis
3. Fundamental
Analysis
4. Financial Game
Theory
5. Financial Power
Laws from Physics and Chaos Theory
6. Investing
Methodologies and Systems and
7. Individual and
Mass Psychology of Investors and Traders
The
workings of the NNES have shown that taking a wider view is a key
component in the optimum strategy for investing.
Imagine yourself standing directly in front of a mile-long billboard.
Although there is a lot of information presented, you can only see a
small part of it because you are much too close. As you step farther
back, much more is revealed. The same is true for stocks.
The
AIQ RESEARCH investment philosophy is based on the premise that raw
market prices, rather than market
fundamental data, are the prime aggregator of information necessary to
make
sound investment decisions. Prices, which may seem random, actually
move through time in complex, but discernible ways.
This
philosophy is based on analysis of almost a century of historical data
that revealed that
market prices will usually form trends that can be profitably
exploited. AIQ RESEARCH believes there is an inherent return
opportunity in participating in price trends that its artificial
Intelligence models have identified. Our programs may
participate
in either rising or falling trends; they do not have an
inherent directional bias
nor do they try to predict price targets. However, these programs
have demonstrated a unique ability to identify market turning
points., that is, trend changes.
As we see it, the behavior of markets is based
on investors’ expectations, which adjust slowly through
time and can manifest in long-term price trends. Our
investment decision tree has been designed to find and exploit these
trends. We maintain that market prices may initially react to
new
information or events, but the aggregate impact on prices may
be extended through a longer period of time. Anything that
could
possibly affect the market price of any financial instrument, including
fundamental, political, or psychological factors, will therefore
eventually be reflected in the price of that
instrument.
AIQ RESEARCH believes that gradual price
adjustments manifest themselves in long-term trends, and that such
market trends can be discerned and best exploited through the use of
artificial intelligence technologies. We consistently employ our
proprietary Neural Network Expert Systems (NNES) to examine market data
for price behavior or relationships that will characterize a
trend, and therefore, an opportunity for profit.
AIQ RESEARCH
considers that price is the combination of the signal plus noise,
where the signal is the trend information and the “noise” is the market
volatility surrounding the trend. Prices may aggregate
market information, but noisy
price signals distort the data and have to be filtered to
discover the underlying price trend. This is one of the key
strengths of our proprietary Neural Network Expert Systems.
Admittedly,
most Trend following systems suffer from a low winning percentage, that
is, they incur many trades with small losses so that they can catch
the big ones.
It is argued that it is better to have winning trades of
large magnitude than it is to have a high winning percentage.
Implicit
in that argument is that you can’t have both. Our work with Artificial
Intelligence has provided sufficient evidence for us to most
emphatically disagree.
|