![]() |
Agents do Securities Investing
|
Since the 1960's, as the field of finance became more scientific, the Efficient Market Hypothesis (EMH) has been a favorite of many academicians. Simply put, the EMH implies that you can't beat the market indices, and the EMH is becoming so popular that companies are doing a booming business selling Indexed Mutual Funds. Of course, it doesn't help the nay sayers that, over the past twenty years, 85% of all mutual funds have underperformed their market indices.
The essence of the EMH is that at any point in time, prices of securities in an efficient market already reflect the assimilation of all information available to investors. In other words, "The other guy already knows what you know, so you can't get an edge.". There are three proposed forms of the EMH, each dealing with a different type of information. The "weak form" considers past price information only. The "semistrong form" considers all publicly available information, and the "strong form" considers all publicly and PRIVATELY available information.
On the surface, the strong form is pretty silly. So many people have made so much money trading on privately available information that the Federal Government has made insider trading illegal for most of this century. We think the strong form of the EMH can be discarded, but the other two forms, ... well they're a lot tougher to discount.
If we were to discount the EMH, the first cracks came, not from computers, but from a number of highly successful investors like Ben Graham, Georg Soros, and Warren Buffett. Even though several of these investors have made billions by systematically beating the market indices, for a number of years many academicians treated them like the Rodney Dangerfields of finance, giving them little respect and attributing their success to "luck". Now, we're a crew of math hacks and computer geeks, so the EMH is a natural for us, but ... well you gotta like a billion dollars.
As more and more scientists address this question, applying powerful modern computer analysis tools, there is an increasing body of evidence indicating the market is "chaotic". Chaotic behavior emerges when systems are composed of random trends which begin and end at random periods and with random duration.
In the absence of modern computing equipment, to the untrained eye, chaotic systems can appear random. But, in fact, chaotic systems are filled with fleeting, changing structure. Much like the surface of the sea, which is not flat, but instead is filled with moving peaks and valleys, called waves, whose position and shape change constantly. To the untrained eye the surface of the sea may appear random; but, in fact, we now know it to be chaotic.
In other words, "The other guy knows what you know; but, not instantaneously. So you can occasionally get an edge; but, only with great difficulty, and you can never count on it persisting".
We believe that the securities market is an example of a chaotic system from which unexpected behavior emerges. The last generation of statistical tools and expert systems were constantly surprised by the market's adaptively. Once a profitable "rule" was discovered and a trading strategy put into play, the market quickly adapted until the old "rule" was no longer profitable.
We believe that the new generation of agent oriented machine intelligence, with its amazing adaptively, is a natural tool for studying chaotic systems, such as the securities market. Inspired by the book, "Society Of Mind", Agent Information Server?; supports a social approach to machine intelligence. Instead of developing a single monolithic opinion about the market, Agent Information Server?; combines and blends opinions from all of its agents into a multifaceted point of view. Agents actually compete with each other as they try to understand emerging behavior in the market. The agents, with the best track records, play an ever increasing role in the final securities recommendations; while Agents who lose their touch, play an ever decreasing role in the final securities recommendations. The key to winning in the market is constant unending change.
Balanced investing combines the safety of bonds with the potential returns of common stocks and options. It is a marriage of two extremes. Bonds represent low risk with historically low profits. Common stocks and options represent high risk with historically high profits. By combining these two extremes, the fund manager hopes to receive most of the high profits from stocks and options while keeping much of the low risk of bonds. In practice, it seems to have worked reasonably well.
Comparing the historical returns of the Standard & Poor's 500 index of stocks, with the average performance of short term Bond mutual funds, with the average performance of Balanced mutual funds, over the last ten years, the average Balanced mutual fund has provided almost 75% of the profits of common stocks with much less risk. Here are the historical results:
NAME | 3 Mo | 1 Yr | 5 Yrs | 10 Yrs |
---|---|---|---|---|
S&P 500 | 17.44% | 34.68% | 19.76% | 14.65% |
BALANCED | 9.85% | 19.99% | 13.20% | 10.94% |
BOND FDS | 2.40% | 6.73% | 5.83% | 7.22% |
Founded in August of 1993, Korns Associates is a privately held applied research and development firm, devoted to research in Artificial Intelligence and next generation Databases. From the beginning, a "killer app" was necessary to prove the capabilities of the technology, Agent Information Server?;. Korns Associates chose to develop a system with domain expertise in securities investing, and to use a business model whereby its research would be self-funding.
Furthermore, hacking the securities markets would be an even greater challenge than hacking chess, and would provide a perfect stress test for the software. So, in late 1993, inspired by what IBM's Deep Blue was doing in the world of chess, Korns Associates started out to write a system to profit from the stock market.
The alpha version of DeepGreen? with its initial StockWizard intelligent agent; was completed in December 1993. DeepGreen? is built upon a database of over 1000 securities, with over 200 quantized attributes per security. The database is updated on a weekly basis and contains information about securities and popular indices. Among the quantized attributes for each security are: industry group, current PE, % debit to assets, return on equity, analyst expected profits for the next four quarters, history of earnings surprises, current price, with historical values for price, volume, and fundamental attributes going back to January 1986.
DeepGreen? is primarily a data mining, knowledge discovery system, based upon a number of machine learning algorithms, which are blended together by a community of "Agents". These algorithms include: rule induction, neural nets, multiple linear regression, and genetic programming.
DeepGreen?'s power comes from the way its agents' use language. Automated investment agents are expressed in a simple, computationally complete, English-like language. Profit measuring agents, for long, short, and option positions, are expressed in a simple, computationally complete, English-like language. The user can easily enter specifications for, and backtest over past historical periods, virtually any conceivable automated stock selection strategy. Additionally, each automated stock selection strategy can be evaluated by a large number of measuring agents which simulate the profits from long, short, or complex derivative positions. The user can very quickly see what works on Wall Street.
As users enter agents, in English-like language, for stock selection and profit measurement, DeepGreen? studies the past historical data and creates agents of its own, using the data mining algorithms listed above. All agents enter a "survival of the fittest pool" and can be sorted by profitability. Even the agents, which have been created by DeepGreen?, are presented to the user in the same English-like language. This goes a long way toward removing the "black box" aspects of the system, making it much easier to select an appropriate investment strategy.
This year, Korns Associates started planning for its next generation computer investing system. The next generation of DeepGreen? will advance beyond the current technology in the following areas: