Financial organizations are starting to recognize that there may be something to this AI stuff and there may be some benefits to be derived. Off setting these positive feelings are the strong negative reactions that are usually associated with change and misunderstanding. Recent comments by two respected financial writers suggest that there is a great deal of misinformation being distributed. The first writer stated, "Probably the worst potential for disaster comes from techniques such as neural networks ...nothing in neural networks is based on financial theory: it is just what seems to fit a set of data in a certain time frame." The second financial writer recently wrote, "despite all the technological advances being made, I still believe success in the market will be the result of good old common sense. Indeed, the market's attempts to ensure gains through these high-tech approaches, in my view, is a tremendous negative. Program trading was an early glimpse of a 'system' gone amok, and recent debacles in the derivatives area show that high-tech advances don't necessarily improve your investment odds. In sum, I'd reiterate my oft-stated advice: You don't need to be glued to your radio or TV to be successful in the stock market. You don't need daily hotlines or online services. And you don't need computers to do your thinking. What you do need is patience, discipline, common sense, and a long-term view. Period. As for financial cybernetics, it's clearly a double-edged sword that warrants our amazement--and fear. "
These two writers are voicing the concerns and fears held by many, perhaps even the majority, of those working in the money management field. Fear of losing ones analysis job to a smart computer or fear of computers taking over the markets tomorrow are just not realistic. That does not mean that could not or will not happen, but it won't be in your life time or mine. People, regardless of how rich or believing they are, are not going to turn their hard earned assets over to a machine to manage.
More realistically, these AI systems, or more correctly, subsystems, will be integrated into money management firms in such a way that they enhance or augment human performance. They will take over task that humans perform poorly or are incapable of. But, before we can start to integrate these intelligent subsystems into the financial community we are going to have to deal with and over come the fears and misunderstanding that exist out there. The best way to accomplish this is through education and trade publications.
As an initial step down that road of understanding lets deal with the comments made earlier by the two financial writers. The first writer suggested that the application of neural networks as a data mining tool is very dangerous. I would agree with that. In statistics 101 they teach you that if you have enough degrees of freedom you can fit any equation to any data. Certainly neural networks have a very large number of degrees of freedom and can, in fact, map just about any set of inputs to a set of outputs. But, than again so can the human brain. The degrees of freedom in the human brain are beyond comprehension and that's what makes it such a magnificent flexible problem solving device.
It is true that we do not know how a neural network solves some problems and it is equally true that we do not know how the human brain solves many problems. We gage the performance of both of these problem solvers by their success and their ability to generalize to new situations.
The first writer went on to suggest that neural networks are devoid of financial theory. This is also true, but it misses the point. Neural networks are computational devices. The only theory they incorporate is how to change their weight structure if they come up with the wrong answer. They are general computational devices that can solve numerous different types of problems. Financial theory, if you think it is important, comes in the selection of the independent variables that are the input to the neural network. I suggest to you, that if you do not know something about financial theory and the markets, you will have a very difficult time making a neural network give you any kind of results that are useful.
Before leaving the subject of financial theory, however, I would like to suggest to you that a lot of financial theory is nothing more than that, theory. Economists tell us that if we hold everything else constant and change supply, demand will change. Well, that is not the way the world works. Everything is changing all the time and the interactions among all these changing events is what makes financial and economic theory so tenuous. In neural networks we finally have a computational device that can start to handle some of the dynamic nonlinear complexity associated with these problems.
I can't let the neural network people completely off the hook, however. Several of the neural net software packages currently available offer an option where one can supposedly do a sensitivity analysis where one can hold all but one input variable constant and vary the one input through its entire input range so that the user can see what impact it has on the output. This is the same type of thinking and theory building we just discussed. All other things are not constant and will never be constant. We live in a dynamic ever changing nonlinear world. If you think you are testing the importance of your neural network input variables with the software's sensitivity function you are sadly mistaken. In our research we have removed variables from our networks that have appeared to have very little impact on the neural nets performance (small weights and little change in output when varied across its entire input range) only to find out that the variables were apparently interacting with three or four other variables in some high order interaction beyond human comprehension and affecting the output and the overall performance of the system. When the variables were added back into the net the performance rebounded. Truly, the whole is greater than the sum of its parts.
Now lets address the second financial writers concerns regarding the potential abuse of AI systems. I happen to share his concern. The potential power of these intelligent systems could significantly alter the financial world as we know it and potentially even shift the power and influence of countries throughout the world. If that is not enough to put the fear of God in you nothing will. For this reason alone it is important for our country to be on the cutting edge of this technology.
Where I take exception to the writer's view is where he likens these systems to program trading and derivatives. The general public seems to be lumping anything that has to do with computers and higher than expected investment returns into a single basket despite the fact that they are all quite different. AI systems, at least as currently used in the equities markets, have been used as decision making aids; aids to enhance and augment human decision making. They are being used to do the very thing that the writer suggested we should be doing; searching out good investments and applying a discipline approach, along with common sense. The problem is that the writer does not think you need a computer to accomplish this task. To even suggest that an intelligent computer system can not enhance or improve an individual's ability to analyze a complex problem like the equities markets is, at the very best, naive. Just like computers are now beating the best chess masters of the world so also can they outperform the best money managers on Wall Street.
Our hybrid intelligent systems at NeuWorld Financial, Inc. have been designed to do the very thing that the second writer implies that they can't do. Using a year ahead time horizon our systems are investment systems, not trading systems. In fact if you don't have patients you should not be using one of our systems. Our systems generally identify good under-valued situations two to three months before the street recognizes the value. This has allowed us to generate returns that are not two hundred to three hundred basis points better than the S&P500, but rather two to three times the S&P500.
Anyone that thinks that AI systems are not going to alter the way we manage money in the future is just not in touch with what is going on in the information sciences. Our two financial writers that have expressed concern and even fear regarding AI systems have taken the first step. They have acknowledged that the technology exist and that we need to be thinking about it. The next step is education. With a greater understanding of what the technology can and can not do, they will come to appreciate the power that lies on our door step.