Computational Intelligence within Finance

Intelligent Agents have been used in the financial industry for a variety of applications, from Simple Network Management Protocol-based (SNMP) monitoring tools for operational infrastructure management to more sophisticated functions like sorting, sifting and digesting the mountains of information that will surround consumers. In the ‘Always On Real Time Access’  Networks of the future, these agents pick out relevant information as they autonomously roam around the always-connected network autonomously.

In computer science, an intelligent agent (IA) is a software agent that exhibits some form of artificial intelligence that assists the user and will act on their behalf, in performing repetitive computer-related tasks. While the working of software agents used for operator assistance or data mining (sometimes referred to as bots) is often based on fixed pre-programmed rules, “intelligent” here implies the ability to adapt and learn.In some literature IAs are also referred to as autonomous intelligent agents, which means they act independently, and will learn and adapt to changing circumstances. –Wikipedia

One of the most notable and interesting use of agents is the Santa Fe Instute’s Artificial Stock Exchange. The Santa Fe Artificial Stock Market’s objective was to use agent-based technology to model an entire market and to get an understanding of the dynamics in relatively traditional economic models. It is these models for which economists often invoke the heroic assumption of convergence to rational expectations equilibrium where agents’ beliefs and behavior have converged to a self-consistent world view. Obviously, this would be a nice place to getto, but the dynamics of this journey are rarely spelled out. Given that financial markets appear to thrive on diverse opinions and behavior, a first level test of rational expectations from a heterogeneous learning perspective was always needed.

The Santa Fe Institute (or SFI) is a non-profit research institute in Santa Fe, New Mexico founded by George Cowan, David Pines, Sterling Colgate, Murray Gell-Mann, Nick Metropolis, Herb Anderson, Peter Caruthers, and Richard Stansky in 1984 to study complex systems.

–Wikipedia

While the Santa Fe Instute’s Artificial Stock Exchange was seminal work on exploring the use of agents to simulate a complex social networks, my scope is a wee bit narrower but I think equally of interest. What I propose is to model a smaller social network; that of an Artificial Trading Room. Imagine, if you will, a hyperspace room filled with autonomous agents buying and selling the same security. the room is filled with posts, similar to specialist stations, but more to just act as the book, and another function as we will see later, but not a market maker. The room is filled with x posts the represent different time horizons that an agent is interested in relative to the stock. Around each post will be ytrading agents that are truly intelligent in the “artificial” sense. They all have a hyper-cortex (Artificial Brains not just rules) that are self-organising neural networks that continuously adapts to incoming market data, recent transactional data, Quote data, and have also been trained on massive amounts of historical market data. The y agents that gather around a trading post are unique entities, they have been trained using a different sub-set of historical data, their neurons were randomized during training using different seed numbers and it is even possible to use different neural network topologies. All these variations will result in agents with differnt dispostions.

This may all seem obvious form a theoretical point of view so far. But what I suggest next is to put a second layer of adaptivity on top of the computational agents hyper-cortex. Most market participants know from empirical “pain” that as quickly as one places a trading strategy into the market its performance begins to decay. Just for the simple fact that the massive social network that make up the other market participants eventually catch on. The computational agents hyper-cortex adapt well but in some cases they may not adapt quickly enough because they have so much historical data to biases them towards a quick change.

I propose that the second layer of adaptivity is a simple strategy that is already employed in the “real trading rooms”. For each trading post we have y agents that buys and sells securities, attempting to out perform the market relative to all others agents;

The Survival of the fittest will drive the best Aston Martin…or more fittingly, in our case,  relegated to the highest performing processing unit!!

A “Post Agent” will monitor the performance of all agents during a fixed period of time, most likely proportional to the trading horizon but not necessarily. At the end of that fixed period the “Post Agent” will rank the agents in terms of market performance. The top performer is left to “his” own business; he is the “cock of the walk” for the day. The bottom performer is “fired”, well actually a little worse, kill -9. More interestingly the “Post Agents” then are responsible to become genetic scientists and mutate the remaining agents in different degrees dependant on their ranking; the over-performers mutate less then the under-performers. This genetic mutation can be implemented by randomly adjusting selected neural connections (Synapses) which will change the disposition of the agent and “may” accelerate the agents adaptivity to sudden market changes. This mutation method is crucial in the success of the evolution. Mutation may also create even poorer performers, but the algorithm of selection is based on survival of the fittest, so it should converge to the dominate spices (Oh yea, Intelligent Agent).

The computational scale and implementation of just one trading room is challenging and probably was not even possible only 1 or 2 years ago. Just the market data alone in terms of level II historical quotes were not available; and even if they were available to store them would be prohibitively expensive. Technologiclly, things have changed and now we have more of the enabling technologies we require.

Let’s look at what would make up a virtual trading room; (i) Many agents acting as traders (ii) several different strategies being implemented by the traders, like pairs, SUE and so on, (iIi) different time horizons for each strategy, (iv) several trading agents for each horizon and (v) and a Post Agent. To be modest, lets say we are using 10 strategies with 1min, 5min, 15min, 30min, 1hour, 2hour, 4, hour and 6 1/2 hour horizons, and we have 10 agents per post. This is just for ONE symbol, not even the entire NASDAQ. The scale becomes massively complex.

Computationally this will be very intense. but with advances in computing and the ability to capture and process massive volumes of meaningful data we have never had a better opportunity. We never before have had the processing power to analysis such a thing. This in conjunction with the advances in computational methods, we have been recently equipped to examine financial data in real-time and more efficiently than anytime previously. With the convergence of four advanced and diverse fields of; (i) model-driven trading, (ii) computational intelligence, (iii) the availability of high frequency market data, and (iv) the evolution of enabling technologies; such as available 64-bit processors, high performance data managers and grid computing we can attemp to model this. 

These rooms will be possible and will be able to analysis the entire NASDAQ cash equity market by loading and committing inhomogeneous time-series market data into cache memory. The market data can then be applied to continuously adaptive and biologically-inspired intelligent agents that will conduct high speed pattern recognition. The resultant patterns will indicate market anomalies and will form stylized facts that in turn can be used to supply a paradigm for model-driven trading. Because of the technology barriers to entry and a high level of domain specific knowledge required the method described here has not been attempted by any known large non-bank entities and is truly ground breaking.

More to Come.