Adaptive Modeler

Adaptive Modeler (also known as Altreva Adaptive Modeler) is a software application for creating financial market simulation models for the purpose of forecasting prices of real world market traded stocks or other securities. The technology used is based on the theory of Agent-Based Computational Economics (ACE), the computational study of economic processes modeled as dynamic systems of interacting heterogeneous agents. The software creates an agent-based model that consist of a population of agents that trade on a virtual market. The agents represent traders or investors that use technical trading rules that evolve through a special adaptive form of genetic programming. The forecasts are based on the behavior of the entire virtual market instead of only the best performing agent. This increases the robustness of the model and its ability to adapt to changing market behavior.

Contrary to many other techniques used in technical trading software (such as repeated optimizing and back-testing of trading rules, genetic algorithms and neural networks), Adaptive Modeler does not optimize or overfit (curve-fit) trading rules to historical training data. Instead, its models evolve incrementally over the available price data. Agents experience every price bar only once (as in the real world). Also there is no difference in the processing of historical and new price data (there is no training phase). Therefore there is no specific reason to expect that a model’s performance that has been demonstrated on historical data is better than its future performance (unlike when optimization or overfitting is used). Although past performance is never indicative of future performance, the results can be considered more significant and reliable with respect to future price data than results demonstrated on historical data by techniques based on optimization or overfitting.

In an example model, Adaptive Modeler shows significant risk-adjusted excess returns after transaction costs over the S&P 500 index. On historical price data covering a period of 60 years (1950-2010) a compounded average annual return of over 22% has been achieved, which is an excess annual return of 15%.

As an example of virtual intelligent life in a complex system (such as a stock market), Adaptive Modeler is said to be an illustration of simple agents interacting in a complex (nonlinear) way to forecast stock prices.

Origins

Adaptive Modeler was created by Jim Witkam and was first released to the public in August 2005. Several updated versions have been released since then.

References

• ACE Comp Labs and Demos. Department of Economics, Iowa State University.
• Financial Markets Show Case - Adaptive Modeler from Altreva. Evil Solutions, Evil Ltd.
• Altreva