Using Fuzzy Inference Systems for the Creation of Forex Market Predictive Models

This paper presents a method for creating Forex market predictive models using multi-agent and fuzzy systems, which have the objective of simulating the interactions that provoke changes in the price. Agents in the system represent traders performing buy and sell orders in a market, and fuzzy systems are used to model the rules followed by traders performing trades in a live market and intuitionistic fuzzy logic to model their decisions' indeterminacy. We use functions to restrict the agents' decisions, which make the agents become specialized at particular market conditions. These “specialization” functions use the grades of membership obtained from an agent's fuzzy system and thresholds obtained from training data sets, to determine if that agent is specialized enough to handle a market's current conditions. We have performed experiments and compared against the state of the art. Results demonstrate that our method obtains predictive errors (using mean absolute error) that are in the same order of magnitude than those errors obtained by models generated using deep learning and models generated by random forest, AdaBoost, XGBoost, and support-vector machines. Furthermore, we performed experiments that show that identifying specialized agents yields better results.