Heterogeneous trading strategies with adaptive fuzzy Actor-Critic reinforcement learning: A behavioral approach

The present study addresses the learning mechanism of boundedly rational agents in the dynamic and noisy environment of financial markets. The main objective is the development of a system that "decodes" the knowledge-acquisition strategy and the decision-making process of technical analysts called "chartists". It advances the literature on heterogeneous learning in speculative markets by introducing a trading system wherein market environment and agent beliefs are represented by fuzzy inference rules. The resulting functionality leads to the derivation of the parameters of the fuzzy rules by means of adaptive training. In technical terms, it expands the literature that has utilized Actor-Critic reinforcement learning and fuzzy systems in agent-based applications, by presenting an adaptive fuzzy reinforcement learning approach that provides with accurate and prompt identification of market turning points and thus higher predictability. The purpose of this paper is to illustrate this concretely through a comparative investigation against other well-established models. The results indicate that with the inclusion of transaction costs, the profitability of the novel system in case of NASDAQ Composite, FTSE100 and NIKKEI255 indices is consistently superior to that of a Recurrent Neural Network, a Markov-switching model and a Buy and Hold strategy. Overall, the proposed system via the reinforcement learning mechanism, the fuzzy rule-based state space modeling and the adaptive action selection policy, leads to superior predictions upon the direction-of-change of the market.