Groupe d’études et de recherche en analyse des décisions

From producer-scrounger game in behavioral ecology to market problem in economy: A reinforcement learning model for decision making in frequency-dependent situation

Mohammad Afshar

Invest in a new market or an established one? How much resources of an investment company should focus on finding or trying new markets? To an investor, these are some of the critical decisions which wrong answers could be fatal. Information on potential markets is the key to answer these questions. But how to find the information? Should I analyze and pilot invest or wait and watch the investment trends of new markets? We show that this problem is closely related to a social phenomena and evolutionary game theoretic problem in behavioral ecology, the Producer-Scrounger (PS) game; and as a solution to PS-game we propose a computational model based on reinforcement learning. The model allows us to analyze the effect of different parameters, including profitability of good markets, number of investor, number of good markets, cost of changing markets, and early bird’s profit, on the market problem. Comparing the results of our model with the real world experiments in behavioral ecology and considering that if a biological problem has an optimal solution, evolution finds that solution, we believe that the optimal solution to the market problem is based on reinforcement learning. The results help to make decision when there is a change in any of the mentioned parameters in the economic system.