An analysis of intervention policies using reinforcement learning in an artificial market approach

摘要

Izumi et al. proposed the artificial market model, a multiagent model for a foreign exchange market, named AGEDASI TOF. And then they construct the support system for the government decides exchange rate scenarios. However in the system, the way of intervention: when the government intervenes and quantity of intervention, needs to be determined before simulations. Namely it is intervention without considering the market situation. First, we improve the way of agents' updating weights of information for forecasting the exchange rate. Second, we incorporate the intervening agent into the model. The intervening agent, which is as the government or the central bank in actual markets, intervenes depending on market situation dynamically and tries to acquire effective intervention policies using reinforcement learning.