Improving classifier agents with order books information

In the study of financial phenomena, agent-based artificial markets are efficient tools for testing economic assumptions of market regulation. While it is easy to populate these virtual worlds with “basic” chartists agents using price history (increase or decrease of prices, moving averages, ...), it is nevertheless necessary, in order to study rationality phenomena and influence between agents, to add some kind of learning agents. Several authors have of course already been interested in adaptive techniques but they mainly take into account price history. But prices are only consequences of orders and therefore reasoning about orders provides a step advance in the deductive process. In this paper we show how to take into account the whole information about the market, including how to leverage information from the order books such as the best limits, size of bid-ask spread or waiting liquidities to accommodate more effectively to market offerings. Like B. Arthur we focus here on the use of LCS agents.