COAST: An architecture based on negotiation among competitive agents for automated asset management

摘要

In order to manage their portfolios in stock markets, often human traders use a set of algorithms and/or indicators, which are based on stock prices series. These algorithms are usually referred to as technical analysis. However, traders prefer to use a combination of various algorithms, rather than choosing a single one: the several signals provided by these algorithms and their own knowledge are combined to determine the orders to buy or sell some stocks. Inspired by the human traders decision processes, our architectural approach composes heterogeneous autonomous trader agents in a competitive multiagent system. This architecture allows the use of various algorithms, based on different technical analysis indexes to manage portfolios. This architecture is named COAST (COmpetitive Agent SocieTy) and it is composed by two kinds of agents: advisors that analyze the market situation autonomously and competing with each other for resources; and coaches that are able to coordinate several advisors and negotiate with each other to define the best money allocation within the society. This negotiation is performed using a negotiation protocol proposed in this work. We have implemented a system using COAST architecture, using a financial market simulator called AgEx. This system was tested using real data from the Brazilian stock exchange. The test results have shown a good performance when compared to an adjusted market index. The negotiation protocol used by coach agents provided a mechanism to easily integrate new trading algorithms, without the notion of a central agent or a centralized decision mechanism, which is a highly desirable feature in scalable multiagent systems.