Agents

The type of the trading agents that will be used in the agent-based financial market model is, without any doubt, the most important design question, and decision when building a model of a market.

Real-World Strategies

There are several design routes that one can take when representing the trading agents that will participate in the market. The simplest way is to model agents in which their trading rules replicate trading strategies that are used in the real financial markets. However, no learning takes place for this type of agent. This direct method results in interesting outcomes but, at the same time, comes with a few costs. One of these is that this type of agent might lack a well-defined objective func­tion. In other words, agents with a utility function can measure and evaluate their performance in the market. A further criticism is the dynamics of interactions in which traders in real markets modify their investment strategy over time.

Zero Intelligence

Another type of agent design is the “zero intel­ligence” agent in which their behaviour is ruled by a simple budget constraint. Although their trading behaviour gives the impression of being a form of learning, these types of agents do not learn. Gode and Sunder (1991) were the first to introduce the concept of the “zero intelligence” agent. The authors used this type of agent to model market transactions in double auctions. Similar are the one done by Sunder (2004), Daniel (2006), and Cliff (2009).

Learning and Adapting Strategies

In relation to the learning mechanism, there are agents that are modelled to learn and adapt to the new conditions in the market.

In the SF ASM, in addition to the learning mechanism, there is another important aspect that distinguishes this type of agent from any others. This is the implementation of forecasting. The prediction is made using a classifier forecasting system in which current market information is mapped into a conditional forecast of future prices and dividends (Lebaron, 2006a). Examples of market models where predication strategies are used are Arifovic (1996), Arthur et al. (1997), Edmonds (1999), Chen and Yeh (2001), LeBaron (2001b), Zimmermann et al. (2001a), Yang (2002), Markose et al. (2003), and Martinez-Jaramillo and Tsang (2009).

Objective Function

A different way to design the agents is to make use of the objective function. Agents could be mod­elled with an objective function either implicitly or explicitly with regard to the agent’s decision­making process. An implicit objective function implies that the agent’s objective is incorporated indirectly into the decision making process. An example of an implicit objective function is maxi­mizing the agent’s outcomes. On the other hand, most ofthe agents in market models have an explicit objective function where the agent’s performance in the market can be measured through the use of the utility function.

Social Learning

There are agents whose strategy is based on ob­serving other individuals ’ trades in the market and adjusting their strategies based on the observed actions of these other traders. This phenomenon is often called “social learning” (LeBaron, 2006a). An example of market models using such an agent are the models introduced in Kirman (1993), Lux (1995), Bak et al. (1997), Lux (1998), Cont and Bouchaud (2000), Alfarano and Lux (2003), Alfarano et al. (2005), and Hott (2009). These market models aim to model the trading agents by incorporating herding behaviour. Hott (2009) showed that price bubbles can be explained in terms of herding behaviour.

We have summarized above the most important design issues regarding the modelling the trad­ing agents. In addition, we have illustrated some examples of agent-based market models, which use different trading agents’ design.

3.2.