Agent-based modeling as an approach to evaluate price discovery process in double auction markets

Abstract

PURPOSE OF THE STUDY

This study investigates how agent-based modeling can be used to evaluate the price discovery process in double auction markets. The study is limited to single-unit continuous double auctions, and especially to constrained zero-intelligence (ZI-C) trader markets first introduced by Gode and Sunder (1993a).

STRUCTURE

First, I evaluate the earlier models and construct an agent-based model using the guidelines from the literature. In particular, the idea is to create an agent-based model as simple as possible, because the earlier literature in agent-based modeling lacks synthesis about the modeling principles used. After having created the model, I compare its results comprehensively against the earlier literature. In addition, I concentrate especially to evaluating the methods of Cliff and Bruten (1997) to analyze ZI-C trader markets as their ideas have influenced literature substantially, but have been recently questioned by Othman (2008).

RESULTS

The results indicate that the methods of Cliff and Bruten (1997) can be improved. Especially, it appears that the probability density functions (PDF) of bids and asks proposed by Cliff and Bruten (1997) have to be constructed in a slightly different manner than what was originally proposed. However, the results also suggest that after refining the ideas of Cliff and Bruten (1997), it is possible to describe the PDF of transaction prices in ZI-C trader markets. Generally, the results suggest that the earlier literature has overlooked the importance of the evolution in the trader population participating in the ZI-C market. In addition, the results indicate that the trading in ZI-C trader markets closely mimics a sequence of trades that would take place on the Marshallian path, which has been previously suggested, but not comprehensively analyzed by Brewer et al. (2002).