Simulation as a research tool for market architects

Abstract

Financial economists have three primary research tools at their disposal: theoretical modeling, statistical analysis, and computer simulation. In this chapter, we focus on using simulation to gain insights into trading and market structure topics, which are growing in importance for practitioners, policy-makers, and academics. We show how simulation can be used to gather data on trading decision behavior and to analyze performance in securities markets under controlled yet competitive conditions. We find that controlled simulations with participants are a flexible and reliable research tool when it comes to studying issues involving traders and market architecture. The role of the discrete event simulation model we have developed is to create a backdrop, or a controlled stochastic environment, for running market experiments with live subjects. Simulations enable us to gather data on trading participants' decision making and to ascertain the ability of incentives and market structures to influence outcomes. The statistical methods we use include experimental design and careful controls over experimental parameters such as the instructions given to participants. Furthermore, results are assessed both at the individual level to understand how participants respond to incentives in a trading setting and also at the market level to know whether the predicted outcomes are achieved and how well the market operated. There are two statistical methods described in the chapter. The first is discrete event simulation and the model of computer-generated trade order flow that we describe in Sect. 4.3. To create a realistic, but not ad hoc, market background, we use draws from a log-normal returns distribution to simulate changes in a stock's fundamental value, or P*. The model uses price-dependent Poisson distributions to generate a realistic flow of computer-generated buy and sell orders whose intensity and supply-demand balance vary over time. The order flow fluctuations depend on the difference between the current market price and the P* value. In Sect. 4.4, we illustrate the second method, which is experimental control to create groupings of participants in our simulations that have the same trading “assignment.” The result is the ability to make valid comparisons of traders' performances in the simulations.​