AGENT-BASED MODELLING FOR FINANCIAL MARKETS WITH HETEROGENEOUS AGENTS: CALIBRATION AND SIMULATION

Abstract

This thesis introduces a framework of using agent-based models (ABM) with heterogeneous agents to produce an artificial financial market. The first ABM develops an order drive market embedded with an endogenous imitation mechanism and a population of noisy agents. The model network structure is established that each noisy agent can be imitated by others with a probability proportional to his wealth. We investigate the influence of imitation behaviour on the statistical properties of model outputs, such as log-return and bid-ask spread. The presence of stylised empirical facts, like volatility clustering and zero auto-correlation in price return, are verified. A variant model is presented by replacing the noisy agents with trending-following ones who can use the Bayesian learning method to track the asset return. Agents place limit orders based on the optimal strategies derived from a portfolio choice problem. A simulated price crash can be caused by a positive feedback loop arising from learning behaviours. Meanwhile, a parameter sensitivity analysis is implemented for both ABMs, which helps choose the initial condition and a further model calibration. We also provide a framework for calibrating an ABM using a genetic algorithm, where we aim to find a region close to an optimal value due to the complexity and stochasticity of an ABM.

Date of Award	1 Aug 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Paul Johnson (Supervisor) & Saraleesan Nadarajah (Supervisor)