Home > Research > Publications & Outputs > Replicating agent-based simulation models of he...

Electronic data

Text available via DOI:

View graph of relations

Replicating agent-based simulation models of herding in financial markets

Research output: ThesisDoctoral Thesis

Publication date2021
Number of pages260
Awarding Institution
  • Lancaster University
<mark>Original language</mark>English


Agent-based simulation of herding in financial markets varies in the herding and market mechanism. Replication studies are a cornerstone of the scientific method although it is not applied very often. The research aims to obtain a greater understanding of herding and the related stylised facts, assess the herding models’ reproducibility, and find ways to improve reproducibility by replicating two herding models.

The first herding mechanism factor (Tedeschi et al., 2012) controls the extent of the neighbour’s influence on the expected returns and hence on the trading decisions. The market mechanism is an artificial market where agents submit either ask or bid orders into the order book, and they trade between themselves. The second replicating herding mechanism (Lux and Marchesi, 2000) is based on transition probabilities to decide whether agents are fundamentalists, optimistic or pessimistic chartists. The market mechanism is demand and supply.

The first replicating study fails to produce the original results, whereas the second does have similar findings to the original paper. The second model’s description is done by following the recent STRESS guidelines for specifying models. The guidelines help to cover everything needed for describing the second model. Then features from the STRESS and ODD guidelines are combined to give a slightly revised guideline with a defined structure. This is considered to give an improvement.

Both models have fat tails, and only the second model has volatility clustering. The behaviour of the second model that gives the volatility clustering is called on-off intermittency. This is analysed in detail to understand how the model enters and leaves periods of high volatility. The conclusions are that randomness causes the model to move into high volatility, which happens when the percentage of noise traders is high, and the price effect in the model soon returns the model back to low volatility.