AGENT-BASED MODELLING

Financial markets are characterized by the interaction of many agents (individual investors, pension funds, hedge funds, market makers…), each having heterogenous objectives and very partial information about others’ strategies. In order to effectively reproduce surprising stylized facts observed in financial markets (e.g., fat-tailed return distribution, autocorrelation of returns), agent-based models are a very flexible approach. They allow to model the behaviour of individual agents and to observe the aggregate effects emerging when many such agents interact in non-trivial environments. Furthermore, agent-based models can be used to gain insights in situations where standard methods fail, for example when studying complex economic and social systems characterized by localized (networked) interaction and/or by knightian (or ontological) uncertainty.

On successful completion of the module, a student will be able to:
- recognize situations in which agent-based models are useful to gain insights about social and economic complex systems.
- conceptualize and develop simple agent-based models to study problems in economics, finance and social sciences.
- know the basic theoretical concepts related to the study of large-scale complex networks.
- characterize and explore the behaviour of social complex systems featuring emerging phenomena and self-organization of many interacting entities.

Elements of Mathematics, analysis and linear algebra (typically provided by 3 university-level modules). Knowledge of Matlab and/or programming languages are useful but not necessary.

The module comprises two main parts a theoretical one and an implementation one.
In the theoretical part, we will first introduce several fundamental Agent-Based Models and discuss how they shed light on interesting emerging phenomena in complex economic, financial, and social systems. Indicatively we will introduce the Shelling Segregation Model, the El Farol Problem (Minority game), the Santa Fe artificial stock exchange model, and cooperation models. These models will allow us to discuss topics such as inductive thinking and economic outcomes, emergent phenomena, and agents' self-organization. Furthermore, in many interesting situations in which agent-based models are useful, interactions among agents are constrained on social networks. For this reason, we will dedicate a few lectures to the study of graph topologies and their functional implications.
In the practical lectures, we will instead learn how to develop an agent-based model using Matlab. The first lectures will introduce general topics of programming (basic elements of programming in Matlab and matrix algebra, conditional constructs, iterative constructs, visualizations), and we will then use them to build an agent-based model.
The actual topics that will be discussed might change a bit depending on the students' knowledge and interests.

Original articles, handouts, and notes.

The student will:
- extend a model presented during the module. This include some coding.
- Analyze the results of the simulations in a short presentaton, in written form.
- Discuss the presentation and the theory of the course in an oral exam.

Frontal lectures, model development tutorial sessions (actual format depending on the number of students).
Active participation is fundamental to fully achieve the learning outcomes.

English

written and oral