Behavioural Biases and Agent-based Asset Price Modelling

Abstract

The price, return and volume series of virtually all traded financial assets share a set of commonly observed statistical characteristics known as the stylized facts of financial data. In the last two decades, a body of literature has developed, attempting to explain these stylized facts as emerging properties from the interaction of a large number of heterogeneous market participants. The
present thesis contributes to the literature on heterogeneous agent-based asset pricing models, that is, the computational study of financial markets as evolving systems of interacting agents.

Taking a prominent agent-based model (Franke and Westerhoff (2012)) as an example, we observe that its price series violates one of the core properties of real financial time series - its non-stationarity. We overcome this problem by extending the original model and drastically reduce the non-stationarity of the price series generated. Next, we estimate the model's parameters and
evaluate the new setting, showing it is able to match a very rich set of stylized facts observed in real financial markets.

Now, a well defined agents-based asset pricing model able to match the widely observed properties of financial time series is valuable for testing the implications of various biases associated with investors' behaviour. In this context, we present two new behavioural asset pricing models. First, we define a setting where agents suffer from the disposition effect and test the implications of this behavioural bias on investors' interactions and price settings. We demonstrate that it has a direct impact on the returns series produced by the model, altering important stylized facts such as its heavy tails and volatility clustering. Moreover, we show that the horizon over which investors compute their wealth has no effect on the dynamics produced by the model.

Second, we present a new behavioural model of asset pricing where the agents are loss averse, and evaluate its implications. On the one hand, measuring how close the simulated time series are to its empirical counterparts, we show that the model with loss aversion better matches and explains the properties of real-world financial data, compared with the base model without the behavioural bias. On the other hand, we assess the impact of different levels of loss aversion not only on the agents' switching mechanism, but also on the properties of the time series generated by the model. We demonstrate how for different levels of the loss aversion parameter, the biased agents tend to be driven out of the market at different points in time. Since even the simplest strategies have been shown to survive the competition in an agent-based setting, we can link our findings with the behavioural finance literature, which states that investors' systematic biases lead to unexpected market behaviour, instabilities and errors.

Finally, we define a further behavioural heterogeneous agent-based asset pricing model with regret and analyse the implications of this behavioural bias on the model's dynamics. We study the coexistence of locally stable attractors of the corresponding nonlinear deterministic system, one of the most common and generic mechanisms for generating important properties observed in real financial markets. By incorporating regret in agents' expectations, we demonstrate that it can destabilise the price series and change a low volatility market regime into a highly volatile one. Consequently, we show that a change in investors' psychology contributes to the emergence of interesting new properties and that regret has the potential to explain key aspects of financial
markets.

More information

Identifiers

ORCID for Enrico Gerding: orcid.org/0000-0001-7200-552X

Catalogue record

Export record