Individual decision making in economic and financial systems with feedback: Theory and Experiments

Heterogeneous agent models (HAMs) are widely used in economics and finance and have proven to be successful in describing economic phenomena. They represent the analytically tractable counterpart to agent-based modelling. In HAMs, people rely on behavioral prediction rules to make their choices, and they switch between the rules based on past performance of these rules. In financial market models, for example, the interaction between stabilizing (fundamental) and destabilizing (trend-extrapolation) prediction rules leads to instability with excess volatility and bubbles that we observe in actual financial markets. The market feedback makes these patterns self-confirming and sustainable, as they ‘feed’ destabilizing rules, making them perform relatively well.

The HAMs have two major ingredients: the prediction rules that people use and the mechanism of switching between rules. The most common approach to modelling switching between prediction rules is through the so-called discrete choice model. This is an important building block of HAMs, because it introduces a nonlinearity in these models and as such is responsible for complex phenomena these models are able to replicate. Recent experiments found support for both building blocks but they also revealed that these assumptions are not complete.

This talk will discuss and critically evaluate this recent experimental evidence in order to suggest ways the model can be extended to better match the decision-making that people use in practice. In particular, it will be shown that the discrete choice specification may need to be adapted to provide a better description of behavior. People use rules with various levels of complexity in terms of efforts needed to gather necessary information and to come to a prediction. Depending on the market dynamics, the rules may have different expected performances. These aspects are largely ignored in the current HAMs, as the parameters that govern the choice of active prediction rules are exogenous and the choice is based on past performance of the rules. However, these parameters may be affected by the complexity of the rules and by the characteristics of the market environment.

Adapting the benchmark model may turn out to be quite relevant for policymakers. Consider, for example, a volatile financial market that is described well by a particular HAM. On the basis of that model, the financial regulator may want to implement a policy that – based upon numerical simulations – stabilizes market dynamics. However, if traders react to the increased stability and predictability of profits in this market by starting to respond more strongly to profit differences – as suggested by experimental results but not reflected in the HAMs – this may substantially mitigate the effect of the policy.

Programme