Modelling Learning as Modelling

2 Criteria for a Good Model of Agent Learning

The function of economic models, for example, is often to demonstrate sufficient conditions for the existence of equilibrium. The introduction of learning schemes into such models is often intended to demonstrate sufficient conditions under which learning makes equilibrium models stable. A classic of this genre is Bray and Savin [3], but see also Binmore and Samuelson [2], Marimon, McGrattan and Sargent [13] and Arifovic [1].

We are not concerned in this paper with equilibrium per se though we are concerned with computer-based simulation models. Such models can be used to generate point or interval forecasts or, alternatively, controlled variations on a single model structure to generate "what-if" analyses of particular business or economic policies.

2.1 - Rigour

2.2 - Incrementality of the Learning Process

2.3 - Examinability of models learnt by the agents

2.4 - The existence of different trade-offs in model search

2.5 - The importance of the form of the agent's models

2.6 - Expressiveness of the Internal Language of Representation

2.7 - Practical to simulate

Generated with CERN WebMaker