Autore
Salmon, Mark

Titolo
Bounded rationality and learning: procedural learning
Periodico
European University Institute of Badia Fiesolana (Fi). Department of Economics - Working papers
Anno: 1994 - Fascicolo: 21 - Pagina iniziale: 1 - Pagina finale: 44

In what follows we consider the question of boundedly rational learning and expectation formation by economic agents. Every-day observation suggests that informational constraints shape an individual’s objectives, their learning activity and ultimately the type of decision rules they adopt. It also suggests that behavioural issues that arise through the interaction of an individual with their economic environment and their perception of that environment might be important when attempting to understand how economic agents actually formulate decision problems. This interaction may induce behaviour which is far from the fully rational model of homo economicus that provides the standard paradigm of economic theory. It also seems important to recognise that a limited knowledge of the environment may affect, not only information sets but also the manner by which people learn and hence behavioural theories of learning would seem to be called for. Herbert Simon, in his outstanding original contributions to the theory of boundedly rational behaviour discussed the distinction between procedural and substantive rationality and a similar distinction could perhaps be usefully drawn between procedural and substantive methods of learning. In this chapter we explore the role of artificial neural networks in providing a conceptually simple, “non-structural”, procedural model of how agents might learn to approximate their true but unknown conditional expectation function and hence form “boundedly rational” expectations. The minimal information required, simply a knowledge of the input and output variables, does not appear to seriously hinder the performance of the approach, either in theory or in practice. Moreover in principle a neural network approximation can evolve in complexity and hence accuracy as knowledge of the environment increases; such adaption to the environment reflects a behavioural aspect of learning which is invariably missing in the standard models of learning assumed in economics. In this chapter we examine the performance of neural network learning, first very briefly in theory and then empirically with two examples drawn from macro-economic policy issues. Our results, while positive regarding the application of neural network learning, lead us to suggest caution against drawing specific conclusions regarding economic behaviour or policy that are dependent on specific and potentially ad hoc assumptions as to how people learn and hence adjust their expectations. The question of how individuals learn in economic environments needs to be considered more deeply and incorporated into a fully integrated theory of boundedly rational economic behaviour.



Testo completo: http://hdl.handle.net/1814/592

Esportazione dati in Refworks (solo per utenti abilitati)