Introduction: the problem of complexity

An economy is a complex system, in the sense that it is a system made up of a large number of parts that interact in a non-simple way (see Simon, 1962, p. 468). The Walrasian programme was an answer to this problem of complexity by setting up a manageable interdependent system of a whole economy. A modern version of this programme is the Cowles approach: a combination of the Walrasian method to construct a mathematical system without empirical content and econometrics to put empirical flesh and blood on this system. The Cowles method to treat complexity was to build more and more comprehensive models.

This development was not justified by its results, more comprehensiveness did not lead to better predictions than very simple univariate naive models (e.g. random walks, low-order autoregressive (AR) models, or simple autoregressive moving average (ARMA) models; see Zellner, 1994 for a brief survey). In interpreting these results, Milton Friedman (1951), who was influential in having such forecasting tests performed, suggested that the Cowles programme of building large scale macroeconomic models was probably faulty and needed reformulation. He saw the disappointing test results as evidence of the prematurity of macromodelling of a whole economy, which sent him in another research direction, namely that of partitioning.