Introduction

Formal models are an essential input to the policy-making process. Most central banks, like other organisations, have a suite of different models designed for different purposes and to answer different questions. The papers in Den Butter and Morgan (2000) examine the use of empirical models in policy-making. Constructing these models involves blending economic theory, local knowledge and econometric methods. A longstanding issue of dispute has been the right way to blend these three ingredients, with different approaches giving each a different weight. In most countries a large amount of ‘tender loving care’ will have been bestowed on the individual equations of the policy model in order to obtain sensible results and to get the model to work. Typically this involves adding a range of country-specific variables, which makes it difficult to compare results across models and means that what should be the same equation applied to different countries can look very different. This variability can prompt the suspicion that the results are the product of data mining and may not be robust. Such criticism has been directed at models such as Project Link that combine highly heterogeneous country models. When standard specifications are estimated for a large number of different countries, they tend to produce a high degree of parameter heterogeneity. This is illustrated by Pesaran, Shin and R.P. Smith (1999), who estimate an error correction consumption function for OECD countries, and Baltagi and Griffin (1997), who estimate a gasoline demand function for OECD countries.