Validating and Calibrating Agent-based Models: a Case Study

In this paper we deal with the validation of an agent-based model and, in particular, with the technical validation process, that is to say all the set of test and methods used to analyze if the results of a simulation agree with reality. Today, thanks to some important studies, validation techniques are more and more complete and reliable: many distributional and goodness-of-fit tests have been developed, while several graphical tools have been studied to give the researcher a quick comprehension of actual and simulated data. In particular, the aim of this paper is to propose a good way to calibrate and validate a simple agent-based model of industrial dynamics we have developed. To achieve our goal we consider actual micro-level data of a sample of Italian manufacturing firms included in the Centrale dei Bilanci's database for the period 1983-2001, with no missing data and reliable values. The sample has been selected on the basis of appropriate requisites we discuss further in this paper. The validation results (both graphical and analytical) are quite promising. As calibration process, we use the method of indirect inference due to Gourieroux and Monfort(1996) to guarantee more accurate parameters, minimizing the differences between simulated and actual data. Even in this case the results we get are promising