All twenty models of the sample have been (or could have been) calibrated using observed data, using readily available computer programmes and following well-established methods and standards. In particular, maximum-likelihood estimation of the ubiquitous logit model has become routine. Yet, while calibration has become easier, the limits to calibrating a model with data of the past have become visible. Calibration of cross-sectional models, as it is practised today, provides the illusion of precision but does little to establish the credibility of models designed to look into the far future. There has been almost no progress in the methodology to calibrate dynamic or quasi-dynamic models. In the face of this dilemma, the insistence of some modellers on ‘estimating’ every model equation appears almost an obsession. It would probably be more effective to concentrate instead on model validation, i.e. the comparison of model results with observed data over a longer period. In the future, the only real test of a model’s performance should be its ability to forecast the essential dynamics of the modelled system over a past period at least as long as the forecasting period. ‎[19]