Spot On or Way Off? Validating Results of the AVID Microsimulation Model Retrospectively

Abstract

A well-known challenge for developers and users of dynamic microsimulation models is the question of credibility, the so-called black box problem. Several strategies have been proposed to validate models. They typically involve comparing the results of a simulation run to other suitable figures. This paper approaches the issue after a considerable time lag and uses empirical data for a retrospective comparison. The validation is concerned with the German AVID model, or more precisely, with the basic employment histories module of the model. The model uses survey data and process-generated data on app. 12 000 Germans to project future employment histories and future old age incomes. After seven years the original respondents were surveyed again so that it is possible to compare their simulated and empirical employment histories. Several aspects regarding desired properties of the simulation results and the characteristics of employment histories are discussed. They include global indicators like cross-sections on the aggregate level, cumulative values on the individual level, and structural indicators on the individual level. The results of the comparison are generally encouraging. On the aggregate and cumulative level there is a very good resemblance. Substantial deviations are only found on the individual, structural level: The empirical employment histories are much more continuous than the simulated ones. This overestimation of complexity can be explained by the developments in the time period used for the estimation of the model parameters, but it still leads to a general underestimation of social inequality in the simulation results.