Abstract:
|
Microsimulation models are commonly used to predict future developments in several societal areas (health, traffic, finances, demographic transition etc.). The quality of such predictions strongly depends on the quality of the empirical input. In addition to the need of high-quality empirical starting data, the rules for updating this data during the microsimulation can also be derived from empirical data. This brings up two main challenges. Firstly, developments on the individual level have to be estimated accurately. This can be done by dynamic panel models in which state dependency (effects of lagged dependent variables) is separated from time-invariant unobserved heterogeneity (modeled by a random intercept). However, it is difficult to obtain unbiased estimates in those models. Some approaches which try to overcome these problems will be presented here. Secondly, it is challenging to transfer the results from these models into a microsimulation model when the starting data does not coincide with the estimation sample (which is typically the case). Strategies are presented for managing this transfer by exploiting empirical information from starting data.
|