Data and model cross-validation to improve accuracy of microsimulation results : estimates for the Polish household budget survey

We conduct detailed analysis of the Polish Household Budget Survey data for the years 2006-2011 with the focus on its representativeness from the point of view of microsimulation analysis. We find important discrepancies between the data weighted with baseline grossing-up weights and official statistics from other sources. A number of re-weighting exercises is examined from the point of view of the accuracy of microsimulation results and we show that using a combination of demographic calibration targets with several economic status variables or tax identifiers from the microsimulation model substantially improves the correspondence of model results and administrative data. While demographic re-weighting is neutral from the point of view of income distribution, calibrating the grossing-up weights to adjust for economic status and tax identifiers significantly increases income inequality. We argue that although data reweighting can substantially improve the accuracy of microsimulation it should be used with caution.