Zusammenfassung:
Development of microsimulation models often requires reweighting some input dataset to reflect the characteristics of a different population of interest. In this paper we explore a machine learning approach whereas a variant of decision trees (Gradient Boosted Machine) is used to replicate the joint distribution of target variables observed in a large commercially available but slightly biased dataset, with an additional raking step to remove the bias and ensure consistency of relevant marginal distributions with official statistics. The method is applied to build a regional variant of UKMOD, an open-source static tax-benefit model for the UK belonging to the EUROMOD family, with an application to the Greater Essex region in the UK.