Events Conference on Foundations and Advances of Machine Learning in Official Statistics, 3rd to 5th April, 2024

Session 1.3 Applied ML 1

Modelling the local housing situation of households based on the multi-sectoral regional microsimulation model (MikroSim)

Sarah Bohnensteffen* 1


In times of increasing computational power and availability of microdata micro-based ex-ante simulation analysis is nowadays used extensively around the world for policy analysis and design in different fields. The development and application of microsimulation models is a statutory task of the Federal Statistical Office (Destatis) in Germany since 2016. Destatis responds to this task as part of the research group “Multi-sectoral Regional Microsimulation Model (MikroSim)”, developing a dynamic microsimulation model based on a synthetic population of Germany. We demonstrate the potential of microsimulation by applying the MikroSim model to analyse housing demand due to population changes. Adequate and affordable housing as a basic need of society is an ongoing political and societal challenge, which affects large parts of the German population. However, the circumstances differ strongly at the local level: while housing scarcity and rising rents are observed in densely populated areas, structurally weak regions are often characterised by high vacancy rates. Using machine learning methods like random forests we estimate housing information for the synthetic German population in MikroSim on a local scale. The model for housing type is based on data from the German Microcensus and its supplementary programme on housing. We show how long-term housing demand can be modelled with microsimulation taking demographic development, household composition, migration and regional mobility patterns into account. Preliminary results include simulations of the housing situation assuming different scenarios such as suburbanisation or rural exodus trends.

*: Speaker

1: Federal Statistical Office - Germany