Editorial

The Winter 2020 issue of the journal contains four articles, plus a book review. The book reviewed is the new book by Cathal O’Donoghue, past President of the International Microsimulation Association. The book offers a broad overview of microsimulation models for the social sciences, with practical examples taken from his long research experience. The book is mostly targeted to PhD students — covering an important gap in the field — but is of interest to any practitioner.

The first research article, by Sarah Kuypers, Jonas Boone, Johannes Derboven, Francesco Figari and Gerlinde Verbist, describes the work done to extend the EUROMOD tax-benefit microsimulation model (Sutherland and Figari, 2013; Sutherland, 2018) to wealth-related policies. Given that wealth is not recorded in the EU-SILC based EUROMOD standard input data, the extension is done by creating new input data using the second wave (2017) of the Eurosystem Household Finance and Consumption Survey (HFCS). The authors motivate the choice of building new input datasets rather than using HFCS as a donor dataset to the standard EUROMOD input data to fully exploit the oversampling of the wealthier population and the multiple imputation procedure used in HFCS to deal with selective item non-response. The extension builds on previous work by some of the same authors (Kuypers et al., 2016) and currently covers 17 EU countries, with policies updated up to 2017. Unfortunately, the extension is for the time being not available for research.

The second paper, by Maria Luisa Maitino, Letizia Ravagli, and Nicola Sciclone, documents a rich dynamic microsimulation model (IrpetDin) that the authors have developed over many years, estimated on both Italian data and —exploiting local administrative data sources— on data for one Italian region, Tuscany. The model simulates demographic events, from ageing to transition to adulthood, partnership formation and dissolution, fertility, and migration; educational attainment at secondary and tertiary-school level; labour market events such as entries to the labour force and career progression; social security policies including pensions, health and long-term care. Labour supply is endogenously determined in the model, while labour demand is obtained from a separate macro model. The matching of labour supply and demand is modelled by sector of activity and education, in order to estimate the quantitative and the qualitative skill mismatch. Validation is performed on the period 2009-2017, with encouraging results. With these features, IrpetDin stands out as one of the most complete dynamic models in the literature. Unfortunately, the model is not openly available, but the authors can be contacted for access.

The third paper is by Bryan Tysinger, and exploits the Future Adult Model to explore obesity and diabetes dynamics in the US, over a ten-year horizon. He also performs a very welcome validation exercise, first estimating the model on 2007-2017 PSID data, and then comparing projections for 2007-2017 with the observed data, with encouraging results. The Future Adult Model —also called Future Americans Model in the official documentation Tysinger refers to in the paper— is a development of the well-known Future Elderly model. We hope to host a description of the model in this journal in the near future.

The fourth and last paper in this issue is a methodological paper by André Decoster, Bram De Rock, Kris De Swerdt, Jason Loughrey, Cathal O’Donoghue, and Dirk Verwerft. They deal with the problem of imputing consumption data into a dataset that contains information on individual incomes. This is relevant for instance when a tax-benefit model is extended to include indirect taxes, as in O’Donoghue et al. (2018), given that consumption and income data are seldom available in a single dataset. The authors compare five different techniques to impute expenditures into income datasets: parametric estimation of Engel curves, non-parametric estimation, both constrained and unconstrained matching using a distance function and grade correspondence. They then test the five methods on artificial data. They find that the parametric and non-parametric estimation yield the best results, generating imputed values that are closest to the pseudo-true values for the budget shares.