Why?

The micro-dataset created at step 1 necessarily represents a moment in the past: the moment at which the data were collected. However, we typically wish to evaluate the impact of policy reforms today or in the near future. To make the data a better representation of the world today or the expected world in the near future, we nowcast or forecast this dataset in function of the economic and demographic projections published by the Federal Planning Bureau. We distinguish two aspects in the currently implemented nowcasting/forecasting of our microdata: uprating the different incomes and prices in function of the economic projections, and adapting the composition of the population in function of the demographic projections.

How?

For updating the different income components and prices, we use the projections of the Federal Planning Bureau’s HERMREG model as well as the projections of the consumption price and health price indices of the Federal Planning Bureau. This information is used to uprate the different household income components as well as consumption prices in our data.

At the same time, the economic evolution of the labor market as well as the demographic evolution of our population will affect the employment rate, household composition and age profile. To integrate these projections, the literature distinguishes between static and dynamic ageing methods. Static methods keep the characteristics of the individual citizens or households fixed but adapt the sample weights in function of the demographic and economic projections. For example, if a population is projected to have more retired households because of an ageing population, then we attribute higher sample weights to these retired individuals in our microdata to reflect this, but we do not change the characteristics of the actual observations in the microdata. Dynamic methods model and predict the evolution of the individual characteristics of each individual in the microdata, and thus have the actual characteristics of all citizens or households evolving over time to match the projected distribution of these characteristics. For example, one models, year after year, for each individual the probability of becoming unemployed, finding a job or retiring, and adapts this information for each individual appropriately in such a way that the population-level characteristics reflect the population-level projections. Within Beamm, we focus on static methods at present. We use the employment and remuneration projections, household composition projections and general demographic projections of the Federal Planning Bureau to reweigh our microdata in function of these projections.

To conclude

Nowcasting and forecasting microdata is difficult. This aspect of Beamm will remain a priority in terms of further development efforts for the coming years, and the use of machine learning techniques will be essential here. For a general introduction to nowcasting and forecasting in microsimulation, see, e.g., “Nowcasting in Microsimulation Models: A Methodological Survey” by O’Donoghue and Loughrey (2014).

NOTE: the now- and forecasting module is not yet implemented in the current online version of Beamm.brussels. We hope to integrate this in the online platform later this year.