This paper presents the results of a micro simulation model designed to make short-term projections of poverty indicators. The unit of observation in the model is the household. In order to project the evolution of household incomes over time, we have specified a model that links the total observed household income to the evolution of a set of macro income indicators that reflect the different ‘micro’ sources of income.
In this paper a micro simulation model with static ageing is presented. Unlike other models of its kind, it explicitly allows individuals to occupy several labour market states and to combine the associated income sources. This possibility seems more realistic than the usual assumption of mutually exclusive states, since household income was used to measure financial well-being. Given that the socioeconomic ‘state’ of a household is hard to define, we have directly specified the link between individual equivalent income (defined as household income, corrected for family composition, attributed to each household member) and the macroeconomic environment. This was achieved by estimating a fixed effects panel regression model with macroeconomic averages for the main income sources as explanatory variables. The dependent variable was obtained from the Panel Survey on Belgian Households (1994-2002), which contains a question on total disposable current monthly income of the household and its main sources. The respondents were offered a choice from one or more of the following income sources: work; independent activity or farming; pension benefits; unemployment benefits or layoff-premiums; other social benefits; rental income, returns from investments or savings; other income. For the first four categories a corresponding macro indicator could be identified; the evolution of the remaining three was modelled using a linear time trend.
The model produces plausible results both from a statistical point of view and based on in-sample simulation. In order to obtain simulated incomes that match the distributional properties of the observations as closely as possible, we have drawn the random disturbances from the empirical (estimated) error distribution, using a kernel density estimator.
The estimated poverty indicators reflect the financial dimension of poverty and are based on poverty thresholds set as 60% of median equivalent income. An individual or family with an income below the threshold level is considered to be ‘at risk of poverty’. Given data and model restrictions, the estimated poverty measures are to be interpreted as indicative numbers rather than precise estimates. They are internally comparable, but do not necessarily coincide with estimates based on other data sources or models.
The simulation results can be summarised as follows. Over the period 1994-2005, between 11% and 12% of individuals lived at risk of poverty. From 1994 to 2002 the overall poverty risk fluctuated around 11%, edging up to 12% afterwards. Women seem to face a higher risk than men, but the difference between the sexes is not statistically significant. This derives from the fact that two partners (usually of the opposite sex) who live in the same household are assigned the same equivalent household income. In a singles-only sample the difference between the sexes would become significant.
The risk of poverty is significantly lower for individuals younger than 65 than for those aged 65 or more. Although for the former the risk of poverty was relatively stable at around 10% during the past decennium, it seems to increase for the latter from 18% in 1994 to 23% in 2005. However, the difference between these age groups is less notable if the average depth of poverty and inequality among the poor are taken into account.
In comparison with other socio-economic categories, employees run the lowest risk of poverty, hovering around 4%. The poverty risk of independent workers is somewhat higher, but the difference is never significant. As for retirees, the poverty risk edged up from around 15% in 1994 to 18% in 2005, with a peak level of 19% in 2001. Finally, the unemployed are the category with the highest risk of poverty, fluctuating rather erratically around 29%, but without a clear trend over the period under study.