Why behavior?

So far, the tax-benefit micro-simulation model only applies the rules of the tax-benefit system: it calculates all the taxes and benefits for each individual citizen or household in the micro-data. As such, it does not account for any behavioral reactions of citizens to tax-benefit reforms. This is not realistic. Moreover, inducing behavioral changes can be one of the motivations of policy reforms (e.g., carbon taxes, kilometer charges…). To accommodate for such behavioral reactions, various behavioral models can be integrated into the microsimulation platform. This is and will be a perpetual and gradual work of our team over the coming years and decades. The integration of behavioral models in Beamm is conceptually relatively simple: the integration occurs principally at the level of the micro-data. This means that a microsimulation platform such as Beamm is in principle quite flexible in the types of behavioral models with which it can link up: econometric models, general equilibrium models, complex systems, agent-based systems, etc. To deal with behavior, tax-benefit system reforms are translated into changes in the determinants of individual behavior featuring in the behavioral model. The behavioral model returns estimations of the behavioral reactions of different profiles of individuals or households in function of these changed determinants. These behavioral predictions are then brought back to our synthetic micro-data: the model reconsiders for each individual in our synthetic micro-dataset the individual behavior in light of these predictions and changes the information in the micro-data accordingly to create a counterfactual micro-dataset. After this, Beamm applies all the rules of the reformed tax benefit system on the counterfactual micro-dataset and computes for each individual all the transfers and taxes. To make this concrete with an example, consider the case of a congestion charge (a tax per kilometer driven by car to reduce road congestion). In a first step, Beamm translates the congestion charge into a change of the determinant of behavior: the general cost per kilometer of driving a car. A transport model will then estimate the probabilities that different profiles of households will change their car use. A household with certain characteristics currently commuting by car may be predicted to continue to do so with a certain probability, change the timing of their trip with another probability or switch to public transport with still another probability, etc. Beamm changes the information about the commuting behavior of this household accordingly in the micro-data, and then calculates all taxes and benefits, including the new congestion charge for all the households on these counterfactual micro-data. Based on these calculations, Beamm then provides an analysis and visualization of the results, this time taking behavioral reactions into account.

Our behavioral models

Modelling economic behavior is complicated and always partial by necessity - the real world is far too complicated to be captured in models. At present, we focus on 3 behavioral models: a model of labor supply, a consumption model and a transport model. The current labor market model is a RuRo model (for Random Utility - Random Opportunity), an econometric discrete choice model that models how people adapt their labor supply in function of taxes, while taking into account differences in job market opportunities for workers with different profiles and in different geographical locations (see, e.g., Dagsvik & Strøm (1992) or Capéau et al. (2015)). The consumption model tries to capture how households change their consumption pattern and savings in function of changes in their disposable income or in prices. An important quality of such a consumption model is that it must be consistent at the level of a household’s aggregate spending. E.g., if a tax reform causes a 5% decrease in a household’s disposable income, the household will typically reduce its consumption of some goods by more than 5% (e.g., luxuries, entertainment, savings…), while barely reducing its spending on some others (e.g., rent, energy). The consumption model needs to predict the changes in spending following this drop in income (or a price change) while ensuring that the total sum of all these changes in spending equals the change in disposable income. At present, we incorporate a standard consumption model (QUAIDS, Quadratic Almost Ideal Demand System, Banks et al. (1997)), but more flexible approaches will be explored in the future. Finally, for the transport model, we will incorporate representations of externally developed transport models.