Abstract for: White-box metamodeling: understanding the effect of aggregation on the use of white-box models for policy analysis

Governments, NGOs, and businesses use increasingly detailed mathematical models to support decision-making. However, detailing a model increases the costs of developing, initialising, running, and analysing a model. Given a budget, this implies that detailed models allow for less experimentation than coarse models. Little experimentation is problematic, as modellers need to (a) evaluate the limitations of their model, (b) understand the effect of uncertainty on the policy performance, and (c) cover the entire solution space. Therefore, we research (a) how to aggregate a detailed model to increase the amount of experimentation and (b) the implications of aggregation for the use of the aggregate model (referred to as the metamodel) for policy analysis. We do this with a case study in the asset management domain. We found that, if the detailed model is well-validated, aggregation decreases the precision of predictions. However, if the metamodel is grounded in theory, we found that it can accurately predict the right behavioural patterns for the right reasons. We conclude that a metamodel is useful to support strategic decision-making by serving as a scanning tool that can highlight areas of interest in the uncertainty- and solution space. This can guide further analysis and lower-level decision-making with a more detailed model.