Aggregate vs micromodels for forecasting the diffusion of new technologies

A key discussion in technology diffusion modeling centers on choosing a robust modeling framework for forecasting and policy analysis. This paper contrasts two conventional and theoretically-backed empirical modeling paradigms, the micromodels and the (aggregate/semi-aggregate) Bass-type diffusion models in the context of solar panel adoption, a technology marked by strong localized social learning and heterogeneous consumer preferences. Using a German household-level dataset, we evaluate each paradigm's predictive performance under varying conditions of social learning localization and household heterogeneity. Our findings reveal a trade-off: micromodels, with appropriate degree of granularity, excel at more localized predictions but may be less accurate at aggregate or semi-aggregate predictions. Semi-aggregate diffusion models offer a better balance of accuracy and computational efficiency for aggregate and segment-level forecasts, with acceptable performance in localized predictions for larger number of segments. We also find that micromodels are much more sensitive to correct model specification, whereas aggregate/semi-aggregate models are relatively more prone to overfitting. Our findings highlight the importance of considering the granularity of data, the degree of social learning localization, and customer heterogeneity when choosing a forecasting framework for technology adoption. Aligning the right model with policy objectives and data availability enables policymakers to develop more effective and targeted interventions.