Due to data limitations and the absence of testable, model-based predictions, theory and evidence on herd behavior are only loosely connected. This paper contributes towards closing this gap in the herding literature. We use numerical simulations of a herd model to derive new, theory-based predictions for aggregate herding intensity. Using high-frequency, investor-specific trading data we confirm the predicted impact of information risk on herding. In contrast, the increase in buy herding measured for the financial crisis period cannot be explained by the herd model.