Marcellino: “Here’s how the economy reacts to financial shocks’”

What happens to the US economy when the financial system is hit by a shock? The answer today is neither linear nor symmetrical. Negative shocks - those that tighten credit or increase perceived risk - provoke much stronger macroeconomic reactions than positive ones. This emerges from a study published in Economics Letters (vol. 250, April 2025, article 112260), entitled Machine learning the macroeconomic effects of financial shocks. The authors of the paper are Massimiliano Marcellino (in photo), director of the BAFFI Centre, together with Niko Hauzenberger (University of Strathclyde), Florian Huber (University of Salzburg), Karin Klieber (Oesterreichische Nationalbank). The authors propose an innovative approach that combines structural econometrics with Bayesian Neural Networks (BNN), an artificial intelligence tool that is particularly well-suited for capturing non-linearities in economic data.

‘The real methodological innovation is that we can ‘machine learn’ impulse responses without specifying a particular functional form a priori. This allows us to discover hidden dynamics, such as sign asymmetries, that traditional linear models often fail to capture,' explains Marcellino, director of Bocconi's BAFFI Centre. Click to learn more