Central Banks’ inflation gamble: Good luck vs. good models.

Principal investigator and first supervisor, with G. Koop and S. McIntyre; PhD student: Petra Gresó
Research Excellence Award (REA), Strathclyde Research Studentship Scheme (SRSS), University of Strathclyde, internally allocated studentship worth GBP 134,525.00, October 2025 - September 2028.

Summary. The surge in inflation after the Covid-19 pandemic and the Russia-Ukraine war, and its subsequent swift return towards central banks' 2% targets, cannot be adequately captured by conventional linear models. Whether bringing inflation back to safe territory reflects good policy or simply good luck is an open question. This project develops and applies machine learning techniques for modelling and forecasting inflation using granular big data: inflation subcomponents are modelled directly rather than only the aggregate, and many macroeconomic covariates are used. The methods build on structural time series frameworks (trend-cycle decompositions and Phillips curve regressions) and embed random forests, Gaussian processes, and neural networks, so that the models remain economically interpretable. The research comprises three strands: (1) nonparametric trend-cycle decompositions of disaggregated inflation series; (2) identifying common inflation drivers across regions and sectors, including the role of international supply chains; and (3) nonlinear Phillips curve modelling of how supply and demand shocks transmit to different inflation subcomponents. The aim is to provide central banks and governments with new tools to track inflation and understand its drivers at a more disaggregated level, and thereby to contribute to the "good luck vs. good policy" debate.