Forecasting inflation rates with multi-level international dependence

We analyze inflation rates in high-income OECD countries employing a multi-level factor structure that is estimated based on canonical correlation analysis (CCA) and sequential least squares (SLS). We show that inflation has a global component, mainly driven by G7 countries, explaining 77% of the variation on average, and a local component that accounts for substantial comovements in a subset of the countries. We demonstrate that this combination of global and local components has outstanding predictive ability, and can improve forecast performance significantly over a global-component-only specification for different policy horizons thus constituting a new benchmark for inflation forecasting.