Forecasting Canadian inflation: A semi-structural NKPC approach

We examine whether alternative versions of the New Keynesian Phillips Curve equation contain useful information for forecasting the inflation process. We notably consider semi-structural specifications which combine, for closed- and open-economy versions of the model, the structural New Keynesian equation with time series features. Estimation and inference are conducted using identification-robust methods to address the concern that NKPC models are generally weakly identified. Applications using Canadian data show that all the considered versions of the NKPC have a forecasting performance that comfortably exceeds that of a random walk equation, and moreover, that some NKPC versions also significantly outperform forecasts from conventional time series models. We conclude that relying on single-equation structural models such as the NKPC is a viable option for policymakers for the purposes of both forecasting and being able to explain to the public structural factors underlying those forecasts.