Trend inflation and inflation expectations in high dimensional vector autoregressions

This paper introduces a new class of large vector autoregression (VAR) models with a hybrid trend structure that explicitly incorporates both trend inflation and inflation expectations, proxied by long-term survey forecasts and statistical filters. We develop efficient Bayesian estimation methods leveraging recent advances in matrix precision algorithms, substantially reducing computational costs and enabling large-scale forecasting exercises. Using a dataset of 20 U.S. macroeconomic variables, we show that incorporating trend inflation and survey-based expectations within a high-dimensional VAR framework markedly improves inflation forecast accuracy relative to widely used large-VAR benchmarks.