Key result
Pooling Vector Autoregression and Local Projections reduces impulse response error vs. either alone.
Why the study?
Pooling VAR and Local Projection impulse response functions may address the bias-variance trade-off in IRF estimation, but methods for inference and performance need evaluation.
Pooling VAR and LP impulse response functions may address the bias-variance trade-off in IRF estimation and yield lower root mean squared error at short and medium horizons.
Pooled VAR-LP estimator cuts RMSE in simulations; leaves open real-data gains and bootstrap validity in macro applications.
I show that pooling Vector Autoregression (VAR) and Local Projection (LP) impulse response functions (IRFs) may address the bias-variance trade-off in IRF estimation. I also propose a sequence-of-block bootstrap method, designed to construct confidence intervals for the pooled IRFs. I document three main findings: (i) Monte Carlo exercises suggest that the pooled estimator generally yields lower root mean squared error than both VAR and LP at short and medium horizons. (ii) Using the sequence-of-block bootstrap, inference based on the pooled estimator delivers coverage rates that are slightly lower than those under LP and more accurate than those based on VAR, while producing substantially shorter average interval lengths than LP. (iii) The pooling approach combined with the sequence-of-block bootstrap produces sensible results in an empirical application to monetary policy.
No takes yet. Share an insight, caveat, or question.
Viet Hoang Dinh (2026) studied this question. Pooling Vector Autoregression (VAR) and Local Projection (LP) impulse response functions vs. VAR and LP alone was evaluated on Root mean squared error and coverage rates. Pooling Vector Autoregression and Local Projection impulse response functions yields lower root mean squared error than either method alone at short and medium horizons.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: