贝叶斯局部投影

Abstract We propose a Bayesian approach to Local Projections (LPs) that optimally addresses the empirical bias-variance trade-off intrinsic in the choice between direct and iterative methods. Bayesian Local Projections (BLPs) regularize LP regressions via informative priors and estimate impulse response functions that capture the properties of the data more accurately than iterative VARs. BLPs preserve the flexibility of LPs while retaining a degree of estimation uncertainty comparable to Bayesian VARs with standard macroeconomic priors. As regularized direct forecasts, BLPs are also a valuable alternative to BVARs for multivariate out-of-sample projections.