🤖 AI Summary
This paper identifies a pervasive Nickell bias in fixed-effects (FE) estimation within panel local projections (LP), arising from the intrinsic dynamic structure of the panel forecasting setup—even when regressors exclude lagged dependent variables—thereby invalidating standard t-tests. We formally establish, for the first time, that this bias affects all regressors, not merely lagged terms. To address it, we propose a split-panel jackknife estimator that asymptotically eliminates the bias and consistently reconstructs standard errors, thereby restoring valid statistical inference. Applying our method to three canonical macro-finance studies, we find that post-crisis output contractions are, on average, revised upward by 30–50%, with substantially improved statistical significance across specifications. Our approach provides a more robust methodological foundation for assessing cross-country financial spillovers and macroeconomic impacts of financial shocks.
📝 Abstract
Local Projection is widely used for impulse response estimation, with the Fixed Effect (FE) estimator being the default for panel data. This paper highlights the presence of Nickell bias for all regressors in the FE estimator, even if lagged dependent variables are absent in the regression. This bias is the consequence of the inherent panel predictive specification. We recommend using the split-panel jackknife estimator to eliminate the asymptotic bias and restore the standard statistical inference. Revisiting three macro-finance studies on the linkage between financial crises and economic contraction, we find that the FE estimator substantially underestimates the post-crisis economic losses.
