How Reliable Are Local Projection Estimators of Impulse Responses?

Abstract

We compare the finite-sample performance of impulse response confidence intervals based on local projections (LPs) and vector autoregressive (VAR) models in linear stationary settings. We find that in small samples, the asymptotic LP interval often is less accurate than the bias-adjusted bootstrap VAR interval, notwithstanding its excessive average length. Although the asymptotic LP interval has adequate coverage in sufficiently large samples, its average length still far exceeds that of bias-adjusted bootstrap VAR intervals with comparable accuracy. Bootstrap LP intervals (with or without bias correction) and asymptotic VAR intervals are shorter on average, but they often lack coverage accuracy in finite samples. © 2011 The President and Fellows of Harvard College and the Massachusetts Institute of Technology.

Measuring Heterogeneous Price Effects for Home Acquisition Programs in At‐Risk Regions

“…Most covariates are indicator variables, which result in four supply curves representing each component of the 2×2 covariate-timing matrix. 3 For any estimates of interest that include nonlinear combinations of coefficients (i.e., marginal effects and the probability of accepting acquisition contracts for various hypothetical property price offers), we use block bootstrapping to estimate standard errors (Efron and Tibshirani 1985, 1986, 1993Cameron et al 2008;Hounkannounon 2008;Kilian and Kim 2011).…”

What are the macroeconomic effects of high‐frequency uncertainty shocks?

“…This would lead to an ordering of the variables that is strictly consistent with their publication lag, since forecasters in the Consensus Economics sample fill out the survey in the first 2 weeks of each month, and the data referring to a given month are published around the middle of that month.15 To ensure a fair comparison across models, both MF-VAR and VAR models have one lag of monthly information; and responses are calculated with the traditional estimator of impulse responses in VAR, since it has been found to be more accurate than the local projection approach for calculating impulse responses with VAR models (seeKilian & Kim, 2011). This would lead to an ordering of the variables that is strictly consistent with their publication lag, since forecasters in the Consensus Economics sample fill out the survey in the first 2 weeks of each month, and the data referring to a given month are published around the middle of that month.15 To ensure a fair comparison across models, both MF-VAR and VAR models have one lag of monthly information; and responses are calculated with the traditional estimator of impulse responses in VAR, since it has been found to be more accurate than the local projection approach for calculating impulse responses with VAR models (seeKilian & Kim, 2011).…”

“…The monthly VAR model includes one lag of monthly information, and we calculate VAR responses with the local projections approach to ensure a fair comparison with MIDAS impulse responses.14 The results are robust to placing the forecast revision variable between the weekly uncertainty of the second and third week of the month. This would lead to an ordering of the variables that is strictly consistent with their publication lag, since forecasters in the Consensus Economics sample fill out the survey in the first 2 weeks of each month, and the data referring to a given month are published around the middle of that month.15 To ensure a fair comparison across models, both MF-VAR and VAR models have one lag of monthly information; and responses are calculated with the traditional estimator of impulse responses in VAR, since it has been found to be more accurate than the local projection approach for calculating impulse responses with VAR models (seeKilian & Kim, 2011).…”

Dynamic Vector Mode Regression

“…IRFs are generally accompanied by the corresponding con…dence intervals. Although asymptotic theory can be used to compute these, it is well-known that asymptotic approximations do not work well for the sample sizes of most applications, and therefore simulation-based methods are often preferred (see Kim, 2011, andWatson, 2018, footnote 11). Obtaining con…dence intervals for the MIRF is also challenging because the model de…ned by (13) is likely to be misspeci…ed and the asymptotic theory provided in this paper will not be valid in this context.…”

“…Obtaining con…dence intervals for the MIRF is also challenging because the model de…ned by (13) is likely to be misspeci…ed and the asymptotic theory provided in this paper will not be valid in this context. To avoid this problem, we use the blocks-of-blocks bootstrap described in Kilian and Kim (2011) (see also Kilian and Lütkepohl, 2017, pp. 351-353) and Efron's percentile method to compute the con…dence intervals for the IRFs and MIRFs (Kim and Kilian, 2011, use the bias-corrected percentile method but note that in the case of IRFs based on local projections the bias correction improves performance only slightly).…”