Targeted Local Projections

Aleksei Nemtyrev, Otilia Boldea · 2026

Read Targeted Local Projections by Aleksei Nemtyrev, Otilia Boldea (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-00248.txt.

Introduction

Nemtyrev and Boldea propose targeted projections, steering local projection fits toward structural vector autoregression versions to cut variance for bias. Local projections carry little bias on enough lags, but variance grows with horizon as shocks build. Structural vector autoregressions carry higher bias on fewer lags but variance shrinks with horizon through exponentiation. This estimator blends the two in a linear mix, chosen for least mean-squared error. Simulations show it cuts projection variance at a distant horizon a lot while keeping near-nominal coverage in small samples with double bootstrapping.

Catalog record

CREATORAleksei Nemtyrev, Otilia Boldea
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62165
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.00248

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Targeted Local Projections Aleksei Nemtyrev, Otilia Boldea Abstract Local projection (LP) and structural vector autoregression (SVAR) are commonly employed to estimate dynamic causal effects of macroeconomic policies at multiple horizons. With enough lags as controls, LP estimators have little bias but their variance can increase with the horizon due to accumulating additional shocks. Because they typically employ fewer lags or suffer from local misspecification, SVAR estimators typically incur higher bias, but their variance decreases with the horizon due to exponentiation. We propose to target the LP estimators towards their SVAR counterparts - constructed with fewer lags than LP at each horizon - to reduce their variance at the cost of incurring some bias. The resulting targeted LP estimator is a linear combination of the LP and SVAR estimators. We propose choosing this linear combination optimally to minimize the mean-squared error of the new estimator. Our simulations show that, under a locally misspecified SVAR model, targeting substantially reduces the LP variance at longer horizons while maintaining near-nominal coverage in small samples when a double bootstrap is employed. Keywords: Targeted Local Projections, Local Projections, Smooth Local Projections, SVAR, Bootstrap, Mean Symmetric Double Bootstrap JEL Codes: C12, C14, C32 1 Introduction Identifying and estimating dynamic causal effects of structural shocks on macroeconomic outcomes is a fundamental goal in macroeconomics. To study these dynamic causal effects, researchers use a variety of tools, including taking to the data theoretical models such as dynamic stochastic general equilibrium models, and other models such as structural vector autoregressions (SVARs) and local projections (LPs). Local projections were popularized by 13 1 1 1 An earlier paper by 5 features a similar idea. and are widely used to estimate the impulse response functions (IRFs) - which coincide with the dynamic causal effects of a structural shock in linear models under appropriate conditioning on lags. A recent overview of the use and inference with LPs is provided in 14 , 12 and 20 . LPs are more robust to misspecification than SVARs, as shown in 21 , and are semi-parametrically efficient under classical assumptions if the number of lags included diverges, as shown in 28 . Therefore, if one is interested in accurate coverage exclusively, either LPs or SVARs with a large number of lags should be employed, as also argued in 20 . In this paper, we are interested in pointwise estimation instead. In that case, it is possible to reduce the variance of the LP estimates at the cost of a small bias. The reason why this may be useful for applied work is that the original LP estimators exhibit variances that increase with the horizon, as shown in 16 , and this is due to the LP regression rather than the true data generating process. The excessive variability of LPs across horizons has prompted researchers to smooth impulse responses across horizons. 23 proposed smoothing the adjacent changes in LP estimators towards each other, after the LP estimators are obtained, and therefore in two steps. 1 also proposed smooth local projection (SLP) estimators that penalize changes across nearby horizons, but the smoothing is done in one step, via a penalized least-squares objective function. However, smoothing nearby horizons does not guarantee an exponential decay similar to that one would expect if the true data generating process was a stationary SVAR of finite or infinite order. In contrast, 7 propose achieving exponential decay by shrinking the LP estimates towards a VAR in a Bayesian framework. To obtain their Bayesian LP (BLP) estimator, they employ a quasi-Bayesian LP methodology where the prior mean of the coefficient is equal to a VAR estimator on a pre-sample, while the prior variance is based on a Minnesota prior, with a shrinkage parameter that maximizes the posterior marginal density of the data. However, due to small samples in macroeconomics, pre-sample VAR estimates are rarely reliable. Additionally, 7 assume that the true data generating process is a finite order SVAR, while in reality, it is likely that the model is misspecified - 21 . To address this issue, 9 propose a similar Bayesian procedure, but the shrinkage parameter is minimizing the prediction error and their prediction error criterion can also be used to choose among LP and VAR estimators. However, their focus is reduced form estimation and on multi-step forecasts rather than structural parameters. For point estimation of target structural parameters, it is unclear that minimizing prediction error would be desirable. In this paper, we introduce a novel frequentist shrinkage estimator - Targeted Local Projections (TLP) . After projecting out controls, this estimator shrinks the scalar LP structural impulse responses of interest at each horizon toward their counterpart IRFs estimated from a SVAR model in the same sample. Since both LP and SVAR estimate the same theoretical IRF - 22 , targeting the lower variance SVAR IRFs reduces the variance of LP estimates, particularly at long horizons, at the cost of introducing some bias. The resulting estimator at each horizon is a linear combination of the LP and the SVAR estimators, with weights depending on the difference in bias and variance across the two estimators. We propose choosing the shrinkage parameter optimally at each horizon, to minimize the asymptotic mean-squared error of the new estimator. Assuming that the SVAR is locally misspecified as in 25 , 11 and 21 , we derive a consistent estimator of this asymptotic mean square error and show that the shrinkage parameter that minimizes it is unique and available in closed-form, which makes it computationally attractive. While, with local misspecification, the implied weights on the LP and SVAR reflect the bias-variance trade-off between the two estimators, with no misspecification, our estimator reduces to SVAR, as desired. Compared to smooth local projections ( 1 ; 23 ) , TLP differs in both its regularization target and its treatment of horizons. SLP penalizes differences across adjacent horizons, shrinking the IRFs toward each other, without a data-driven target. TLP instead shrinks toward the SVAR IRFs, which exhibit the exponential decay consistent with stationary autoregressive dynamics. SLP also applies a single smoothing parameter across all horizons and creates distortions at boundary horizons; TLP allows shrinkage to vary by horizon and does not impose cross-horizon smoothness. Compared to Bayesian local projections ( 7 ) , TLP uses the full sample for both LP and SVAR estimation, whereas BLP relies on a pre-sample VAR as the prior. Additionally, BLP assumes correct VAR specification and our simulations show that under misspecification, it exhibits severe coverage distortions. Finally, TLP is a linear combination of LP and SVAR with transparent, data-driven weights, while BLP is not a linear combination of the two, making the latter harder to interpret. In the case of local misspecification, the SVAR estimates will exhibit asymptotic bias, distorting coverage of TLP. To remove these distortions, we rely on double bootstrap, which was shown to restore coverage in the presence of asymptotic bias by 3 . Therefore, for inference, we propose a Mean Symmetric Double Bootstrap (MSDB) procedure that provides near-nominal coverage for LP, SVAR, SLP, and TLP even when the SVAR is misspecified, as shown in our simulations. The procedure uses a nested bootstrap structure: the first level generates bootstrap samples via moving-block resampling of SVAR residuals,

Read the complete text