Empirical Methods
Paper Session
Saturday, Jan. 8, 2022 3:45 PM - 5:45 PM (EST)
- Chair: Harold D. Chiang, University of Wisconsin
Broken Instruments
Abstract
Repeated use of the same potentially related instrumental variables by a literature can "collectively invalidate" these instruments. This paper examines two ways in which this can happen. First, when instruments sharing significant sources of variation are used to instrument multiple distinct covariates, it is increasingly likely the exclusion restriction was not satisfied in any individual specification from the outset. Second, when a variable is documented to affect many outcomes that are likely to be highly or even mildly persistent, using lagged values of that variable as an instrument is likely to violate the exclusion condition. This paper produces a dataset of approximately 960 instrumental variables papers from 1995-2019 in highly-ranked economics general interest and field journals. We find six groups of commonly-used instruments whose literatures, taken together, suggest they are likely to fail the strict exogeneity condition: (i) elevation and bodies of water (ii) sibling structure (iii) ethnicity/ethnolinguistic fractionalization (iv) religion (v) weather and (vi) immigrant enclaves. Taken together, these potentially related instruments have been used in 86 “top five" publications and 317 well-ranked field or general interest journals, with 189 total uses cataloged from 2011 onwards. We propose a Hausman-like test for suspect regressions and discuss its asymptotic properties. We then apply it to two IV papers, finding little reason to be concerned about one, and tentative evidence to be concerned about the other.When Should We (Not) Interpret Linear IV Estimands as LATE?
Abstract
In this paper I revisit the interpretation of the linear instrumental variables (IV) estimand as a weighted average of conditional local average treatment effects (LATEs). I focus on a practically relevant situation in which additional covariates are required for identification while the reduced-form and first-stage regressions implicitly restrict the effects of the instrument to be homogeneous, and are thus possibly misspecified. I show that the weights on some conditional LATEs are negative and the IV estimand is no longer interpretable as a causal effect under a weaker version of monotonicity, i.e. when there are compliers but no defiers at some covariate values and defiers but no compliers elsewhere. The problem of negative weights disappears in the overidentified specification of Angrist and Imbens (1995) and in an alternative method, termed "reordered IV," that I also develop. Even if all weights are positive, the IV estimand in the just identified specification is not interpretable as the unconditional LATE parameter unless the groups with different values of the instrument are roughly equal sized. I illustrate my findings in an application to causal effects of college education using the college proximity instrument. The benchmark estimates suggest that college attendance yields earnings gains of about 60 log points, which is well outside the range of estimates in the recent literature. I demonstrate that this result is driven by the existence of defiers and the presence of negative weights. Corrected estimates indicate that attending college causes earnings to be roughly 20% higher.Local Projections versus VARs: Lessons from Thousands of DGPs
Abstract
We conduct a simulation study of Local Projection (LP) and Vector Autoregression (VAR) estimators of structural impulse responses across thousands of data generating processes (DGPs), designed to mimic the properties of the universe of U.S. macroeconomic data. Our analysis considers various structural identification schemes and several variants of LP and VAR estimators, and we pay particular attention to the role of the researcher's loss function. A clear bias-variance trade-off emerges: Because our DGPs are not exactly finite-order VAR models, LPs have lower bias than VAR estimators; however, the variance of LPs is substantially higher than that of VARs at intermediate or long horizons. Unless researchers are overwhelmingly concerned with bias, shrinkage via Bayesian VARs or penalized LPs is attractive.JEL Classifications
- C1 - Econometric and Statistical Methods and Methodology: General