Journal of Econometrics最新文献

Attrition is a common and potentially important threat to internal validity in treatment effect studies. We extend the changes-in-changes approach to identify the average treatment effect for respondents and the entire study population in the presence of attrition. Our method, which exploits baseline outcome data, can be applied to randomized experiments as well as quasi-experimental difference-in-difference designs. A formal comparison highlights that while widely used corrections typically impose restrictions on whether or how response depends on treatment, our proposed attrition correction exploits restrictions on the outcome model. We further show that the conditions required for our correction can accommodate a broad class of response models that depend on treatment in an arbitrary way. We illustrate the implementation of the proposed corrections in an application to a large-scale randomized experiment.

We study the wild bootstrap inference for instrumental variable regressions under an alternative asymptotic framework that the number of independent clusters is fixed, the size of each cluster diverges to infinity, and the within cluster dependence is sufficiently weak. We first show that the wild bootstrap Wald test controls size asymptotically up to a small error as long as the parameters of endogenous variables are strongly identified in at least one of the clusters. Second, we establish the conditions for the bootstrap tests to have power against local alternatives. We further develop a wild bootstrap Anderson–Rubin test for the full-vector inference and show that it controls size asymptotically even under weak identification in all clusters. We illustrate their good performance using simulations and provide an empirical application to a well-known dataset about US local labor markets.

Consider a panel data setting where repeated observations on individuals are available. Often it is reasonable to assume that there exist groups of individuals that share similar effects of observed characteristics, but the grouping is typically unknown in advance. We first conduct a local analysis which reveals that the variances of the individual coefficient estimates contain useful information for the estimation of group structure. We then propose a method to estimate unobserved groupings for general panel data models that explicitly accounts for the variance information. Our proposed method remains computationally feasible with a large number of individuals and/or repeated measurements on each individual. The developed ideas can also be applied even when individual-level data are not available and only parameter estimates together with some quantification of estimation uncertainty are given to the researcher. A thorough simulation study demonstrates superior performance of our method than existing methods and we apply the method to two empirical applications.

This paper provides novel tests for comparing out-of-sample predictive ability of two or more competing models that are possibly overlapping. The tests do not require pre-testing, they allow for dynamic misspecification and are valid under different estimation schemes and loss functions. In pairwise model comparisons, the test is constructed by adding a random perturbation to both the numerator and denominator of a standard Diebold–Mariano test statistic. This prevents degeneracy in the presence of overlapping models but becomes asymptotically negligible otherwise. The test is shown to control the Type I error probability asymptotically at the nominal level, uniformly over all null data generating processes. A similar idea is used to develop a superior predictive ability test for the comparison of multiple models against a benchmark. Monte Carlo simulations demonstrate that our tests exhibit very good size control in finite samples reducing both over- and under-rejection relative to its competitors. Finally, an application to forecasting U.S. excess bond returns provides evidence in favour of models using macroeconomic factors.

Randomized experiments balance all covariates on average and are considered the gold standard for estimating treatment effects. Chance imbalances are nonetheless common in realized treatment allocations. To inform readers of the comparability of treatment groups at baseline, contemporary scientific publications often report covariate balance tables with not only covariate means by treatment group but also the associated $p$-values from significance tests of their differences. The practical need to avoid small $p$-values as indicators of poor balance motivates balance check and rerandomization based on these $p$-values from covariate balance tests (ReP) as an attractive tool for improving covariate balance in designing randomized experiments. Despite the intuitiveness of such strategy and its possibly already widespread use in practice, the literature lacks results about its implications on subsequent inference, subjecting many effectively rerandomized experiments to possibly inefficient analyses. To fill this gap, we examine a variety of potentially useful schemes for ReP and quantify their impact on subsequent inference. Specifically, we focus on three estimators of the average treatment effect from the unadjusted, additive, and interacted linear regressions of the outcome on treatment, respectively, and derive their asymptotic sampling properties under ReP. The main findings are threefold. First, the estimator from the interacted regression is asymptotically the most efficient under all ReP schemes examined, and permits convenient regression-assisted inference identical to that under complete randomization. Second, ReP, in contrast to complete randomization, improves the asymptotic efficiency of the estimators from the unadjusted and additive regressions. Standard regression analyses are accordingly still valid but in general overconservative. Third, ReP reduces the asymptotic conditional biases of the three estimators and improves their coherence in terms of mean squared difference. These results establish ReP as a convenient tool for improving covariate balance in designing randomized experiments, and we recommend using the interacted regression for analyzing data from ReP designs.

We develop a flexible Bayesian model for cluster covariance matrices in large dimensions where the number of clusters and the assignment of cross-sectional units to a cluster are a-priori unknown and estimated from the data. In a cluster covariance matrix, the variances and covariances are equal within each diagonal block, while the covariances are equal in each off-diagonal block. This reduces the number of parameters by pooling those parameters together that are in the same cluster. In order to treat the number of clusters and the cluster assignments as unknowns, we build a random partition model which assigns a prior distribution over the space of partitions of the data into clusters. Sampling from the posterior over the space of partitions creates a flexible estimator because it averages across a wide set of cluster covariance matrices. We illustrate our methods on linear factor models and large vector autoregressions.

We revisit the finite-sample behavior of single-variable just-identified instrumental variables (just-ID IV) estimators, arguing that in most microeconometric applications, the usual inference strategies are likely reliable. Three widely-cited applications are used to explain why this is so. We then consider pretesting strategies of the form $t_1>c$, where $t_1$ is the first-stage $t$-statistic, and the first-stage sign is given. Although pervasive in empirical practice, pretesting on the first-stage $F$-statistic exacerbates bias and distorts inference. We show, however, that median bias is both minimized and roughly halved by setting $c=0$, that is by screening on the sign of the estimated first stage. This bias reduction is a free lunch: conventional confidence interval coverage is unchanged by screening on the estimated first-stage sign. To the extent that IV analysts sign-screen already, these results strengthen the case for a sanguine view of the finite-sample behavior of just-ID IV.