
on Econometrics 
By:  Alexander Chudik; M. Hashem Pesaran 
Abstract:  This paper contributes to the GMM literature by introducing the idea of selfinstrumenting target variables instead of searching for instruments that are uncorrelated with the errors, in cases where the correlation between the target variables and the errors can be derived. The advantage of the proposed approach lies in the fact that, by construction, the instruments have maximum correlation with the target variables and the problem of weak instrument is thus avoided. The proposed approach can be applied to estimation of a variety of models such as spatial and dynamic panel data models. In this paper we focus on the latter and consider both univariate and multivariate panel data models with short time dimension. Simple Biascorrected Methods of Moments (BMM) estimators are proposed and shown to be consistent and asymptotically normal, under very general conditions on the initialization of the processes, individualspeci.c e¤ects, and error variances allowing for heteroscedasticity over time as well as crosssectionally. Monte Carlo evidence document BMM.s good small sample performance across di¤erent experimental designs and sample sizes, including in the case of experiments where the system GMM estimators are inconsistent. We also .nd that the proposed estimator does not su¤er size distortions and has satisfactory power performance as compared to other estimators. 
Keywords:  shortt dynamic panels, GMM, weak instrument problem, quadratic moment conditions, panel VARs, Monte Carlo evidence 
JEL:  C12 C13 C23 
Date:  2017 
URL:  http://d.repec.org/n?u=RePEc:ces:ceswps:_6688&r=ecm 
By:  Taras Bodnar; Solomiia Dmytriv; Nestor Parolya; Wolfgang Schmid 
Abstract:  In this paper we construct two tests for the weights of the global minimum variance portfolio (GMVP) in a highdimensional setting, namely when the number of assets $p$ depends on the sample size $n$ such that $\frac{p}{n}\to c \in (0,1)$ as $n$ tends to infinity. The considered tests are based on the sample estimator and on the shrinkage estimator of the GMVP weights. The asymptotic distributions of both test statistics under the null and alternative hypotheses are derived. Moreover, we provide a simulation study where the power functions of the proposed tests are compared with other existing approaches. A good performance of the test based on the shrinkage estimator is observed even for values of $c$ close to 1. 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1710.09587&r=ecm 
By:  Susan Athey; Guido W. Imbens; Stefan Wager 
Abstract:  There are many settings where researchers are interested in estimating average treatment effects and are willing to rely on the unconfoundedness assumption, which requires that the treatment assignment be as good as random conditional on pretreatment variables. The unconfoundedness assumption is often more plausible if a large number of pretreatment variables are included in the analysis, but this can worsen the performance of standard approaches to treatment effect estimation. In this paper, we develop a method for debiasing penalized regression adjustments to allow sparse regression methods like the lasso to be used for sqrt{n}consistent inference of average treatment effects in highdimensional linear models. Given linearity, we do not need to assume that the treatment propensities are estimable, or that the average treatment effect is a sparse contrast of the outcome model parameters. Rather, in addition standard assumptions used to make lasso regression on the outcome model consistent under 1norm error, we only require overlap, i.e., that the propensity score be uniformly bounded away from 0 and 1. Procedurally, our method combines balancing weights with a regularized regression adjustment. 
Date:  2016–04 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1604.07125&r=ecm 
By:  Susan Athey; Julie Tibshirani; Stefan Wager 
Abstract:  We propose generalized random forests, a method for nonparametric statistical estimation based on random forests (Breiman, 2001) that can be used to fit any quantity of interest identified as the solution to a set of local moment equations. Following the literature on local maximum likelihood estimation, our method operates at a particular point in covariate space by considering a weighted set of nearby training examples; however, instead of using classical kernel weighting functions that are prone to a strong curse of dimensionality, we use an adaptive weighting function derived from a forest designed to express heterogeneity in the specified quantity of interest. We propose a flexible, computationally efficient algorithm for growing generalized random forests, develop a large sample theory for our method showing that our estimates are consistent and asymptotically Gaussian, and provide an estimator for their asymptotic variance that enables valid confidence intervals. We use our approach to develop new methods for three statistical tasks: nonparametric quantile regression, conditional average partial effect estimation, and heterogeneous treatment effect estimation via instrumental variables. A software implementation, grf for R and C++, is available from CRAN. 
Date:  2016–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1610.01271&r=ecm 
By:  Fabian Dunker; Konstantin Eckle; Katharina Proksch; Johannes SchmidtHieber 
Abstract:  The random coefficients model is an extension of the linear regression model which allows for additional heterogeneity in the population by modeling the regression coefficients as random variables. Given data from this model, the statistical challenge is to recover information about the joint density of the random coefficients which is a multivariate and illposed problem. Because of the curse of dimensionality and the illposedness, pointwise nonparametric estimation of the joint density is difficult and suffers from slow convergence rates. Larger features, such as an increase of the density along some direction or a wellaccentuated mode can, however, be much easier detected from data by means of statistical tests. In this article, we follow this strategy and construct tests and confidence statements for qualitative features of the joint density, such as increases, decreases and modes. We propose a multiple testing approach based on aggregating single tests which are designed to extract shape information on fixed scales and directions. Using recent tools for Gaussian approximations of multivariate empirical processes, we derive expressions for the critical value. We apply our method to simulated and real data. 
Date:  2017–04 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1704.01066&r=ecm 
By:  Badi H. Baltagi (Center for Policy Research, Maxwell School, Syracuse University, 426 Eggers Hall, Syracuse, NY 13244); Georges Bresson (Université Paris II, Sorbonne Universités); Anoop Chaturvedi (University of Allahabad); Guy Lacroix (Départment d’économique, Université Laval) 
Abstract:  The paper develops a general Bayesian framework for robust linear static panel data models using εcontamination. A twostep approach is employed to derive the conditional typeII maximum likelihood (MLII) posterior distribution of the coefficients and individual effects. The MLII posterior means are weighted averages of the Bayes estimator under a base prior and the datadependent empirical Bayes estimator. Twostage and three stage hierarchy estimators are developed and their finite sample performance is investigated through a series of Monte Carlo experiments. These include standard random effects as well as Mundlaktype, Chamberlaintype and HausmanTaylortype models. The simulation results underscore the relatively good performance of the threestage hierarchy estimator. Within a single theoretical framework, our Bayesian approach encompasses a variety of specifications while conventional methods require separate estimators for each case. 
Keywords:  εContamination, Hyper gPriors, TypeII Maximum Likelihood Posterior Density, Panel Data, Robust Bayesian Estimator, ThreeStage Hierarchy 
JEL:  C11 C23 C26 
Date:  2017–09 
URL:  http://d.repec.org/n?u=RePEc:max:cprwps:208&r=ecm 
By:  Susan Athey; Guido Imbens 
Abstract:  In this review, we present econometric and statistical methods for analyzing randomized experiments. For basic experiments we stress randomizationbased inference as opposed to samplingbased inference. In randomizationbased inference, uncertainty in estimates arises naturally from the random assignment of the treatments, rather than from hypothesized sampling from a large population. We show how this perspective relates to regression analyses for randomized experiments. We discuss the analyses of stratified, paired, and clustered randomized experiments, and we stress the general efficiency gains from stratification. We also discuss complications in randomized experiments such as noncompliance. In the presence of noncompliance we contrast intentiontotreat analyses with instrumental variables analyses allowing for general treatment effect heterogeneity. We consider in detail estimation and inference for heterogeneous treatment effects in settings with (possibly many) covariates. These methods allow researchers to explore heterogeneity by identifying subpopulations with different treatment effects while maintaining the ability to construct valid confidence intervals. We also discuss optimal assignment to treatment based on covariates in such settings. Finally, we discuss estimation and inference in experiments in settings with interactions between units, both in general network settings and in settings where the population is partitioned into groups with all interactions contained within these groups. 
Date:  2016–07 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1607.00698&r=ecm 
By:  Matthew A. Masten; Alexandre Poirier 
Abstract:  Conditional independence of treatment assignment from potential outcomes is a commonly used but nonrefutable assumption. We derive identified sets for various treatment effect parameters under nonparametric deviations from this conditional independence assumption. These deviations are defined via a conditional treatment assignment probability, which makes it straightforward to interpret. Our results can be used to assess the robustness of empirical conclusions obtained under the baseline conditional independence assumption. 
Date:  2017–07 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1707.09563&r=ecm 
By:  Antoine A. Djogbenou (Queen's University) 
Abstract:  This paper proposes two consistent model selection procedures for factoraugmented regressions in finite samples. We first demonstrate that the usual crossvalidation is inconsistent, but that a generalization, leavedout crossvalidation, selects the smallest basis for the space spanned by the true factors. The second proposed criterion is a generalization of the bootstrap approximation of the squared error of prediction of Shao (1996) to factoraugmented regressions. We show that this procedure is consistent. Simulation evidence documents improvements in the probability of selecting the smallest set of estimated factors than the usually available methods. An illustrative empirical application that analyzes the relationship between expected stock returns and factors extracted from a large panel of United States macroeconomic and financial data is conducted. Our new procedures select factors that correlate heavily with interest rate spreads and with the FamaFrench factors. These factors have strong predictive power for excess returns. 
Keywords:  factor model, consistent model selection, crossvalidation, bootstrap, excess returns, macroeconomic and financial factors 
JEL:  C52 C53 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:qed:wpaper:1391&r=ecm 
By:  Conrad, Christian 
Abstract:  We show that the consensus forecast can be biased if some forecasters minimize an asymmetric loss function and the DGP features conditional heteroscedasticity. The timevarying bias depends on the variance of the process. As a consequence, the information from the exante variation of forecasts can be used to improve the predictive accuracy of the combined forecast. Forecast survey data from the Euro area and the U.S. confirm the implications of the theoretical model. 
JEL:  C51 C53 
Date:  2017 
URL:  http://d.repec.org/n?u=RePEc:zbw:vfsc17:168200&r=ecm 
By:  Ben Jann (University of Bern) 
Abstract:  In their paper titled Why Propensity Scores Should Not Be Used for Matching, Gary King and Richard Nielsen suggest that propensityscore matching (PSM) is inferior to other matching procedures such as Mahalanobis matching (King and Nielsen 2016). They argue that PSM approximates complete randomization, whereas other techniques approximate fully blocked randomization, and that fully blocked randomization dominates complete randomization in terms of statistical efficiency. They illustrate their argument using constructed examples, simulations, and applications to real data. Overall, their results suggest that PSM has dramatic deficiencies and should best be discarded. Although the claim about the superior efficiency of a fully blocked design over complete randomization is true (given a specific sample size), the problems King and Nielsen identify apply only under certain conditions. First, the complete randomization argument is valid only with respect to covariates that are not related to the treatment. Second, and more importantly, King and Nielsen's "PSM paradox" occurs only for specific variants of PSM. I will explain why this is the case, and I will show that other variants of PSM compare favorably with blocking procedures such as Mahalanobis matching. I will illustrate my arguments using a new matching software called "kmatch". 
Date:  2017–09–20 
URL:  http://d.repec.org/n?u=RePEc:boc:dsug17:01&r=ecm 
By:  Hajime Seya; Takahiro Yoshida 
Abstract:  This study proposes a simple technique for propensity score matching for multiple treatment levels under the strong unconfoundedness assumption with the help of the Aitchison distance proposed in the field of compositional data analysis (CODA). 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1710.08558&r=ecm 
By:  Fernando Linardi (University of Amsterdam, The Netherlands; Central Bank of Brazil, Brazil); Cees (C.G.H.) Diks (University of Amsterdam, The Netherlands; Tinbergen Institute, The Netherlands); Marco (M.J.) van der Leij (University of Amsterdam, The Netherlands; De Nederlandsche Bank, The Netherlands); Iuri Lazier (Central Bank of Brazil, Brazil) 
Abstract:  Longitudinal network data are increasingly available, allowing researchers to model how networks evolve over time and to make inference on their dependence structure. In this paper, a dynamic latent space approach is used to model directed networks of monthly interbank exposures. In this model, each node has an unobserved temporal trajectory in a lowdimensional Euclidean space. Model parameters and latent banks' positions are estimated within a Bayesian framework. We apply this methodology to analyze two different datasets: the unsecured and the secured (repo) interbank lending networks. We show that the model that incorporates a latent space performs much better than the model in which the probability of a tie depends only on observed characteristics; the latent space model is able to capture some features of the dyadic data such as transitivity that the model without a latent space is not able to. 
Keywords:  network dynamics; latent position model; interbank network; Bayesian inference 
JEL:  C11 D85 G21 
Date:  2017–10–26 
URL:  http://d.repec.org/n?u=RePEc:tin:wpaper:20170101&r=ecm 
By:  Susan Athey; Guido Imbens; Thai Pham; Stefan Wager 
Abstract:  There is a large literature on semiparametric estimation of average treatment effects under unconfounded treatment assignment in settings with a fixed number of covariates. More recently attention has focused on settings with a large number of covariates. In this paper we extend lessons from the earlier literature to this new setting. We propose that in addition to reporting point estimates and standard errors, researchers report results from a number of supplementary analyses to assist in assessing the credibility of their estimates. 
Date:  2017–02 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1702.01250&r=ecm 
By:  Ryan Cumings 
Abstract:  Optimal transportation is used to define a nonparametric density estimator as the solution of a convex optimization problem. The framework allows for density estimation subject to a variety of shape constraints, including $\rho$concavity and Myerson's (1981) regularity condition. The mean integrated squared error for the density estimator of a random variable in $\mathbb{R}^{d}$ achieves an asymptotic rate of convergence of $O_{p}(N^{4/(d+4)}).$ After deriving algorithms for finding the density estimate, the framework is applied to data from the California Department of Transportation to explore whether their choice of awarding construction contracts using a first price auction is cost minimizing. 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1710.09069&r=ecm 
By:  Federico Belotti (University of Rome Tor Vergata); Giuseppe Ilardi (Bank of Italy) 
Abstract:  The classical stochastic frontier panel data models provide no mechanism for disentangling individual timeinvariant unobserved heterogeneity from inefficiency. Greene (2005a, b) proposed the ‘true’ fixedeffects specification, which distinguishes these two latent components while allowing for timevariant inefficiency. However, due to the incidental parameters problem, the maximum likelihood estimator proposed by Greene may lead to biased variance estimates. We propose two alternative estimation procedures that, by relying on a firstdifference data transformation, achieve consistency when n goes to infinity with fixed T. Furthermore, we extend the approach of Chen et al. (2014) by providing a computationally feasible solution for estimating models in which inefficiency can be heteroskedastic and may follow a firstorder autoregressive process. We investigate the finite sample behavior of the proposed estimators through a set of Monte Carlo experiments. Our results show good finite sample properties, especially in small samples. We illustrate the usefulness of the new approach by applying it to the technical efficiency of hospitals. 
Keywords:  stochastic frontiers, fixedeffects, panel data, pairwise differencing 
JEL:  C13 C16 C23 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:bdi:wptemi:td_1147_17&r=ecm 
By:  Susan Athey; Raj Chetty; Guido Imbens; Hyunseung Kang 
Abstract:  Estimating the longterm effects of treatments is of interest in many fields. A common challenge in estimating such treatment effects is that longterm outcomes are unobserved in the time frame needed to make policy decisions. One approach to overcome this missing data problem is to analyze treatments effects on an intermediate outcome, often called a statistical surrogate, if it satisfies the condition that treatment and outcome are independent conditional on the statistical surrogate. The validity of the surrogacy condition is often controversial. Here we exploit that fact that in modern datasets, researchers often observe a large number, possibly hundreds or thousands, of intermediate outcomes, thought to lie on or close to the causal chain between the treatment and the longterm outcome of interest. Even if none of the individual proxies satisfies the statistical surrogacy criterion by itself, using multiple proxies can be useful in causal inference. We focus primarily on a setting with two samples, an experimental sample containing data about the treatment indicator and the surrogates and an observational sample containing information about the surrogates and the primary outcome. We state assumptions under which the average treatment effect be identified and estimated with a highdimensional vector of proxies that collectively satisfy the surrogacy assumption, and derive the bias from violations of the surrogacy assumption, and show that even if the primary outcome is also observed in the experimental sample, there is still information to be gained from using surrogates. 
Date:  2016–03 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1603.09326&r=ecm 
By:  Susan Athey; Stefan Wager 
Abstract:  We consider the problem of using observational data to learn treatment assignment policies that satisfy certain constraints specified by a practitioner, such as budget, fairness, or functional form constraints. This problem has previously been studied in economics, statistics, and computer science, and several regretconsistent methods have been proposed. However, several key analytical components are missing, including a characterization of optimal methods for policy learning, and sharp bounds for minimax regret. In this paper, we derive lower bounds for the minimax regret of policy learning under constraints, and propose a method that attains this bound asymptotically up to a constant factor. Whenever the class of policies under consideration has a bounded VapnikChervonenkis dimension, we show that the problem of minimaxregret policy learning can be asymptotically reduced to first efficiently evaluating how much each candidate policy improves over a randomized baseline, and then maximizing this value estimate. Our analysis relies on uniform generalizations of classical semiparametric efficiency results for average treatment effect estimation, paired with sharp concentration bounds for weighted empirical risk minimization that may be of independent interest. 
Date:  2017–02 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1702.02896&r=ecm 
By:  Alberto Abadie; Susan Athey; Guido W. Imbens; Jeffrey M. Wooldridge 
Abstract:  Consider a researcher estimating the parameters of a regression function based on data for all 50 states in the United States or on data for all visits to a website. What is the interpretation of the estimated parameters and the standard errors? In practice, researchers typically assume that the sample is randomly drawn from a large population of interest and report standard errors that are designed to capture sampling variation. This is common practice, even in applications where it is difficult to articulate what that population of interest is, and how it differs from the sample. In this article, we explore an alternative approach to inference, which is partly designbased. In a designbased setting, the values of some of the regressors can be manipulated, perhaps through a policy intervention. Designbased uncertainty emanates from lack of knowledge about the values that the regression outcome would have taken under alternative interventions. We derive standard errors that account for designbased uncertainty instead of, or in addition to, samplingbased uncertainty. We show that our standard errors in general are smaller than the infinitepopulation samplingbased standard errors and provide conditions under which they coincide. 
Date:  2017–06 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1706.01778&r=ecm 
By:  Fabian Dunker; Stephan Klasen; Tatyana Krivobokova 
Abstract:  Ratio of medians or other suitable quantiles of two distributions is widely used in medical research to compare treatment and control groups or in economics to compare various economic variables when repeated crosssectional data are available. Inspired by the socalled growth incidence curves introduced in poverty research, we argue that the ratio of quantile functions is a more appropriate and informative tool to compare two distributions. We present an estimator for the ratio of quantile functions and develop corresponding simultaneous confidence bands, which allow to assess significance of certain features of the quantile functions ratio. Derived simultaneous confidence bands rely on the asymptotic distribution of the quantile functions ratio and do not require resampling techniques. The performance of the simultaneous confidence bands is demonstrated in simulations. Analysis of the expenditure data from Uganda in years 1999, 2002 and 2005 illustrates the relevance of our approach. 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1710.09009&r=ecm 
By:  VÁZQUEZALCOCER, Alan; GOOS, Peter; SCHOEN, Eric D. 
Abstract:  Definitive screening designs are increasingly used for studying the impact of many quantitative factors on one or more responses in relatively few experimental runs. In practical applications, researchers often require a design for m quantitative factors, construct a definitive screening design for more than m factors and drop the super uous columns. This is done when the number of runs in the standard mfactor definitive screening design is considered too limited or when no standard definitive screening design exists for m factors. In these cases, it is common practice to arbitrarily drop the last column of the larger definitive screening design. In this paper, we show that certain statistical properties of the resulting experimental design depend on which columns are dropped and that other properties are insensitive to the exact columns dropped. We perform a complete search for the best sets of 18 columns to drop from standard definitive screening designs with up to 24 factors. We observed the largest differences in statistical properties when dropping four columns from 8 and 10factor definitive screening designs. In other cases, the differences are moderate or small, or even nonexistent. Our search for optimal columns to drop necessitated a detailed study of the properties of definitive screening designs. This allows us to present some new analytical and numerical results concerning definitive screening designs. 
Keywords:  Conference matrix, Defficiency, Isomorphism, Projection, Secondorder model, Twofactor interaction 
Date:  2017–09 
URL:  http://d.repec.org/n?u=RePEc:ant:wpaper:2017010&r=ecm 
By:  Søren Johansen (University of Copenhagen and CREATES); Morten Ørregaard Nielsen (Queen's University and CREATES) 
Abstract:  We consider the fractional cointegrated vector autoregressive (CVAR) model of Johansen and Nielsen (2012a) and show that the test statistic for the ususal CVAR model is asymptotically chisquared distributed. Because the usual CVAR model lies on the boundary of the parameter space for the fractional CVAR in Johansen and Nielsen (2012a), the analysis requires the study of the fractional CVAR model on a slightly larger parameter space so that the CVAR model lies in the interior. This in turn implies some further analysis of the asymptotic properties of the fractional CVAR model. 
Keywords:  cointegration, fractional integration, likelihood inference, vector autoregressive model 
JEL:  C32 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:qed:wpaper:1394&r=ecm 
By:  Benjamin Wong 
Abstract:  The historical decomposition is standard within the vector autogression (VAR) toolkit. It provides an interpretation of historical fluctuations in the modelled time series through the lens of the identified structural shocks. The proliferation of nonlinear VAR models naturally leads to extending the historical decomposition into nonlinear settings. This article discusses how to calculate an exact historical decomposition for a large class of popular nonlinear VAR models. In particular, the standard historical decomposition one obtains from a linear VAR is nested within the nonlinear case. The approach discussed in this article is sufficiently general to be relevant for many popular variants of nonlinear VAR models. 
Keywords:  Historical Decomposition, Innovation Accounting, Nonlinear VAR models 
JEL:  C32 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:een:camaaa:201762&r=ecm 
By:  Knüppel, Malte; Krüger, Fabian 
Abstract:  In many empirical applications, a combined density forecast is constructed using the linear pool which aggregates several individual density forecasts. We analyze the linear pool in a mean/variance prediction space setup. Our theoretical results indicate that a wellknown 'disagreement' term can be detrimental to the linear pool's assessment of forecast uncertainty. We demonstrate this argument in macroeconomic and financial forecasting case studies. 
JEL:  C53 
Date:  2017 
URL:  http://d.repec.org/n?u=RePEc:zbw:vfsc17:168294&r=ecm 
By:  Antoine A. Djogbenou (Queen's University) 
Abstract:  This paper proposes a statistical test for the decoupling of the comovements in the real activity of developed and emerging economies. Although globalization has shaped the world economy in recent decades, emerging economies such as China have experienced impressive growth compared to developed economies, suggesting a decoupling between developed and emerging business cycles. Using economy activity variables measured by the log differences of the gross domestic product and the industrial production of developed and emerging countries, we investigate whether the latter assertion can be supported by observed data. Based on a twolevel factor model, we assume these activity variables can be decomposed into a global component, emerging or developed common component and idiosyncratic national shocks. Furthermore, we propose a statistic that tests the null hypothesis of a onelevel specification, where it is irrelevant to distinguish between emerging and developed latent factors against the twolevel alternative. This paper provides a theoretical justification and simulation evidence that documents the testing procedure. An application of the test to our panel of developed and emerging countries leads to strong statistical evidence against the null hypothesis of coupling developed and emerging activity factors. Using a sequential principal component method, we identity global, developed and emerging economy activity factors from a twolevel factor model. We find that these factors are able to track major economic events since 1996. 
Keywords:  latent factor, decoupling, emerging and developed countries, global economy activity, test statistic 
JEL:  C12 F44 O47 
Date:  2017–01 
URL:  http://d.repec.org/n?u=RePEc:qed:wpaper:1392&r=ecm 
By:  Susan Athey; Guido Imbens 
Abstract:  In this paper we discuss recent developments in econometrics that we view as important for empirical researchers working on policy evaluation questions. We focus on three main areas, where in each case we highlight recommendations for applied work. First, we discuss new research on identification strategies in program evaluation, with particular focus on synthetic control methods, regression discontinuity, external validity, and the causal interpretation of regression methods. Second, we discuss various forms of supplementary analyses to make the identification strategies more credible. These include placebo analyses as well as sensitivity and robustness analyses. Third, we discuss recent advances in machine learning methods for causal effects. These advances include methods to adjust for differences between treated and control units in highdimensional settings, and methods for identifying and estimating heterogeneous treatment effects. 
Date:  2016–07 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1607.00699&r=ecm 
By:  Alberto Abadie; Susan Athey; Guido Imbens; Jeffrey Wooldridge 
Abstract:  In empirical work in economics it is common to report standard errors that account for clustering of units. Typically, the motivation given for the clustering adjustments is that unobserved components in outcomes for units within clusters are correlated. However, because correlation may occur across more than one dimension, this motivation makes it difficult to justify why researchers use clustering in some dimensions, such as geographic, but not others, such as age cohorts or gender. It also makes it difficult to explain why one should not cluster with data from a randomized experiment. In this paper, we argue that clustering is in essence a design problem, either a sampling design or an experimental design issue. It is a sampling design issue if sampling follows a two stage process where in the first stage, a subset of clusters were sampled randomly from a population of clusters, while in the second stage, units were sampled randomly from the sampled clusters. In this case the clustering adjustment is justified by the fact that there are clusters in the population that we do not see in the sample. Clustering is an experimental design issue if the assignment is correlated within the clusters. We take the view that this second perspective best fits the typical setting in economics where clustering adjustments are used. This perspective allows us to shed new light on three questions: (i) when should one adjust the standard errors for clustering, (ii) when is the conventional adjustment for clustering appropriate, and (iii) when does the conventional adjustment of the standard errors matter. 
Date:  2017–10 
URL:  http://d.repec.org/n?u=RePEc:arx:papers:1710.02926&r=ecm 