Shuowen Chen

The R Journal, 12(1): 131–146, 2020. [Package]

Abstract: Chernozhukov, Fernandez-Val and Luo (2018) proposed the sorted effect method for nonlinear regression models. This method consists of reporting percentiles of the partial effects in addition to the average commonly used to summarize the heterogeneity in the partial effects. They also proposed to use the sorted effects to carry out classification analysis where the observational units are classified as most and least affected if their causal effects are above or below some tail sorted effects. The R package SortedEffects implements the estimation and inference methods therein and provides tools to visualize the results. This vignette serves as an introduction to the package and displays basic functionality of the functions within.

American Economic Association Papers and Proceedings, 109: 77 - 82, 2019

Abstract: Fixed effects estimation for dynamic linear panel models suffers from incidental parameters problem. Arellano-Bond estimator is subject to the many instrument problem. This note shows that sample splitting alleviates both problems.

Abstract: Fixed effect estimators of nonlinear panel models suffer from the incidental parameter problem. This leads to two undesirable consequences in applied research: (1) point estimates are subject to large biases, and (2) confidence intervals have incorrect coverages. This paper proposes a simulation-based method for bias reduction. The method simulates data using the model with estimated individual effects, and finds value of parameters by equating fixed effect estimates obtained from observed and simulated data. The asymptotic framework provides consistency, bias correction, and asymptotic normality results. An application and simulations to female labor force participation illustrate the finite-sample performance of the method.

Abstract: We study a class of discrete outcome models that can permit multiple outcome values. Whether a model makes such an incomplete predict often depends on its policy-relevant features and can be examined by testing restrictions on the underlying structural parameters. We provide a new test of model incompleteness using a score-based statistic. Our test statistic is computationally tractable even with a moderate number of nuisance parameters because they only need to be estimated in the restricted model that permits a unique prediction. A Monte Carlo experiment shows the proposed test outperforms existing ones in terms of local power. An empirical application to a model of market entry in the airline industry illustrates the computational feasibility of the new method.

Abstract: Why is the U.S. industry-level productivity dispersion countercyclical? Theoretically, we build a duopoly model in which heterogeneous R&D costs determine firms' optimal behaviors and the equilibrium technology gap after a negative profit shock. Quantitatively, we calibrate a paramterized model, simulate firms' post-shock responses and predict that productivity dispersion is due to the low-cost firm increasing R&D efforts and the high-cost firm doing the opposite. Empirically, we construct an index of negative profit shocks and provide two reduced-form tests for this mechanism.

Abstract: Fixed effects estimators suffer from the incidental parameter problem in dynamic or nonlinear panel models with unobserved effects. Hahn and Newey (2004) and Dhaene and Jochmans (2015) proposed convenient jackknife bias corrections, which require that all the variables in the panel be stationary over time. Many covariates of interest in panel and difference-in-differences applications such as policy indicators, age or cohort are not stationary over time. We propose a jackknife bias correction for fixed effects estimators that does not rely on stationarity. We name the new correction as crossover jackknife as it is based on partitioning the panel in two halves, each including half of the time series observations for each cross sectional unit. Numerical examples show that crossover jackknife improves over the existing jackknife corrections, which are not even applicable under some common forms of non-stationarity such as a policy intervention that starts in the middle of the time dimension for some of the cross sectional units.

Abstract: Incorporating unobserved heterogeneity into dynamic discrete choice models (DDCM) is an active area of research. The popular approach uses random effects that treats unobserved heterogeneity as an unobserved state variable with a discrete distribution. The fixed effects approach does not impose distributional assumptions on the unobserved heterogeneity or relationship with the other explanatory variables, but estimation is challenging because (1) DDCM yield criterion functions that lead to the incidental parameters problem; (2) individual effects enter both the current payoff and the continuation value. This paper considers identification, estimation and inference of DDCM with time--invariant individual fixed effects.

Abstract: Discretization is used in applied research for various purposes. A common feature is that some components of the model that enter the criterion function for estimation are approximated by discretization. The validity of estimation methods involving discretization is established by asymptotic theory in which the number of grids increase with sample size. Many papers also recommend how to specify the grids and illustrate through Monte Carlo simulations. These stylized simulations, however, may be of limited applicability to practitioners' actual research questions. Since practitioners often do not know in advance how fine a discretization suffices a satistfactory result, they face a trade-off between accuracy of estimation and ease of computation. This paper aims to provide a method that allows practitioners to visualize the impact of a marginal change in the number of grids on the bias and variance properties of the resulting estimators.