Robust Standard Error Estimation in Fixed-Effects Panel Models

By G. Kézdi. 2003

403 citations recorded by the source corpus.

The paper focuses on standard error estimation in FE models if there is serial correlation in the error process. Applied researchers have often ignored the problem, probably because major statistical packages do not estimate robust standard errors in FE models. Not surprisingly, this can lead to severe bias in the standard error estimates, both in hypothetical and real-life situations. The paper gives a systematic overview of the different standard error estimators and the assumptions under which they are consistent (in the usual large N, small T asymptotics). One of the possible reasons why the robust estimators are not used often is a fear of their bad finite sample properties. The most important results of the paper, based on an extensive Monte Carlo study, show that those fears are in general unwarranted. I also present evidence that it is the abolute size of the cross-sectional sample that primarily affects the finite-sample behavior, not the relative size compared to the time-series dimension. That indicates good small-sample behavior even when N and T are of similar magnitude. I introduce a simple direct test analogous to that of White (1980) for the restrictive assumptions behind the estimators. Its finite sample properties are fine except for low power in very small samples.

Subfields: Economics and Econometrics

View Paper: https://www.semanticscholar.org/paper/f8f2a83a0c70c499b21cd0f0ec86ee3bc3584dd1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

Citations in the corpus

Good Principals or Good Peers? Parental Valuation of School Characteristics, Tiebout Equilibrium, and the Incentive Effects of Competition among Jurisdictions (2006, background): “I report clustered standard errors that allow for the error structure implied by (5B) (Ga´bor Ke´zdi, 2004).”
Religious Social Norms and Corporate Financial Reporting (2012, positive): “…theory related to clustering standard errors suggests that the standard errors become unbiased as the number of clusters approaches infinity, and standard errors may become unreliable if the number of variables in the model exceed the number of clusters (Cameron and Trivedi, 2005; and Kezdi, 2004).”
t-Statistic Based Correlation and Heterogeneity Robust Inference (2010, methodology): “In contrast, the approach of Kiefer and Vogelsang (2002, 2005) requires Υ(·) and h(·) to be constant. There is substantial empirical evidence for persistent instabilities in the second moment of macroeconomic and financial time series: see, for instance, Bollerslev, Engle, and Nelson (1994), Kim and”
Difference in Difference Meets Generalized Least Squares: Higher Order Properties of Hypotheses Tests (2008, methodology): “However, in many applications of DID, N is unlikely to be large enough in relation to the number of time periods T to permit the first order asymptotic approximation to be sufficiently accurate to provide ac- curate inference.”
Gendered key events in the life course: effects on changes in travel mode choice over time (2014, background): “However, the estimator converges to the true standard error as the number of clusters (not the number of observations) approaches infinity (Kézdi, 2004; Nichols and Schaffer, 2007).”
X-DIFFERENCING AND DYNAMIC PANEL MODEL ESTIMATION (2013, methodology): “In implementing the sequential asymptotic tests, the asymptotic variances are estimated by the generalized heteroskedasticity-robust formula (Arellano, 1987; Kezdi, 2002) −1  b −1  , where  is defined in Theorem 2 and b  is found in (19).”
Shock-Based Causal Inference in Corporate Finance and Accounting Research (2015, methodology): “On two-way clustering on both firm and time, see Kezdi (2004) (simulations suggest that two-way clustering can be appropriate with as few as 10 observations in the shorter dimension); Thompson (2010) (formulas); Cameron et al. (2011) (Stata code cgmreg.ado available on Colin Cameron’s website).”
Teacher Quality in Educational Production. Tracking, Decay, and Student Achievement. NBER Working Paper No. 14442. (2008, positive): “Unfortunately, it is not possible to allow for dependence, as clustered variance-covariance matrices are consistent only if the number of clusters grows with the number of parameters fixed (Kezdi, 2004) and in my application, the number of parameters grows with the number of clusters.”
A comprehensive study of life course, cohort, and period effects on changes in travel mode use (2013, background): “However, the clusterrobust standard error estimator converges to the true standard error as the number of clusters (not the number of observations) approaches infinity (Kézdi, 2004; Nichols and Schaffer, 2007).”