Asymptotically Unbiased Inference for a Dynamic Panel Model with Fixed Effects when Both N and T are Large

Abstract

We consider a dynamic panel AR(1) model with fixed effects when both n and T are large. Under the "T fixed n large" asymptotic approximation, the maximum likelihood estimator is known to be inconsistent due to the well-known incidental peirameter problem. We consider an alternative asymptotic approximation where n and T grow at the same rate. It is shown that, although the MLE is asymptotically biased, a relatively simple fix to the MLE results in an asymptotically unbiased estimator. The bias corrected MLE is shown to be asymptotically efficient by a Hajek type convolution theorem.

Markups and fiscal transmission in a panel of OECD countries

“…Lütkepohl (2006, p. 15 and p. 194) and also Hahn andKuersteiner (2002, p. 1640). We therefore use an extended version of the Hahn and Kuersteiner (2002) estimator that allows for linear constraints.…”

“…Lütkepohl (2006, p. 15 and p. 194) and also Hahn andKuersteiner (2002, p. 1640). We therefore use an extended version of the Hahn and Kuersteiner (2002) estimator that allows for linear constraints. To control for time e¤ects, we use a projection matrix to average the observations over individuals and then use the transformed data in the estimations (which is equivalent to including the matrix of time dummies as regressors).…”

“…To make the Hahn and Kuersteiner (2002) estimator suitable for models with higher order VAR dynamics (we use two lags), one can use the fact that any VAR(p) process can be written in VAR(1) form by imposing blockwise zero and identity restrictions on the VAR slope coe¢ cients, see e.g. Lütkepohl (2006, p. 15 and p. 194) and also Hahn andKuersteiner (2002, p. 1640).…”

Improved Inference in the Evaluation of Mutual Fund Performance Using Panel Bootstrap Methods

“…It is static so does not face the problems of a dynamic panel with lagged dependent variables. 9 It also has a large number of funds (N = 561) and a large number of time series observations (T = 129 months), so does not suffer from the fixed effects estimator being asymptotically biased which would result in the computed confidence intervals being unreliable because the coverage rate is below the nominal level (Neyman and Scott, 1948;Nickell, 1981;Beran, 1987Beran, , 1990Martin, 1990;Hahn and Kuersteiner, 2002). …”

Estimation of random coefficients logit demand models with interactive fixed effects

“…Following some recent panel data literature, e.g. Hahn and Kuersteiner (2002; and Hahn and Newey (2004), we handle this problem by considering asymptotics where both J and T become large. Under this alternative asymptotic, the incidental parameter problem is transformed into the issue of asymptotic bias in the limiting distribution of the estimators of the parameters of interest.…”