Fama-Macbeth Test Under Firm Dynamics

Abstract

We analyze the bias in the standard error caused by the correlated variance-covariance error matrix from using the Fama-McBeth method with panel data. We propose a three step Generalized Fama-McBeth method to model the correlated errors. In the first step we estimate the betas using cross-sectional regressions. Next, the betas are fitted with a a stationary ARMA process. We then test the drift parameter of the ARMA process of the betas. We show that this is equivalent to testing for the beta being equal to zero. We also model a stochastic entry/exit process for firms in a dataset which helps estimate and test the betas with this method for a more general class of error process. This approach is much simpler and easier to implement than the methods commonly used to correct for the OLS standard errors for large data sets.