In recent years there has been a rapid growth in the number of studies that have used the GMM estimator to decompose the earnings covariance structure into its permanent and transitory parts. Using a heterogeneous growth model of earnings, we consider the performance of the estimator in this context. We use Monte Carlo simulations to examine the sensitivity of parameter identification to key features such as panel length, sample size, the degree of persistence of earnings shocks and the specification of the earnings model. We show that long panels allow the identification of the model, even when persistence in transitory shocks is high. Short panels, on the other hand, are insufficient to identify individual parameters of the model even with moderate levels of persistence.