Title: EM estimation of dynamic panels with heteroskedastic random coefficients
Authors: Andrea Nocera - Birkbeck, University of London (United Kingdom) [presenting]
Abstract: How to combine the EM algorithm with the Restricted Maximum Likelihood method to estimate dynamic heterogeneous panels is shown. The EM-REML approach allows us to estimate iteratively both the average effects and the unit-specific coefficients. Compared to existing methods, it leads to an unbiased estimation of the variances of the random coefficients. Second, our approach allows the random coefficients residuals to have heteroskedasticity of unknown functional form and thus can be seen as a generalization of the one-way error component models where both the random effects and the regression disturbances are heteroskedastic. The estimation procedure can also be adapted to allow for cross-section dependence. An interesting feature of the EM algorithm is that it allows us to make inference on the random coefficients population. Monte Carlo simulations reveal that the proposed estimator has good properties even in small samples and therefore, should be regarded as a valid alternative to Bayesian estimation whenever the researcher wishes to make inference on the coefficients distribution while having little knowledge on what a sensible prior might be. Finally, a novel approach to investigate heterogeneity of the sensitivity of sovereign spreads to government debt is presented.