A decomposition of labor earnings growth: Recovering Gaussianity?