In what follows, I show that the standard parametric Galton approach to the study of earnings dynamics is not an adequate method for accounting for non-linearities in the transition process. Instead, I propose an alternative flexible semi-parametricbi-normality. Using this new approach, I find that the estimated inter-temporal earnings elasticity is extremely heterogeneous by previous earnings level and that estimates based on the standard approach are mis-specified for most of the range of reported earnings. Persistence is especially high over the middle range of earnings whereas the tails exhibit considerable regression to the mean. Moreover, the majority of these changes in earnings appear to have little to do with permanent individual characteristics, but rather are strongly influenced through transitory factors.

The paper is proceeds as follows: in section 1, I discuss pertinent issues concerning the measurement of earnings persistence, beginning with the Galton-Markov model, and then moving on to a motivation of a semi-parametric adaptation of this approach; section 2 provides background information on the dataset as well a discussion of how I deal with the potential confounding effects arising from measurement error, life-cycle effects, and attrition biases; section 3 then turns to the key empirical results; section 4 presents evidence of the permanent and transitory components associated with non-linear earnings persistence; finally, concluding remarks are offered in section 5.