Life Cycle Human Capital Investment: New Evidence on an Old Issue*

Tests of the life cycle human capital investment theory using the implied restrictions on the covariance structure of earnings have not been favorable to the theory. The theory implies that earnings follow individual specific growth paths, which has been modeled as a random coefficient on experience, or “random growth.” The random growth model implies an individual effect in earnings changes. An alternative model for the covariance structure of earnings is the “random walk” model which implies that there are not individual-specific earnings growth paths, and there is not an individual effect in earnings changes. MaCurdy (1982) and Abowd and Card (1989) showed that there is not an individual effect in earnings changes, and therefore the random walk model provides a better fit to the covariance structure. Baker (1997) has revisited this issue and finds regression evidence in favor of random growth, but also reports conflicting statistical tests rejecting random growth. In this paper, we examine the implications of the life cycle human capital investment theory for earnings changes and find that the random growth specification does not capture the decline in the variance of earnings changes that results from the transversality condition necessary to solve for the optimal investment path. We develop and estimate a model for the covariance structure that allows for increased curvature for the wage profile for workers with a higher rate of growth. In earnings changes, the model allows for a declining variance in individual rates over the life cycle. We fit the model using the Panel Study of Income Dynamics and find significant variation among individuals in growth profiles that decline with experience. We also estimate correlations in earnings profiles within families and find that brothers have similar earnings growth paths, suggesting that the variation in earnings changes is systematic and not random. Finally, we estimate the model on regression residuals from replications of the data used by MaCurdy and by Abowd and Card and obtain results similar to those using our data. These results are supportive for the existence of individual specific growth paths for earnings, and therefore provide evidence in favor of the life cycle human capital investment model.