This paper shows that repeated cross-section data with multiple skill measures
(one continuous and repeated) available each period are sufficient to nonparametrically
identify the evolution of skill returns and cross-sectional skill distributions. With panel
data and the same available measurements, the dynamics of skills can also be identified.
Our identification strategy motivates a multi-step nonparametric estimation strategy.
We further show that if any continuous repeated measurement is shown to be linear in
skills, a much simpler GMM estimator can be used.
Using HRS data on men ages 52+ from 1996-2016, we show that one of the available
(continuous and repeated) skill measures is linear in skills and implement our GMM
estimation approach. Our estimates suggest that the returns to skill were fairly stable
from the mid-1990s to the Great Recession, rising thereafter. We document considerable differences in skills and lifecycle skill profiles over ages 52–70 across cohorts, with
more recent cohorts possessing lower skills in their mid-50s but experiencing much
weaker skill declines with age. We also document skill differences by education and
race, which are stable across ages and explain roughly one-third and one-half, respectively, of the corresponding differences in wages. We observe substantial differences
in skills for men in their mid-50s choosing to retire at different ages, but no clear evidence of sharp declines in skills surrounding retirement ages. Finally, we show that
individual fixed effects account for more than a third of all skill variation at age 60,
with considerable persistence in year-to-year skill innovations.