In summary, as a result of comparative advantage in the assignment of workers to job levels,
the model can explain that wage increases predict promotions while explaining only a fraction
of the difference in average wages across job levels. Individual heterogeneity in human capital
accumulation or the growth in effective ability the model also explains the observed serial
correlation in wage increases and promotion rates. The introduction of learning allows for the
possibility of real wage decreases and that average wage increases are higher upon promotion
than before and after promotion. With some restrictions on the parameters of the model, η'
and (η" — η') are such that the model also predicts an absence of demotions. Thus, the model
can explain the stylized facts highlighted in the literature on wages and intra-firm mobility.
2.2 Econometric Specification
The model of Gibbons and Waldman emphasizes the importance of endogenous choice of job
levels or self-selection of workers into the rungs of the firm’s job ladder as well as endogenous
mobility across job levels both driven by the evolution of an unmeasured ability term. The
purpose of this Section is to present an econometric specification of the wage dynamics implied
by the model of Gibbons and Waldman where these endogeneity problems can be accounted
for and the relative importance of the effects of comparative advantage and learning on the
dynamics of wages can be estimated.
In the general case of comparative advantage and learning the process for wages given in
equation (3) can be written using the expectation of workers’ ability, θiet .
wijt = dj + cj θiet f (xit)
(5)
Employing dummies, Dijt , indicating the rank j of individual i at time t, the equation to
be estimated can be written as:
JJ J
Wijt = ∑ Dijtdj + ∑ DijtXitβj + ∑ DijtCj θetf (xit) + μit (6)
j=1 j=1 j=1
where μit is a measurement error independent of rank assignment, and Xit corresponds to
individual characteristics to control for the measurable part of human capital. Comparative
advantage is characterized by the fact that the coefficients βj and cj vary by rank and learning
is represented by the conditional expectation θiet . In the model with perfect information about
innate ability, θiet is a time invariant term θi , unmeasurable by the econometrician.