A final consideration in the specification of the model involves the choice of a functional
form for ffXxχit)), the ratio of accumulated experience in t compared to t — 1, which appears in
the estimation equation under both the perfect and imperfect information cases. The Mincer
wage equation specifies log wages as a polynomial function of experience, implying that the
level of wages is an exponential function of experience. Since wages here are in levels, it is
reasonable to assume an exponential function of this same polynomial in experience. This
leads to the following functional form for the ratio g(xit) =
f (xit)
f (xit-1 )
12
g(xit) = b0e-b1xit
(14)
This ratio links to the model’s predictions of serial correlation in wage increases and promo-
tions in the following way. According to the wage equation (5), it is the experience accumulation
term f which, interacted with ability θ, drives the results on serial correlation in wage increases
and promotions (low and high ability workers accumulate experience at different rates). In
terms of the ratio (14), an estimated coefficient b1 different from 0 and b0 different from unity
shows evidence of experience accumulation (or a non constant function f) and as a result,
evidence of serial correlation in wage increases and promotions. On the other hand, a constant
function f (corresponding to an estimated ratio of one) implies that individual unobserved
(or unmeasured) ability does not affect the rate of human capital accumulation. This in turn
implies an absence of serial correlation in wage increases and promotions.
In terms of the interpretation of the remaining parameters, the Gibbons and Waldman
model predicts that if comparative advantage based on unmeasured ability matters, the slope
parameters cj will be significantly different from one another. Because unmeasured ability is
likely to be correlated with measured ability, one expects the same result for the βj . One also
expects the magnitude of these parameters to increase from lowest for the lower job level to
highest for the top job level, reflecting the differences in sensitivity of the different job levels
to ability. The constant terms dj should also be significant from one another and, due to the
characterization of the technology, should rank from higher in the lowest job rank to lower in
the highest one.
3 The Data
The data for the analysis come from the German Socio-Economic Panel. The GSOEP is a
representative longitudinal study of private households conducted every year in Germany since
12Assuming f(xit) = eα0+α1xit-α2xi2t and given that xit = xit-1 + 1 then g(xit) = eα1+α2-2α2xit .
12