over time and across individuals. This implies that serial correlation in wage increases and
promotion is not a prediction supported by the data. These results are in line with other
studies using survey data. 28
Summarizing the results, the dynamics of wages within German firms reflect the importance
of non random selection of workers into the different rungs of the firm’s job ladder. Measured
and unmeasured ability both play an important role in the workers’ assignment into ranks with
unmeasured ability being more important at higher levels of the hierarchical job structure.
There is no evidence of learning effects generating workers mobility across ranks. Neither
is there evidence of serial correlation in wage increases and promotion. These results are not
surprising given the few episodes of mobility of German workers observed in the data. This in
turn may result from the importance of the apprenticeship system in Germany, in which firms
and individuals can learn about the quality of the employment relationship before individuals
finish school and enter the job market, reducing the need to experience job mobility to learn
about individual ability. It may also result from collective bargaining agreements which may
regulate the workers’ career progression.
To assess the robustness of the preceding results, the next Section presents the results of
tests performed to establish the validity of the instruments as well as their predictive power in
explaining the variables instrumented.
4.2 Instruments
In the estimation of the perfect information model with comparative advantage, the variables
used to instrument previous period wage (other than the exogenous right hand side variables
of the wage equation) correspond to the interactions in job rank at t-1 and t. This choice of
instrument is based on the idea that effective ability, as defined in (1)according to the Gibbons
and Waldman model, and therefore wages differ among workers because of innate ability. As
a result, the way to capture differences in innate ability and therefore in wages is through the
observation of the worker’s career path.
When learning is considered, current job rank and previous period wage have to be in-
strumented. Rank affiliation in t - 1 and t - 2 helps predict current affiliation using the same
argument about the informativeness of the worker’s career path to capture differences in innate
ability. As mentioned previously, differences in innate ability help predict wages and, in the
28 Abowd and Card (1989) and Topel and Ward (1992)
22