likely to be fired for poor performance”.
By showing that younger players react more strongly to their countries’ Euro Cup
qualifications, we provide additional evidence consistent with career concerns. Impor-
tantly, the situation we consider allows us to compare not only young and old players, but
also players of the same age in the treatment and the control group. This adds confidence
to the conclusion that findings are due to age, and not driven by an unobserved factor
correlated with age (such as systematic differences in educational backgrounds).
The next section develops a theory of nomination contests and derives comparative
statics results that will provide the background for our empirical tests. Section 3 describes
the data, our choice of output measures, and the institutional context. Section 4 explains
and discusses our empirical strategy. Section 5 contains the empirical results, and section
6 concludes.
2 Theory
This section develops a simple model of nomination contests. Suppose that there are
two agents (for example, soccer players of the same nationality) and that exactly one of
them will be selected for an attractive post at the end of the nomination period. The
nomination decision is taken by a principal (the national team coach) whose objective is
to select the most skillful agent. Hence, unlike in a tournament `a la Lazear and Rosen
(1981), it is not the agents’ relative performances but rather the principal’s beliefs about
the agents’ skills that determine which agent wins the contest. The principal’s prior beliefs
about the agents’ skills will have an important impact on expected winning probabilities
and on incentives.
In modeling each agent’s reputation formation we follow Holmstrom’s (1982) seminal
paper on career concerns. Let ηj denote agent j’s (j ∈ {1, 2}) skill level, which is assumed
to be fixed. At the beginning of the nomination season, the agents and the principal share
the same prior beliefs about each ηj . Specifically, we assume that the prior of ηj follows a
normal distribution with mean mj and precision (equal to the inverse of the variance) hj .
The prior distributions of η1 and η2 are independent. Over time, learning about ηj will
occur through the observation of j ’s performance. For simplicity, we consider learning in
a single time period, called the nomination period. Agent j’s output in the nomination
period is given by
yj = ηj + aj + εj ,