that the amount of insurance offered within the firm, parameterized by b, varies in pre-
dictable ways with workers’ risk aversion, their propensity to shirk and the variance of the
shocks to firm performance. Our aim is to recover an estimate of b and check whether the
factors that theory indicates as relevant to determining the extent of insurance are impor-
tant empirically. Note that in both the frictionless spot model with infinitely elastic labor
supply curve and the implicit contract model of the labor market, b = O: the compensation
of workers who remain with the firm is unaffected by idiosyncratic shocks to the firm’s
performance. The distinction between the models is in employment dynamics; our study,
however focuses only on the wage relationship.3
3 Shocks and insurance: modelling the stochastic structure
of firms’ performance and workers’ earnings
The principal-agent model of the previous section offers a highly stylized characterization
of wage contracts. When taken to the data, however, various adaptations are needed.
First, firm performance can be measured in several ways. The market value of the firm is
perhaps the best measure, but it is only available for listed firms. In this respect, sales, prof-
its and value added are more appealing proxies; here, we elect value added, but we examine
the sensitivity of results using the alternative gauges. Second, the firm-worker relationship
is dynamic. In dynamic agency models there may be an additional source of variability in
output: the agent’s ability or permanent component, assumed to be time-invariant (i.e., a
random effect) in the simple principal agent model. This is utterly restrictive, as ability
may evolve stochastically over time (say, due to learning on both sides of the employment
relationship), which suggests a more appealing random walk specification. We incorporate
this important feature in our empirical model. Third, in its bare form the agency model
3The optimal static model has no place for terminations, and, accordingly, we examine the insurance role
of the firm without considering them. In reality, the risk of losing one’s job is a major source of uncertainty,
implying that firings should play a role in an optimal incentive scheme. Theoretical results on this issue
are still limited, given the analytical challenges that an optimal dynamic incentive model with dismissals
presents. Using computational techniques, Sleet and Yeltekin (2000) show that, even in the presence of
dismissals, wages will be linked to performance and thus used as an incentive device; they show that the
basic predictions of the simple static principle-agent model carry over to this more general setup.