Insurance within the firm

the amount of risk faced by workers. Larger firms will therefore find it optimal to forgo
risk-sharing in favor of more powerful incentives. This discussion indicates that size might
be important for insurance provision, but that its effect can only be signed empirically.

The last prediction we consider is based on an extension of the basic model that em-
phasizes the importance of comparative performance to optimize the wage contract (see,
e.g., Holmstrom and Milgrom, 1987). If shocks to performance are correlated across firms,
then the uncertainty about the unobservable component of output growth can be reduced
by looking at the performance of similar firms. An interesting feature of the Italian econ-
omy is the widespread presence of industrial districts. These are groupings of small- and
medium-sized firms specializing in a particular product and located in a circumscribed area,
ordinarily within a few miles one another. Firms within a district should be better able
to extract valuable information from the performance of closely related firms compared to
those outside a district, and therefore rely more on incentives.

To check all the above implications we modify our IV estimation strategy allowing the
sensitivity coefficients b_u and b_v to depend on observable worker and firm characteristics.
Thus we estimate by IV:

Δ^iJt — β ^(Xij't) Δ-jt + βijt                            ⁽²⁹⁾

with β (X_i3't) being linear in the set of individual-firm variables Xij_t (risk aversion, job type,
tenure, firm size, location in an industrial district, historical variability of firm performance,
and a constant term). This amounts to including interactions of such variables with value
added growth shocks, ∆ε_jt, and augmenting the set of instruments by the interactions of
the original instruments with the relevant worker and firm characteristics.³⁵

As far as worker characteristics are concerned, we create dummies for production work-
ers, clerical workers and managers, and use age as an (imperfect) indicator of seniority.
Risk aversion, however, is not available in the CAD data set. To classify individuals by
risk aversion, we use outside information on a measure of relative risk aversion obtained

^3a0ur estimates of b (X⅛_t) are not affected by the relationship between ∆ω⅛∙t and Xi₃1 that may happen
to exist in the cross-section. Including main effects has virtually no effect on the estimates of b (X⅛_t) (results
available on request).

28

<<   26   27   28   29   30   31   32   >>

More intriguing information