significant. This is consistent with the idea that district firms can rely on their neighbors
to improve the precision of their inference and according to the agency model this should
make it easier to motivate workers and thus result in less insurance, other things being
equal. Size also matters. The pattern of incentives is increasing and concave with the size
of the firm, a finding that is not novel to our study (see Gibbons and Murphy, 1997, for
similar evidence).40 This goes against the idea that incentive mechanisms are less effective
in large production units (perhaps because of free riding effects), and that insurance is more
costly to small firms.
To get a sense of the results contained in Table 8, consider a 30 years-old, highly risk-
averse production worker employed in a medium-sized firm (50 employees) located in a
district, and with a historical performance variability of 20 percent. For this worker, the
sensitivity to firm permanent shocks is 0.06. For an individual with the same characteristics
—except risk aversion— the coefficient becomes as high as 0.14. For a firm with the same
characteristics —but a larger standard deviation of, say, 50 percent— the coefficient declines
to 0.05. In line with the predictions of the agency model, changes in worker and firm
characteristics may thus impart a wide range of variability in bu.
Note finally that the p-value of the J-test does not point to misspecification of the model
(11 percent), and that in all cases the power of the instruments (as measured by the partial
K2) is high enough to allow identification of the relevant parameters and to dismiss the
possibility of finite sample bias and inconsistency.
In column (2) we repeat the estimation exercise for the sensitivity of earnings to tran-
sitory shocks. In accordance with the results reported in Table 7, neither worker nor firm
characteristics appear to be statistically significant. This implies that insurance of transi-
tory shocks to value added is pervasive involving all types of workers and all types of firms.
The J-test has a high p-value of 36 percent, which suggests that it is unlikely that the
model is misspecified due to measurement error. The high values of the partial R2 for the
reduced-form regressions, on the other hand, suggest that the lack of a relation between
40Note that the pay-performance elasticity with respect to permanent shocks () equals b,4^1∙s) ψ
2t„ ((ln S)2] —4-. Given the empirical estimates of bu (In S) and bu ((ln S)2], the implied sensitivity increases
with firm size at a decreasing rate.
31