dummy should reduce misclassification error due to the imputation procedure.39
As for firm characteristics, the historical performance variability is measured by the
standard deviation of log real value added over the period observed (from a minimum of
5 to a maximum of 13 years). We also construct a dummy for location in an industrial
district and use a quadratic in log firm size (number of employees). Using a polynomial in
log firm size has the advantage that when evaluated at the minimum size observed in our
sample (one employee) the wage-performance sensitivity is zero, which can be thought of
as the baseline. Table 8 reports the results. Column (1) shows the effects of worker and
firm characteristics on the sensitivity of wages to permanent shocks to performance; column
(2), transitory shocks. To check the power of the instruments excluded in the reduced-form
regressions, we report the partial K2 measure suggested by Shea (1997) in the context of
multivariate models with multiple endogenous variables.
We first comment on the results reported in column (1). The indicator for high risk
aversion is associated with a statistically significant lower sensitivity of wages to permanent
shocks to performance (i.e., more insurance and a lower value of bu). Overall, there is
a quite sizable sensitivity differential due to risk aversion (-0.07). In the same direction,
managers have less insurance than either white collar or blue collar workers (the excluded
category). However, standard errors are high and prevent reliable inference, arguably due
to the small number of observations on such workers (a little more that 1 per cent of the
sample). Finally, we find no solid evidence of a relation between tenure (age) and incentive
schemes.
In terms of firms, consistently with the predictions of the basic agency model, those
with higher variability in performance provide more insurance and less incentives: the
coefficient is negative (-0.0338) and highly significant. We interpret this as evidence that
incentive schemes are less effective the noisier the relation between effort and performance,
supporting one of the fundamental implications of the theory. Predictions coming from
extensions of the basic model also find empirical support. Firms located within a district
provide less insurance and more incentives than others, and the difference is statistically
39Direct use of the imputed risk aversion variable in levels or logs gives qualitatively similar results,
although somewhat less precisely measured.
30