Insurance within the firm

(initial) worker - would permanently lower the earnings of the continuing workers by 203
euro.

Table 7, Panel B, also reports the estimated value of the relevant moments of the shocks
to output and wages. One can notice that while these are lower than for the full sample (see
Tables 3 and 5), they are not dramatically different.³⁰ Consistently with the estimates of b_u
and b_υ, as seen before we Hnd that the estimate of E {∆ωi_j't∆εjt) is positive and statistically
significant, while that of E (∆ω_lj't∆sjt-ι) is economically minuscule and insignificant.

To allow for an evaluation of the amount of insurance involved, we use equally weighted
minimum distance methods (EWMD) to estimate the variances of idiosyncratic shocks to
value added and (conditioning on these and the estimated insurance parameters b_u and b_υ)
the variances of idiosyncratic shocks to earnings.³¹ The estimate of the variance of the
permanent shock to value added, σ²_a, is 0.0229 (with a standard error of 0.0035), while the
estimated variance of the transitory shock, σ², is 0.0334 (with a standard error of 0.0048).
These are both sizeable and imply standard deviations of 15 and 18 percent, respectively.

Next, we estimate the parameters of the idiosyncratic part of the earnings process, i.e.
after filtering the variability that is due to the amount of insurance∕incentives provided by
the firm. Following the discussion in Section 4, we assume that this idiosyncratic part of
the earnings process can be written as:

^δ^j∙t ≡     t — ^buUjt — ^bv^δOt = ξijt + ^ρζijt-1 + ^δ∣^jIjt + ^ρ∆Цijt-l        ⁽²⁸⁾

i.e., ∆ω_jt follows a composite MA(2) process. In part, the coefficient ρ will reflect the legacy
of the autoregressive process of the value added (see above); in part, however, it will be
related to an idiosyncratic moving average component in earnings. The EWMD-estimated
variances of idiosyncratic shocks to wages are smaller than the firm counterpart: σ∣, the
variance of permanent shocks, is 0.0058 (standard error 0.0015), while σ² is 0.0034 (s.e.
0.0011). The MA coefficient ρ in the stochastic process of earnings is negative (-0.16) and

³⁰For example, E (∆^ι_l∙t∆ω⅛_t) is 0.0139 in the matched sample and 0.0165 in the full sample;
E (∆u¾_t∆ι¾_f_i) is -0.0055 in the matched sample and -0.0065 in the full sample.

³¹An alternative would be to use a generalized least squares procedure (optimal minimum distance, or
OMD). Our choice is dictated by the evidence presented in Altonji and Segal (1996), who show that EWMD
dominates OMD even for moderately large sample sizes.

25

<<   23   24   25   26   27   28   29   >>

More intriguing information