Experience, Innovation and Productivity - Empirical Evidence from Italy's Slowdown

innovative and non innovative firms.⁵ Innovative firms had introduced a product or process

innovation (or both) in the three-year period considered by the survey (2001-2003). Non innovative
firms declared to never introduce innovations in the period. This test is at first carried out without
allowing for endogenous regime switching.

We perform the test by comparing the estimates of equation (4) on the two subgroups
(unconstrained model) and on the whole sample (constrained model). The null hypothesis is that the
constrained model is valid. The F-test when the firms are grouped as innovative and non innovative
is F(31,7915)=15.99 p-value=0.0, thus rejecting the null. P-values are reported in Table 4 and refer
to the three innovation modes, depending on whether the group discriminant refers to whether the
firm undertakes either product or process innovations, product innovation only, or process
innovation only. The Chow test always rejects the null hypothesis, thus indicating that the partial
correlation between age and the share of temporary workers, on one side, and the dependent
variable on the other differs across the two groups of firms.

Consistently with the test results, we allow the parameters of interest of the experience variables to
vary across groups, according to the specification equation (5):

к y ] Y 1 к κ ( K1                                    к mp ( Temporary _

^δ 2 ^lnl L I = ^ 2 ^lnl L I ^{+ β}1 D1 ^{+ β}2 D2 ⁺ Y1 Age_l* D1 ⁺ Y2 Age_i * ^D2 ⁺ ¼1 ----L---- I   * ^D1 ⁺

∖ L J it           ∖ L J it                                                          ∖ L     Z it-2             (5)

+μ2 ( Tempoarù * D2 + εit

∖ L     J it -2

The dummies D1 and D2 identify the two groups of firms (D1 = 1 if the firm is innovative and D2 =
1 if it is non innovative). D2 will be omitted in the regression because of collinearity. Notice that the
innovation dummy captures the impact of technological progress on labor productivity as in a
standard Cobb-Douglas production function approach.

In Table 4 we also report the Wald tests (and the relative p values) of parameter instability for each
parameter, when the null hypothesis is:

⁵ In the robustness checks section we applies the Chow test also to α, assuming that the growth rate of the capital-labor
ratio differs across groups as well. In the benchmark model, we assume that the effect of K/L on labor productivity
does not differ across types of firms.

10

<<   9   10   11   12   13   14   15   >>

More intriguing information