reject the null. The imprecise estimates of the capital per worker growth in the non innovative group
might be due to its potential endogeneity.
To confirm our results, Table 4 reports p-values of the Wald tests of parameter instability for the H0
hypotheses:
Y1 = Y2
H о: U = λ2
μ = μ2
The test for age and the capital-labor ratio coefficients rejects the null of equal parameters (p-value
of the test is equal to zero). In general the three Wald tests reveal that the constrained model has to
be rejected.
5.2 Endogenous innovation decision
Table 5 and 6 show the results from the two-stage method of estimation of the parameters in (7) and
(9’), which take into account the endogenous formation of the two groups. The two-stage analysis
allows us to perfect our understanding of the relation between our variables of interest and
productivity growth.
Table 5 shows the estimates of the first stage, namely the decision to innovate. The probit
estimations of Table 5 are used as a first stage in the switching regression with endogenous
switching (eq. (7)). The probit results for product innovation appear in column 3 and 4 and those for
process innovation appear in column 5 and 6. These latter regressions are useful to understand the
importance of the different instruments in determining innovation decisions. The given set of
instruments predicts much better the probability to undertake a product innovation than a process
innovation. Moreover, engaging in R&D activity and hiring R&D workers has a much stronger
effect on the probability of undertaking product innovation than on the one of process innovation. In
contrast, per-worker current cash flow seems to be an important pre-condition for introducing
19