Following the method suggested by Lokshin and Sajaia (2004) we estimate the equation system (9’)
first by including the growth rate of the capital stock as it is (analogously to the OLS method in
Table 6), second by substituting the variable with its predicted value obtained through a first-stage
regression with instruments (analogously to the 2SLS method used in Table 6). The results of the
ML Endogenous Switching Model are shown in Table 7 below.
The results for innovative firms largely confirm those of Table 6. The estimated share of the capital
stock is equal to 0.196 when the capital stock is predicted. The parameters for mean age are
estimated to be equal to -0.40 (overestimating with respect to OLS and 2SLS). The temporary
worker share coefficient is higher both in ML and ML predicted (-0.19 and -0.22 respectively). The
value of ρ1 greater than zero means that the correlation between the residuals of the second-stage
equation and the selection equation is positive. This means that an innovative firm does better in
increasing productivity than a firm which randomly chooses to be in either regime.
The capital share for non innovative firms is zero both with ML and with ML predicted. This value
is similar to the OLS result of Table 4.
Our estimates also indicate that the impact of average board age for non innovative firms is zero.
This result is the same as that of Table 6, where the 2SLS coefficient is zero as well.
The impact of the share of temporary workers is instead negative and significant (μ2,ML ≡ μ2,MLpred =
-0.24). Finally, the correlation between selecting the regime and the second-stage equation is zero.
The Wald test of independence across the equations (8) and (8’) rejects the null of independence.
Together with the positive and statistically significant Mills ratios, this means that the error
correction-switching method is appropriate.
23