Technological change and employment: some micro evidence from Italy

Abstract

Consistently with previous evidence of the microeconomic literature on the subject and applying GMM-SYS on a unique Italian panel dataset, this study finds a significant – although small in size – positive relationship between innovation and employment at the firm's level in the 1990s.

Acknowledgement

Financial support by Cofinanziamento MIUR 2001/2001133591-002-Bologna Unit: ‘Patterns and determinants of skill-biased technical change in a medium-technology country’ (Responsible: Prof. Enrico Santarelli) is gratefully acknowledged.

Notes

¹ If the generic x_i, t variable is assumed to be endogenous in the sense that x_i, t is correlated with ν _i, t and earlier shocks, but not with ν _{i, t + 1} and subsequent shocks, the lagged values x_{i, t− 2} and longer lags are valid instrumental variables in the first differenced equations for periods t = 3, … , T. If the variable is predetermined in the sense that x _i, t may still be correlated with ν _{i ,  t − 1} and earlier shocks, then x_{i, t− 1} is additionally available as a valid instrument in the first differenced equation. If a stronger assumption is made, that x_i, t is strictly exogenous, then the complete time series of the x_i, t or the contemporaneous first difference are valid instrumental variables on each of first-differenced equations (Bond, 2002).

² The rule linking the two estimators is the following: if x_{i, t− j} is a good instrument for the equation in the first difference, hence Δx _{i, t− ( j− 1)} will be a good instrument for the equation in levels if Δx_{i, t− ( j− 1)} is not correlated with ε _i .

³ AR(1) estimated coefficients of employment and innovative investment are respectively 0.98 and 0.89, both of them significant at the 99% level of confidence.

⁴ In the second Italian Community Innovation Survey (CIS) covering 1994–1996, 46.7% of the 19 157 innovative firms declared they had innovated through machinery and equipment, meanwhile only 27% used internal R&D activities (ISTAT, 1997, Table 11).

⁵ The GMM-SYS estimates are presented in the one-step version, consistent with possible heteroscedasticity and more reliable than the two-step estimates needed for obtaining the validity tests of the model (Blundell and Bond, 1998; Doornik et al., 1999).

⁶ Since OLS tends to overestimate the coefficient of the lagged dependent variable and within estimator tends to underestimate it, the GMM-SYS estimate of it should assume an intermediate value.

⁷ Results from the Differenced Sargan test give a further support to the implementation of the GMM-SYS methodology with respect to the GMM-DIF. Differenced Sargan test can be used to test the hypothesis concerning the validity of some instruments. The full set of instruments under H₀ is tested against a strict subset under H₁. The test is asymptotically χ² distributed with degrees of freedom equal to the difference between the usual Sargan test under H₀ and H₁. In this case, the test is the validity of H₀ GMM-SYS against H₁ GMM-DIF and the value is 19.66 with 13 degrees of freedom suggesting not to reject the null hypothesis.