New digital technologies and firm performance in the Italian economy

ABSTRACT

New digital technologies can generate substantial gains for adopting businesses. In this paper we analyse the impact of new technologies associated with the Industry 4.0 paradigm on labour productivity, average wages and sales growth. The analysis is based on microdata produced by the Italian National Institute for the Analysis of Public Policies (INAPP) on a large representative sample of Italian firms. We merge INAPP data with Orbis data covering the period 2010–2014-2018. By applying a Diff-in-Diff methodology, we show that the economic size of the effect of new technologies on productivity and sales is approximately twice as large as the effect on average wages. The positive impact is stronger for small and medium-size firms, even though the effects appear to be concentrated among more mature rather than younger firms and are heterogeneaous along the distributions. Results are robust to unobserved heterogeneity and endogeneity issues.

KEYWORDS:

• Digital technologies
• industry 4.0
• firm performance
• labour productivity
• wages
• Diff-in-Diff and quantile estimations

SUBJECT CLASSIFICATION CODES:

• D20
• L23
• O33

Acknowledgement

Andrea Mina acknowledges support from the Italian Ministry of Education, University and Research, PRIN-2017 project 201799ZJSN: “Technological change, industry evolution and employment dynamics.”

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ We merged Orbis 2014 with RIL-Inapp 2015 since the latter collects firms’ information referring to 2014.

² See also Jovanovic and Rousseau (2005) and Bresnahan (2010), for discussions of ICT as a general purpose technology.

³ Several contributions have focused on the impact of new digital technologies on employment and have devoted special attention to the effect of the automation on both the task content of occupations and aggregate outcomes in terms of job creation or destruction. The specific focus of this paper is on the effects of digital technology adoption on firm performance. As we cannot adequately cover here all existing literature on the relationship between new technologies (including robotics and artificial intelligence) on jobs, we refer the reader to Goos, Manning, and Salomons (2014), Autor (2015), Brynjolfsson and McAfee (2016), Frey and Osborne (2017), Brynjolfsson, Rock, and Syverson (2017), Felten, Raj, and Seamans (2018), Balsmeier and Woerter (2019), Acemoglu, Lelarge, and Restrepo (2020), Domini et al. (2021), Cirillo et al. (2020a).

⁴ One may be tempted to overemphasise the role of Artificial Intelligence in modern manufacturing, even though its application is still quite limited (Martinelli et al. 2021).

⁵ The RIL-Inapp survey sample is stratified by size, sector, geographical area and the legal form of firms. Inclusion depends on firm size, measured by the total number of employees. This choice has required the construction of a ‘direct estimator’ which is defined for each sample unit (firm) as the inverse of the probability of inclusion in the sample. For more details on RIL questionnaire, sample design and methodological issues see: <http://www.inapp.org/it/ril>.

⁶ It should be emphasised that RIL-Inapp 2015 collects firms’ information referring to 2014, therefore we merged RIL-Inapp 2015 with 2014 Orbis data.

⁷ Results are robust to changes in the size cut-off point and do not change if we include in the sample all firms that have at least one employee. These results are available from the authors upon request.

⁸ In what follows we refer to ‘RIL 2015’ when considering RIL alone (since this is the name of the survey) and to 2014 when considering the RIL-Orbis merged database.

⁹ For the interested reader, Table A1 in the appendix shows the incidence of I4.0 investments by firms’ size, macro-region, sector of activity and age separately for the cross-section and for the panel component. In terms of coverage by size, age and sector, the statistics show satisfactory coverage and good balance between smaller vs. larger firms, older vs. younger firms and across sectors of activities. Overall, figures of Table A1 describe a larger diffusion of investments in new enabling technologies among large and young manufacturers located in Northern Italy.

¹⁰ See Table A2 in Appendix for further details. Additional information on the policy and its implementation can be found in the 2018 Report on the Competitiveness of Productive Sectors, produced by Italian National Institute of Statistics (ISTAT 2018, Rapporto sulla competitività dei settori produttivi, Rome).

¹¹ The inclusion of controls for managerial and corporate governance characteristics, workforce composition, firm production, sectors of activity, nuts 2 regions, and industrial relations does not raise any multicollinearity concerns, as evidenced by Variance Inflation Factors tests performed after OLS estimates.

¹² Figure A1 in the Appendix shows the conditional trends of the two groups (treated and untreated) before and after treatment for the three outcomes plotting coefficients from Table 2.

¹³ More in detail, we test the robustness of the effect of Industry 4.0 technologies on firm performance by adopting a two-step procedure. First, we estimate a propensity score matching (PSM) enabling to control for sample selection into the decision of I4.0 investment by adjusting for “observable” variables. Indeed, the PSM requires to combine a group of “treated” firms investing in I4.0 technologies with a group of “untreated” firms having similar observable characteristics which did not invest in these technologies. In a second step, we use this “restricted” control group to estimate the counterfactual effects of the I4.0 investment on our three outcomes through a Diff-in-Diff approach. The variables used for matching the two sample of firms (treated and untreated) are the same included in the specification of the main equation [1]. To adjust for observable differences between treated and untreated firms, the matching procedure is run on 2010 selecting the longitudinal component of the RIL-Orbis database that allows to collect information on firms operating all three sample years. To assess the quality of the matching, Table A4 in the Appendix presents the differences between the mean value of a large subset of the variables used to match treatment and control groups for both productivity and wage equations. Overall, the figures in Table A4 confirm that the two groups, though initially different, appear to be rather similar after matching, with no statistical differences in the means of the reported values with very few exceptions. In other words, the matching is successful both for labor productivity and average wages, even though matching on some variables falls below conventional significant values.

¹⁴ Unfortunately, we cannot test this speculative argument because we do not have any information about the type of productive assets firms have at birth.

¹⁵ Note as well that further explorations of heterogeneity effects by firm size and age could in theory be carried through the application of a triple-DiD methodology. Unfortunately, this would raise difficult identification concerns because of biases in the common trend assumption. The reader will find in Table A5, Tabe A6 and Table A7 of the Appendix supplementary evidence on industry vs. services sectors. Regarding the problem of technological heterogeneity, it would be extremely interesting to study technology-specific substitution and complementary effects, ideally with detailed information about wages, and the skills/tasks profiles of new hires and separations. The observed rates of adoption, however, are not high enough to provide sufficient statistical power for these estimations, and we do not have the fine-grained matched employer-employee data that would be necessary to test for specific effects on individual wages and skills.

¹⁶ The first step is used to estimate unobserved fixed effects using a standard within fixed effects estimator. In the second step, the consistently estimated fixed effects are used to reduce the (log of) labour productivity (or alternatively the log of wages/sales per employee) and this transformed (adjusted) measure is used as the dependent variable to conduct standard conditional quantile regression with in our framework the inclusion of the interaction term for time effects (diff-in-diff). For a similar application see Cirillo and Ricci (2020).

¹⁷ For further details on the fiscal incentives available for Italian firms, see the RIL questionnaire 2018 (http://www.inapp.org/it/ril).

¹⁸ See Bratta et al. (2020) for a specific investigation on the impact of hyper-depreciation on I4.0 investment in Italy.

¹⁹ The only exclusion relates to the following sectors or activities: ‘Financial and Insurance Activity’ (K) and export-related activities or aid contingent upon the use of domestic over imported goods..