ICT and resilience in times of crisis: evidence from cross-country micro moments data

Bertschek, Irene; Polder, Michael; Schulte, Patrick Article — Published Version ICT and resilience in times of crisis: evidence from crosscountry micro moments data Economics of Innovation and New Technology Provided in Cooperation with: ZEW - Leibniz Centre for European Economic Research Suggested Citation: Bertschek, Irene; Polder, Michael; Schulte, Patrick (2019) : ICT and resilience in times of crisis: evidence from cross-country micro moments data, Economics of Innovation and New Technology, ISSN 1476-8364, Routledge, London, Vol. 28, Iss. 8, pp. 759-774, https://doi.org/10.1080/10438599.2018.1557417 This Version is available at: https://hdl.handle.net/10419/225070 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ ICT and resilience in times of crisis: evidence from cross-country micro moments data Irene Bertschek a , Michael Polder b and Patrick Schulte c a ZEW Mannheim & University of Giessen, Mannheim, Germany; b Statistics Netherlands, The Hague, Netherlands; c ZEW Mannheim & Deutsche Bundesbank, Frankfurt, Germany ABSTRACT ICT-intensive firms are often found to have a better performance than their non-ICT-intensive counterparts. Along with investing in ICT capital they have to adapt their production and business processes in order to reap the potentials implied by the use of ICT. Are these firms also more resilient in times of crisis? We study this question by exploiting a novel and unique data set from the Micro Moments Database. Covering 12 countries, 7 industries and the period from 2001 to 2010, the data allow us to distinguish between ICT-intensive and non-ICT-intensive firms within industries. We find evidence that indeed during the crisis in 2008 and 2009, ICT-intensive firms were hit less hard with respect to their productivity. This holds in particular for firms from service industries. Moreover, ICT-intensive firms were also more successful in introducing process innovations during that period which could explain their better productivity performance compared to non-ICT intensive firms. ARTICLE HISTORY Received 18 December 2017 Accepted 19 November 2018 KEYWORDS ICT; innovation; productivity; economic crisis; resilience; meso-level data JEL CLASSIFICATION H12; J24; O31; O47 1. Introduction An economic crisis comes with huge economic and social costs such as firm exits, unemployment and lower productivity. To reduce these costs it is important to understand what makes firms, industries and countries more resilient, i.e. what fosters their ability to resist to shocks and to recover quickly afterwards. 1 ICT may be one, especially potent, source of firms’resilience. It has been shown to be a general purpose technology (Bresnahan and Trajtenberg 1995; Jovanovic and Rousseau 2005 as well as Van Ark and Inklaar 2005) facilitating both product and process innovations as well as spurring (longer-term) productivity growth. Specifically, firms using ICT in a clever way may be able to deal with economic shocks more flexibly through easier reorganization of their production processes, i.e. through process and organizational innovations, and that way possibly can achieve a higher productivity and competitiveness, while firms lagging in the adoption of new technology face the risk of being driven out of the market. To provide evidence on the potential role of ICT for firms’resilience, here rather narrowly defined as firms’ability to keep productivity up during times of crisis, we exploit a novel and unique dataset: the Micro Moments Database (MMD). It comprises information at the meso-level for 12 countries and 7 industries and allows us going beyond the traditional growth accounting setting where industries – rather than firms –are classified according to their intensity of ICT usage. Instead, the MMD focuses more on the firm level distribution and on heterogeneity within industries, allowing us to study more detailed aggregates by firm characteristics such as ICT intensity. CONTACT Patrick Schulte [email protected] ZEW Mannheim & Deutsche Bundesbank, Wilhelm-Epstein- Straße 14, D-60431 Frankfurt am Main, Germany ECONOMICS OF INNOVATION AND NEW TECHNOLOGY 2019, VOL. 28, NO. 8, 759–774 https://doi.org/10.1080/10438599.2018.1557417 © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Our analysis comprises three steps. In a first step, the data are analysed descriptively comparing productivity levels and growth rates as well as innovation rates during the economic crisis with the pre-crisis period. In a second step, we apply the analytical framework of production functions relating firms’productivity in levels and growth rates to ICT intensity. A difference-in-differences approach (see for instance McGuckin and Stiroh 2001 and Stiroh 2002) allows identifying significant differences with respect to the productivity contribution of firms’ICT intensity during the economic crisis compared to the period before the crisis. In a third step, we analyse the contribution of ICT intensity to firms’innovation output measured as product innovation and process innovation. The empirical results support our hypothesis that ICT-intensive firms, especially from service industries, indeed were hit less hard by the economic crisis. Their productivity level and growth hardly decreased during the period 2008–2009 whereas non-ICT-intensive firms experienced a strong reduction in productivity. Also, ICT-intensive firms became relatively more innovative in terms of realizing process innovations. This latter finding fits to the hypothesis that ICT allow firms to adjust their production processes in times of crisis and in that way allow them to cope better with the increased competitive pressure during an economic crisis. Our paper contributes to the literature in two ways: Firstly, we provide first evidence on the role of ICT for the resilience of firms and countries in times of an economic crisis. Such knowledge is important for economic policy that is in search for strategies to improve resilience towards potential future crises. Secondly, we contribute to the wider literature on ICT and productivity by illustrating the benefits of micro-aggregated industry data, which so far has not been used to study the ICT-productivity relationship. The paper is structured as follows: Section 2summarizes the related literature. Section 3intro- duces the MMD database, describes our estimation sample and presents descriptive evidence on the relationship between ICT and performance during the crisis. Section 4lays out our analytical framework and presents our estimation results. Finally, Section 5concludes. 2. Literature The link between productivity and ICT has been studied extensively both at the macro- and microlevel. At the macro- or industry-level, there exist studies using either single- or multiple-country data (e.g. Jorgenson and Stiroh 1999; Stiroh 2002; Van Ark, O’Mahony, and Timmer 2008; Jorgenson and Timmer 2011). In contrast, at the micro-level so far most studies are conducted based on data for a single country (see the surveys by Draca, Sadun, and Van Reenen 2007; Bertschek 2012; Cardona, Kretschmer, and Strobel 2013). The reason for this gap is the scarcity of data which would allow for cross-country micro-level studies. 2 Finally, more recently, studies, such as ours, started using mesolevel data sets, like the Micro Moments Database (see e.g. Bartelsman, van Leeuwen, and Polder 2016), which allow in a cross-country setting focusing more on the firm level and on heterogeneity within industries and which allow studying more detailed aggregates by firm characteristics such as ICT intensity. The ICT-productivity literature shows that the measured size of the contributions of ICT to productivity seems to depend on the available data and on the methodology used. In particular, the evidence on excess returns to ICT compared to other capital is mixed (O’Mahony and Vecchi 2005; Draca, Sadun, and Van Reenen 2007). Moreover, there are different routes how ICT may affect firm performance. An important link between ICT and productivity is innovation. As so-called general purpose technologies (Bresnahan and Trajtenberg 1995; Jovanovic and Rousseau 2005), ICT diffuse throughout the whole economy and enable innovation in adopting firms and sectors (see, e.g. Brynjolfsson and Saunders 2010; Brynjolfsson and McAfee 2011) leading to higher productivity. Hall, Lotti, and Mairesse (2013), for example, consider investment in ICT and in research and development (R&D) as potential sources of innovation which in turn may enhance labor productivity. Based on Italian firm-level data they show that R&D and ICT contribute directly to labor productivity but also indirectly through enabling innovation, the latter result confirming the findings in Polder, van Leeuwen, and Mohnen (2010). Empirical evidence also shows that the relationship between ICT and firm 760 I. BERTSCHEK ET AL. performance is heterogeneous with respect to firms and industries, i.e. some firms or industries are more successful in exploiting ICT than others. Chun et al. (2008), for instance, analyse the relationship between ICT intensity and performance heterogeneity for a panel of U.S. firms from 1971 to 2000. They find that heterogeneity in firm performance (i.e. variability in stock return and sales growth) is positively and significantly correlated with ICT intensity (measured as ICT capital relative to total capital) and that firm heterogeneity is associated with faster productivity growth at the industry level. They consider this result as evidence of creative destruction (i.e. increased competition) at the firm level. That is, through their use of ICT, more productive firms displace less productive firms. In a more recent and related paper, Chun, Ha, and Kim (2014) demonstrate that firm heterogeneity leads to more R&D investment which in turn leads to higher long-run growth. Besides leading to increased heterogeneity in firm performance, ICT itself is a heterogeneous concept. Some studies take account of this fact and analyse the effects of ICT infrastructure or of specific types of ICT. For the case of broadband infrastructure, for example, Grimes, Ren, and Stevens (2012)find for New Zealand that firms with broadband internet have a significantly higher labor productivity. By contrast, Bertschek, Cerquera, and Klein (2013)find a positive and significant effect of broadband internet on German firms’innovation activity but not on their labor productivity. The results by Polder, van Leeuwen, and Mohnen (2010) for the Netherlands suggest that broadband internet is particularly important for the innovative activities of services firms. Engelstaetter (2013), using German firm-level data, reveals that productivity gains based on enterprise system usage can be maximized by jointly employing three widely established enterprise software systems, i.e. Enterprise Resource Planning (ERP), Supply Chain Management (SCM), and Customer Relationship Management (CRM). Up until now, there is no analysis explaining whether ICT-induced productivity gains help firms enduring or surviving the recent European economic and financial crisis. Given that ICT capital deepening was responsible for up to 27 % of the increase in European labor productivity in the beginning of this century (Van Ark and Inklaar 2005) it seems natural to consider its potential for fostering recovery. Our work aims at providing evidence on this issue based on micro-aggregated data. 3. Data This section describes the MMD, the sample we derive from it and presents first descriptive evidence on the role of ICT for the resilience of firms during a period of economic crisis. 3.1. The MMD The data used in this study come from the MMD. 3 The MMD has been created through a series of international collaborative projects of national statistical offices. 4 In each country the following firm-level data sets were linked: the Community Innovation Surveys, (below called IS), the Survey on ICT usage and e-commerce in enterprises (EC), the Structural Business Survey or Production Survey (PS) and the underlying business register (BR). Using the linked firm-level data a common computer code was used to create aggregated data at the industry level and combined it into a harmonized cross-country database. The database includes measures of ICT usage and innovative activity together with measures of business performance, various firm characteristics and industry dynamics. A big advantage and unique feature of the database is that it not only provides industry-level aggregates but also contains aggregates created for subsets of firms within an industry, such as e.g. firms using ICT more or less intensively. 5 Although other breakdowns are available, for the empirical applications in this paper we mainly exploit this distinction between ICT-intensive and non-ICT-intensive firms and compare their developments over time. Our empirical analysis consists of two parts: in the first part, we compare firms’productivity developments, in the second part we compare firms’innovation activity. The productivity analysis is based on a combined sample of firms surveyed both within the Production Survey (PS) and the Survey on ECONOMICS OF INNOVATION AND NEW TECHNOLOGY 761 ICT usage and e-commerce in enterprises (EC). Our sample covers up to 12 countries (AT, DE, DK, FI, FR, IE, IT, NL, NO, SE, SI, UK), seven industries, namely ‘Electrical machinery, post and communication services’(Elecom), ‘Consumer manufacturing’(ConsG), ‘Intermediate manufacturing’(IntmdG), ‘Investment goods, excluding hightech’(InvesG), ‘Distribution’(Distr), ’Finance and business, except real estate’(FinBu), ‘Personal services’(Pers), two subgroups (ICT-intensive and non-ICT-inten- sive firms) and 10 years (2001–2010). This results in an unbalanced panel of 866 observations. Basic continuous variables considered include ‘nominal value added’and ‘gross output’,‘full-time employment’, the ‘total wage-bill’,‘nominal expenditures on intermediates’and a ‘capital service measure’.In addition, it includes derived variables such as ‘gross output based labor productivity’. Firms are classified as ICT-intensive and non-ICT-intensive according to an indicator, which is equal to zero (non-ICT-intensive firm) if the firm has no broadband access or if less than 40 % of the workers have access to broadband, otherwise, it is equal to one and the firm is classified as ICT-intensive. We use this measure of ICT intensity since common alternatives like PC use or internet access in general might be not very informative anymore, since their use and adoption is meanwhile nearly universal, whereas with respect to the access to fast internet and the share of workers having internet access there is still a substantial heterogeneity. 6 For the innovation analysis the same classifier for the distinction between ICT-intensive and non-ICT-intensive firms is used. However, the sample here consists of firms surveyed both for the Community Innovation Surveys (IS) and the Survey on ICT usage and e-commerce in enterprises (EC). It covers only 10 countries (since for Germany and the UK no data are available) but the same set of industries, subcategories and years. 7 The main outcome variables for this analysis are ‘product innovations’(inpd) and ‘process innovations’(inps). 8 Further control variables, which reduce the sample size, include a measure of ‘total innovation expenditures’(rtot) and a measure of ’cooperation arrangements on innovation activities’(co). Table A1 in the Appendix provides further details. 3.2. Descriptive evidence Table 1 presents summary statistics for the productivity growth of both ICT-intensive and non-ICT- intensive firms and compares their development for the pre-crisis and crisis period. It confirms the dramatic overall productivity decline during the economic crisis, showing that productivity for the 12 countries covered was increasing by 3.5 % annually during the pre-crisis period (2002–2007), but was negative during the economic crisis, resulting in a productivity decrease of 1.1 % a year. Comparing ICT-intensive and non-ICT-intensive firms, the results show that during the pre-crisis period (2002–2007) the average annual productivity growth of the former ones was positive (0.4 percentage points) but lower than that of non-ICT-intensive firms. The latter group increased its productivity on average by 1.3 percentage points a year. Thus, non-ICT-intensive firms increased their relative productivity by 0.9 % annually in the years preceeding the crisis, compared to their ICT-intensive counterparts. Given the productivity-enhancing character of ICT this might be surprising. However, as Table A2 in the Appendix shows, ICT-intensive firms are more productive than their non-ICT counterparts in both the pre-crisis and the crisis period. Also, during the crisis this development was reversed, since ICT-intensive firms experienced a much lower reduction in their productivity growth rate. The productivity of ICT-intensive firms decreased by only 0.5 percentage points a year during the crisis, whereas that of non-ICT-intensive firms decreased by 2.3 percentage points. This suggests that ICT-intensive firms, despite a lower productivity growth trend and higher levels of productivity, were hit less hard during the crisis, compared to non-ICT-intensive firms. Their productivity growth rate during the crisis was 0.9 percentage points lower than that of the pre-crisis period whereas that of non-ICT-intensive firms decreased by 3.6 percentage points. In addition, productivity growth, though negative, was 1.8 percentage point higher for the ICT-intensive group. This finding holds not only for the total economy but also for most of the industries examined. Only for ‘Investment goods, excluding hightech’(InvesG) and ’Finance and business, except real estate’(FinBu) the 762 I. BERTSCHEK ET AL. relative productivity of non-ICT-intensive firms has improved during the crisis compared to the precrisis period. 4. Is resilience linked to ICT use? This section examines econometrically the link between firms’productivity growth during the crisis and their ICT intensity. If ICT are a driving force behind firms’resilience in times of crisis, then those firms using ICT more intensively should exhibit smaller productivity decreases than those using ICT less intensively. To study this relationship, we exploit the unique feature of the MMD, which allows comparing groups of firms differing in ICT intensity within the same industry. Previous studies were either only able to compare ICT-intensive industries with less ICT-intensive industries, thereby ignoring within-industry heterogeneity, or had to rely on firm-level data, which typically precludes cross-country comparisons. 4.1. Difference-in-difference estimates Following McGuckin and Stiroh (2001) and Stiroh (2002) the main approach we apply is a differencein-difference methodology, which uses a binary classification of ICT-intensive firms and compares their relative productivity level or development to that of other firms. Using the binary information is useful, as Stiroh (2002) suggests, because it is less sensitive to how ICT are measured, but is limited in that it does miss variation in ICT intensity across firms. Difference-in-difference estimation extends the test for differences in means with an additional constant, an indicator for the crisis period, and an interaction term for the group of ICT-intensive firms. In short, it tests for whether the difference in productivity (growth) between the intensive and non-intensive groups is significantly different during the crisis, compared to the productivity differential in the other years. The main estimation equation, in growth rates, is: Dln LPGO ijt = a + b DCrisis t+ g DICT ij + d DCrisis t×DICT ij + u Dln Xijt + e ijt, (1) where Dln LPGO ijt represents gross output-based labor productivity growth of country–industry combination i,jdenotes the group of ICT- and non-ICT-intensive firms within iand trepresents the respective year. In addition, DCrisis t=1 if 2007 ,t,2010 and DCrisis t=0 otherwise, whereas DICT ij =1 if a subgroup contains the ICT-intensive firms and DICT ij =0 otherwise. The vector Xijt contains a set of control variables such as capital, labor and intermediate inputs. αis the mean growth rate for non-ICT-intensive firms in the non-crisis period (before 2008 and in 2010), whereas a + g equals the growth rate of ICT-intensive firms during that time. βcaptures the change in productivity growth during the crisis for non-ICT-intensive firms, whereas b + d is the change for ICT-intensive firms. Table 1. Labor productivity growth by ICT-intensity –before and during crisis. Pre-crisis (2002–2007) Crisis (2008–2010) Productivity growth All Non-IT IT ΔAll Non-IT IT Δ Consumer manufacturing 2.9 0.6 −0.7 −1.3 2.8 −0.1 3.2 3.3 Investment goods, excl. hightech 4.7 3.8 1.2 −2.6 −6.9 −3.3 −7.5 −4.2 Intermediate manufacturing 4.1 1.9 1.3 −0.6 1.1 −3.4 0.3 3.7 Finance and Business, except real estate −0.4 −3.5 0.8 4.4 −2.2 −2.5 −1.5 1.1 Distribution 4.6 0.5 0.8 0.3 −0.7 −3.7 −0.4 3.4 Personal Services 2.7 1.9 −4.3 −6.2 −2.3 1.9 −0.9 −2.7 Electrical machinery, post and telecom services 5.1 3.8 1.1 −2.7 1.1 −4.3 4.6 8.9 All industries 3.5 1.3 0.4 −0.9 −1.1 −2.3 −0.5 1.8 Notes: The table contains unweighted average annual percentage labor productivity growth rates by industry for the full sample of countries covered (12 countries) as well as for two periods (2002–2007 and 2008–2010). For both periods it displays the average growth rate of the group of all firms (All) within an industry, of non-ICT-intensive firms (Non-IT) and of ICT-intensive firms (IT). In addition it shows the differences in growth rates between ICT-intensive and non-ICT-intensive firms, i.e. (D=IT minus Non-IT). Also, values for the full sample (row All industries) are given. ECONOMICS OF INNOVATION AND NEW TECHNOLOGY 763 The coefficient of main interest is δ, since it represents the difference in the adjustment during the crisis, i.e. it can be considered a measure of differences in the resilience between ICT-intensive and non-ICT-intensive firms. We also estimate a specification with productivity levels rather than growth as dependent variable. Standard errors are corrected for heteroscedasticity and for correlation across industries within countries. Table 2 contains our baseline results describing the link between ICT intensity and productivity during the times of crisis. It exhibits eight columns, where the first two columns contain specifications with labor productivity levels as the dependent variable. In column 1 the productivity level is explained by the crisis dummy, the ICT dummy, their interaction and our three inputs: capital, labor and materials. In addition, we include country and industry dummies. All variables, except the capital coefficient, are significant. The crisis dummy, as expected, shows a strongly negative and significant coefficient. In contrast, the interaction term, the variable of interest, is positive and significant, equal to 0.157, indicating that during the crisis in 2008 and 2009 the difference in productivity between ICT-intensive and non-ICT-intensive firms relative to that of the other firms has increased by around 15.7 %. More precisely, the results indicate that during the crisis non-ICT-inten- sive firms suffered from a productivity decline of around 10.2 %, whereas ICT-intensive firms experienced an improvement in productivity levels on average. In specification 2, we include country– industry-subcategory fixed effects to control for the differences in mean productivity levels of each group. The ICT dummy has to be dropped in this specification to avoid multicollinearity. Doing so results in very similar findings. Columns 3–8, by applying specifications in growth rates, also largely confirm these findings. Compared to the specifications in levels, which control for permanent differences in labor productivity levels between groups, e.g. among ICT- and non-ICT-intensive firms, specifications in growth rates control, through fixed effects, for differences in growth trends between groups. Such differences in productivity trends, if they exist and are not controlled for, could drive the findings of the levels specification. If e.g. ICT-intensive firms have on average a higher productivity growth rate, then a specification in levels would indicate that ICT-intensive firms compared to non-ICT-intensive ones have in later periods a relatively higher productivity level, and thus would indicate that they were hit less strongly by the crisis just because of the difference in growth trends and not because of a higher resilience during the crisis. Column 3 provides a simple specification without country–industry-subcate- gory fixed effects, which results again in a negative significant coefficient for the crisis-dummy, indicating that the productivity growth of non-ICT-intensive firms during the crisis was lower than in the pre-crisis period. The ICT-dummy is insignificant which indicates that there is on average no difference in the productivity growth rate between those two groups of firms. However, the interaction term is significant and positive, which suggests, despite the insignificant difference in the pre-crisis growth rates of ICT- and non-ICT-intensive firms, that the relative growth rate of ICT-inten- sive firms during the crisis increased. Just as in the levels specifications, this suggests that ICT-inten- sive firms were hit less strongly during the crisis. Column 4, controlling for country–industry subindustry average labor productivity growth rates, capital, labor and intermediate input growth rates, results in similar findings, where again the crisis dummy is, as can be expected, negative significant, whereas the interaction term is positive and significant. The labor and intermediate input variables show significant signs in line with values which can be expected from production theory. In contrast, for capital we find a negative value, whereas theory suggests that it should be positive. 9 However, as the following robustness checks show, the negative capital coefficient becomes insignificant and disappears in several alternative specifications (columns 5 and 6). In column 5 we add the share of workers with higher formal education as a control variable, which is important to capture potential sources of omitted variables related to complementary inputs to ICT, such as high-skilled labor. 10 Column 6 restricts the production technology to constant returns to scale by replacing the input levels with capital and materials to labor ratios (i.e. capital and intermediate input levels divided by the employment level). Doing so does not change our conclusion with respect to the role of ICT during the crisis. Column 7 studies the robustness of our results towards the exclusion 764 I. BERTSCHEK ET AL. Table 2. Labor productivity and IT –baseline results. Labor productivity Labor productivity growth 12345678 DCrisis −0.102*** −0.119*** −0.129*** −0.141*** −0.124** −0.133*** −0.141*** −0.108*** (−6.83) (−6.61) (−4.54) (−3.88) (−2.83) (−4.86) (−4.02) (−3.89) DCrisis ×DICT 0.157*** 0.144*** 0.100* 0.114* 0.098* 0.105*** 0.119* 0.054** (5.41) (5.11) (2.26) (2.03) (1.95) (4.36) ( 2.21) (2.95) DICT 0.212*** −0.012 (4.01) (−0.44) ln k avg −0.022 −0.011 (−0.46) (−0.58) ln e avg −0.471*** −0.297*** (−7.38) (−6.73) ln nm avg 0.533*** 0.390*** (6.22) (9.35) dln k avg −0.035*** −0.037*** −0.012 −0.039*** −0.041*** (−4.34) (−4.58) (−0.51) (−3.58) (−7.31) dln e avg −0.116*** −0.101*** −0.133** −0.141*** −0.108*** (−3.86) (−4.03) (−2.84) (−3.74) (−4.57) dln m avg 0.278*** 0.262*** 0.254** 0.308*** 0.252*** (6.43) (7.88) (2.95) (7.63) (7.39) dln hkpct 0.138 (1.00) dln k e 0.013 (0.75) dln m e 0.144*** (5.85) Country–subindustry FE No Yes No Yes Yes Yes Yes Yes Country FE Yes No Yes No No No No No Industry FE Yes No Yes No No No No No Sample Full Full Full Full Full CRS No Elecom No FRA R20.88 0.35 0.28 0.27 0.29 0.20 0.30 0.25 N824 824 824 824 627 645 705 779 Notes: This table analyzes the relationship between ICT-intensity and labor productivity levels/growth rates during and before the crisis. In the first two columns the dependent variable is the log labor productivity level, whereas in columns 3–8 it is labor productivity growth. Specifications 1 and 3 are estimated using OLS, whereas the remaining specifications are estimated using an FE-estimator. The sample covers in specification 1 to 6 all countries, industries, subgroups and years, whereas specification 7 excludes observations from the industry Elecom whereas specification 8 excludes France. In column 6 we assume constant returns to scale by using capital and material to labor ratios (i.e. capital and materials divided by employment) as input variables. Heteroscedasticityrobust t-values are provided in parentheses, which are corrected for clustering of observations at the country level. ***, **, *: Significantly different from 0 at the 1%, 5%, and 10% levels, respectively. All regressions are weighted by the country–industry-subgroup specific average full-time equivalent employment. ECONOMICS OF INNOVATION AND NEW TECHNOLOGY 765 of the Elecom industry, i.e. the ICT-producing industry. The results confirm our previous findings and thus suggest that our findings are not driven by the ICT-producing sector. Column 8 excludes France, since for this country there is only data available from 2007 onwards, and including it makes the sample highly unbalanced. 11 All specifications confirm the baseline findings from column 4, indicating that ICT-intensive firms seem to be hit less during the crisis and thus seem to be more resilient, at least in terms of their productivity, to an economic crisis. The next subsection aims at providing evidence on a potential explanation for these findings, namely on potential differences of ICT- and non-ICT-intensive firms with regard to their innovation behavior during the crisis. Following that, Subsection 4.3 provides further robustness checks aimed at establishing the main results more rigorously. 4.2. A potential explanation: process innovations ICT have been shown to improve firms’capacity to innovate (see e.g. Brynjolfsson and Saunders 2010; Brynjolfsson and McAfee 2011 as well as the literature review in Section 2of the paper). This section aims at providing evidence on whether ICT-intensive and non-ICT-intensive firms were differing with respect to their innovation behavior during the crisis and whether their relative innovativeness changed in that time. If this was the case, it could help explaining why ICT-intensive firms were hit less strongly with respect to productivity during the crisis. 12 Indeed as Table 3 shows, there are differences between ICT-intensive and non-ICT-intensive firms with respect to product and process innovativeness. During the pre-crisis period ICT-intensive firms were more innovative, i.e. ICT-intensive firms have implemented more new processes and introduced more new or significantly improved products or services. Half of the ICT-intensive firms had introduced product innovations during the pre-crisis period, whereas only 35 % of non-ICT-intensive firms did so. For process innovations the difference is smaller: 42 % of ICT-intensive firms did introduce innovative processes, whereas 36 % of non-ICT-intensive firms had a process innovation. These differences are consistent across industries, and moreover were amplified during the crisis period. For product innovations, the innovation gap of 14 percentage points between ICT-intensive and non-ICT-intensive firms increased to 16 percentage points. For process innovations this change was more pronounced, whereas before the crisis Table 3. Product and process innovation by IT-intensity before and during crisis. Pre-crisis (2002–2007) Crisis (2008–2010) Product innovation All Non-IT IT ΔAll Non-IT IT Δ Personal services 0.16 0.15 0.23 0.08 0.20 0.18 0.21 0.04 Investment goods, excl. hightech 0.59 0.53 0.70 0.17 0.54 0.46 0.62 0.17 Intermediate manufacturing 0.44 0.41 0.51 0.11 0.43 0.36 0.51 0.15 Finance and business, except real estate 0.36 0.29 0.39 0.10 0.37 0.26 0.40 0.13 Electrical machinery, post and telecom services 0.58 0.47 0.66 0.19 0.61 0.45 0.69 0.23 Distribution 0.23 0.16 0.31 0.15 0.28 0.20 0.33 0.13 Consumer manufacturing 0.46 0.42 0.59 0.17 0.46 0.39 0.58 0.19 All industries 0.41 0.35 0.50 0.14 0.42 0.33 0.49 0.16 Process Innovation All Non-IT IT ΔAll Non-IT IT Δ Personal services 0.17 0.16 0.23 0.08 0.26 0.21 0.33 0.12 Investment goods, excl. hightech 0.47 0.46 0.50 0.04 0.46 0.39 0.52 0.13 Intermediate manufacturing 0.47 0.45 0.50 0.05 0.44 0.39 0.49 0.09 Finance and business, except real estate 0.32 0.29 0.34 0.05 0.34 0.24 0.36 0.12 Electrical machinery, post and telecom services 0.46 0.43 0.47 0.04 0.47 0.43 0.49 0.05 Distribution 0.29 0.24 0.34 0.09 0.30 0.26 0.33 0.08 Consumer manufacturing 0.43 0.40 0.49 0.09 0.41 0.37 0.49 0.12 All industries 0.39 0.36 0.42 0.06 0.39 0.34 0.43 0.10 Notes: The table contains average annual shares of firms having introduced a new product or process innovation. It covers the full innovation sample of countries (10 countries) for two periods (2002–2007 and 2008–2010). For both periods it displays the average shares for all firms (All), for non-ICT-intensive firms (Non-IT) and for ICT-intensive firms (IT). In addition it provides differences between ICT-intensive and non-ICT-intensive firms with respect to their innovation shares (D=IT minus Non-IT). Also, values for the full sample (row All) and for each industry covered are given. 766 I. BERTSCHEK ET AL. Table A3. Labor productivity growth –manufacturing and services. Manufacturing Services 12345678 DCrisis −0.104* −0.141** −0.101 −0.063 −0.163*** −0.142*** −0.155*** −0.166*** (−1.83) (−2.31) (−1.55) (−1.18) (−5.24) (−5.41) (−4.48) (−5.28) DCrisis ×DICT 0.029 0.061 0.022 −0.004 0.171* 0.132* 0.226** 0.169* (0.63) (1.45) (0.39) (−0.10) (2.20) (1.98) (3.02) (2.18) dln k avg −0.054 −0.055 −0.073* −0.030 −0.030*** −0.059*** 0.008 −0.031*** (−1.80) (−1.64) (−2.21) (−1.04) (−3.61) (−3.90) (0.54) (−4.02) dln e avg −0.189*** −0.207*** −0.167** −0.176*** −0.148** −0.233** −0.111 −0.135** (−5.82) (−4.05) (−2.83) (−4.86) (−2.95) (−2.88) (−1.23) (−2.43) dln nm avg 0.302*** 0.301*** 0.293*** 0.311*** 0.346*** 0.528*** 0.199** 0.335*** (7.21) (5.63) (5.20) (5.54) (3.91) (4.64) (2.76) (3.73) Country–subindustry FE Yes Yes Yes Yes Yes Yes Yes Yes Country FE No No No No No No No No Industry FE No No No No No No No No Weighted Yes Yes Yes Yes Yes Yes Yes Yes ConsG ✓✓✓ IntmdG ✓✓ ✓ InvesG ✓✓✓ Distr ✓✓ ✓ FinBu ✓✓✓ Pers ✓✓✓ R20.37 0.44 0.35 0.33 0.29 0.42 0.13 0.28 N364 262 221 245 341 227 208 247 Notes: This table analyzes the relationship between ICT-intensity and labor productivity growth rates during and before the crisis for manufacturing and service industries separately. Columns 1–4 contain only observations from manufacturing industries whereas columns 5–8 contain only observations from service industries. Each specification contains a different (sub)set of industries. Heteroscedasticity-robust t-values are provided in parentheses, which are corrected for clustering of observations at the country level. ***, **, *: Significantly different from 0 at the 1%, 5%, and 10% levels, respectively. All regressions are weighted by the country–industry-subgroup specific average full-time equivalent employment. ECONOMICS OF INNOVATION AND NEW TECHNOLOGY 773 Table A4. Labor productivity growth –alternative measures of the crisis. 123456 DCrisis09 −0.157*** (−5.72) DCrisis09 ×DICT 0.035** (2.53) DCrisis0810 −0.114*** (−3.88) DCrisis0810 ×DICT 0.102* (1.88) DCrisis0710 −0.024 (−1.59) DCrisis0710 ×DICT 0.006 (0.27) DCrisis0610 0.008 (0.59) DCrisis0610 ×DICT 0.007 (0.25) DCrisis −0.057* −0.058* (−1.98) (−2.10) DCrisis fl −0.128** (−3.19) DCrisis ×DICT 0.052* 0.046 (2.04) (1.79) DCrisis fl ×DICT 0.093 (1.35) DCrisis fl2 −0.129** (−3.19) DCrisis fl2 ×DICT 0.102 (1.38) dln k avg −0.035*** −0.034*** −0.032** −0.028** −0.039*** −0.039*** (−3.35) (−3.31) (−3.16) (−2.90) (−5.44) (−5.59) dln e avg −0.086*** −0.122*** −0.132*** −0.137*** −0.106*** −0.106*** (−3.70) (−4.72) (−4.85) (−4.58) (−4.09) (−4.12) dln nm avg 0.240*** 0.278*** 0.295*** 0.296*** 0.256*** 0.255*** (6.97) (8.16) (6.21) (6.06) (7.12) (7.13) Country–subindustry FE Yes Yes Yes Yes Yes Yes Country FE No No No No No No Industry FE No No No No No No Weighted Yes Yes Yes Yes Yes Yes Sample Full Full Full Full Full Full R20.24 0.27 0.18 0.18 0.29 0.29 N824 824 824 824 824 824 Notes: This table analyzes the relationship between ICT-intensity and labor productivity growth during and before the crisis. The dependent variable is gross-output-based labor productivity growth. In column 1 the crisis is defined as the year 2009. In column 2itisdefined as the years 2008 to 2010. In column 3 the crisis also includes 2007, whereas in column 4 it even includes 2006. In column 5 and 6 the crisis is defined as the years 2008 and 2009. In column 5 we add a dummy equal to one if a country experienced an average productivity decline larger than 1% in 2008 or 2009 (DCrisis fl). In column 6 we add a dummy equal to one if a country experienced an average productivity decline of more than 5% in 2008 or 2009. All specifications are estimated using a FE- estimator. In all specifications, the sample covers all countries, industries, subgroups and years. Heteroscedasticity-robust t-values are provided in parentheses, which are corrected for clustering of observations at the country level. ***, **, *: Significantly different from 0 at the 1%, 5%, and 10% levels, respectively. All regressions are weighted by the country–industry-subgroup specific average full-time equivalent employment. 774 I. BERTSCHEK ET AL.