The productivity impact of innovation on industry in Argentina

Abstract

This paper assesses the innovative process of Argentinian manufacturing firms and its impact on labour productivity. Applying a CDM model, we combined firms’ innovative decisions with innovation results and their impacts on labour productivity. We used recent data from Argentina’s National Survey on Employment and Innovation Dynamics (ENDEI in Spanish) from 2010-2012 and 2014-2016. Our findings verify the innovative process which links innovation with productivity regardless of prevailing macroeconomic and industrial conditions.

Keywords:

• Innovation
• productivity
• CDM model
• manufacturing industry
• Argentina

SUBJECT CLASSIFICATION CODES:

• J24
• O12
• O14
• O33

Disclosure statement

No potential conflict of interest was reported by the authors.

Acknowledgements

The authors would like to acknowledge the thoughtful comments received from Gustavo Atilio Crespi and Pablo Ortiz, as well as two anonymous referees.

Notes

1 Source: WDI Indicators. GDP per capita at constant LCU. Accessed: April 2021 https://data.worldbank.org/indicator/NY.GDP.PCAP.KN?locations=AR.

2 See data from Maddison Project Database, available at https://www.rug.nl/ggdc/historicaldevelopment/maddison/releases/

3 According to data from the Industrial Monthly Estimator (base 2012 = 100, trend-cycle), manufacturing production averaged for the 2010–2012 period a maximum level, surpassing by 62% what was recorded at the beginning of the series (1994). After that, manufacturing production decreased by 5% in 2014 (this is the last available estimate).

4 Concerning the decision to innovate and their links with certain performance variables of firms, in a previous study, the authors – using methods of quantile regression and ordinary least squares – concluded that the innovation activities of Argentinian firms are significantly associated with higher levels of productivity, wages and job skills, in a magnitude that differs at the sectoral and firm levels due to the structural heterogeneity of the industry.

5 Among the newly industrialised countries, a positive association between R&D, innovation and productivity has been found for South Korea (Lee and Kang 2007), Malaysia (Hegde and Shapira 2007), Taiwan (Aw et al. 2008) and China (Jefferson et al. 2007). By investing in R&D and human capital, these countries have managed to narrow their distance from the best practices.

6 Crespi and Zúñiga (2012) reported that productivity gaps in the manufacturing sector between innovative and non-innovative firms are much higher in Latin America than in industrialised countries. For a typical country in the European Union, the productivity gap is 20%, while it is 70% in a typical Latin American country. Thus, Latin America has great potential to benefit from investment and policies that foster innovation.

7 The data collection method for the ENDEI II consisted of the application of two forms: an online self-administered questionnaire and a questionnaire in the form of a face-to-face interview with an official surveyor. The web-based questionnaire collected balance sheet information from enterprises and was completed autonomously by the respondent. It had automated consistency criteria that allowed the respondent to review the information uploaded and rectify it if necessary. The face-to-face questionnaire was designed to be completed through a notebook application. The information it collected was qualitative and structured and was completed by the interviewer through face-to-face interviews. Therefore, the design of the questionnaire was directed and participatory. Both questionnaires were semi-structured as they featured both pre-coded and open-ended responses.

8 The inclusion of the acquisition of machinery and equipment could potentially present bias towards the innovation intensity variable. As an embodied innovation effort, it is not possible to disaggregate and discount the annual depreciation rate.

9 It is necessary to link innovative activity in developing countries with the reconversion efforts that firms face in response to the new conditions generated by openness and globalisation, where the organisational dimension is an essential activity (Jaramillo et al. 2000). It is possible to regard decisions to innovate as investment decisions with objectives focused on productivity and competitiveness. In the Frascatti Manual, the most common measure of input (R&D) has limitations as a measure of innovation effort, ignoring other relevant innovative activities. Kline and Rosenberg (1986) and Albornoz (2009) have discussed this in relation to the political implications of these methodological issues.

10 All equations included control variables of unobserved heterogeneity, described at the end of this section.

11 Certain variables intervened both in the stages of innovation efforts and innovation output stages, as indicated in the following section.

12 In this period (2014–2016), this condition was defined as 9 years or more.

13 This strategy was equivalent for human capital, cooperation and access to financing variables, following Crespi and Zúñiga (2012) and Crespi, Tacsir and Vargas (2016).

14 We could argue that the variable considers an excessively low threshold to define whether the firm is integrated with foreign capital (1% of total capital). However, previous studies share this type of construction for other analyses on innovation data from Argentina (Lugones, Suárez and Gregorini 2007; Arza and López 2021).

15 As patents stand both as a determinant of innovation efforts and an indicator of outputs – although, more frequently in developed countries – their presence may have generated an endogeneity bias due to their high correlation with innovation efforts. As in Crespi and Zúñiga (2012), we assumed exogeneity, considering that bureaucratic processes for obtaining a patent are lengthy and it is likely that innovations patented by the firm are older than the coverage period of the ENDEI.

16 It is assumed that this kind of support has a strong impact on both the probability to innovate and the amount invested. Such high correlation may generate an upward bias on the effects of this factor if the observed variable is considered - whether the firm assessed public financial support or not – (Raffo et al. 2008, 236). Thus, the strategy was to measure P_sup using the percentage of firms of the same size and in the same sector to avoid potential endogeneity with the dependent variable (Tello 2017).

17 In the innovation intensity expression, while the expected value may be estimated both conditionally on reporting positive values (as an innovative firm) and non-conditionally, we adopted the latter, as this reflects the dependent variable prediction considering the innovative status of firms.

18 Outliers on value-added and innovation activities were eliminated and firms which reported growth in real sales higher than 500% in the period were filtered, along with those that reported innovation expenditure to be higher than 50% of their sales and those that did not declare personnel.

19 There are two reasons why innovation results reported statistics similar to those of innovation efforts in certain cases. The first one is that over 90% of the innovative firms obtained at least one innovative result. The other is related to the construction of the innovation results variables, which were constructed in the database for the entire periods (2010–2014), while the innovation expenditure only accounted for 2010 and 2014, respectively. See the variable definitions in Table A2.

20 Raffo et al. (2008), Arza and López (2010) and Crespi, Tacsir and Vargas (2016).

21 The differences in firms’ age may be related to the different construction of the variable (see the variable definition in Table A2).

22 Williams (2012) expressed the average marginal effects (AME) as a proper alternative when computing predicted values, particularly when the objective is to compare two hypothetical populations that differ in the specific values of the variable of interest and have the same values of the other independent variables in the model. As AME uses all of the data instead of just the mean values, a majority of authors prefer this method to measure impacts.

23 Previous studies have reported mixed results. Raffo et al. (2008) did not find a significant effect over the decision to engage in R&D activities nor the intensity of innovation. However, significant and positive effects were reported by Crespi and Zúñiga (2012).

24 Both in innovation results and productivity equations, the instruments used to control endogeneity were tested, rejecting the null hypothesis of a weak instrument.

25 As in previous studies for Argentina, we assume that all kinds of innovative activities exert an influence on innovation outputs. As in this study, in Crespi and Zúñiga (2012) and Crespi, Tacsir and Vargas (2016) the innovation efforts are estimated by the aggregate innovation expenditure. In Arza and López (2010), innovation expenditures are classified according to certain activity categories, though they are all inserted in the knowledge production function.

26 Williams (2016, 2017) and other authors in the Statalist forum stated that the differences that could arise between the original and the marginal effect coefficients in terms of statistical significance is related to the fact that they are the result of testing different hypotheses and the non-linearity of these models produces these seemingly ‘paradoxical’ results. The consensus is – when these differences are reported – to follow the sign and p-value of the original coefficients.

27 As in the exporting condition, we followed the original probit coefficient for product/process innovation in the 2014–2016 period.