Does Innovation Lead to Firm Growth? Explorative versus Exploitative Innovations

ABSTRACT

In this article, we examine the relationship between innovation and firm growth. Drawing on previous research, we implement a classification of innovations based on whether they are explorative or exploitative. Access to Swedish register data comprising the entire private sector from 1997 to 2012 allows us to construct innovation patterns for more than 480,000 firms. GMM-estimations confirm a significant and positive effect of both exploitative and explorative innovation on firms’ employment growth. More radical explorative innovations are shown to have a more persistent growth effect, while exploitative innovation increases labour demand in the short run.

KEYWORDS:

• Innovation
• firm growth
• exploration innovation
• exploitation innovation

JEL CLASSIFICATION:

• O31
• L25

I. Introduction

Understanding how innovation influences firm performance, in particular employment, is high on the agenda for policy-makers. As firms convert knowledge into innovations and strengthen their market position, they are likely to contribute to economic growth and welfare. A large number of countries have also stressed knowledge upgrading and innovation policies as a means to promote long-term, sustainable growth. The combination of globalization and rapid technological change has meant a stiffening in competition, further emphasizing the importance of firms’ innovation capacities.

Here, we examine how different types of innovation influence employment growth at the firm level, an area where knowledge is scarce and ambiguous as compared to other performance measures (Coad 2009; Harrison et al. 2014; Vivarelli 2014). We make a distinction depending on whether innovation strategies can be characterized as explorative or exploitative (March 1991), where the former can be characterized as having a search scope in order to invent new products that presently are not available on the market. The latter strategy emphasizes improvement rather than novelty of products or processes.

Combining patent application with patent classification, we create a knowledge profile for each firm at the two-digit level over a period of 15 years. If a firm did not apply for a patent in the same patent class during the last 5 years, it is defined as an explorative innovation; otherwise, it is an exploitative innovation.¹ Altogether, our dataset includes more than 480,000 firms from the manufacturing and service industries, and we estimate the effects innovation has on firm-growth using the system GMM-model.

Comparing the two strategies, we find that firms engaged in explorative innovations enjoy a stronger employment growth, but the differences are not drastic. The results support a persistent employment growth effect, however, only in the case of explorative innovation, while displacement effects seem to prevail for exploitative innovations.

II. Hypothesis and econometric strategy

Based on previous research findings we test the following hypotheses: i) employment is positively related to both explorative and exploitative innovative activities, ii) employment effects are more pronounced for explorative innovative activities iii) employment effects to persist over several periods.

We embark on a standard log-linear employment equation (Layard and Nickell 1986), modified to first differences in order to eliminate firm fixed effects and to incorporate both current and lagged measures of our two innovation measures,

(1) $\begin{array}{r}\mathrm{\Delta }{n}_{it}=\alpha {\hspace{0.17em}}_1\mathrm{\Delta }{n}_{i\left(t-1\right)}+{\alpha }_2\mathrm{\Delta }{n}_{i\left(t-2\right)}+{\beta }_1\mathrm{\Delta }{w}_{it}+{\beta }_2\mathrm{\Delta }{k}_{it}+{\beta }_3\mathrm{\Delta }y{s}_{it}\\ +{\beta }_{4}Exploitativ{e}_{it}+{\beta }_{5}Explorativ{e}_{it}+{\beta }_{6}Exploitativ{e}_{i\left(t-1\right)}\\ +{\beta }_{7}Explorativ{e}_{i\left(t-1\right)}+{\beta }_{8}Exploitativ{e}_{i\left(t-2\right)}\\ +{\beta }_{9}Explorativ{e}_{i\left(t-2\right)}+{\mathbf{X}}_{\mathbf{i}\mathbf{t}}^{\prime }\delta +\mathrm{\Delta }{\epsilon }_{it}\end{array}$(1)

where $\mathrm{\Delta }{n}_{it}=n_{it}-n_{i\left(t-1\right)}$ is the first difference in the logarithm of employment of firm i at year t. All other continuous variables are defined in the same way. We control for the wage rate $w_{it}$, gross fixed capital $k_{it}$ and industry demand $y{s}_{it}$² Vector X contains dummy variables for ownership structure,³ sub-industries,⁴ years and regions.⁵ Finally, ${\epsilon }_{it}$ is the error term, expected to exhibit standard properties.

Equation (1)(1) $\begin{array}{r}\mathrm{\Delta }{n}_{it}=\alpha {\hspace{0.17em}}_1\mathrm{\Delta }{n}_{i\left(t-1\right)}+{\alpha }_2\mathrm{\Delta }{n}_{i\left(t-2\right)}+{\beta }_1\mathrm{\Delta }{w}_{it}+{\beta }_2\mathrm{\Delta }{k}_{it}+{\beta }_3\mathrm{\Delta }y{s}_{it}\\ +{\beta }_{4}Exploitativ{e}_{it}+{\beta }_{5}Explorativ{e}_{it}+{\beta }_{6}Exploitativ{e}_{i\left(t-1\right)}\\ +{\beta }_{7}Explorativ{e}_{i\left(t-1\right)}+{\beta }_{8}Exploitativ{e}_{i\left(t-2\right)}\\ +{\beta }_{9}Explorativ{e}_{i\left(t-2\right)}+{\mathbf{X}}_{\mathbf{i}\mathbf{t}}^{\prime }\delta +\mathrm{\Delta }{\epsilon }_{it}\end{array}$(1) is estimated using system-GMM techniques, which implies that lagged variables are used as instruments to control for potential endogeneity. First, the lagged dependent variable $\mathrm{\Delta }{n}_{i\left(t-1\right)}$ is potentially correlated with the error term $\mathrm{\Delta }{\epsilon }_{it}$ and therefore risk introduces endogeneity in the estimations. We use $n_{i\left(t-3\right)}$ and earlier realizations of $n_{it}$ as instruments for the first difference lagged dependent variable $\mathrm{\Delta }{n}_{i\left(t-1\right)}$ (Lachenmeier and Rottman 2011).

Second, one might consider the endogeneity of our two innovation variables. As suggested by Lachenmeier and Rottman (2011), innovation decisions are often based on long-term considerations, while employment decisions are based on more short-term considerations. Hence, we assume that innovation decisions are made at least one period before employment decisions and instrument our innovation variables with one-period lagged level values, assumed uncorrelated with the error term.⁶

III. Data and descriptive statistics

Data are provided by Statistics Sweden’s Business Register and covers all registered firms and establishments in Sweden since 1987 and the European Patent Office’s PATSTAT database, supplemented with data from the Swedish Patent Office. Pooling firm-level data and patent application data leave us with a sample of 2,159,666 observations for 482,513 firms across 20 industries for the period 1997 to 2012.

For all firms, a patent history profile is created based on the patents the firm applied for during a five-year moving window⁷ prior to any given year⁸ which enables us to categorize innovations in the following way:

• Exploitative innovation: A dummy variable is equal to one if a firm applies for a patent in year t in a patent class where it has already applied for a patent for the last 5 years, zero otherwise.

• Explorative innovation: A dummy variable is equal to one if a firm applies for a patent in year t within a patent class where it has not applied for a patent for the last 5 years, zero otherwise.

Table 1 provides descriptive statistics.⁹

Table 1. Descriptive statistics.

Variables	Mean	Std. Dev.	Min	Max
Employment	11.3	124.9	1	23,588
Employment growth	0.027	0.305	–0.916	1.099
Explorative innovation	0.0012	0.0345	0	1
Exploitative innovation	0.0018	0.0429	0	1
Annual wage	202	142	0	33,100
Physical capital	7,778	218,000	0	72,800,000
Sector gross value added	139,000,000	119,000,000	437	447,000,000

Number of observations: 2,159,666. All monetary variables are expressed in deflated thousands of SEK.

IV. Empirical results

The results from the regressions are shown in Table 2. Based on specifications 1 and 2, we find that both exploitative and explorative innovations have positive and significant effects on subsequent firm employment growth. The strength, however, decreases over time for both types of innovations. For exploitative innovations, it turns negative after 2 years, albeit insignificant, indicating that firms may enjoy efficiency gains already in the short term, while it may take time before it results in a reduced workforce.¹⁰

Table 2. Results, regression coefficients.

	(1)	(2)	(3)
L1. Δ log employment	0.126	0.103	0.0963
	(0.90)	(0.98)	(0.90)
L2. Δ log employment	0.0162	0.0212	0.0220
	(0.93)	(1.49)	(1.53)
Exploitative innovation	0.0232*		0.00309
	(1.92)		(0.24)
L1. Exploitative innovation	0.0170*		0.00405
	(1.75)		(0.38)
L2. Exploitative innovation	–0.00135		–0.00597
	(–0.15)		(–0.65)
Explorative innovation		0.0347***	0.0336***
		(3.19)	(2.86)
L1. Explorative innovation		0.0262***	0.0245**
		(2.77)	(2.25)
L2. Explorative innovation		0.0218**	0.0220**
		(2.49)	(2.19)
Δ Log wage	–0.0275***	–0.0540***	–0.0537***
	(–6.93)	(–11.92)	(–11.64)
Δ Log physical capital	0.00327***	0.00213***	0.00213***
	(18.77)	(22.50)	(22.52)
Δ Log sector gross value added	0.00464***	0.00304*	0.00306*
	(3.06)	(1.88)	(1.90)
Constant	–0.0354***	–0.0514***	–0.0515***
	(–6.84)	(–9.58)	(–9.63)
No. of observations	1,173,311	1,173,311	1,173,311
Sargan value	52.17	83.88	84.5
Sargan p-value	0.833	0.998	0.996
AR(1)	–4.42	–5.72	–5.57
AR(1) p-value	0.000	0.000	0.000
AR(2)	1.33	1.52	1.43
AR(2) p-value	0.183	0.129	0.154

Note: *** denotes 0.1% significance; ** denotes 1% significance; * denotes 5% significance. L. denotes the lag operator and, hence, L1. stands for one-year lag etc. All specifications include dummy variables for ownership, year, region and industry.

Explorative innovations are more likely to come up with new products and processes and generate an increase in labour demand. While both explorative and exploitative innovations are included in the regressions simultaneously, only the former remains statistically significant.

All control variables (wage rate, capital stock and sector gross value added) have the expected signs and are highly significant. The test statistics support the validity of the system-GMM method. The Sargan test does not reject the null hypothesis that our instruments are exogenous, while the AR(2) test does not reject the null hypothesis of no second-order autocorrelation at the 5% significance level.¹¹

V. Conclusion

Using patent data, we have shown that explorative innovations have a more pronounced and persistent effect on employment growth than exploitative innovations.¹² This type of innovation adheres more closely to Schumpeter’s early view on the role of the entrepreneur in initiating creative destruction processes.

The different impact of exploitative and explorative innovations on firm growth has important implications for government policies. Policies encouraging novelty may be rewarding for society in terms of long-term employment growth. An essential part of such policies is acceptance of failure and exit rules that allow for a second chance. Well-developed financial markets and access to venture capital may be another pillar. Moreover, as shown in numerous previous studies, only stimulating R&D may not be an optimal policy instrument and may even deter high-tech entry (Acemoglu et al. 2013).

We end this paper with two suggestions for future research on the link between innovation and employment growth. First, to test if and how the results are affected by excluding firms that never apply for a patent. Second, assess how productivity might play a mediating role in linking innovation and employment growth.

Acknowledgments

We would like to thank Ding Ding for excellent support with the econometric part of the study.

Disclosure statement

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

Data availability statement

The data that support the findings of this study are available from Statistics Sweden (SCB). Restrictions are applied to the availability of these data, which are used under licence for this study. Further information on how to access these data can be found at:

https://www.scb.se/en/services/ordering-data-and-statistics/guidance-for-researchers-and-universities/

Additional information

Funding

This work was supported by Marianne and Marcus Wallenberg’s Foundation.

Notes

¹ Akcigit and Kerr (2018), introducing similar concepts in a theoretical model, provide some correlations.

² Industry demand is measured as industry-aggregated gross value added. All monetary variables are measured in fixed prices..

³ The influence of types of ownership on employment growth and innovation are inconclusive (see e.g. Barba Navaretti 2004; Dachs and Peters 2014).

⁴ See Herstad (2018). We use the first level of NACE Rev. 2 to identify 20 industries.

⁵ We use 72 labour market regions (FA-regions) as our spatial unit of measurement.

⁶ The validity of this assumption is subsequently tested using the Sargan test.

⁷ The depreciation rate for knowledge capital is five years (Griliches 2007).

⁸ The patent applications classes (121) are determined at the two-digit level of International Patent Classification (IPC). See Bloom, Schankerman, and Van Reenen (2013) for a similar approach.

⁹ See Daunfeldt and Halvarsson (2015) for details regarding the employment-growth distribution of Swedish firms.

¹⁰ Labour regulations may also deter laying off employees.

¹¹ The results are virtually unchanged when using a three-year window of innovation or a Heckman selection model to control for survival effects. We expect them to be robust to a separation on patenting and non-patenting firms, see Balasubramanian and Sivadasan (2011).

¹² For a brief discussion of alternative innovation measures, see Colombelli, Krafft, and Quatraro (2014).