Internationalisation of firms and their innovation and productivity

^†The views expressed in this paper are purely those of the authors may not in any circumstances be regarded as stating an official position of the European Commission.

Abstract

This paper examines the links between the internationalisation of firms and their innovation and productivity performance using data from Ireland over the period 2004–2008. Our econometric results indicate that, relative to firms that served the domestic market only, firms with international activities were more likely to invest in innovation, they were more likely to be successful in terms of innovation output, and they had a higher labour productivity. In line with the most recent literature on international trade with heterogeneous firms, our empirical evidence shows that, among firms with international activities, labour productivity was higher in foreign affiliates in comparison to domestic exporters.

Keywords:

• multinational firms
• exporting
• knowledge production
• productivity

JEL Classification::

• F10
• F23
• O31

1. Introduction

The most recent international trade and investment literature has established that firms with international activities have a higher productivity in comparison to firms that serve only the domestic markets. Existing empirical evidence shows that foreign-owned firms are more productive than domestic firms (Driffield 1997; Doms and Jensen 1998; Griffith 1999; Griffith and Simpson 2001; Ruane and Ugur 2005; Girma and Görg 2007). More recent studies have found that a large part of this productivity differential is between multinationals and non-multinationals (Temouri, Driffield, and Higon 2008). Recent theoretical models of international trade with firm heterogeneity demonstrate that, given fixed costs associated to entry on export markets, only firms with high productivity self-select into exporting (Bernard and Jensen 1999; Melitz 2003). While this literature has assumed that firm productivity is exogenous, more recent theoretical contributions allow for the possibility of firms to increase their productivity through innovation activities (Yeaple 2005; Bustos 2011).

Empirical evidence on the sources of the productivity advantage of firms with international activities relative to firms serving only domestic markets is still scarce. To fill this gap, this paper focuses on innovation as the source of the productivity advantage of firms with international activities.

A second strand of literature to which this paper is related is the endogenous growth theory which points to the role of knowledge and innovation on productivity growth (Griliches 1979; Aghion and Howitt 1998). Following the analytical framework proposed by Griliches (1979), Crépon, Duguet, and Mairesse (1998) have put forward an econometric model (known as the CDM model) to estimate the relationships between innovation input, innovation output and productivity at firm level. A number of analyses have followed using data from the Community Innovation Survey (CIS),¹ to estimate the CDM model in different countries:² Janz, Lööf, and Peters (2004) for Germany and Sweden; Benavente (2006) for Chile; Chudnovsky, López, and Pupato (2006) for Argentina; Griffith et al. (2006) for France, Germany, Spain and the UK; Jefferson et al. (2006) for China; Lööf, and Heshmati (2006) for Sweden; van Leeuwen and Klomp (2006) for the Netherlands; Raffo, L’ Huillery, and Miotti (2008) for Latin-American and European countries; Masso and Vahter (2008) for Estonia. Castellacci (2011a) estimated an augmented CDM model to analyse the effects of industry-level competition on firm-level innovation and productivity in Norway. A cross-country analysis including 18 OECD countries has been reported in OECD (2009).

However, these studies do not distinguish between firms with international activities and firms that serve only the domestic market. This paper aims to contribute to filling this gap by linking the productivity of firms with international activities to their innovation performance. To analyse the role of international linkages on innovation and productivity performance, we consider inward foreign direct investment and exporting. We use data for Ireland from the CIS conducted in 2006 and 2008. Given its high openness to international trade and investment, Ireland is a suitable country case for this analysis.

Table 1 shows key indicators of innovation performance in Ireland over the period 2004–2008.

Table 1. Innovation performance in Ireland 2004–2008.

	% of firms with innovation activities
All firms	47.2
10–49 employees	42.7
50–249 employees	62.5
250+ employees	74.9
Foreign-owned	64.7
Irish-owned	43.0
2004–2006
% of total turnover that is	% of total persons engaged who		% of firms engaged in
generated by firms with innovation activities	work in firms with innovation activities	% of firms with innovation expenditure	co-operation for innovation activities	% of firms with innovation activities
75.1	68.0	34.1	27.0	44.9
39.3	46.4	29.2	22.1	40.1
74.4	65.5	49.7	35.7	61.4
90.0	80.1	67.8	53.8	76.0
84.4	76.3	43.6	35.2	60.7
59.5	62.5	31.7	24.0	41.3
2006–2008
% of total turnover that is	% of total persons engaged who work in	% of firms	% of firms engaged
generated by firms with innovation activities	firms with innovation activities	with innovation expenditure	in co-operation for innovation activities
64.6	65.9	31.4	24.1
36.8	42.5	26.5	20.0
71.1	63.6	47.8	30.2
72.3	79.0	64.3	48.6
76.8	79.8	39.0	32.6
47.9	56.8	29.6	21.2

Source: CIS 2004–2006, CSO Ireland (2009) and CIS 2006-2008, CSO Ireland (2010).

Over this period, Ireland ranked 7th among the EU-27 countries with respect to firms’ innovation rates (CSO Ireland, 2009, 2010). Foreign-owned firms had higher innovation expenditures and innovation rates compared to Irish-owned firms. Innovation rates for all firms were 47.2% over 2004–2006 and 44.9% over 2006–2008. Innovation rates for foreign-owned firms were 64.7% and 60.7%, respectively, while innovation rates for Irish-owned firms were lower, 43.0% and 41.3%, respectively. As expected, larger firms had higher innovation rates in both foreign-owned and Irish-owned firms. The predominant innovation type for all firms was process innovation followed by organisational innovation. Innovation rates for product innovation were lower.

Total innovation expenditures were estimated at ≠uro4.6 billion in 2006 – of which 38% was spent on in-house R&D, and ≠uro5.3 billion in 2008 – of which just over 30% were spent on in-house R&D. Foreign-owned firms accounted for more than two thirds of all innovation expenditure (77% in 2006 and 68% in 2008).

Specifically, in this paper we ask the following research questions. Are firms with international activities more likely to invest in innovation and do they have a higher innovation expenditure intensity? Do firms with international activities innovate more than firms serving only the domestic market? Are firms with international activities more productive?

To answer these questions, we estimate an augmented structural model which builds on and expands previous research by Crépon, Duguet, and Mairesse (1998) and Griffith et al. (2006). This approach allows us to account for the role of international activities in explaining the innovation and productivity performance of firms in Ireland. In contrast with these two studies that are based on cross-sectional data, we use panel data from two waves of the CIS for Ireland for the period 2004–2008. The panel data allows us to account for unobserved heterogeneity and possible endogeneity in the analysed relationships and thus provide more robust evidence on the links between innovation input, innovation output and productivity.

In contrast to Crépon, Duguet, and Mairesse (1998) and many of other subsequent empirical studies, and similar to Griffith et al. (2006), we estimate the model for all firms and not only for innovative firms. In this way, we can account for the selection bias which arises from the fact that, while it is likely that all firms have some innovative effort, not all firms report innovation investment. In addition to using panel data, we go beyond Crépon, Duguet, and Mairesse (1998) as well as Griffith et al. (2006) in three ways. First, we add to the model explanatory variables which capture international activities. In particular, we estimate whether and to what extent foreign affiliates and domestic exporters have a different innovation and productivity performance in comparison to firms that serve only the domestic market. Second, we consider three types of innovation, namely product, process and organisational innovation,³ as well as complementarities among them. Third, we estimate an improved econometric model to account for three econometric issues: (i) selection bias due to the fact that the set of firms which report innovation investment might be non-random; (ii) endogeneity, due to innovation investment, innovation output and productivity being endogenously determined; in addition, we account for the fact that firms’ international activities, innovation and productivity might be determined jointly; and (iii) omitted variable bias.

Our research relates to Criscuolo, Haskel, and Slaughter (2010) who estimate a knowledge production function to analyse the role of firms’ global engagement on their innovation performance in the UK using data from two waves of the CIS over the period 1994–2000. In contrast to Criscuolo, Haskel, and Slaughter (2010), we model in addition to knowledge production two other distinct stages which are part of the innovation behaviour of firms: the decision to invest in innovation and the effect of innovation output on productivity. Castellani and Zanfei (2007) show that firms with international activities in Italy had better productivity and innovation performance in comparison to purely domestic firms. However, they use cross-sectional data and cannot account for the fact that productivity and innovation output may be simultaneously determined. Finally, our analysis goes beyond Doran and O'Leary (2011) who use cross-sectional data from the CIS for Ireland over the period 2004–2006 to estimate the relationships between innovation investment, innovation output and productivity without modelling the role of international activities.

Our econometric results indicate that, relative to firms that served the domestic market only, firms with international activities were more likely to invest in innovation, they were more likely to be successful in terms of innovation output, and they had a higher labour productivity. In line with the most recent literature on international trade with heterogeneous firms, our empirical evidence shows that, among firms with international activities, labour productivity was higher in foreign affiliates in comparison to domestic exporters.

The rest of the paper is organised as follows. Section 2 reviews the relevant theoretical and empirical literature and derives testable hypotheses. Section 3 discusses our empirical methodology and econometric issues. Section 4 describes the data set and summary statistics. Econometric results are discussed in Section 5. Section 6 concludes.

2. Theoretical and empirical background

The starting point for our analysis is the theoretical and empirical literature on international trade and industry growth with heterogeneous firms. Two classes of theoretical models could be distinguished. In the first group of models, firm's internationalisation activities are linked to exogenous static productivity. In these models, industry growth is the result of selection and reallocation mechanisms. The second group of models account for the possibility of productivity to be influenced by past investments in R&D.

Bernard and Jensen (1995, 1999) started a large literature focused on international trade and firm heterogeneity which has established that exporters tend to be more productive than non-exporters and that a large number of firms export only a small fraction of their output. Subsequently, Bernard et al. (2003) formalised these stylised facts into a theoretical model that links firm productivity, size, and export participation to the underlying variation in producer efficiency.

This trade model with heterogeneous firms is set up in a Ricardian framework with Bertrand competition. Industries have imperfect competition with variable mark-ups. Firms have different efficiency levels determined stochastically as a random draw from a Pareto distribution. Producers with higher efficiency/productivity charge a lower price and sell more in the domestic market as well as abroad. Openness and international trade affect industry productivity growth through three channels: (i) price competitiveness due to the availability of cheaper imports (lower prices of intermediate inputs); (ii) entry (exit) of plants with productivity higher (lower) than the industry average; (iii) reallocation of production among incumbents with different efficiency levels and related changes in firms’ market shares – more productive firms get larger market shares.

Melitz (2003) demonstrates that the outcomes in an industry context of monopolistic competition with heterogeneous producers are determined by a combination of firm-specific productivity levels and fixed (sunk) costs. A firm productivity level is determined as a random draw from a probability distribution and it is assumed time-invariant. Given the productivity distribution and the level of sunk costs, firms are classified in three groups: (i) those with productivity levels below the fixed production costs exit; (ii) those with productivity below a threshold required to export produce only for the domestic market; (iii) those with productivity above the threshold for exporting produce and sell both for the domestic and the international market. Industry dynamics is driven by reallocation and selection mechanisms and there is no productivity growth or technological change following firms’ innovation or imitation activities.

Further, the theoretical model by Melitz and Ottaviano (2008) demonstrates that the competition and selection mechanism driving industry growth is endogenously dependent on market size. The entry of foreign firms in the domestic market increases competition, the less productive firms exit the market, and average industry productivity increases. Market reallocation and growth are driven by increased product competition. A larger market leads to a lower industry cut-off level, higher aggregate productivity and product variety and lower mark-ups and prices.

Helpman, Melitz, and Yeaple (2004) demonstrate that, when foreign direct investment is motivated by market size, the most productive firms become multinationals, the next most productive export, the less productive serve only the domestic market, and the least productive, exit. Head and Ries (2003) show that when firms invest abroad for efficiency related reasons (factor prices), the least productive firms locate abroad in small countries whereas the more productive produce at home.

A large empirical literature⁴ provides evidence for these theoretical predictions. These model set ups ignore productivity dynamics. Another class of theoretical models, known as ‘active learning models’,⁵ allow the possibility that firms invest actively in R&D to improve their productivity (Ericson and Pakes 1995; Luttmer 2007).

There is less, but growing empirical evidence for this class of theoretical models. Bernard and Wagner (1997) and Bernard and Jensen (1999) provide evidence showing that export participation leads to future higher productivity. In addition, Aw, Roberts, and Winston (2007) show that this export effect on productivity is larger if the firm had previously invested in R&D. Constantini and Melitz (2008) and Iacovone and Javorcik (2010) show that both innovation performance and export activity are linked to previous firm's decisions to invest in R&D. Damijan, Kostevc, and Polanec (2010) find evidence that in the case of medium and large firms in Slovenia, past exporting led to process innovation and productivity growth. In contrast, they find no impact of previous exporting on the probability to introduce product innovations. However, Bratti and Felice (2012) uncover a robust positive relationship between the export status and the probability to introduce product innovations in Italian manufacturing firms. Halpern and Muraközi (2012) find evidence for Hungary showing that innovative firms are more productive and they are more likely to engage in exporting.

On the basis of these theoretical and empirical findings we can derive three testable hypotheses:

• (i) Productivity is positively linked to previous investment in R&D and innovation;

• (ii) Past exporting is positively associated with the probability to innovate;

• (iii) Among internationalisation modes, foreign affiliates have higher productivity than firms engaged in exporting.

The CDM model provides an unified econometric framework and it is well suited to capture the complex relationships between internationalisation activities of firms, innovation and productivity discussed above. We present below our augmented version of CDM model specifications to analyse these links.

3. Model specification and econometric issues

To explain the innovation and productivity performance differential of firms with international linkages we estimate an augmented structural model by extending the econometric framework proposed by Crépon, Duguet, and Mairesse (1998) and Griffith et al. (2006). This modelling framework accounts for the following firm behaviour: in the first stage, firms decide whether and how much to invest in innovation; in the next stage, firms produce knowledge (innovation outputs) using innovation inputs; finally, knowledge (innovation outputs) is used together with other inputs to produce final output.

This model consists of the following equations:

The first equation models the decision of firm i to invest in innovation: (1) where is an unobserved latent variable measuring the predicted utility of engaging in innovation, x_1it is a vector of firm-level characteristics, β is the related vector of coefficients, λ_j is a vector of industry fixed effects, μ_t is a vector of time fixed effects and the error term.

To account for the fact that we only observe what the firms report as innovation effort, we estimate the following selection equation which describes the propensity of firms to invest in innovation: (2) where y_it is a binary variable equal to one for firms reporting innovation expenditure and zero for firms without innovation expenditure. We explain the propensity of firms to invest in innovation as a function of firm characteristics (international linkages, firm size, human capital intensity, obstacles to invest in innovation⁶ as well as time specific and industry specific effects.

Further, conditional on investing in innovation we estimate the innovation expenditure intensity (w_it) as follows: (3)

We assume that and ϕ_it are bivariate normal with zero mean, variances and and the correlation coefficient . Following this assumption, we estimate the system of Equations (2) and (3) as a generalised Tobit model by maximum likelihood using the Heckman two-step procedure. We explain innovation expenditure intensity measured as innovation expenditure per employee as a function of firm characteristics (international linkages, firm size, human capital intensity) as well time specific and industry specific effects.⁷

To alleviate potential endogeneity concerns related to the simultaneous determination of the decision to invest in innovation, internationalisation activities and firm performance, all Equations (2) and (3) are estimated with lagged values of the main explanatory variables (foreign ownership, exporting, size, human capital intensity).

Further, to analyse innovation outputs, we estimate the following knowledge production function: (4) where z_it is a measure of innovation output (product, process, organisational innovations, the innovative turnover share). We explain innovation outcomes as a function of predicted innovation expenditure intensity , firm characteristics (international linkages, firm size), external knowledge flows from co-operations for innovation with other enterprises and institutions⁸ as well as time-invariant unobserved industry specific effects (λ_j) and common time specific effects (μ_t). is obtained on the basis of Equations (2) and (3) for all firms, and hence Equation (4) is estimated for all firms in the sample. This procedure allows the estimates to be free from selection bias. In addition, by using the predicted innovation input as an explanatory variable in the innovation output equation we alleviate the endogeneity arising from the fact that innovation investment and innovation output may be determined simultaneously. For example, innovation investment may be correlated with the error term if part of this innovation input is attributed to unobserved firm-specific effects. To alleviate potential endogeneity concerns related to the simultaneous determination of innovation output, internationalisation activities and firm performance, we estimate Equation (4) with lagged values of the main explanatory variables (foreign ownership, exporting, size, human capital intensity).

The last equation is an augmented Cobb-Douglas production function with constant returns to scale: (5) π_it is a measure of labour productivity (turnover per employee) in firm i at time t and is explained as a function of predicted innovation (knowledge) output , firm characteristics (international linkages and firm size) as well as time-invariant unobserved industry specific effects (λ_j), and common time specific effects (μ_t). Predicted innovation output is obtained from Equation (4). We thus account for potential endogeneity issues arising from the fact that labour productivity and innovation output might be determined simultaneously. To alleviate potential endogeneity concerns related to the simultaneous determination of productivity, internationalisation activities and firm performance, we estimate Equation (5) with lagged values of the main explanatory variables (foreign ownership, exporting, size).⁹

The model is estimated as a recursive system consisting of Equations (2)–(5). Given that not all firms are surveyed in both periods, we estimate weighted regressions, with weights calculated using the distribution of employment across industries.¹⁰ In addition, we estimate standard errors that are clustered at industry level to account for the fact that error terms may be correlated within industries. For example, it is likely that firms belonging to the same industry share a common part of the utility (or production) functions described by Equations (2)–(5). Usually, this common part is unobservable and it enters the error term in each equation. The consequence is that error terms are correlated within industries. As shown by Moulton (1986, 1990) this correlation leads to downward biased standard errors and thus spurious statistical significance. To account for this bias, we follow Pepper (2002) and compute standard errors clustered at NACE two-digit industry level.

4. Data and summary statistics

We use data from two waves of the Irish CIS 2006 and CIS 2008. Ireland's CIS 2006 and CIS 2008 were jointly conducted by Forfás¹¹ and the Central Statistics Office (CSO) of Ireland in 2006 and 2008. For these two surveys, information on 1974 and 2181 firms, respectively, were obtained, yielding response rates of 47.6% and 46.9%, respectively. CIS 2006 covers the innovation activities of firms from 2004 to 2006 and CIS 2008 covers those from 2006 to 2008. We construct a balanced panel of firms in the manufacturing and service sectors¹² appearing in both surveys. This panel data consists of 723 firms. With respect to international linkages, we identify the following groups of firms: 245 foreign-owned firms (34%), 282 domestic exporters (39%) and 196 domestic non-exporters (27%).

In the CIS, firms report whether they are in an enterprise group and whether they sell goods or services to local, national or foreign markets. Further, firms are asked whether they have introduced product, process or organisational innovation, or have on-going innovation or abandoned innovation during a previous three-year period. Only those firms that had successful, on-going or abandoned innovation activities (termed as innovators) were asked to answer more questions in relation to their R&D and other innovation expenditures, and co-operation for innovation activities with other enterprises or institutions over the three-year period. Information on ownership, turnover and the number of employees were added to the dataset from other surveys conducted by Forfás and the CSO.

Detailed definitions and data sources for each variable are given in Table A1. Table 2 reports summary statistics for all firms and separately for foreign-owned, domestic exporters and domestic non-exporters.

Table 2. Summary statistics of the panel sample: manufacturing and services.

CIS 2006 and 2008
	Innovators and non-innovators
Types of firms	N=723 All firms	N=245 Foreign-owned firms	N=282 Domestic exporters	N=196 non-exporters
1. Innovation input
Engagement in in-house R&D (per cent)	30.8	39.1	38.8	8.9
In-house R&D expenditure per employee (euro in 2004 price)	2722.7	4915.5	2504.6	307.3
Engagement in innovation (per cent)	44.5	53.0	52.3	22.9
Innovation expenditure per employee (euro in 2004 price)	6953.3	12,405.1	5484.8	2277.5
2. Innovation output
Any type of innovation (per cent)	64.7	77.3	68.6	43.3
Product innovation (per cent)	41.6	53.0	47.3	19.3
Process innovation (per cent)	41.6	50.9	45.0	24.9
Organisational innovation (per cent)	47.0	57.3	51.2	28.2
All types of innovation (per cent)	22.8	29.4	27.1	8.4
Product and process innovation (per cent)	27.6	36.0	31.9	10.9
Product and organisational innovation (per cent)	29.3	37.8	34.4	11.5
Process and organisational innovation (per cent)	31.5	39.5	35.8	15.3
Innovative turnover share (per cent)	10.0	13.4	11.8	3.1
3. Knowledge flows
Other enterprises within the same enterprise group (per cent)	11.6	24.7	6.0	3.3
Suppliers (per cent)	12.7	18.8	12.9	4.8
Clients or customers (per cent)	10.4	15.1	11.7	2.5
Competitors (per cent)	4.1	5.5	3.7	2.8
Consultants, commercial labs or private R&D institutes (per cent)	8.3	12.3	9.2	2.0
Universities or other higher education institutions (per cent)	8.2	13.9	8.0	1.5
Government or public research institutes (per cent)	4.6	5.7	6.4	0.8
4. Hampering factors
Lack of internal funds	13.8	10.4	17.9	12.0
Lack of external funds	8.7	5.1	11.9	8.7
Innovation costs too high	11.2	8.4	14.2	10.4
Lack of qualified personnel	7.1	6.1	9.0	5.6
Lack of technology information	2.8	1.6	3.9	2.8
Lack of market information	3.7	1.4	6.6	2.3
Difficulty to find co-operation partners for innovation	3.7	2.5	5.0	3.6
Market dominated by incumbents	13.9	1.4	15.8	10.7
Uncertain demand for innovative goods or services	11.6	8.8	15.2	9.9
Need to meet government regulation	8.9	7.6	8.3	11.2
Excessive economic risks	6.8	5.3	8.5	6.4
5. Other firm characteristics
Labour productivity (turnover per employee, 2004 prices)	738,219.0	1,430,688.0	459,804.0	276,155.2
Employees	153.3	238.9	130.9	79.1

Notes: Innovators are firms that report having at least one of the following types of innovation: product, process or organisational innovation. Firms reporting no innovation are considered non-innovators. Types of firms include: foreign-owned firms (as indicated in the original survey), domestic exporters (Irish-owned firms that export to Northern Ireland, the UK, the EU or other countries in the world) and domestic non-exporters (Irish-owned firms that do not export).

With respect to innovation inputs, on average, 30.8% of all firms reported positive spending on in-house R&D. The average R&D expenditure per employee in 2004 prices was ≠uro2723. Foreign-owned firms had the highest propensity to invest in innovation as well as innovation expenditure intensity, while domestic non-exporters had the lowest figures, as expected. About 44.5% of all firms reported innovation expenditure and the intensity of innovation expenditures was more than twice that of in-house R&D expenditures. These descriptive statistics suggest that a large proportion of innovation expenditures was spent on obtaining external knowledge, such as purchase of external R&D, acquisition of machinery, equipment and software and other external knowledge.

Turning to innovation output, it appears that 64.7% of firms in our sample implemented innovations over the period 2004–2008. Firms with international activities had higher innovation rates than firms that served only the domestic market. While 77.3% of foreign-owned firms and 68.6% of domestic exporters reported innovation outputs, the innovation rate for domestic non-exporters was 43.3%. The predominant innovation type was organisational innovation, reported by 47.0% of firms, while 22.8% of firms implemented all three innovation types.

It is noteworthy that, while 64.7% of firms indicated that they had innovation output (any innovation type), only 44.5% of firms reported innovation expenditures. This fact can be explained by two situations: (i) innovation output in a number of firms, in particular foreign-owned firms, is linked to knowledge produced outside Ireland; (ii) some firms tend not to report innovation expenditure if this was below a certain threshold, as suggested by Griffith et al. (2006). With respect to external knowledge flows, firms with international activities were much more likely to engage in co-operation for innovation in comparison with domestic non-exporters. On average, 11.6% of firms reported co-operation for innovation activities with other enterprises in the same enterprise group and the share is much higher for foreign-owned firms, 24.7%. This can be seen as evidence of the advantage of being in an international enterprise group, in terms of giving a firm more chances to access external knowledge. In comparison, a lower proportion of domestic firms co-operate for innovation activities with enterprises within the same group. Foreign-owned firms rank first with respect to all the other external knowledge flows with the exception of government and public research institutes-sourced knowledge. Finally, it appears that on average foreign-owned firms are more productive than the other types of firms.

Foreign-owned firms appear to face to a less extent obstacles to innovation and innovation activities relative to domestic-owned firms. Lack of internal funds appears to be the most important factor preventing enterprises from innovating. This obstacle was of high importance for 18% of domestic exporters, 12% of domestic non-exporters and 10% of foreign-owned enterprises.

5. Empirical results

5.1. Innovation input

Table 3 shows the estimates for the propensity to invest in innovation and the innovation expenditure intensity obtained with a Heckman two-step estimator by using information available for all firms. The figures shown in the table are marginal effects obtained from weighted regressions with standard errors clustered at industry level. The dependent variable in the selection equation is a binary variable which is equal to one for firms with reported innovation expenditures¹³ and zero for firms with no innovation expenditures.

Table 3. Innovation input: manufacturing and services.

		Intensity of
	Propensity of investing	innovation input
	in innovation	Log of innovation
Dependent variable	Innovation expenditure >0	expenditure per employee
Estimator	Heckman first stage: Probit	Heckman second stage: OLS
Foreign-owned firm (lagged)	0.123*** (0.034)	0.345* (0.210)
Domestic exporter (lagged)	0.207*** (0.047)	−0.002 (0.183)
Size (lagged)	0.105*** (0.012)	−0.107** (0.051)
Human capital intensity (lagged)	0.116** (0.046)	0.413* (0.240)
Time fixed effect	0.038 (0.026)	−0.269* (0.140)
Lack of internal funds	0.151** (0.073)
Lack of external funds	−0.081 (0.077)
Costs too high	−0.093 (0.065)
Lack of qualified personnel	0.085 (0.071)
Lack of technology information	0.063 (0.096)
Lack of market information	−0.014 (0.086)
Difficulty to find co-operation partners	0.156* (0.090)
Market dominated by incumbents	0.043 (0.046)
Uncertain demand	0.078 (0.069)
Need to meet government regulation	0.149* (0.087)
Excessive economic risks	−0.022 (0.063)
Industry fixed effects	yes	yes
Wald test for joint significance (p-values)	0.000	0.000
Constant	yes	yes
Log likelihood	−3491.4
λ	0.039 (0.726)
ρ	0.027 (0.501)
Wald test for H0: ρ=0 (p-value)	0.000
N	1359

Notes: Figures reported in the table are marginal effects. Estimates are obtained from weighted regressions. Standard errors shown in parentheses are clustered at industry level (NACE rev. 1.1, 2-digit). The dependent variable of the propensity to invest in innovation equation is equal to 1 if a firm reports positive innovation expenditures and 0 otherwise. Foreign-owned firm, domestic exporter and obstacles to innovation are dummy variables. Size is measured as numbers of employees. Human capital intensity is defined as wages per employee. λ is the coefficient of the inverse mills ratio. ρ is the coefficient of the correlation of the error terms of the two equations measuring the interdependence of the two equations. OLS, ordinary least squares.

*Significant at the 10% level.

**Significant at the 5% level.

***Significant at the 1% level.

As shown in column 1, firms with international activities were more likely to invest in innovation. More specifically, on average, in comparison to firms that served only the Irish market, foreign-owned firms were more likely to invest in innovation by 12 percentage points and domestic exporters by 21 percentage points, respectively. In addition, in line with the relevant literature discussed in Section 2, we find that the propensity to invest in innovation increased with firm size and human capital intensity. With respect to obstacles to innovation activities, we find that the perceived lack of internal funds, difficulty to find co-operation partners and the need to meet government regulations were positively and significantly related to the propensity of firms to invest in innovation. The first two positive associations could be explained by the fact that perceptions about such obstacles to innovation are likely to be encountered by innovative firms that have invested in innovation. The third positive association suggests that government regulations act as a demand-related driver for innovation. None of the other hampering factors were statistically significant in relation to the propensity to invest in innovation. Further, we find that foreign-owned firms had a significantly higher innovation expenditure intensity in comparison to firms that served only the domestic market. Innovation expenditure intensity increased with human capital intensity and decreased with firm size. On average, innovation expenditure intensity was lower in the second survey period.

5.2. Innovation output

Table 4 (panel A) shows the probit estimates of the determinants of firms’ propensity to implement innovations. The figures shown in the table are marginal effects obtained from weighted regressions with standard errors clustered at industry level. The three columns show the results obtained from separate probit regressions for product, process and organisational innovation, respectively.

Table 4. Innovation output: manufacturing and services.

Dependent variable	Product innovation	Process innovation	Organisational innovation
Estimator	Probit	Probit	Probit
Panel A
Foreign-owned firm (lagged)	0.143** (0.066)	0.036 (0.056)	0.041(0.051)
Domestic exporter (lagged)	0.228*** (0.032)	0.053 (0.050)	0.051 (0.040)
Predicted innovation expenditure per employee	0.043 (0.131)	0.033 (0.094)	0.100 (0.075)
Size (lagged)	0.067*** (0.013)	0.093*** (0.015)	0.066*** (0.015)
Co-operation with other enterprises	0.089 (0.072)	0.090* (0.048)	0.132* (0.078)
Co-operation with suppliers	0.176*** (0.047)	0.200*** (0.077)	0.202*** (0.062)
Co-operation with customers	0.315*** (0.065)	0.101 (0.074)	−0.026 (0.064)
Co-operation with competitors	0.127 (0.140)	0.168* (0.102)	−0.002 (0.110)
Co-operation with consultants	−0.001 (0.098)	0.194** (0.090)	0.249*** (0.081)
Co-operation with universities	0.287*** (0.060)	0.277*** (0.094)	0.179** (0.084)
Co-operation with government	0.163 (0.138)	−0.068 (0.105)	0.251*** (0.093)
Time fixed effect	−0.063 (0.039)	0.039 (0.032)	0.016 (0.030)
Industry fixed effects	yes	yes	yes
Wald test for joint significance (p-values)	0.000	0.000	0.000
Constant	yes	yes	yes
Log-likelihood	−1413.8	−1488.2	−1512.9
Pseudo R^2	0.222	0.174	0.163
N	1356	1356	1356

In comparison to firms serving only the domestic market, foreign-owned firms and domestic exporters were more likely to implement product innovations. On average, foreign-owned firms were more likely to have product innovations by 14 percentage points, and domestic exporters by 23 percentage points, respectively. It appears that innovation expenditure intensity had no significant effect on the innovation output over and above other determinants such as international activities, firm size and external knowledge flows, as well as unobserved industry, and time specific effects. Larger firms were more likely to implement product, process or organisational innovations. Engagement in co-operation for innovation activities increased the probability to innovate. For all three types of innovation, knowledge flows from co-operations with suppliers, and with universities or other higher education institutions were positively associated with the probability to implement innovations over and above other determinants. Process and organisational innovations were more likely when firms engaged in co-operation with other enterprises within the same enterprise group, and with consultants, commercial labs or private R&D institutes. Further, firms engaged in co-operation for innovation with customers were more likely to implement product innovations, while co-operation with the government and public institutes increased the probability to implement organisational innovations.

Table 4 (panel B) reports the results of estimates of the innovation output separately for various combinations of innovation outputs: product and process innovations; product and organisational innovations; process and organisational innovations; product, process and organisational innovations. The fourth column reports the probit estimates for the implementation of all three types of innovations and the last column reports estimates of regressions using a measure of innovation intensity, namely, the innovative turnover share. To account for the fact that this latter variable is bounded between 0 and 1, we use in regressions a modified version of this dependent variable obtained from a logit transformation.

Foreign-owned firms and domestic exporters were more likely than firms serving only the domestic market to implement product combined with process innovations as well as all three types of innovation combined. Larger firms and firms engaged in co-operation for innovation activities with suppliers and with universities were more likely to have any of the three combinations innovation type innovations as well as all three innovation types together. Co-operation for innovation activities with enterprises within the same group, and co-operation with the government and other public institutes increased the probability to implement product and organisational innovations, while the probability to implement product and process innovation was higher for firms engaged in co-operation with customers and with competitors.

As shown in the last column in Table 4 (panel B), firms with international activities had a higher innovative turnover share in comparison to firms that served only the domestic market. Over and above foreign ownership and exporting, the innovation intensity increased also with firm size, and with engagement in co-operation for innovation activities with suppliers, with customers, with competitors, and with universities.

5.3. Productivity

Table 5 shows the estimates of the last equation of the model which explains labour productivity (log of turnover per employee) as a function of predicted innovation output, international activities, labour input, as well as unobserved industry and time specific effects. The figures shown in the table are marginal effects obtained from weighted regressions and standard errors clustered at industry level. Each column reports results from an OLS regression estimator obtained for each of the innovation output measures.

Table 5. Labour productivity: manufacturing and services.

	Log of turnover per employee
	Product	Process	Organisational				All types of	Innovative
Dependent variable	innovation	innovation	innovation	innovation	innovation	innovation	innovation	turnover share
Estimator	OLS	OLS	OLS	OLS	OLS	OLS	OLS	OLS
Foreign-owned firm (lagged)	0.551*** (0.092)	0.558*** (0.089)	0.538*** (0.088)	0.576*** (0.088)	0.560*** (0.087)	0.546*** (0.086)	0.568*** (0.087)	0.527*** (0.093)
Domestic exporter (lagged)	0.163** (0.072)	0.180*** (0.064)	0.171**0.065)	0.194*** (0.065)	0.172** (0.064)	0.172** (0.063)	0.183*** (0.064)	0.147** (0.068)
Predicted innovation output	0.214 (0.151)	0.358* (0.194)	0.448** (0.190)	0.154(0.172)	0.360** (0.165)	0.488** (0.193)	0.328* (0.185)	0.055 (0.034)
Size (lagged)	0.138*** (0.030)	0.119*** (0.035)	0.121*** (0.034)	0.141*** (0.034)	0.127*** (0.033)	0.116*** (0.034)	0.131*** (0.033)	0.136*** (0.031)
Time fixed effect	0.019 (0.043)	−0.003 (0.046)	0.010 (0.045)	0.005 (0.045)	0.011 (0.045)	−0.002 (0.045)	0.001 (0.045)	0.025 (0.045)
Industry fixed effects	yes	yes	yes	yes	yes	yes	yes	yes
F-test for joint significance (p-values)	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
Constant	yes	yes	yes	yes	yes	yes	yes	yes
Log-likelihood	−1716.0	−1714.7	−1713.5	−1716.4	−1714.2	−1712.8	−1714.9	−1715.0
R^2	0.285	0.286	0.288	0.285	0.287	0.288	0.286	0.286
N	1330	1330	1330	1330	1330	1330	1330	1330

Notes: Figures reported in the table are marginal effects. Estimates are obtained from weighted regressions. Standard errors in parentheses are clustered at industry level (NACE rev. 1.1, 2-digit).

*Significant at the 10% level.

**Significant at the 5% level.

***Significant at the 1% level.

Labour productivity in firms with international linkages was higher in comparison to firms serving only the domestic market. In addition, as predicted by the international trade and investment literature, we find that foreign-owned firms had a higher labour productivity than domestic exporters. Firms with international activities had also a higher innovative turnover share in comparison to firms serving only the domestic market. We uncover a positive link between innovation output and labour productivity for all types of innovation with the exceptions of firms with product innovation only and with product and process innovations combined. The effect of the innovation output on productivity is the highest in the case of firms with process and organisational innovation combined. We find no significant association between the innovation turnover share and labour productivity. Larger firms had a higher labour productivity.

6. Conclusions

This paper examined the effects of the internationalisation of firms on their innovation and productivity performance. We used micro data from two waves of the CIS in Ireland covering the period 2004–2008, and estimated a structural model to analyse the role of foreign ownership and exporting on firms’ innovation and productivity performance. The analysis distinguishes different types of innovation outputs such as product, process and organisational innovations as well as combinations of these innovation outputs.

Our econometric analysis uncovers the following key findings. Relative to firms that serve only domestic markets, firms with international activities were more likely to invest in innovation, and they were more likely to have innovation outputs. Furthermore, innovation intensity – measured as the share of turnover related to product innovations, was higher in firms with international activities relative to domestic firms without international markets. As predicted by the most recent literature on international trade with heterogeneous firms discussed in Section 2, we find that firms operating in international markets had a higher productivity than firms that served only the domestic market. Furthermore, foreign affiliates had a higher productivity than domestic exporters.

In addition, our analysis indicates that the propensity to invest in innovation increased with firm size and human capital intensity. In contrast, the innovation expenditure intensity was higher in smaller firms. Larger firms were more likely to introduce innovations. Firms engaged in co-operation for innovation activities were more likely to innovate successfully. Over and above internationalisation of firms, firm size and co-operation for innovation activities, innovation expenditure intensity was not significantly linked to innovation outputs. Finally, innovative firms had a higher productivity, over and above other firms characteristics such as international activities, size as well as unobserved industry characteristics. This relationship is found for all innovation types with the exception of product innovation and product combined with process innovation. The strongest innovation-productivity link appears to be for firms with process and organisational innovations.

Our research results suggest a number of policy implications. First, policies to incentivise firms’ innovation and productivity should be designed in connection to measures to assist firms to internationalise their production and services. Second, enabling and fostering co-operation with other enterprises and institutions is an important way to source knowledge in order to generate innovation output. Third, innovation expenditure per se does not translate into innovation output. It appears that in the case of Ireland, access to international markets and to external knowledge played a bigger role in the innovation performance of firms in comparison to investment in innovation. However, there might be lagged effects of innovation investment on the innovation output which are not captured in this analysis due to data limitations. Furthermore, our results might reflect innovation failures and the lack of absorptive capacity. Fourth, given the increased internationalisation of production as well as of innovation and R&D activities, innovation policies need to be designed in an international context.

Acknowledgements

This research was part of the IRCHSS-funded project: ‘Turning Globalisation to National Advantage: Economic Policy Lessons from Ireland's Experience’. This research uses statistical data from Forfás and the Central Statistics Office (CSO) of Ireland. The permission for controlled access to confidential micro data sets has been granted in line with the Statistics Act, 1993. The use of these statistical data does not imply the endorsement of Forfás or the CSO in relation to the analysis or interpretation of the statistical data. We thank Brian Cahill for his assistance to prepare the data set for this analysis. We also wish to thank for their helpful comments and suggestions on earlier versions of this paper two anonymous referees, the Editor, Cristiano Antonelli, Antoine Berthou, Helena Connellan, Jonathan Healy, Ian Hughes, Yessica Lagos, Claude Mathieu, Debbie Quinn, Frances Ruane, Andrew Stockman, Pierre Therrien, and participants in research presentations at the workshop on ‘Innovation – Developing Credible, Timely and Appropriate Metrics’ in Dublin, the European Trade Study Group conference in Lausanne, the Summer School at the University of Cambridge, the Economic and Social Research Institute (ESRI), Dublin, Forfás, the Leuven Centre for Irish Studies, Katholieke Universiteit Leuven, the ESRI-TCD Conference on Globalisation, in Dublin, the International Workshop on ‘Global and Local Firm Linkages’ at the Institute for Applied Economic Research, in Tübingen, the 3rd Central Statistical Office's Business Statistics Seminar in Dublin, and the European Economic Association Congress in Oslo.

Notes

^† The views expressed in this paper are purely those of the authors may not in any circumstances be regarded as stating an official position of the European Commission.

1. This survey is part of a harmonised framework across European Union's countries co-ordinated by EUROSTAT – the statistical office of the European Union – for the purpose of investigating the innovation performance of firms, and providing a cross-country comparison.

2. These studies are reviewed by Hall and Mairesse (2006) and Mairesse and Mohnen (2010).

3. While marketing innovations were covered in the CIS 2008, they were not covered in the CIS 2006.

4. Recent micro-econometric evidence has been surveyed by Helpman (2006), Bernard et al. (2007), Greenaway and Kneller (2007), and Wagner (2007).

5. Castellacci (2011b) reviews this class of models.

6. We distinguish the following obstacles to invest in innovation: lack of funds within the enterprise or group; lack of external funds; innovation costs too high; lack of qualified personnel; lack of information on technology; difficulty to find co-operation partners for innovation; market dominated by established enterprises; uncertain demand for innovative goods or services; need to meet government regulations; excessive perceived economic risks.

7. The excluded variables in the second stage are the variables which account for various obstacles to innovation.

8. We distinguish the following types of co-operation for innovation: with enterprises within the same enterprise group; with suppliers of equipment, materials, components or software; with clients or customers; with competitors, commercial labs or private R&D institutes; with universities or other higher education institutions; with Government of public research institutes.

9. While capital intensity impacts on productivity, firm-level data on capital stocks are not available.

10. Weights are calculated separately for each of the following employment size classes: 1–19 employees; 20–49 employees; 50–249 employees; 250–499 employees; equal or greater than 500 employees.

11. Forfás is Ireland's policy Advisory Board for Enterprise, Trade, Science, Technology and Innovation.

12. The industry classification used in this paper is NACE Rev. 1.1. Manufacturing firms are classified in NACE 15–NACE 37 industries, while the service sector includes firms classified in NACE 50–NACE 74.

13. Innovation expenditures include the following: in-house R&D expenditures, purchase of external R&D, acquisition of machinery, equipment and software (excluding expenditures on equipment for R&D), acquisition on other external knowledge.

Appendix

Table A1. List of variables.

Variable	Definition	Source
1. International activities
Foreign-owned firm	Dummy variable indicating whether a firm is foreign-owned (1) or not (0)	CIS
Domestic exporter	Dummy variable indicating whether a non-foreign-owned firm exports to Northern Ireland, the UK or other countries (1) or not (0)	CIS
2. Innovation input
Engagement in in-house R&D	Dummy variable indicating whether a firm reported a positive value of in-house R&D expenditure (1) or not (0)	CIS
In-house R&D expenditure per employee	The ratio of in-house R&D expenditure in constant prices over the number of employees	CIS
Engagement in innovation	Dummy variable indicating whether a firm reported a positive value of total innovation expenditure (1) or not (0)	CIS
Innovation expenditure per employee	The ratio of innovation expenditure in constant prices over the number of employees	CIS
3. Innovation output
Any type of innovation	Dummy variable indicating whether a firm reports having any type of innovation, namely product, process or organisational innovation (1) or not (0)	CIS
Product innovation	Dummy variable indicating whether a firm reports having product innovation (1) or not (0)	CIS
Process innovation	Dummy variable indicating whether a firm reports having process innovation (1) or not (0)	CIS
Organisational innovation	Dummy variable indicating whether a firm reports having organisational innovation (1) or not (0)	CIS
Product and process innovation	Dummy variable indicating whether a firm reports having both product and process innovation (1) or not (0)	CIS
Product and organisational innovation	Dummy variable indicating whether a firm reports having both product and organisational innovation (1) or not (0)	CIS
Process and organisational innovation	Dummy variable indicating whether a firm reports having both process and organisational innovation (1) or not (0)	CIS
All three types of innovation	Dummy variable indicating whether a firm reports having all three types of innovation, namely product, process and organisational innovation (1) or not (0)	CIS
Innovative turnover share	Percentage of turnover of innovative products in total turnover	CIS
4. Knowledge flows
Co-operation with other enterprises	Dummy variable indicating whether a firm co-operated with other enterprises within the same enterprise group (1) or not (0)	CIS
Co-operation with suppliers	Dummy variable indicating whether a firm co-operated with suppliers of equipment, material, components or software (1) or not (0)	CIS
Co-operation with customers	Dummy variable indicating whether a firm co-operated with clients or customers (1) or not (0)	CIS
Co-operation with competitors	Dummy variable indicating whether a firm co-operated with competitors or other enterprises in the same sector (1) or not (0)	CIS
Co-operation with consultants	Dummy variable indicating whether a firm co-operated with consultants, commercial labs or private R&D institutes (1) or not (0)	CIS
Co-operation with universities	Dummy variable indicating whether a firm co-operated with universities or other higher education institutions (1) or not (0)	CIS
Co-operation with government	Dummy variable indicating whether a firm co-operated with government or public research institutes (1) or not (0)	CIS
5. Other firm and industry characteristics
Labour productivity	Turnover per employee in constant prices	CSO
Employees	Number of employees	CSO
Human capital intensity	Wages per employee	CSO
6. Hampering factors to innovation
Dummy variable indicating whether a firm perceived the following factors as		CIS
having a high degree of importance in relation to innovation (1) or not (0):
Cost factors
Lack of internal funds
Lack of external funds
Innovation costs too high
Knowledge factors
Lack of qualified personnel
Lack of technology information
Lack of market information
Difficulty to find co-operation partners for innovation
Market factors
Market is dominated by incumbents
Uncertain demand
Need to meet government regulation
Excessive economic risks