Overcoming Obstacles: The Internet's Contribution to Firm Development

Abstract

Based on 49,610 firm observations across 117 developing and emerging countries for 2006–2011, we show that their respective industries’ adoption of the Internet had positive spillover effects on firms’ productivity across world regions and economies at different development stages. We find that even firms facing financial constraints, frequent power outages, skills shortages, corruption and cumbersome labor regulations gained. Quantile regression results confirm these conclusions hold across different levels of productivity. They also show that the most productive firms benefited much more than the least productive firms. This suggests absorptive capacities are critical for the Internet to support firm development.

I. Introduction

The uptake of the Internet in developing economies has been ubiquitous, bringing about one of the most fundamental changes in firms’ business environments (ITU 2014). There are multiple ways in which the Internet can help boost firms’ performance in developing countries: by improving access to relevant market information also for smaller and informal businesses, by facilitating more effective coordination of firms’ production and delivery chains, and by creating new business opportunities (Kaushik and Singh 2004; Aker and Mbiti 2010). Moreover, Paunov and Rollo (2014) identify positive impacts from Internet-enabled knowledge spillovers: the Internet provides a more effective means for disseminating new knowledge and in this way expands opportunities for firms to access relevant knowledge produced by others independently of their own investments in generating new knowledge.¹ Such evidence supports optimistic conclusions about the Internet's potential for improving firm performance in developing economies. However, cumbersome business framework conditions, which include shortcomings in physical infrastructures, weakly developed financial markets, and an often insufficiently skilled labor force, challenge efforts to boost enterprise development. Those obstacles might also reduce possible contributions of the Internet to firm productivity and require difficult-to-address framework conditions are resolved for firms’ development.

This paper investigates how widespread gains from using the Internet on firm's productivity effectively are and what effects cumbersome business framework conditions have. It builds on the work by Paunov and Rollo (2014), who demonstrate how industries’ adoption of the Internet affected the performance of firms belonging to such industries. It analyses whether firms' productivity gains that result from Internet-driven knowledge spillovers differ across world regions, across economies at different stages of development, as well as across the productivity distribution. Moreover, it tests whether business framework conditions constrained productivity gains. Our analysis is based on 49,610 firm observations across 117 developing and emerging countries for 2006–2011 from the World Bank Enterprise Surveys. The impacts of an industry's use of the Internet on firms’ labor productivity are identified using within country-year differences in the adoption of the Internet across sectors. An industry's use of the Internet is unlikely to be affected by its individual firms’ productivity performance, and therefore the risk of reverse causality is low. We employ ordinary least squares regressions as well as quantile regressions.

We find that positive impacts of industries’ Internet adoption on firms’ labor productivity were widespread: firms in different world regions and firms at different development stages benefitted equally. We also show that firms facing cumbersome business environments saw their labor productivity increase in response. However, benefits were lower where labor market regulations were more cumbersome, where financial markets were less developed, and where electricity outages were more prevalent. Also, we find our general conclusions are confirmed across firms' productivity distribution. Only cumbersome labor regulations significantly affected the least productive firms more. Moreover, we find that benefits from the Internet are higher for the most productive firms, possibly a result of these firms’ larger absorptive capabilities. Overall, our results support optimistic conclusions regarding the potential of the Internet to boost firm productivity even where business framework conditions are difficult. Thus, expanding firms’ use of the Internet is valuable in support of their development. Investments in firm's capabilities to take advantage of the Internet will help ensure benefits to be more widespread across firms at different levels of productivity.

The paper contributes to the work on the impact of information and communication technology (ICT) investments on firms’ productivity. Extensive research found positive impacts of ICTs (Jorgenson and Vu 2005; Bartel et al. 2007). Several studies have identified positive association between firms’ ICT use and their productivity in developing countries (World Bank 2006; Commander et al. 2011; Paunov and Rollo 2014). Our paper also relates to the literature that has documented the impacts of business framework conditions on firm development (Tybout 2000; Dollar et al. 2006).

The remainder of the paper is organized as follows. Section 2 describes the empirical framework, while Section 3 gives an overview of the data we use in our analysis. Section 4 describes the results of the analysis. The last section concludes.

II. Analytical Framework

In order to study the impact of industries’ adoption of the Internet on firms’ productivity performance in different country contexts, we adopt the following baseline estimation model:

$$Pro{d}_{ict}=\alpha +{\beta }_I\ast IC{T}_{sct}+\gamma \ast X_{ict}+{\lambda }_{st}+{\lambda }_{ct}+{ϵ}_{ict}$$

(1)

where Prod_ict is a measure of firm i's labor productivity. ICT_sct is an indicator of sector s's uptake of using email to communicate with clients and suppliers for country c at year t, whereas X_ict is a full set of firm-level control variables. Coefficient β₁ is our parameter of interest to identify spillover effects.²Seker (2012) and Dollar et al. (2006) apply a similar approach to identify impacts of different business conditions on firm's performance. We also add λ_st and λ_ct, respectively a set of sector-year and country-year dummies. That is, our identification strategy exploits differences in sectors' adoption of the Internet across countries controlling for differences characterizing specific industries or countries in specific time periods.

Two major challenges affect the analysis of the impacts of firms' ICT use on firm's performance: i) endogeneity (i.e. the fact that while ICT might support productivity performance, it could also be the case that more productive firms rely more on ICT), and ii) omitted variable biases (i.e. the fact that there might be other unaccounted unrelated factors that affect the estimated β₁). Endogeneity is less of a challenge for our analysis, which focuses on the adoption of the Internet at the sector level: it is unlikely that firms' innovation and productivity performance has a direct impact on their sector's adoption of the Internet. Furthermore, to avoid potential endogeneity concerns firm i's own use of the Internet is excluded from the industry average we compute. The use of an aggregate measure also reduces risks of measurement error. Moreover, we address omitted variable biases by introducing sector-year and country-year fixed effects. The inclusion of country-year fixed effects allows isolating the potential differences across countries including also specific government policies that might affect the firms' productivity and innovation performance. Controlling for industries in given years is also important because certain industries are more technology-intensive than others, so that allowing for the variation of control variables across industries would lead to spurious results. In addition, we introduce a set of firm-level controls which include: firms' employment and age, indicators of public ownership and of multi-plant establishments, controls for whether the firm has connections abroad (i.e. foreign-ownership and exporter status), proxies for managerial quality, access to finance, and whether the firm owned a website.³

In order to assess differential impacts across world regions and different income levels, we obtain separate coefficients for the different regions of the world and economies' different income levels.

Moreover, we test for differential effects of business framework conditions by using the following estimation approach:

$$\begin{array}{rl}Pro{d}_{ict}=& \alpha +{\beta }_{ADV1}\ast [IC{T}_{sct}\ast AD{V}_{ict}]+{\beta }_{DIS1}\ast [IC{T}_{sct}\ast DI{S}_{ict}]+\\ & \gamma \ast X_{ict}+{\lambda }_{st}+{\lambda }_{ct}+{ϵ}_{ict}\end{array}$$

(2)

where ADV_ict and DIS_ict are dichotomous variables indicating whether firms face cumbersome business framework conditions or not. We focus on five business framework conditions: power outages, corruption, financing constraints, skills shortages, and cumbersome labor regulations.

We apply ordinary least squares regressions as well as quantile regressions to assess whether impacts differ. Quantile regressions can be expressed in the general form Prod_ict = x_ict’β + ε_ict with Q_θ(Prod_ict/z_isct) = z_isct’β_θ, where z_isct includes all explanatory variables including our variable of interest as in (1) and (2) (Koenker and Basett 1978). Estimating θ from 0 to 1 gives the entire distribution of Prod_ict conditional on z_isct. Finally, robust standard errors clustered at the country-sector-year level are applied systematically following the procedure suggested by Moulton (1990).

III. Data

Our analysis makes use of the World Bank Enterprise Surveys (WBES), which collect comparable information on a representative sample of formal firms in the nonagricultural sector. Our estimating sample consists of 49,610 firm observations across 117 countries in 2006–2011. As to the composition of our dataset, 40% of our observations are from Latin America or the Caribbean, 27% from Africa, 22% from Eastern Europe, Central Asia or the Middle East, and 11% from the East Asia Pacific and South Asia regions. The sector coverage is diversified; 53% of firms are from the manufacturing sector, while the remaining 47% are firms in the services (including construction) sector. About 73% of the firms in the sample have fewer than 50 employees. Further detail is provided in Paunov and Rollo (2014).

Interestingly for our purposes, the WBES include information on whether firms used email services to communicate with suppliers and customers. The indicator is suitable for our analysis since it relates to the exchange of knowledge with clients and suppliers, which are both critical sources for firms' acquisition of relevant knowledge for their business activities. Evidence from our dataset, shown in Figure 1, indicates that in 2006–2011 a large share of firms used the Internet to communicate with clients and suppliers. Even among firms in low-income economies, 45.2% of firms had adopted this communication tool. Also, while small firms were less active users than their larger counterparts, even among the smallest uptake was 44.5%.

Figure 1.

Share of Firms that Communicate with Clients and Suppliers by Email (in Percentages)

Notes: Statistics provided are obtained for the 49,610 firms included in our baseline sample.

Moreover, the questionnaire provides rich information on business climate conditions that firms face. This allows for a more detailed analysis on how these conditions affect firms' benefits from the Internet. The correlation between the five different business conditions we analyze (as described above) is weak. We can therefore assess how these distinct factors affect the Internet's contribution to firms productivity. Finally, the data provide rich information on firm characteristics, which allows computing firm productivity and a set of firm level controls for our analysis. Detailed descriptions of variables used are provided in the appendix.

IV. Results

First, we investigate how widespread gains are across world regions and economies at different development stages. As a starting point, we report baseline results from Paunov and Rollo (2014) in column (1) of Table 1. These results show positive significant spillover effects of the Internet's adoption on firms' labor productivity. Regarding the magnitude of the estimated effects, all else equal, our findings indicate that an increase in the intensity of a firm's industries' use of the Internet by 1 standard deviation would raise its labor productivity by what is equivalent to an increase from the 50th to the 55th percentile of the distribution. However, as shown in Figure 2, benefits differ across the productivity distribution with the most productive firms benefitting about three times more than their less productive counterparts.

Table 1.

Impacts of Internet Adoption across World Regions and Stages of Economic Development

	(1)	(2)	(3)
Industry Email Usesct	0.007***
	(0.001)
Industry Email Usesct		0.008***
* Africa		(0.002)
Industry Email Usesct		0.002
* East Asia Pacific and South Asia		(0.003)
Industry Email Usesct		0.006**
* Eastern Europe, Central Asia and Middle East		(0.002)
Industry Email Usesct		0.011***
* Latin America and the Caribbean		(0.002)
Industry Email Usesct			−0.004
* High-Income Economies			(0.007)
Industry Email Usesct			0.007***
* Upper-Middle-Income Economies			(0.002)
Industry Email Usesct			0.007***
* Lower-Middle-Income Economies			(0.002)
Industry Email Usesct			0.007***
* Low-Income Economies			(0.002)
Firm Level Controls	Yes	Yes	Yes
Sector-Year Fixed Effects	Yes	Yes	Yes
Country-Year Fixed Effects	Yes	Yes	Yes
Observations	49,610	49,610	49,610
R-squared	0.81	0.81	0.81

Notes: The dependent variable is labor productivity. Firm controls included are described in the appendix. Robust standard errors clustered at country-sector-year level are shown in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively.

Figure 2.

Estimated Coefficients of Labor Productivity for Quantile Regressions

Notes: The dependent variable is labor productivity. The figure plots the coefficient of the impact of Internet adoption (β₁) of quantile regressions for each decile.

Turning to the question of gains across world regions, column (2) of Table 1 shows positive effects for firms in Africa, Eastern Europe, Central Asia, and the Middle East as well as Latin America and the Caribbean. Only for East Asia Pacific and South Asia we do not identify any impacts. These findings shows that the benefits of the Internet for firm's development were effectively widespread. Table 2 shows that these effects differ depending on firms' levels of productivity: while for firms in Latin America and the Caribbean effects are positive and significant across the distribution, in Table 2 we find that for African firms effects are low for less productive firms but high for more productive firms. For firms in East Asia Pacific and South Asia we find positive significant effects for median performers while for Eastern European, Central Asia, and the Middle East it is the least productive firms that benefit most.

Table 2.

Quantile Regression Results of the Impacts of the Internet by World Region

	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Industry Email Usesct	0.001	0.002	0.004**	0.005***	0.007***	0.008***	0.008***	0.010***	0.010***
* Africa	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.003)
Industry Email Usesct	0.004	0.003	0.004*	0.005*	0.006*	0.004	0.007	0.005	−0.002
* East Asia Pacific and South Asia	(0.003)	(0.003)	(0.002)	(0.003)	(0.003)	(0.004)	(0.004)	(0.005)	(0.007)
Industry Email Usesct	0.007***	0.007***	0.004	0.005	0.004	0.003	0.001	0.001	0.008
* Eastern Europe, Central Asia and Middle East	(0.002)	(0.003)	(0.004)	(0.003)	(0.003)	(0.004)	(0.003)	(0.004)	(0.005)
Industry Email Usesct	0.009***	0.007***	0.008***	0.010***	0.009***	0.011***	0.012***	0.012***	0.008
* Latin America and the Caribbean	(0.003)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.003)	(0.002)	(0.006)
Firm Level Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Sector-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Country-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	49,610	49,610	49,610	49,610	49,610	49,610	49,610	49,610	49,610
R-squared	0.78	0.80	0.81	0.81	0.81	0.81	0.81	0.80	0.79

Notes: The dependent variable is labor productivity. Firm controls included are described in the appendix. Robust standard errors clustered at country-sector-year level are shown in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively.

We also test whether we have widespread effects for firms in economies at different stages of development. Results, reported in column (3) of Table 1, confirm that estimated effects are the same across middle-income and lower-income economies. Only for the small number of firms from high-income economies we do not identify significant effects. Unreported results from quantile regressions show greater impacts for firms with higher levels of productivity for different income levels. There are no statistically significant differences across firms at different levels of development across the productivity distribution.⁴ Findings from quantile regressions confirm results reported in column (3) of Table 1.

Second, we analyze whether cumbersome business conditions affected impacts on firm productivity. We evaluate five challenges: i) electrical power outages, ii) corruption, iii) financing constraints, iv) skills shortages, and v) difficult labor regulations. In Table 3 we find that positive effects from the Internet on firm productivity are maintained even where business conditions posed challenges: first, findings show that corruption (column 2) and skills shortages (column 4) are not reducing benefits. Second, although obstacles faced in terms of power (column 1), financing (column 3) and labor regulations (column 5) reduce the positive impact of ICTs, we find that firms still reap significant positive productivity effects. However, it might be the case that business conditions affect mostly the least productive businesses. Unreported results from quantile regressions reject this hypothesis: while generally impacts rise with higher levels of productivity, the difference between firms facing cumbersome and less problematic business conditions does not change along the productivity distribution. The only exception is labor regulations. Results, provided in Table 4, show that difficult relations reduce any benefits from spillovers for the least productive businesses but have less of a differential impact at higher levels of productivity.

Table 3.

Impacts of the Internet Depending on Different Business Conditions

	Power Outages	Corruption Challenges	Financing Constraints	Skills Shortages	Labor Regulations
	(1)	(2)	(3)	(4)	(5)
Industry Email Usesct	0.008***
* Few Power Outages	(0.001)
Industry Email Usesct	0.007***
* More Power Outages	(0.001)
Industry Email Usesct		0.007***
* Low Corruption Incidence		(0.001)
Industry Email Usesct		0.007***
* High Corruption Incidence		(0.001)
Industry Email Usesct			0.008***
* Low Financial Constraints			(0.001)
Industry Email Usesct			0.006***
* High Financial Constraints			(0.001)
Industry Email Usesct				0.007***
* Low Skills Shortages				(0.001)
Industry Email Usesct				0.007***
* High Skills Shortages				(0.001)
Industry Email Usesct					0.007***
* Less Difficult Labor Regulations					(0.001)
Industry Email Usesct					0.005***
* Difficult Labor Regulations					(0.002)
P-Value of the Difference in Coefficients	0.00	0.30	0.01	0.56	0.05
Firm Level Controls	Yes	Yes	Yes	Yes	Yes
Sector-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes
Country-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes
Observations	49,610	49,610	49,610	48,959	49,610
R-squared	0.81	0.81	0.81	0.814	0.81

Notes: The dependent variable is labor productivity. Firm controls included are described in the appendix. Robust standard errors clustered at country-sector-year level are shown in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively.

Table 4.

Quantile Regression Results on the Impacts of the Internet Depending on Labor Regulations

	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Industry Email Usesct	0.004**	0.004***	0.005***	0.006***	0.007***	0.007***	0.008***	0.009***	0.007***
* Less Difficult Labor Regulation	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.002)
Industry Email Usesct	0.001	0.001	0.001	0.002	0.004*	0.005**	0.005**	0.008***	0.007***
* Difficult Labor Regulation	(0.003)	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)	(0.003)	(0.003)	(0.002)
P-Value of the Difference in Coefficients	0.10	0.04	0.00	0.02	0.12	0.10	0.30	0.56	0.85
Firm Level Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Sector-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Country-Year Fixed Effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	49,610	49,610	49,610	49,610	49,610	49,610	49,610	49,610	49,610
R-squared	0.78	0.80	0.81	0.81	0.81	0.81	0.81	0.80	0.79

Notes: The dependent variable is labor productivity. Firm controls included are described in the appendix. Robust standard errors clustered at country-sector-year level are shown in parentheses. ***, **, and * indicate significance at 1%, 5%, and 10% confidence levels, respectively.

V. Conclusion

This paper documents the capacity of the Internet to support firm productivity in spite of multiple obstacles that are well known to affect enterprise development. We show that the Internet had positive impacts on firm's productivity across world regions and across different stages of development. Even firms facing financial constraints, frequent power outages, skills shortages, corruption, and cumbersome labor regulations gained. However, our results also indicate that the most productive firms benefitted much more than their less productive counterparts. This result points to the critical role of boosting firms' absorptive capacities for the Internet to benefit a wider group of firms.

Appendix: Definitions of Variables Used

A. Business Conditions

Power outages: The variable is defined as the share of firms reporting power outages for sector s in location l in country y at year t. Environments with a high (low) incidence of power outages are defined as those above (below) the median value.

Corruption: The variable is defined as the share of firms which report corruption to be a major obstacle for their operations for sector s in location l in a country y at year t. High (low) corruption environments are defined as those where more (less) than half of the firms report corruption to be a major obstacle.

Financing constraints: The variable is defined as the share of firms reporting to have a credit line for sector s in country y and year t. Environments with high financial constraints are defined as those where the share is above (below) the median value.

Skills shortages: The variable is defined as the share of skilled workers in total employment in a given sector s in county y and year t. Environments with high (with low) skills shortages are defined as those where with below (above) the median value.

Cumbersome labor regulations: The variable is defined as the share of firms reporting labor regulations to be a major obstacle for operations in sector s in country y at year t. Environments with cumbersome (less problematic) labor regulations are defined as those where less (more) than one third of firms reported them to be a major challenge.

B. Country Classification by Income Level

Countries covered (using the World Bank country classification for distinct categories):

• High-income economies: The Bahamas, Barbados, Croatia, Czech Republic, Estonia, Hungary, Latvia, Poland, Slovak Republic, Slovenia, Trinidad and Tobago.

• Upper-middle-income economies: Albania, Antigua and Barbuda, Argentina, Azerbaijan, Belarus, Bosnia and Herzegovina, Botswana, Brazil, Bulgaria, Chile, Colombia, Costa Rica, Dominica, Dominican Republic, Fiji, Gabon, Grenada, Jamaica, Kazakhstan, Lithuania, Macedonia, Mauritius, Mexico, Montenegro, Namibia, Panama, Peru, Romania, The Russian Federation, Serbia, South Africa, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname, Turkey, Uruguay, Venezuela.

• Lower-middle-income economies: Angola, Armenia, Belize, Bhutan, Bolivia, Cameroon, Cape Verde, Republic of Congo, Cote d'Ivoire, Ecuador, El Salvador, Georgia, Guatemala, Guyana, Honduras, Indonesia, Iraq, Kosovo, Lesotho, Federal States of Micronesia, Moldova, Mongola, Nicaragua, Nigeria, Pakistan, Paraguay, The Philippines, Samoa, Senegal, Sri Lanka, Swaziland, Timor-Leste, Tonga, Ukraine, Uzbekistan, Vanuatu, Vietnam, Yemen.

• Low-income economies: Benin, Burkina Faso, Burundi, Central African Republic, Chad, Democratic Republic of Congo, Eritrea, The Gambia, Ghana, Guinea, Guinea-Bissau, Kenya, Kyrgyz Republic, Laos, Liberia, Madagascar, Malawi, Mali, Mauritania, Mozambique, Nepal, Niger, Rwanda, Sierra Leone, Tajikistan, Tanzania, Togo, Uganda, Zambia, Zimbabwe.

C. Firm-Level Variables

Labor Productivity: Logarithm of the ratio of total annual sales over full time employment windsorized at the top and bottom 1% for any country-year, reported in thousand USD, value in parenthesis.

Employment: Logarithm of the firm's full time employment.

Age: Logarithm of the difference between the year the survey was conducted and the year the firm was created.

Public ownership: A dummy equal to one if the government or state owns a share of 40% or more of the firm and zero otherwise.

Multi-plant firm: A dummy equal to one if the firm belongs to at least one other business and zero otherwise.

Foreign ownership: A dummy equal to one if the share of foreign ownership is bigger or equal to 40% and zero otherwise.

Exporter status: An indicator that is equal to one if the firm has exporter activities (includes both direct and indirect activities).

Credit access: Dummy variable that is equal to one if the firm has a line of credit or loan from a financial institution and zero otherwise.

Managerial expertise: Logarithm of years of the manager's experience.

Website: An indicator that is equal to one if the firm has its own website and zero otherwise.

Sectors: A variable indicating in which of the following sectors the firm is operating: i) food, ii) wood and furniture, iii) textiles, iv) garments, v) leather, vi) non-metallic and plastic materials, vii) chemicals and pharmaceuticals, viii) electronics, ix) metals and machinery, x) auto and auto rvices.

1

Knowledge spillovers in turn have been identified as critical for economic growth (Romer 1986).

2

The set-up is similar to that of the prior literature on knowledge spillovers from industries' R&D or foreign direct investment (FDI) as, for example, in Acs et al. (1994) or Haskel et al. (2007) to provide but one example for each.

3

We introduce the latter variable to control for firm investment in ICTs so as to identify spillover effects from the Internet.

4

All unreported results are available from the authors upon request.

Author notes

*

Caroline Paunov (corresponding author) is Senior Economist at the Directorate for Science, Technology and Industry, OECD, 2, rue André Pascal, 75 775 Paris Cedex 16, France; her email is and . Phone: [phone]. Valentina Rollo is PhD candidate, Graduate Institute of International and Development Studies, Maison de la Paix, Chemin Eugène-Rigot 2, Ch1202, Geneva, Switzerland; her e-mail is . Valentina Rollo gratefully acknowledges support from the SNF, Project Number PDFMP1_135148. The authors would like to thank Dominique Guellec, Richard Baldwin, Nicolas Berman, Ana Margarida Fernandes and participants of the 2014 ABCDE Conference for valuable comments. The findings expressed in this paper are those of the authors and do not necessarily represent the views of the OECD or its member countries.