The influence of product innovation on exports of entrepreneurs: a case study at the national level

(1)

The influence of product innovation on exports of

entrepreneurs: a case study at the national level

Annelies Stemfoort S2697017

a.stemfoort@student.rug.nl

University of Groningen

Faculty of Economics and Business MSc International Economics & Business Date: January 8, 2019

(2)

2

Abstract

This paper investigates whether product innovation influences exports for entrepreneurs at the national level, using the Global Entrepreneurship Monitor database. The main finding is that product innovation by entrepreneurs positively relates to exports. Some level of product innovation is relevant to start exporting, while the highest level of product innovation is more important for the highest level of exports. Splitting the sample by development level shows that the effect is larger and statistically more significant within developed countries. Adding lags shows that it takes longer to observe the effect within developing countries.

(3)

3

Content

1. Introduction ... 4

2. Literature review ... 6

2.1 Entrepreneurs(hip) ... 7

2.2 Relation between innovation and exports of entrepreneurs ... 9

3. Data and methods ... 10

3.1 Source data and final sample ... 10

3.2 Estimation method ... 11

3.2.1 Dependent and independent variables ... 11

3.2.2 Control variables ... 12 3.2.3 Additional tests ... 14 3.2.4 Interview ... 15 3.3 Descriptive statistics ... 16 4. Empirical results ... 21 4.1 First results ... 21

4.2 Results with controls ... 23

4.3 Results with controls and lags ... 24

4.4 Results for developed and developing countries ... 26

4.5 Robustness... 31

4.6 Interview ... 34

5. Conclusions ... 35

6. References ... 38

7. Appendix ... 41

A. Results random effects... 41

B. Countries included in sample per year ... 42

C. Robustness ... 44

(4)

4

1. Introduction

A startup is a major buzz word that shapes the 21st_{century. These startups are created by}

entrepreneurs. Although entrepreneurs have existed for many more years, the appeal for the new ideas these people bring has grown substantially over the last two decades. Everyone aims to connect to these new ideas and innovations that automatically connect to firms from entrepreneurs. Governments reach out to them with a helping hand and large firms want to select the best startups for cooperation (Ministry of Foreign Affairs Netherlands, 2018; Prats & Siota, 2018). Startups have something that no one else has and that attracts attention from the public and the private sectors. It has been proven by scientific research that these firms from entrepreneurs are unique in their very nature. Innovation is seen as one of the key elements making entrepreneurship a unique phenomenon (Schumpeter & Elliott, 1983; Davidsson, 2004). Why are startups so innovative? How does this relate to exports? And why do these firms matter? These questions provide the starting point and motivation of this paper.

One of the key attributes of entrepreneurs is their flexibility and ability to respond more easily to opportunities (Rosenbusch, Brinckmann & Bausch, 2011). With a rigid structure in large organisations, it is hard to move forward quickly. Additionally, entrepreneurs often have large motivation and energy which they use to create new products in new markets (Schumpeter & Elliott, 1983). Whenever this paper uses the term entrepreneurs, this will refer to entrepreneurs with young firms (younger than 3.5 years). Using this specific population sample ensures us that the key attributes of entrepreneurs are still integrated. For instance, that entrepreneurs are still flexible and have not yet grown into rigid organisations. These attributes are essential for creating value through innovations. Consequently, these innovations can potentially contribute to firm growth, because innovative products might lead to selling more products (Rosenbusch et al., 2011). Furthermore, Schumpeter and Elliott (1983) did not only talk about new products, but also about new markets. Since entrepreneurs are quickly moving forward, they often seek new market opportunities. One way of entering new markets is by expanding business to foreign countries. Hence, innovation could contribute to the first step in internationalisation for entrepreneurs: exports. Therefore, this paper investigates what the influence of product innovation is on the level of exports of entrepreneurs with young firms at the national level.

If this relation between innovation and exports proves to exist, it gives policy advisors a reason to focus their policy and the use of their resources on only those entrepreneurs that provide the highest benefits for society. In the end, policy advisors will only focus on entrepreneurs if they believe that their firms provide clear public benefits. These benefits are that stimulating entrepreneurship could enhance employment opportunities and innovation in the future. As a result, entrepreneurs can contribute to economic growth (Van Stel, Carree & Thurik, 2005; Wong, Ho & Autio, 2005; Shane, 2009), which shows that this research has important implications.

(5)

5 & Vettas, 2018; Urbano, Guerrero, Ferreira & Fernandes, 2018). The literature indicates that product innovation is the most relevant type of innovation for entrepreneurs and has a positive effect on exports (e.g. Cassiman, Golovko & Martinez-Ros, 2010; Cassiman & Golovko, 2011; Becker & Egger, 2013).

This paper will extend the literature in several ways. Firstly, by taking a broad country perspective over a period of nine years (2006 to 2014) to make results more universal, as opposed to a large part of the literature that merely regards one country (Van Beveren & Vandenbussche, 2010; Golovko & Valentini, 2011; Giotopoulos & Vettas, 2018) or one single year (Van Stel et al., 2005). Such large panel dataset also allows for inclusion of country and time fixed effects. Secondly, it will contribute to the existing literature by using a database (the Global Entrepreneurship Monitor) that is developed specifically for entrepreneurs. As a result, questions in the survey are specifically designed for entrepreneurs and therefore fit our population sample (entrepreneurs) better than conventional measures, such as research & development (R&D) investments for innovation. Thirdly, the use of this database also enables us to distinguish between different levels of innovation and exports, as proposed by Golovko and Valentini (2011). Differentiating between these levels can give more detailed insight into the relation, which is clearly needed based on the mixed results found by the literature. Lastly, new insights are also found by splitting the sample between developed and developing countries. Hence, this paper contributes to the literature by allowing for differences across both individual entrepreneurs (by allowing different levels of innovation and exports) and countries (by using a broad country range). It hypothesizes that entrepreneurs with higher levels of product innovation export relatively more than entrepreneurs with low levels of product innovation, at the national level.

The main finding of this paper is that innovation has a positive effect on exports of entrepreneurs. The estimation equations are built on four dependent variables for exports and three independent variables for product innovation. First results indicate that the highest level of innovation is most relevant for the highest level of exports, while a lower level of innovation is more relevant to start exporting in the first place. Splitting the sample in developed and developing countries provides the new insight that the effect of product innovation on exports is larger within developed countries. Although it cannot be stated with certainty that innovation influences exports, it seems likely that innovation is translated into an effect on exports within one year for developed countries. Within developing countries it takes longer before the effects of product innovation are observed.

(6)

6

2. Literature review

Literature on the relation between innovation and exports for other types of firms can help us understand the relation, which could potentially hold truth for entrepreneurs as well. Recall that the term entrepreneurs should be read as entrepreneurs with young firms throughout this paper. A firm can choose to produce only for the domestic market or to internationalise in different ways. Expanding sales to foreign markets is a way to increase profits, but it cannot be achieved by all firms due to differing productivity levels. There are four steps of internationalisation (Helpman, Melitz & Yeaple, 2004). Export is the first step to internationalisation and is found to go hand in hand with higher productivity levels (Melitz, 2003). This paper focusses on exports as internationalisation mode, because this first step is easier to reach for entrepreneurs than the alternative modes of internationalisation. Previously, productivity levels were taken as exogenous variables in the literature (Melitz, 2003). Currently, there is a growing consensus in the literature that firm-level productivity is endogenous, meaning that firms can engage in activities to improve their productivity levels (Constantini & Melitz, 2008). If entrepreneurs can also influence their productivity levels, they can prepare themselves for internationalisation. Alvarez and Lopez (2005) introduced the concept of conscious self-selection, which means that firms consciously choose to perform activities that increase their productivity, making it possible to enter export markets.1 Innovation may be one of those activities that drives entrepreneurs’ self-selection into export markets. Hence, innovation is an activity that might lead to productivity and possibly to exports. Other papers already confirmed that innovation leads to increased productivity (Cassiman et al., 2010; Cassiman & Golovko, 2011; Altomonte, Aquilante, Békés & Ottaviano, 2013), but until now the connection as proposed by Alvarez and Lopez (2005) is missing. Therefore, this paper examines whether innovation could influence exports.

The link between innovation and exports has been researched extensively in the literature. Most research is devoted to large firms, which provides the starting point for this paper. Although most researchers agree that there is a relation between the two, there is no full consensus on the direction and causality of this relation. The majority of literature shows that innovation has a positive effect on exports (Cassiman & Martinez-Ros, 2007; Cassiman et al., 2010; Cassiman & Golovko, 2011; Altomonte et al., 2013; Becker & Egger, 2013), while there are also a few studies that argue that exports lead to innovation (Damijan et al., 2010; Bratti & Felice, 2012; Damijan & Kostevc, 2015). Literature generally states that product innovation is more important in the early stage to start exporting (e.g. Klepper, 1996; Cassiman et al., 2010; Becker & Egger, 2013). Consequently, firms learn from exporting and start process innovation when they can reap more advantages from process innovation due to the larger production (Klepper, 1996; Becker & Egger, 2013). In addition to this view, Klepper (1996) states that younger firms are more innovative in this respect. Thus, product innovation is the type of

1_{Alvarez and Lopez (2005) investigated three possible explanations for why exporters are more productive than}

(7)

7 innovation that is most relevant for young firms, which is why this paper will specifically concentrate on product innovation.

In the early papers regarding innovation, the terms creativity and innovation are often mentioned together in one breath (Kuhn, 1985; Urabe & Child, 1988). At first, the generation of an idea (creativity) is seen as a part of innovation. Later, creativity is distinct from innovation by Cumming (1998), who states that creativity is the foundation for innovation, but something can only be called an innovation when an idea is successfully implemented and becomes tangible. The process from creativity to innovation often demands means like technological progress to make it possible to turn creativity into innovation (Cumming, 1998).

Product innovation can become tangible in the form of a new or changed product. This paper will follow the definition of Damanpour (1996) who defined product innovation as “the introduction of new products or services to meet an external market or user need” (Salavou, 2004, p. 39).2_{Damanpour (1996) focuses on how these new products are perceived by different}

players in the marketplace. Since product innovation for entrepreneurs often concerns their first product offered to the market, how this new product is perceived by the external environment is most important to them. This will determine if they have the potential to become successful. Entrepreneurs have received limited attention in the academic literature on the relation between innovation and exports so far. This seems strange, since exports and product innovation are relevant and applicable to entrepreneurs. The following section provides more details on entrepreneurs. The final section extends the previous discussion with an overview of the existing literature on the relation between innovation and exports, specifically for Small and Medium Enterprises (SME’s) and entrepreneurs. At the end, the hypothesis of this paper is formed.

2.1 Entrepreneurs(hip)

A limited amount of literature has been devoted to entrepreneurs, because it has been perceived difficult to define and measure entrepreneurship for several reasons (Van Stel et al., 2005). Firstly, entrepreneurship cannot be identified by one indicator only, whereas in case of companies, a firm size determines whether you are a small, medium, or a large enterprise. Secondly, the definition is not set in stone and how you define entrepreneurship determines how you measure it. This paper uses the Global Entrepreneurship Monitor (GEM) database’s definition of entrepreneurship: "Any attempt at new business or new venture creation, such as self-employment, a new business organisation, or the expansion of an existing business, by an individual, a team of individuals, or an established business" (Global Entrepreneurship Monitor, n.d.). The existence of this database, together with increasing interest in entrepreneurship, create a solid foundation for focusing on entrepreneurs. Furthermore, entrepreneurs deserve their own academic focus since they differ greatly from larger multinational firms. Notably, they have a different mindset, organisational structure and decision-making, which makes them unique (Schumpeter & Elliott, 1983; Rosenbusch et al., 2011). Consequentially, there are

2_{Multiple definitions exist in the literature. For instance, Bessant and Tidd (2007) define product innovation as}

(8)

8 several reasons to focus specifically on entrepreneurs in this research on the influence of product innovation on exports.

First of all, entrepreneurs are often seen as the gateway to new innovations. Several researchers have identified innovation as a key attribute of entrepreneurship that distinguishes it from other disciplines (Schumpeter & Elliott, 1983; Davidsson, 2004). Schumpeter and Elliott (1983) describe how entrepreneurs can create value by adding will and action to the equation, which they can use to create new goods for new markets. Therefore, it can be observed that newness is essential in the field of entrepreneurship. Recently, Rosenbusch et al. (2011) also finds that innovation has a positive effect on the performance of SME’s and that the innovation impact is strongest for young firms. These young firms possess unique capabilities to create and appropriate value through innovations. Young firms are less rigid in their organisation processes and can therefore more easily jump on opportunities.

Secondly, the impact of entrepreneurs on the economy and society has been on the rise. The number of entrepreneurs increases and they receive substantially more attention from policy makers, who hope to increase innovation and economic growth (Ministry of Foreign Affairs Netherlands, 2018). A growing amount of literature points towards an impact of entrepreneurship on economic growth or development. For instance, Van Stel et al. (2005) finds that the influence of entrepreneurship on economic growth is positive in rich countries. In addition, Wong et al. (2005) and Shane (2009) state that high growth potential entrepreneurs have an impact on economic growth, which also points to an increase in importance.

However, literature also provides restrictions that entrepreneurship does not always have a positive effect on economic growth, since the impact these entrepreneurs can have on the economy might depend on differences across countries. Van Stel et al. (2005) find a negative effect in poor countries, meaning that the impact depends on the level of development of a country. Moreover, Peris-Ortiz, Ferreira and Fernandes (2018) come to a similar conclusion and state that the relation between Total early-stage Entrepreneurial Activity (TEA) and innovative practices in OECD countries depends on the development stage of the economy (e.g. efficiency driven or innovation driven). These results are also consistent with the statement of Global Entrepreneurship Monitor (2018) that more developed countries have higher innovation levels. This might be due to better access to education or protection of property rights. Van Stel et al. (2005) provide an additional explanation, stating that there might not be enough successful large firms to make the transition towards a developed country, which also impacts the productivity levels of entrepreneurs.

Furthermore, the impact on economic growth might also depend on differences among entrepreneurs. The findings of Wong et al. (2005) suggest that not all entrepreneurs are truly innovative and only a small part is likely to engage in technological innovation. This shows that the level of innovation among entrepreneurs will most likely differ and that it is important to take these differences among entrepreneurs into account. Van Stel et al. (2005) do not consider these differences at the individual level, which might influence their findings. Knowing which entrepreneurs are more likely to positively impact the economy is valuable information for policy makers. If they want to support entrepreneurs, they ought to focus their resources on the ones adding the highest value to the economy.

(9)

9 country. Previous research often does not simultaneously take into account differences across countries and differences among entrepreneurs, while both have an impact. This means that, respectively, no country level determinants and/or fixed effects have been taken into account, or that all entrepreneurs in the sample are held as equal. Based on the literature discussion in this section, one would expect to identify differences in innovations among entrepreneurs and among countries.

To conclude, entrepreneurs matter in the economy nowadays and should be researched more as a unique group. Entrepreneurs are important for innovation which makes them suitable for investigating the influence of innovation on exports.

2.2 Relation between innovation and exports of entrepreneurs

Empirical evidence mainly focuses on larger firms, since small firms were previously not considered ready for internationalisation. More recently, the literature on small and medium enterprises has grown substantially, as researchers point out that new firms can start exporting right away (Knight & Cavusgil, 2004). Love and Roper (2015), just like Golovko and Valentini (2011), find that innovation plays an important role for firms that start exporting. Consequently, this leads to higher SME growth. Golovko and Valentini (2011) distinguish themselves from Love and Roper (2015) by stating that innovation and exporting should be seen as complementary to each other, instead of solely one causing the other.

More importantly, considering research on entrepreneurs, several scholars find a positive effect of innovation on exports as well (O’Cass & Weerawardena, 2009; Saeedikiya, Aeeni, Motavaseli & Farsi, 2017; Giotopoulos & Vettas, 2018). On the contrary, Urbano et al. (2018) finds that export market orientation has a positive effect on entrepreneurial innovations. These results seem contradictory, but then again might be valid, since several papers also state that innovation and export reinforce each other. On the other hand, there could also be an issue of reverse causality, meaning that it is not clear why the relation could not run from the development of new technologies to export market orientation, instead of the claimed relation by Urbano et al. (2018). Therefore, it is important to account for reverse causality problems.

Even though, the literature shows the relevance of researching this relation, some limitations are identified regarding their focus and methodologies. Some papers (Van Beveren & Vandenbussche, 2010; Golovko & Valentini, 2011; Giotopoulos & Vettas, 2018) merely focus on one specific country, while a broader perspective could identify different determinants and make results more universal. O’Cass and Weerawardena (2009) do take a broader country perspective, but subsequently do not allow for different levels of innovation or exports across firms. However, it is important to do so, since the hypothesized effect might depend on a certain threshold level or differ in strength across levels. Golovko & Valentini (2011) state in their discussion that allowing for different levels of innovation and export would be valuable in future research.

(10)

10 entrepreneurs that export or innovate do so in the same proportion. Since product innovation is found to be the most relevant type of innovation for young firms, the effect of product innovation on exports will be investigated. This leads to the following hypothesis:

Hypothesis: entrepreneurs with higher levels of product innovation will export relatively more than entrepreneurs with low levels of product innovation, at the national level.

The relation is researched at the national level to account for differences across countries. Therefore, this paper extends the literature by taking both differences across countries and differences among entrepreneurs into account.

The next section explains how this hypothesis is tested. It outlines which data is used in this paper and how the final sample is constructed. Furthermore, several estimation equations are developed based on the theory and data that is used to investigate the research question.

3. Data and methods

3.1 Source data and final sample

The Global Entrepreneurship Monitor (GEM) is a database that is built on surveys of individuals, conducted across many countries and years. It consists of a database focused on the individuals starting a business (the Adult Population Survey) and of another database that focuses on the national context in which a business is started (the National Expert Survey). Both databases are available at the individual level and the national level. In this paper, the Adult Population Survey (APS) with data at the national level is used, because it includes data on innovation and exports that is needed to analyse the hypothesis (Global Entrepreneurship Monitor Database, 2018). This database is unique, because of its focus on entrepreneurship, its reliance on survey data, and the scope and comparability across countries of this survey.

The export variables are included in the database from 2006 to 2014. Before 2006, there are no questions about exports included in the survey. For 20 countries (none in Africa), the data is available over the whole time period. To lose the least amount of observations and still be able to perform analyses over time across countries, all countries with at least two observations during the time period are included. As a result, 19 countries with one year of observations are excluded from the dataset. These are excluded, because it is not possible to apply a fixed effects model and lags for countries with only one observation. Fixed effects rely on multiple observations to account for within-country variation and lags refer to the values in the previous year.

(11)

11 3.2 Estimation method

3.2.1 Dependent and independent variables

The population sample of this paper is entrepreneurs with young firms, which is referred to as entrepreneurs. The GEM created the term Total early-stage Entrepreneurial Activity (TEA), which is the sum of two types of young firms, namely startups and baby business owners. These groups of entrepreneurs are identified in the individual survey by asking questions concerning ownership of a firm, how long it exists and whether wages are earned. Startups are defined in the GEM as people that are (part) owner of a firm that has been active past year and has not earned wages yet. Baby business owners are defined as people that own-manage a business with an income for less than 3.5 years. Thus, the measure TEA includes entrepreneurs that are either setting up a firm or are the (partial) owner of a firm that is less than 3.5 years old. This is how entrepreneurs are defined in this paper, in accordance with the literature on entrepreneurs (Giotopoulos & Vettas, 2018; Peris-Ortiz et al., 2018; Urbano et al., 2018). TEA is either measured by the count of entrepreneurs that are identified as startup or baby business owner, or by the percentage of the 18-64 years old population that is setting up a firm or is owner of a young firm. The percentage measure is used in this paper, because it reveals more about the relative importance of TEA in the population sample of each country. This paper does not distinguish between startups and baby business owners, but instead focusses on TEA as a whole due to data limitations on these distinct categories for our variables of interest.

The dependent variable exports is measured in the GEM database as the percentage within TEA that has a certain percentage of customers outside the home country. Four levels of exports are identified: the percentage of entrepreneurs with no customers outside the home country, 1-25% of customers outside the home country, 25-75% of customers outside the home country or 75-100% customers outside the home country. This makes it possible to distinguish between four levels of exports, giving us four continuous dependent variables. For each country and each year, the percentage of TEA that belongs to each of the four export levels is given. For instance, the database tells us that in 2014, 34.68% of entrepreneurs (TEA) in the Netherlands had 1-25% of customers outside the home country (Global Entrepreneurship Monitor Database, 2018). The same data is available for the remaining export levels.

(12)

12 Combining the measurements of the variables mentioned above, the first estimation equation of this paper is presented below:

(1) 𝐸𝑥𝑝𝑜𝑟𝑡𝑠_𝑁𝑢𝑙𝑙𝑖,𝑡 = 𝛼𝑖,𝑡+ 𝛽1𝑁𝑒𝑤_𝐴𝑙𝑙𝑖,𝑡+ 𝛽2𝑁𝑒𝑤_𝑆𝑜𝑚𝑒𝑖,𝑡+

𝛽₃𝑁𝑒𝑤_𝑁𝑜𝑛𝑒_𝑖,𝑡+ 𝛾_𝑡+ 𝑢_𝑖+ 𝜀_𝑖,𝑡

Here, Exports_Null stands for the first export level in the GEM database. Each model explained in this section is executed for the four different levels, meaning that there are four dependent variables. Exports_Null represents entrepreneurs with no customers outside the country, Exports_25 represents entrepreneurs with 1-25% of customers outside the country, Exports_75 represents entrepreneurs with 25-75% of customers outside the country and Exports_100 represents entrepreneurs that have 75-100% of customers outside the country. This approach is similar to quantile regression, since the four dependent variables represent the level of exports at certain percentiles (0, 25, 75, 100). However, the levels are not divided into equal portions, meaning that they are not quantiles by definition. Therefore, the data structure of the GEM database makes the approach shown in estimation equation (1) more suitable than quantile regression.

The product innovation measure newness to customers is divided in three continuous independent variables that give the percentage of entrepreneurs (TEA) whose product is new to all customers (New_All), new to some customers (New_Some) or new to none customers (New_None). New_All represents the highest level of product innovation here, since customers have never seen something like it before.

Data for the variables is given across countries (i) and over time (t), which allows us to add different fixed effects. Firstly, time effects are included to control for aggregate trends over time that apply to all countries, which are represented by 𝛾𝑡. Secondly, country effects, as

represented by 𝑢𝑖, are added to absorb time-invariant trends within countries that cannot be

observed but affect the variables. This enables us to extract the effect of innovation on exports over time within countries. The error term is represented by 𝜀. Standard errors are clustered by country to account for heteroskedasticity.

Following the hypothesis, this research expects to find a positive sign for New_All and New_Some in regressions on the dependent variables involving exports (Exports_25, Exports_75, Exports_100) and a negative sign for Exports_Null. The coefficient of New_None is expected to show a negative sign for the dependent variables involving exports and a positive sign for Exports_Null. Since we hypothesize that higher levels of innovation lead to higher exports levels, we expect a larger (absolute) value for the coefficient of New_All relative to the value for New_Some and we expect the effect to be strongest for the highest export level (Exports_100).

3.2.2 Control variables

(13)

13 (2) 𝐸𝑥𝑝𝑜𝑟𝑡𝑠_𝑁𝑢𝑙𝑙_𝑖,𝑡 = 𝛼_𝑖,𝑡+ 𝛽₁𝑁𝑒𝑤_𝐴𝑙𝑙_𝑖,𝑡+ 𝛽₂𝑁𝑒𝑤_𝑆𝑜𝑚𝑒_𝑖,𝑡+

𝛽3𝑁𝑒𝑤_𝑁𝑜𝑛𝑒𝑖,𝑡+ 𝜷𝟒𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒔𝒊,𝒕+ 𝛾𝑡+ 𝑢𝑖+ 𝜀𝑖,𝑡

The first control variable that is included is technology. As identified by Bloomer, Jagoda and Landry (2010), the use of advanced technologies can help firms compete in international markets. Therefore, technology is expected to influence exports. Two different control variables for technology are included, one from the GEM database and one from the Global Competitiveness Index (GCI). In the GEM database, entrepreneurs are asked the question whether they use the very latest technology (only available since last year), new technology (1 to 5 years) or no new technology. This paper combines the first two questions into one variable (GEM_Technology) that measures the percentage of TEA that uses new technology, defined as technologies that are available since last year up to 5 years. The second control variable for technology is pillar 9 of the GCI that gives countries a value of 1 to 7 (best) based on their technological readiness (Tech_Readiness) (World Economic Forum, 2018). This pillar combines several indicators connected to ICT use and technological adoption into one value. Both control variables are included in the analysis, since they focus on different aspects of technology. The GCI focuses on availability and absorption of technologies for countries as a whole, whereas the GEM focuses on the usage of technology by entrepreneurs specifically. The coefficients of both variables are expected to be positive in regressions on the dependent variables involving exports (Exports_25, Exports_75, Exports_100) and negative for Exports_Null.

The second control variable is income, which might have an effect on exports within a country. Since it matters in this case how much can be bought by individuals, the measure Gross Domestic Product per capita based on Purchasing Power Parity (PPP) at constant 2011 international US dollars is taken from the World Bank as a measure for income (GDPpcPPP) (World Bank, 2018). The coefficient of this variable is expected to have a negative sign. Following the gravity model, countries with similar levels of income often trade with each other in differential goods. If income changes within a country, it might influence to which markets the country exports (Bergstrand, 1985). An increase in income might make a country less competitive or might decrease the amount of countries with the same level of income that it can trade with.

Education might also influence the exports of entrepreneurs within countries. Giotopoulos and Vettas (2018) also include the education level in their analysis, but from an individual perspective instead of a country perspective. Since this paper takes a country perspective, it will use the education level within countries as control variable. A measure from the GCI on educational quantity is chosen to take gross enrolment in secondary and tertiary education into account. Education_Quantity combines the two previously mentioned measures into a value between 1 and 7 (best) for each country and year. Education_Quantity is expected to have a positive effect on exports over time within countries (Giotopoulos & Vettas, 2018). Thus, a positive sign is expected for Exports_25, Exports_75 and Exports_100 and a negative sign is expected for Exports_Null.

(14)

14 financial market has a positive impact on export shares for countries. Therefore, financial market development from the GCI is included as control variable. The 8th pillar of the GCI gives countries a score between 1 and 7 (best) for financial market development, which is based on several indicators that show the efficiency and trustworthiness of the financial market. This control variable Financial_Market is expected to have a positive effect on exports over time within countries, because better financing could enable firms to expand their business.

In accordance with the literature, competition is expected to influence exports (Cassiman & Martinez-Ros, 2007; Giotopoulos & Vettas, 2018). Where domestic competition matters for the ability to start a firm and become part of TEA, foreign competition might influence the decision and ability to export. Therefore, it is included in this paper. The GCI has a measure on foreign competition that gives countries, as seen before, a score between 1 and 7 (best). It includes, among others, indicators on trade barriers and trade tariffs. Less rules or barriers contribute to a higher score of foreign competition. A higher score on Foreign_Competition is therefore expected to have a positive effect on the level of exports over time within countries.

Lastly, changes in the macroeconomic environment might impact the level of exports. Factors like inflation, national savings and government debt could potentially have an influence on how attractive products are to foreign countries and how much there is to gain from exporting as entrepreneur. For instance, high inflation is associated with low exports (Gylfason, 1999). The GCI measures the macroeconomic environment in its 3rd pillar where it gives each country a value between 1 and 7 (best) for each year. A higher score might lead to higher exports over time, when foreign buyers see that a country has a stable macroeconomic environment. Thus, we expect a positive sign for the coefficient of Macro_Environment in regressions on the dependent variables that have some level of exports.

3.2.3 Additional tests

As mentioned earlier, it is important to account for reverse causality in this research field. Reverse causality could play a role through learning-by-exporting. This is less likely to exist for this population sample, because it is unlikely that these young firms (maximum 3.5 years old) have learned from past exporting history which makes them more innovative. However, it cannot be ruled out that entrepreneurs in the database have export experience from other firms, so reverse causality issues cannot be completely eliminated.3 Additionally, it could also be the case that firms decide simultaneously on innovation and exports, meaning that it is not the case that one influences the other (Van Beveren & Vandenbussche, 2010). To account for the simultaneity in decisions and reverse causality that could exist within the sample, one year lags of the independent variables are included. In the literature, lags are regularly used to solve these problems (Cassiman & Martinez-Ros, 2007; Cassiman et al., 2010; Van Beveren & Vandenbussche, 2010). A lag length of one year is chosen, because the youngest firms in our sample are only one year old. The following estimation equation includes the lagged independent variables:

3_{The issue of export experience from other firms can only be accounted for by including an instrumental}

(15)

15 (3) 𝐸𝑥𝑝𝑜𝑟𝑡𝑠_𝑁𝑢𝑙𝑙_𝑖,𝑡 = 𝛼_𝑖,𝑡+ 𝛽₁𝑁𝑒𝑤_𝐴𝑙𝑙_{𝑖,𝒕−𝟏}+ 𝛽₂𝑁𝑒𝑤_𝑆𝑜𝑚𝑒_{𝑖,𝒕−𝟏}+

𝛽₃𝑁𝑒𝑤_𝑁𝑜𝑛𝑒_{𝑖,𝒕−𝟏}+ 𝛽₄𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_𝑖,𝑡+ 𝛾_𝑡+ 𝑢_𝑖+ 𝜀_𝑖,𝑡

where t-1 represents the independent variables with a year lag. These lagged independent variables are expected to show the same sign for the coefficients as the independent variables without lags.

Building onto estimation equation (3), the sample is split into countries with different levels of development. Following the literature, the relation between innovation and exports of entrepreneurs might depend on the level of development of a country (Van Stel et al., 2005; Peris-Ortiz et al., 2018). Therefore, splitting the sample into developed and developing countries might have an impact on the results of the regressions and identify differences between these groups of countries. Splitting the sample is based on the division of the United Nations (United Nations, 2014). This means that changes in the division over time will not be taken into account. However, it is unlikely that many countries in the sample change position over a period of 9 years, so therefore this will not impact the results substantially. The United Nations have composed three categories (developed economies, transition economies and developing economies) that are ‘intended to reflect basic economic country conditions’ (United Nations, 2014, p.143). For simplification, transition economies are recorded as developing economies for several reasons. Firstly, the sample for these transition economies is too small to base solid conclusions on. Secondly, it could be argued that these countries have not fully transitioned into developed economies yet and could therefore be categorized as developing economies. Following Global Entrepreneurship Monitor Global Report (2018) and Peris-Ortiz et al. (2018), entrepreneurs in developed countries are expected to have relatively higher levels of innovation. Therefore, we expect to find the strongest effect of innovation on exports within developed countries.

3.2.4 Interview

In extension of the quantitative analysis, it is valuable to see what policy implications are. Therefore, an interview is conducted with Robbie Peeters as expert on entrepreneurs from a government perspective. He is policy offer and part of the startup team within the International Enterprise Department, at the Ministry of Foreign Affairs in the Netherlands. Mr. Peeters explains what the Dutch government does with and for startups, why they do so and what the implications of this research are for them. A transcript of the interview is provided in the Appendix (Appendix D). The term startups can be read as entrepreneurs in this interview, but is preserved to maintain authenticity of statements from the interviewee.

(16)

16 avoid mistakes in translating. Even though both speakers (interviewer and interviewee) are native Dutch, both are used to speaking in English on a regular basis, which limits the constraints of speaking in a non-native language.

3.3 Descriptive statistics

As seen in Figure 1 below, the percentage of the population (18-64 years old) that engages in early-stage entrepreneurial activity has risen substantially over time in our sample. This means that overall, the importance of entrepreneurship has increased globally. In Figure 2 the same statistic is shown, but this time after dividing the sample into developed countries and developing countries. The graph shows that in developing countries the largest share of the population engages in early-stage entrepreneurial activity. One could speculate that this result stems from the fact that it is harder for people in developing countries to obtain a job and that necessity to obtain certain living standards gives them additional motivation to become entrepreneur. However, these numbers do not provide details on what kind of entrepreneurship people are involved in. Therefore, it could still be the case that entrepreneurs in developed countries are more successful or add more value to the economy, since these countries have higher innovation levels (Global Entrepreneurship Monitor, 2018).

Figure 1 Percentage of population involved in TEA, mean over time

(17)

17

Figure 2 Percentage of population involved in TEA by stage of development, mean over time

Note: TEA stands for Total early-stage Entrepreneurial Activity.

Table 1 shows the summary statistics of all variables. First, the four dependent variables (Exports_Null, Exports_25, Exports_75 and Exports_100) are examined. The relatively low values of standard deviation within tell us that the values do not fluctuate much over time within countries. As could be expected, following Helpman et al. (2004), the higher the percentage of exports, the lower the percentage of entrepreneurs within a country that belongs to this level. The mean value of Exports_Null is highest with 53.63% across the sample, followed by 30.31% for Exports_25, decreasing to 9.88% for Exports_75 and merely 6.18% for Exports_100. This means that more than half of the sample does not export at all and only a minority of firms is able to achieve the highest level of exports. This might indicate that it is harder to achieve (higher levels of) exports than to sell products in the domestic market only (no exports), which is consistent with the literature. Second, the independent variables are considered, which show a similar pattern. The highest level of innovation (New_All) has the lowest mean, indicating that it might be harder to achieve this than a lower level of innovation (New_Some).

(18)

18 to normalize the distribution. Thirdly, the variable Education_Quantity has a mean value of 5.13, which tells us that we have relatively many countries in the database with high enrolment rates in secondary and tertiary education. The standard deviation is also high, meaning that there is still a large variation in the scores of countries. Lastly, for the other GCI control variables, the mean scores are lower than for Education_Quantity (ranging from 4.38 to 4.92). This indicates that countries on average score worse on the other control variables. However, the variables are difficult to compare with Education_Quantity, because their range and standard deviation are also smaller. The countries in the sample reach the minimum and maximum value of the GCI for Education_Quantity, but score less extreme on the others.

Table 1 Summary statistics

A correlation table of the relevant variables is given below (Table 2). It shows a statistically significant correlation between the dependent and independent variables. Important to note is that the dependent variables have high correlations with each other as well, since their values within each country sum up to 100%. These correlations are negative for Exports_Null and positive for the remaining levels. Overall, correlations between the independent and dependent variables are as expected based on the literature (e.g. Cassiman & Golovko, 2011; Altomonte et al., 2013; Becker & Egger, 2013). If a product is not new to consumers (New_None), this is positively correlated with Exports_Null and negatively correlated with the other export levels. So, the percentage of entrepreneurs in a country that does not export is expected to be higher if its product is not new to consumers. On the contrary, the correlation between these independent variables and the percentage of entrepreneurs that does export is negative, meaning that less firms export if the independent variable New_None and is high.

Between the different variables for newness to consumers, it is observed that the sign of the correlations with the different dependent variables switches for New_None. Whereas New_None has a positive correlation with Exports_Null and negative correlations with Exports_25, Exports_75 and Exports_100, the opposite is true for New_All and New_Some. Therefore, this seems to indicate that more newness, and thus innovation, is positively associated with the percentage of firms that exports. Correlations of New_All with the

Obs Mean SD overall SD within Min Max

Exports_Null 480 53.6159 22.8112 9.2059 0 99.0950 Exports_25 480 30.3619 16.4159 7.9400 0.4616 75.3370 Exports_75 480 9.8277 6.37295 3.2985 0 30.4761 Exports_100 480 6.1945 4.7676 2.7002 0 23.4821 New_All 480 16.3400 9.8737 5.4067 0 55.3709 New_Some 480 27.7196 10.0895 5.7805 1.2092 60.2594 New_None 480 55.9404 15.9442 8.7063 9.8203 98.7908 GEM_Technology 480 31.8792 13.1664 7.0110 0 92.8341 GDPpcPPP 475 25911.3 16635.65 1686.03 1039.76 93655.33 Education_Quantity 469 5.1359 1.2694 0.3745 1.0 7.0 Financial_Market 470 4.6529 0.7174 0.3072 2.4930 6.6209 Tech_Readiness 470 4.4361 0.7387 0.2913 2.3928 6.4003 Foreign_Competition 470 4.3770 1.0335 0.1952 2.1475 6.3649 Macro_Environment 470 4.9221 0.7985 0.3682 2.2549 6.8349

(19)

19 dependent variables are small and insignificant, indicating that the effect of New_All on the dependent variables will most likely be small. A reason for this might be that this variable demonstrates relatively less variation over time within countries compared to the other variables for newness to customers (small SD within, see Table 1).

(20)

(21)

21

4. Empirical results

4.1 First results

The results for estimation equation 1 are listed in Table 3, where each column represents the results for one of the four dependent variables. The four columns include country and time fixed effects, which means that the findings represent the effect of innovation on exports over time within countries. The variable New_None is omitted due to exact collinearity. Cluster robust standard errors are used to absorb within-country error correlations that might be present in the data.

Table 3 Regression of export on innovation based on estimation equation 1

(1) (2) (3) (4)

Variables Exports_Null Exports_25 Exports_75 Exports_100

New_All -0.157 0.0647 0.0257 0.0670** (0.0967) (0.0802) (0.0405) (0.0270) New_Some -0.355*** 0.256*** 0.0788* 0.0204 (0.106) (0.0767) (0.0416) (0.0252) Constant 67.07*** 21.00*** 7.208*** 4.725*** (4.475) (3.636) (1.346) (0.975) Observations 480 480 480 480 R-squared within 0.089 0.064 0.065 0.048 R-squared overall 0.068 0.081 0.048 0.019 Number of countries 85 85 85 85

Country FE Yes Yes Yes Yes

Time FE Yes Yes Yes Yes

Note: standard errors clustered by country in parentheses. FE stands for fixed effects. Significance level: *** p<0.01, ** p<0.05, * p<0.1

Table 3 shows that the coefficients have the highest absolute values in column 1 for Exports_Null. All columns except column 1 show positive coefficients for both independent variables, which means that an increase in any level of innovation (New_Some or New_All) leads to a higher percentage of entrepreneurs that export over time within countries. These signs are consistent with the hypothesis. Following the hypothesis, one would expect a higher level of innovation (New_All) to have a stronger effect on exports which is not the case for Exports_Null. The absolute value of the coefficients is smaller for New_All than for New_Some. The coefficient of New_Some in column 1 tells us that an increase of one percentage point in New_Some leads to a decrease in Exports_Null of 0.355 percentage point. A decrease in Exports_Null means that entrepreneurs started exporting. Furthermore, only New_Some is highly significant at the 1% level, while New_All is not statistically significant. However, to reach more solid conclusions, it is important to determine whether the same is true for the other dependent variables that identify different levels of exports.

(22)

22 that offer products that are new to customers. The coefficient of New_Some in column 2 shows that an increase of one percentage point in New_Some leads to an increase of 0.256 percentage points in Exports_25. New_Some is still statistically significant at the 1% level for Exports_25. Thus, a higher percentage of entrepreneurs that offer new products to customers leads to a higher percentage of firms that export between the 1 and 25% over time within countries. Although the coefficient of New_Some remains positive, it becomes smaller and less significant in column 3, eventually not being statistically significant anymore in column 4. On the contrary, the coefficient of New_All becomes larger and turns statistically significant at the 5% level in column 4. This suggests that the highest level of innovation is most relevant for the highest level of exports, while a lower level of innovation is more relevant to start exporting in the first place. A potential explanation for this finding could be that having a product that is new to some customers is enough to start selling in foreign markets. Subsequently, a product needs to add more value to customers if an entrepreneur wants to be able to sell more products abroad. The coefficient for New_All shows that an increase of one percentage point in New_All leads to an increase of 0.067 percentage point in Exports_100 within countries. Thus, if a higher percentage of entrepreneurs sell products that are new to all consumers, the percentage of entrepreneurs that export between the 75 and 100% also increases. This value of the coefficient for New_All is small, indicating a rather limited influence on changes in the level of exports over time within countries. The limited relation could potentially be explained by the relatively small values of New_All and Exports_100 and their low standard deviation within (see Table 1).

It should be noted that the within and overall R-squared in these estimations are low, for all dependent variables under 0.09. For instance, the within R-squared value of column 1 shows that 8.9% of the variation in Exports_Null over time within countries is explained by the estimation method. The overall R-squared shows the weighted average of within and between country variation and tells us that 6.8% of the overall variation in Exports_Null is explained by the estimation model. In this fixed effects model, we are most interested in the within R-squared. The highest within R-squared is estimated for Exports_Null, while the highest overall R-squared is given for Exports_25. The lowest values are given for Exports_100, which might be the result of this dependent variable having the smallest variation within countries (see SD within, Table 1). All R-squared values are relatively low, which is not uncommon for panel data. Another explanation for the low values could be that only a limited number of independent variables is taken into account. Additionally, it might also be low because all dependent variables do not variate much over time (see SD within, Table 1).

To conclude, based on the results in Table 3, innovation is associated with higher exports, supporting the hypothesis. The independent variables of innovation show that the signs are negative for no exports (Exports_Null) and positive for all exports levels, in accordance with expectations. Having a product that is at least new to some customers matters most for the lowest exports level (1-25% exports), while having a product that is new to all customers becomes relatively more important for higher exports levels. This is consistent with the hypothesis that higher levels of innovation lead to higher levels of exports. However, the results should be interpreted with care, because the within R-squared is so low.

(23)

23 between the fixed effects model and the random effects model are small for Exports_100. The largest difference is that the coefficient of New_Some is slightly larger and statistically significant at the 10% level for the random effects regression. The random effects results seem to overstate the effect and significance levels of New_Some.

4.2 Results with controls

Table 4 shows results based on estimation equation 2, which adds control variables. Comparing Table 4 with Table 3, it is observed that including control variables alters the estimation results in several ways. Firstly, the values of the coefficients for New_All and New_Some change slightly for some regressions. However, these differences are very small and the signs remain the same. Secondly, the coefficient of New_All in the regression on the dependent variable Exports_100 has become less statistically significant. The coefficient of New_All is statistically significant at the 5% level in Table 3, whereas it is only marginally significant (at the 10% level) in Table 4.

Some control variables have an impact on the dependent variables. First of all, the coefficient for Financial_Market is positive and has a high value in the regression on Exports_Null. The coefficient tells us that a one unit increase in the value of Financial_Market, increases the value of Exports_Null with 4.037 percentage points over time within countries. This value is statistically significant at the 10% level and shows us that a stronger financial market will increase the percentage of firms not exporting over time within countries, which contradicts expectations. Potentially, it could be the case that stronger financial markets in the home country motivate firms to use financing for the domestic market instead of exports. Additionally, firms might need less contacts abroad to seek financing somewhere else if this has already been arranged in the home country, leading to less exports.

Secondly, two control variables have an influence on Exports_100, as is observed in column 4. For a start, the coefficient for LN_GDPpcPPP has a large negative value. This coefficient tells us that a 1% increase in GDPpcPPP leads to a 0.0657 percentage point decrease in Exports_100 over time within countries. This value is also statistically significant at the 10% level. Additionally, the coefficient of Tech_Readiness shows a positive and large impact on the dependent variable Exports_100, which is consistent with expectations. The coefficient tells us that an increase of one unit value in Tech_Readiness, leads to an increase of 1.72 percentage points in Exports_100. In other words, the more ready a country becomes over time for new technologies, the higher the percentage of entrepreneurs that export most (75 to 100%) is.

(24)

24

Table 4 Regression of export on innovation with control variables based on estimation equation 2

(1) (2) (3) (4)

New_All -0.142 0.0763 0.0140 0.0516* (0.117) (0.0945) (0.0445) (0.0288) New_Some -0.383*** 0.297*** 0.0774* 0.00864 (0.115) (0.0846) (0.0430) (0.0261) GEM_Technology 0.0570 -0.106 0.0258 0.0233 (0.112) (0.0903) (0.0287) (0.0222) LN_GDPpcPPP -0.672 9.570 -2.327 -6.570** (13.23) (11.42) (3.426) (2.893) Education_Quantity 0.702 -1.174 0.339 0.133 (2.106) (1.837) (0.608) (0.503) Financial_Market 4.037* -3.034 -0.578 -0.425 (2.345) (2.224) (0.711) (0.541) Tech_Readiness -0.141 -1.389 -0.190 1.720** (2.502) (2.229) (0.821) (0.702) Foreign_Competition 0.0340 1.344 -0.768 -0.610 (3.816) (3.793) (1.393) (1.175) Macro_Environment -2.783 1.765 0.917 0.100 (2.420) (2.027) (0.631) (0.485) Constant 64.96 -61.34 30.14 66.23** (126.3) (109.6) (33.17) (26.80) Observations 464 464 464 464 R-squared within 0.107 0.093 0.082 0.089 R-squared overall 0.003 0.156 0.039 0.124 Number of countries 83 83 83 83

4.3 Results with controls and lags

Now, we have come to estimation equation 3 with one year lags for the independent variables. As a result of including these lags, some observations that do not have a previous year for certain countries are excluded. Therefore, one should be careful when comparing Table 4 and Table 5. The results for estimation equation 3 are shown in Table 5. Let us first identify how adding lagged independent variables changes the results compared to the last estimation equation.

(25)

25 coefficient for New_All in column 4 has even turned negative, although close to zero. Whereas the results in Table 4 showed us that having a higher level of product innovation becomes relatively more important for higher levels of exports, the opposite is now observed in Table 5. Here, having a higher level of product innovation becomes relatively less important for higher levels of exports. Potentially, this indicates that the effects of a higher level of innovation are observed faster (within one year) than the effects of a low level of innovation. This explanation seems plausible, because more innovative products typically find their way to foreign customers faster. The highest level of innovation (New_All) has a stronger effect on Exports_25 than New_Some which is in line with the hypothesis. However, in regressions on Exports_75 and Exports_100, the coefficients of New_Some are larger than those of New_All which is opposed to previous results and expectations. Important to note is that none of the coefficients for the lagged independent variables are statistically significant anymore in Table 5. This means that it cannot be concluded that innovation influences exports. It might be the case that exports influence innovation or that both happen simultaneously.

Most coefficients of control variables remain roughly the same in Table 5 compared to Table 4. However, some differences are described here. Firstly, the most striking difference is observed for Education_Quantity which switches sign in all four regressions. Secondly, the coefficients of LN_GDPpcPPP change substantially. Thirdly, the Financial_Market coefficient for the regression on Exports_75 shows that an increase of one unit value in Financial_Market leads to a 1.695 percentage point decrease in Exports_75. This coefficient has now turned statistically significant at the 10% level, in addition to the coefficient of Financial_Market in the regression on Exports_Null that is already statistically significant in Table 4. Fourth, Tech_Readiness has also turned statistically significant at the 10% level in column 2, in addition to column 4. These changes could be the result of the large change in the number of observations and countries.

(26)

26

Table 5 Regression of export on innovation with controls and lags based on estimation equation 3

(1) (2) (3) (4)

New_All_L1 -0.162 0.165 0.00575 -0.00869 (0.138) (0.116) (0.0368) (0.0330) New_Some_L1 -0.163 0.0993 0.0245 0.0394 (0.137) (0.118) (0.0387) (0.0249) GEM_Technology -0.111 0.0478 0.0499 0.0131 (0.138) (0.107) (0.0330) (0.0254) LN_GDPpcPPP -11.89 16.07 0.902 -5.087 (17.06) (14.07) (3.709) (3.819) Education_Quantity -0.294 1.448 -0.306 -0.848 (2.586) (1.991) (0.649) (0.651) Financial_Market 3.990* -1.510 -1.695* -0.786 (2.043) (1.968) (0.950) (0.654) Tech_Readiness 2.082 -3.852* -0.239 2.009** (2.707) (2.293) (0.855) (0.865) Foreign_Competition 2.051 0.989 -2.013 -1.026 (5.078) (4.097) (1.908) (1.182) Macro_Environment -1.750 0.724 0.911 0.115 (3.056) (2.383) (0.775) (0.678) Constant 159.8 -132.9 14.04 59.08* (159.1) (131.7) (33.78) (34.77) Observations 331 331 331 331 R-squared within 0.089 0.095 0.085 0.090 R-squared overall 0.105 0.186 0.031 0.107 Number of countries 74 74 74 74

4.4 Results for developed and developing countries

As explained in the methodology, the sample is divided into developed and developing countries. Although only the inclusion of lags can tell us that results represent the influence of innovation on exports, we cannot fully neglect the results without lags since it is plausible that the effects of innovation can be observed within one year (Interview, Appendix D). Therefore, splitting the sample is executed for both estimation equation 2 and 3 for comparison. The results for estimation equation 2 are given in Table 6 and the results for estimation equation 3 are given in Table 7. For both estimation equations, there are relatively more developed countries than developing countries, but both groups are sufficiently large to base conclusions on. The sample in Table 7 consists of less observations and countries due to the inclusion of lags. In both tables, each first column (1, 3, 5, 7) per dependent variable depicts the results for developing countries, and the second (2, 4, 6, 8) depicts the results for developing countries.

(27)

27 seems to be larger within developed countries. For instance, when New_Some increases with 1 percentage point, Exports_25 increases with 0.518 percentage point. This coefficient is statistically significant at the 1% level. However, using this estimation equation, it cannot be stated with certainty that innovation influences exports. Therefore, Table 7 includes lagged independent variables. At first sight, the results in this table seem strange, because they are the opposite of the findings in Table 6. In Table 7, innovation actually has a larger impact on exports within developing countries than within developed countries. However, only a few coefficients in Table 7 are statistically significant and the results of both tables cannot be compared directly due to a different number of observations and countries. Speculating about other reasons for these contradicting findings, a potential explanation might be that the effect of product innovation on exports is observed faster in developed countries than in developing countries. That would mean that innovation has a positive effect on exports within both groups of countries, but that the effect is observed stronger and faster within developed countries. It cannot be stated with absolute certainty that innovation influences exports in Table 6 without lags, but it seems likely that innovation is translated into an effect on exports within one year within developed countries (Interview, Appendix D). Since the GEM database only provides yearly data, it is not possible to research the effect on a smaller time scale.

Additionally, differences are observed between the two independent variables of product innovation. In Table 6, the effect of New_Some on exports is larger than the effect of New_All, with the largest difference between the two observed within developed countries. Towards the highest exports level, the effect of New_All increases within developed countries, but never turns statistically significant for export levels. This suggests that, as stated earlier, the highest level of innovation is most relevant for the highest level of exports, while a lower level of innovation is more relevant to start exporting in the first place, in line with the hypothesis. In Table 7, the influence of New_All on exports is larger than the influence of New_Some on exports within developing countries, which is also in line with the hypothesis. Column 1 shows that an increase of one unit in New_All leads to a decrease of 0.318 percentage point in Exports_Null within developing countries. A potential explanation is that the effect of some product innovation (New_Some) is more quickly translated than the effect of radical product innovation (New_All) within developing countries. On the contrary, the effect of New_Some on exports seems to be larger than the effect of New_All within developed countries. It seems likely that the effect of product innovation on exports within developed countries is transmitted faster, which is why the effects after one year in Table 7 are small.

(28)

28 countries, because the financial markets in the latter are usually more advanced while they try to catch up with that level in developing countries. The negative influence of changes in the domestic financial market on exports within developing countries potentially indicates that a stronger financial market at home gives less motive to seek financing opportunities elsewhere. The within R-squared values have increase substantially, compared to previous tables. Comparing both tables, the within R-squared values for developing countries are higher in Table 7 than in Table 6, while generally speaking the opposite is the case for developed countries. Thus, a larger share of the variance in exports within developed countries is explained by estimation equation 2, while a larger share of the variance in exports within developing countries is explained by estimation equation 3. The highest within R-squared value is given in column 5 of Table 7 which shows that this estimation equation explains 28.7% of the variance in Exports_75 within developing countries over time. The within R-squared values are in line with the previous statement that product innovation is probably more quickly translated into an increase in exports within developed countries than within developing countries.

(29)

29

Table 6 Regression of export on innovation based on estimation equation 2, sample split according to development stage (1) (2) (3) (4) (5) (6) (7) (8) Variables Exports_ Null Exports_ Null Exports_ 25 Exports_ 25 Exports_ 75 Exports_ 75 Exports_ 100 Exports_ 100 New_All -0.170 -0.370* 0.154 0.152 -0.00329 0.122 0.0198 0.0960 (0.155) (0.186) (0.115) (0.181) (0.0587) (0.0728) (0.0340) (0.0578) New_Some -0.234 -0.646*** 0.166 0.518*** 0.0309 0.137* 0.0365 -0.00914 (0.141) (0.163) (0.104) (0.0978) (0.0524) (0.0746) (0.0283) (0.0503) GEM_Technology 0.0965 -0.00715 -0.109 -0.0994 0.00550 0.0410 0.00667 0.0656 (0.160) (0.130) (0.125) (0.116) (0.0368) (0.0506) (0.0230) (0.0403) LN_GDPpcPPP 14.60 2.355 -6.345 5.623 -2.168 -1.339 -6.088 -6.638 (11.90) (26.05) (9.643) (24.49) (3.356) (6.190) (3.668) (4.103) Education_ Quantity -1.155 4.515 0.0775 -4.022 0.791 -0.580 0.287 0.0876 (2.457) (2.892) (1.856) (3.276) (0.848) (0.944) (0.481) (1.130) Financial_Market 6.739 2.020 -7.982* -0.573 0.597 -1.488 0.646 0.0407 (5.142) (2.780) (4.437) (3.002) (1.123) (1.311) (0.844) (0.815) Tech_Readiness 1.500 -0.335 -1.531 -1.791 -0.420 -0.840 0.451 2.966** (3.599) (4.439) (3.009) (4.364) (1.087) (1.249) (0.815) (1.353) Foreign_ Competition 1.018 -6.599 -1.318 9.649 -0.00583 -1.045 0.306 -2.005 (5.509) (5.663) (4.550) (7.563) (1.564) (3.113) (1.243) (2.716) Macro_ Environment -5.373** 1.395 3.713** -2.332 0.657 1.461 1.003 -0.525 (2.224) (3.979) (1.638) (3.489) (0.650) (1.158) (0.640) (0.750) Constant -67.53 28.56 99.27 -32.57 19.45 31.72 48.81 72.30* (115.3) (250.3) (93.39) (236.4) (32.17) (55.62) (34.01) (38.65) Developed (=1) 0 1 0 1 0 1 0 1 Observations 245 219 245 219 245 219 245 219 R-squared within 0.135 0.229 0.158 0.173 0.145 0.138 0.094 0.148 R-squared overall 0.129 0.123 0.052 0.050 0.000 0.195 0.074 0.037 Number of countries 51 32 51 32 51 32 51 32

Country FE Yes Yes Yes Yes Yes Yes Yes Yes

Time FE Yes Yes Yes Yes Yes Yes Yes Yes

(30)

30

Table 7 Regression of export on innovation based on estimation equation 3, sample split according to development stage (1) (2) (3) (4) (5) (6) (7) (8) Variables Exports_ Null Exports_ Null Exports_ 25 Exports_ 25 Exports_ 75 Exports_ 75 Exports_ 100 Exports_ 100 New_All_L1 -0.318* 0.0172 0.234 0.0410 0.0679 -0.00795 0.0170 -0.0503 (0.173) (0.190) (0.150) (0.143) (0.0410) (0.0694) (0.0353) (0.0629) New_Some_L1 -0.0627 -0.210 0.163 0.0558 -0.0834** 0.0847 -0.0169 0.0693 (0.202) (0.163) (0.165) (0.156) (0.0380) (0.0609) (0.0282) (0.0410) GEM_ Technology 0.0139 -0.310** -0.0539 0.237** 0.0255 0.0512 0.0145 0.0220 (0.197) (0.142) (0.163) (0.116) (0.0422) (0.0599) (0.0319) (0.0504) LN_GDPpcPPP -18.10 -12.52 18.14 22.22 -0.966 2.129 0.925 -11.83* (16.81) (27.26) (14.77) (24.54) (4.500) (7.051) (6.006) (5.970) Education_ Quantity -5.175 5.528* 3.737 -1.577 1.240 -2.018 0.198 -1.934* (3.667) (3.074) (2.750) (2.163) (0.957) (1.254) (0.845) (1.051) Financial_Market 12.53* -0.121 -10.50* 1.683 -1.789 -1.696 -0.237 0.135 (6.571) (1.872) (5.430) (2.235) (1.434) (1.519) (1.380) (0.708) Tech_Readiness 6.458** 0.217 -5.999* -1.760 -1.597* -0.0973 1.137 1.641 (3.107) (3.692) (3.358) (3.690) (0.874) (1.377) (1.214) (1.566) Foreign_ Competition 2.156 -0.493 -3.536 6.472 -0.187 -2.973 1.567 -3.005 (7.932) (7.020) (6.364) (5.399) (2.137) (3.154) (1.446) (2.170) Macro_ Environment -2.938 1.788 1.330 -1.460 0.395 0.709 1.213 -1.037 (3.013) (5.234) (2.373) (4.071) (0.643) (1.382) (1.071) (0.945) Constant 199.6 152.5 -102.1 -222.9 22.65 18.60 -20.10 151.9** (157.4) (258.8) (136.2) (231.4) (42.63) (71.53) (54.71) (63.59) Developed (=1) 0 1 0 1 0 1 0 1 Observations 159 172 159 172 159 172 159 172 R-squared within 0.188 0.126 0.205 0.113 0.287 0.118 0.086 0.193 R-squared overall 0.002 0.093 0.000 0.031 0.006 0.097 0.142 0.058 Number of countries 44 30 44 30 44 30 44 30

Country FE Yes Yes Yes Yes Yes Yes Yes Yes

Time FE Yes Yes Yes Yes Yes Yes Yes Yes