1 Introduction

It is now well established that small and medium-sized enterprises (SMEs) play a vital role in the process of transition to a market economy. As the large firm sector, the prevalent form of organisation under central planning, underwent restructuring and decline, thousands of new SMEs took advantage of liberalised entry conditions and entered the market.Footnote 1 They responded rapidly to systemic shocks, produced goods and services demanded by the population and, in the process, contributed to the generation of new jobs and incomes.Footnote 2 While the contribution of SMEs to domestic output and employment has been studied by many authors, and, for many TCs, their role in cross-border trade and their contribution to exports has not been studied widely. The aim of this paper is to develop the research in this area by investigating the factors influencing the export behaviour of SMEs and by providing empirical evidence for TCs.Footnote 3

An important constraint on our analysis is the absence of a well-developed theory on the behaviour of SMEs and, in particular, on SMEs and international trade (Brock and Evans 1989; Dunning 1988, 1993, 2001; Axinn and Matthyssens 2002; and others). However, a recent strand of the international trade literature linking firm heterogeneity and participation in foreign markets has been developed. This approach, initiated by the pioneering work of Melitz (2003), argues that firm’s export entry and exit decisions are determined by the interplay of two factors: firm-level variation in productivity, and sunk costs. As Greenaway and Kneller (2007) explain, as a result of the interaction between these two factors, high-productivity firms self-select into export markets compared to less productive firms, which resort to the domestic market. Because of capacity constraints, SMEs may be expected to be less represented in export markets relative to their large counterparts due to economies of scale and the fixed costs involved in exporting activities. Furthermore, as a result of the latter, there is a high level of persistence in firms’ exporting activities, which additionally leads to a higher representation of larger firms in exporting activities.

Many of the variables we employ in our empirical investigation fall within the Melitz framework. Factors related to higher quality labour, physical capital, R&D and innovation activities, learning-by-doing, firm and industry spillovers, and others, are all productivity enhancing factors (for a comprehensive survey on the sources of productivity, see Syverson 2011). However, the concerns noted earlier with regards to the limitations of the existing theories of the firm’s internationalisation process still echo within academic circles (see Spence and Crick 2006). Hence, we draw upon a number of additional strands of thought either to substantiate or to complement Melitz’s approach to internationalisation.

Although the core of our theoretical framework is based on the Melitz (2003) approach, it is augmented with a variety of supplementary hypotheses in order to allow for other influences that may be important in the transition process or derived from the empirical literature. Transition is a process whereby countries increasingly acquire the institutional and economic characteristics of market economies. In addition, the passage from central planning to market system requires a change and upgrading of human capital and investment in physical capital as well as technology. The technology spillovers resulting from FDI were crucial in speeding up the transition process in these countries. Although these factors are important in all countries, the transition process was heavily reliant on these factors. In principle, we expect firms’ export behaviour in TCs to be mainly and increasingly influenced by similar variables to those that influence firms’ export behaviour in developed market economies. However, the study also draws on the literature in transition economics by including a number of variables to address transition specific influences and the institutional and cultural heterogeneity amongst transition countries: ownership variables (especially foreign ownership), capital city effects, which may be more important in TCs, and country dummies.

This paper employs large firm-level datasets drawn from the Business Environment and Enterprise Performance Surveys (BEEPS) conducted jointly by the World Bank and EBRD, which have remained underutilised in the area of cross-border trade.Footnote 4 Tobit estimation is used to account for influences both on the likelihood that firms will decide to export (propensity) and on the export decisions of existing exporters (intensity). Three features of our empirical strategy help to ensure the robustness of the results and subsequent inferences. First, we investigate six datasets with corresponding variations in model specification: each of the three comparable rounds of BEEPS separately, the three waves pooled, a panel of firms surveyed in all three rounds, and a 2-year panel of firms surveyed in the final two waves. Secondly, we compare the results from alternative—cross-section and panel—estimators. Thirdly, as a robustness check, we re-estimate the models using each of the six datasets made complete by the imputation of missing values.

Our estimates show, ceteris paribus, that the accumulation of human capital and technology are important sources of international competitiveness for SMEs. Consistent with this, companies with a greater percentage of highly educated workers in the workforce export more, while gross investment and new and upgraded technologies also promote exports. In addition, we find some evidence that productivity-enhancing spillovers promote SME exports. With regards to the firm-specific variables, the bigger the size of the firm, the larger the share of sales generated in export markets. Companies with a foreign capital share have better prospects for exports; the same applies to firms engaging in production activities, which are more proficient in foreign markets than are non-production companies. The availability of external finance appears to be a significant determinant of the export behaviour of SMEs in TCs, and so does membership of business associations, which enhances the networking ability of firms. Finally, period dummies highlight 2005 as the best performing year in terms of export performance, and, as expected, country dummies capture major differences in firms’ export behaviour in transition countries.

Using Jones and Coviello’s (2005) language, our study puts in place a few ‘pieces of the puzzle’ in the firm’s internationalisation process in TCs. The contribution of this study can be viewed from different perspectives. First, we fill an important gap in SME internationalisation literature by studying SME export behaviour—by which we mean both export propensity and export intensity—in TCs. Although there are many studies of internationalisation of firms’ entrepreneurial activities, those relating to international activities of SMEs in transition are very scarce,Footnote 5 and lagging well behind investigations linking SMEs with other developments in the economy, such as growth and employment. The second contribution relates to the large number of mainly supply-side factors included in the analysis, reflecting a comprehensively specified model of export behaviour. The third contribution concerns research practice. We apply multiple imputation techniques to the BEEPS datasets, because—we argue—this enables us to utilise these data more fully, which is an issue ignored by previous authors using these surveys.Footnote 6

The paper proceeds with Section 2 where we present the theoretical reasoning informing the investigation. Section 3 presents our empirical strategy and the datasets. Section 4 reports and discusses the econometric results. The final section concludes.

2 Determinants of export behaviour

An important constraint on our analysis is the lack of well-established theories explaining the behaviour of SMEs in the economy, specifically their internationalisation decisions. In a series of studies, Dunning (1988, 1993, 1995, 2001) argues that there is no single theory of international trade able to fully explain a firm’s international expansion. Dunning explains (1995, p. 165):

… the nature and character of international transactions have so much changed in recent years, that the traditional intellectual apparatus of the international economist is, by itself, no longer adequate to explain real-world phenomena, and only by drawing upon the tools of other branches of economics, notably, industrial, institutional, and techno-growth economics, can contemporary cross-border flow of goods, services, and assets be properly understood.

Dunning’s position on the incompleteness of international trade theory continues to resonate with his peers to this day. More recent studies (Axinn and Matthyssens 2002; Jones and Coviello 2005; Crick and Spence 2005; Spence and Crick 2006; Thai and Chong 2011) share the same concerns. Axinn and Matthyssens (2002) provide a review of the existing theories of internationalisation, which include: industrial organisation theory, the resource-based view, transaction costs theory, the amalgamation of these three in the form of Dunning’s eclectic paradigm, the Uppsala model of internationalisation, innovation-based models network theory, and other approaches. In the same vein as Dunning, they argue that current internationalisation theories fail to explain and predict the behaviour of firms in the global marketplace, primarily because theoretical developments have been unable to keep pace with the rapidly changing, hyper-competitive global environment. Further, it is argued that each theory explains a specific aspect of firm behaviours in a specific environmental context. If the latter holds true, then attempts to develop an approach to explain and predict firms’ internationalisation process in the transition context, the subject of this study, are virtually non-existent. Yet, a transition country setting, according to Thai and Chong (2011), provides a unique backdrop characterised by distorted information, weak market structures, poorly specified property rights, and institutional uncertainty, making existing explanations of firms’ internationalisation process less convincing.

However, a stream of recent studies (Melitz 2003; Melitz and Ottaviano 2003; Bernard et al. 2003; Helpman et al. 2004; Arnold and Hussinger 2005; Aw et al. 2007, 2008; among others) on firm heterogeneity and participation in international markets has provided a comprehensive framework for analysing factors influencing firms’ decisions to internationalise.Footnote 7 According to this line of thinking, export entry and exit decisions are determined by differences in firm productivity levels and incurred sunk costs. According to Melitz’s (2003) dynamic industry model of heterogeneous firms, high-productivity firms self-select into export markets. An important ingredient of the theory is the recognition that entering export markets incurs sunk costs. As Greenaway and Kneller (2007) explain, firms have to engage in market research, train people, modify products to respond to local requirements, establish new distribution networks, etc. The importance of sunk costs has been recognised for some time (see Dixit 1989; Dixit and Pindyck 1994), and their effect on export entry has been demonstrated empirically (Roberts and Tybout 1997; Bernard and Jensen 2004; etc.). These studies argue that the presence and the magnitude of sunk costs generate large hysteresis effects.

To guide the specification of the empirical model, we draw upon the Melitz dynamic model of export participation as well as on other lines of thought arguing for the inclusion of human-related factors, technology-related factors, industry linkages, and other firm characteristics. Most variables included in the empirical model share a common characteristic, namely, that they are supply-side variables in that they relate directly to the ability of firms to produce. In line with previous research, we use export intensity (foreign sales as a percentage of total sales) to measure the degree of firms’ involvement in foreign markets (Bonaccorsi 1992; Calof 1993; Wakelin 1998; White et al. 1998; Becchetti and Rossi 2000; Wagner 2001; Gorodnichenko et al. 2010; and others). The theory that export behaviour is determined essentially by the interplay of productivity levels and the fixed costs of exporting suggests that the same factors will affect both the firm’s propensity to export and, if it exports at all, the firm’s export intensity (see Melitz 2003, pp. 1695–1696 and Greenaway and Kneller 2007). This influences our empirical strategy discussed below. To anticipate, the estimated effects of our independent variables represent the combined effects of two channels of influence on our dependent variable: namely, influences on the likelihood that firms will decide to export (propensity); and influences on the export decisions of existing exporters (intensity). We refer to these estimates as influences on export behaviour. We do not refer to export performance, because our dependent variable does not directly correspond to conventional efficiency measures.Footnote 8

2.1 Human capital related factors

The importance of human resources has been extensively examined at the country and firm level. These studies systematically highlight the supremacy of human capital for the sustained comparative and competitive advantage of nations and firms. Human capital is at the core of the New Growth Theory, which argues that the creation and diffusion of knowledge is the primary engine of economic growth (Grossman and Helpman 1994). At the micro-level, human capital factors affect firms’ export propensity and intensity through increases in productivity (Arnold and Hussinger 2005; Bryan 2006; Kleynhans 2006; Kagochi and Jolly 2010). The connection between the quality of labour variables such as education, training, overall experience, and tenure at the firm and the firm’s productivity has been investigated in a growing body of work (see Syverson 2011). For instance, Chevalier et al. (2004) and Carlin et al. (2001a) argue that higher levels of education or skill acquisition signal or enhance productivity. In addition, according to Bryan (2006), training helps to sustain higher levels of productivity. In our model of export behaviour, we measure the impact of human capital accumulation through several proxies: (1) the education of the workforce, (2) on-the-job-training, (3) the presence of highly skilled workers within the firm, which also includes the managerial staff and other professionals, (4) changes in organisational structureFootnote 9, and (5) the general manager’s education.

First, a number of studies (Keeble et al. 1991; Wood 1991; Dex and Scheibl 2001; Dex 2002; Power and Reid 2005; etc.) argue that SMEs are more inclined to have flexible organisational arrangements than are larger firms, because of their limited scope of operations, well-understood relationships within the firm, relatively simple organisational structures, ease of accessing networks of firms, etc. Conversely, Meijaard et al. (2005) argue that organisational structures within SMEs are much more complex than is argued by transaction costs and agency costs theories. We investigate whether or not organisational flexibility translates into higher export intensity and propensity by introducing a dummy variable indicating whether or not a firm underwent any organisational transformation (from minor reallocations to adoption of completely new organisational arrangements) in the previous 3-year period. Second, the link between firm management and firm productivity is well established. However, as Syverson (2011) argues, the literature has yet to dig deeper into the role of managers in productivity gains. This study aims to shed some light on this question by employing a variable that depicts the level of education of the general manager and its influence on firm’s export behaviour. The above discussion leads us to the following hypothesis:

Hypothesis 1

The quality of labour in SMEs in transition countries is positively related both to the propensity of firms to export (i.e. the likelihood of exporting at all) and to the intensity of exports by those firms that do export.

2.2 Technology-related factors

A significant body of literature has concentrated on explaining the productivity–export relationship through firm-level investments in productivity enhancing activities. Most of these studies have focused on R&D investment (Aw et al. 2007, 2008, 2011; Esteve-Perez and Rodriguez 2012; among others). Less attention has been paid to the impact of investment in physical capital on increasing firm’s productivity levels (see Syverson 2011). We start with the latter; first, following Carlin et al. (2001a), we use gross investment in capital goods as a proxy for embodied technological change, and expect it to have a positive impact on the firm’s productivity, leading to better export behaviour of the firms under consideration. Second, R&D expenditure can be used as an indicator of innovation activities (an input measure of innovation) to investigate its effect on the export behaviour of firms.Footnote 10 Third, the introduction of new or upgraded technology or new or upgraded products can be used as another, broader indicator of the innovation process, expected to have positively affected the firm’s export behaviour. Finally, a firm’s level of technology relative to its main rivals may also be used as an indication of technological progress, with positive impact on export behaviour. These indicators are expected to translate into similar changes in export behaviour, i.e. a higher propensity to export and a greater intensity of exporting.

The estimated relationship between the technology-related variables and export behaviour is potentially flawed by endogeneity, caused by reverse causation. Aw et al. (2011) summarises recent work on the firm’s investments in technology adaptation and the latter’s impact on the productivity–export link. Their survey shows that exporting and technology-related investments are interdependent firm decisions, and both may endogenously affect the firm’s future productivity. In our case, because of the way in which the above variables are defined in some waves of the survey, such endogeneity would be precluded.Footnote 11 We hypothesise that past technical progress may influence current export intensity. However, we have no such reasons for hypothesising that current export intensity could affect past technical progress (see Table 1 for the description of variables). In such cases, the activities captured by these questions substantially lag current export intensity, our dependent variable. Finally, the dummy variable modelling the firm’s technology level relative to its competitors reflects repondents’ judgements that can only arise from past experience and corresponding accretion of knowledge. In this case, this variable too refers to a period preceding the one in which respondents estimate their current export intensity.

The above discussion informs the following hypothesis:

Hypothesis 2

The physical capital, technological capabilities—R&D expenditures and innovativeness—and technological sophistication of SMEs in transition countries are positively related both to the decision to export and to the intensity of exporting.

2.3 Productivity-enhancing spillovers

We investigate two types of productivity-enhancing spillovers that occur when the activities of a firm lead to improvements in the technology or productivity of other firms, namely, economic externalities that may arise from agglomeration, and/or industry linkages, especially vertical linkages through input-markets in intermediate goods (see World Bank 2009).Footnote 12 We consider these in turn.

We start from the view that localisation and urbanisation economies generated by the concentration of firms in cities play an important role in the overall performance of firms, including their export activities (Audretsch and Stephan 1999; Fujita and Thisse 2002; Fujita et al. 1999; etc.). In our specifications, we concentrate on the impact of the location of SMEs in capital cities on firms’ export behaviour. The process of transition can be most markedly observed in the capital cities, which have by far overtaken other parts of these countries (even in those countries that have joined the EU). Therefore, we expect firms located in these cities to be able to benefit from agglomeration economies (specifically resulting from a favourable environment for identifying and exploiting synergies between previously unrelated industries, knowledge spillovers, university and research institutions, access to the pool of higher quality work force, etc.).Footnote 13

We also investigate the impact of industry spillovers derived from sales to multinationals and large domestic firms. The recent literature, including those on TCs, has emphasised the positive effects of knowledge spillovers from MNEs on domestic firms (Greenaway et al. 2004; Kneller and Pisu 2007; Sutton 2007). Greenaway et al. (2004) is the first study to empirically test export spillover effects from MNEs to domestic firms. Among the channels of export spillovers they identify are two that might be facilitated by relationships arising from sales to MNEs; namely: (1) information externalities—a transfer of knowledge and experience from MNEs operating in the host market, and (2) demonstration and imitation effects—domestic firms acquiring new technologies and management techniques used by MNEs. However, little attention, especially in the transition literature, has been given to the spillovers from large domestic companies to SMEs and import–export transformation. With regards to the connection between export behaviour and spillovers from large domestic companies, we rely on the same reasoning as that used for sales to MNEs due to a number of theoretical and empirical considerations related to large firms, domestic or foreign. First, there is overwhelming evidence that large firms are more export-oriented than small firms (see the discussion on firm size below); hence, any form of linkage between SMEs and large firms is likely to produce information spillovers. Linked to the first, a second observation is that larger firms are better resource endowed than SMEs, meaning that they are more likely to be at the frontier of technology adoption and adaptation. As a result, there is a likelihood that SMEs will also benefit from imitation and demonstration effects from large domestic firms.

In addition, the literature on international trade establishes the importance of imports for the development of domestic firms and industries through a number of channels. One channel concerns the importance of imports for firms’ capital formation. Firms benefit from imported technology, but also from the possibility that firms will make simultaneous investment to assimilate this technology (see Aw et al. 2007). Second, imports are used as inputs in export production processes (Arize 2002). Accordingly, we integrate the import intensity factor as one of the determinants of exports.

Hence, in the context of productivity-enhancing spillovers, the hypothesis states:

Hypothesis 3

The propensity as well as the intensity of exporting of SMEs in TCs are positively affected by: (Hypothesis 3a) agglomeration-induced externalities, and (Hypothesis 3b) industry linkages, especially vertical linkages through input-markets.

In practice, it is difficult to identify the separate effects of externalities associated in large part with, respectively, geographic proximity and vertical linkages; for example, both may be promoted by institutions promoting workforce (re)training and/or knowledge creation and transfer. To capture as distinctly as possible—given the available data—the influence of these two different types of externalities, we use different indicators for the two types of externality. To capture spillover effects arising from industrial linkages, especially vertical linkages through input markets in intermediate goods, we estimate the influence of sales to MNEs and large domestic firms as well as of import intensity (Hypothesis 3b). Within the framework of our multivariate model, the inclusion of these three variables should minimize the extent to which our capital city variable also captures spillover effects arising from vertical links, thereby maximising the extent to which the estimated capital city effect captures agglomeration economies (Hypothesis 3a).

2.4 Other firm characteristics

Discussion on the firm size–export relationship has produced a number of hypotheses. First of all, the literature mainly supports the export proficiency of larger firms relative to smaller firms on the grounds of resource availability and lower transaction costs (Brock and Evans 1989; Kim et al. 1997; Acs et al. 1997; Wakelin 1998; Bleaney and Wakelin 2002; etc.). However, a number of studies support the idea that smaller firms perform better in export markets due to their inherent flexibility (Mills 1984; Mills and Schumann 1985; etc.). Moreover, firm size as measured by the number of employees may be potentially endogenous. As a result, we use a lagged size variable, that is, the number of employees working for the company 3 years earlier. Finally, the size variable may have non-linear effects on the degree of a firm’s export involvement. Accordingly, we transform the model into a quadratic form, by including both the natural logarithm of the total number of employees for each firm and its squared value.

Firms’ business experience influences their productivity level (learning-by-doing effects). Syverson (2011, p. 344) argues that ‘experience allows producers to identify opportunities for process improvements’. The importance of firm experience in export markets has been widely acknowledged. To test for the latter, we rely on Learning Theory—rooted in the behavioural theory of the firm—which argues that development of knowledge may have an impact on perceptions of opportunities offered by greater internationalisation (Clercq et al. 2005). In addition, studies have identified a non-linear relationship between business experience and export growth, which we also test. Everett and Watson (1998) argue that the rate of failure among younger firms is higher than among experienced ones, due to the greater variability in their cost functions when they begin operations. Everett and Watson (1998) concentrate on firms’ experience in the domestic market. However, this effect may be more pronounced in foreign markets, where cost variability is likely to be higher to the extent that foreign markets are unfamiliar and entrepreneurs face lack of information and different systems as well as different languages and cultures. In a similar vein, Arnold and Hussinger (2005) argue that experience may be important for younger firms but not for older ones. They argue that there is a certain threshold age, beyond which a firm is unlikely to gain more experience. To test for this non-linear effect, we specify our models with both firm age and its squared value.

The Industrial Economics literature and the literature on transition economies have established the impact of ownership structure on firm performance. Demsetz (1997, p. 429), for example, argues that wealth and its distribution among different stakeholders matters to society’s productivity. In this paper, ownership structure refers to the ‘type of dominant owner’ of a company. The BEEPS dataset allows us to identify the largest shareholder of a company (thus the dominant shareholder) as a foreign entity, the state or a private (domestic) individual or company. There is already a well-established literature on the importance of ownership structure for firm performance in transition economies. There is almost complete unanimity in the transition literature that dominant foreign ownership has a positive and significant impact on the performance of firms. Private (domestic) ownership is expected to be the next, i.e. it is also expected to have a positive effect on the performance of firms. Firms owned by the state are expected to be least well performing given the problems of state ownership and the shortage of resources needed to restructure state-owned companies. Yet the empirical work on the impact of private ownership is rather mixed. Private ownership does not immediately improve the performance of privatised firms: it takes time for the new owners to be able to engage in strategic restructuring and gradually improve the firm’s performance. Private owners taking over in the course of privatisation do not always find it easy to obtain the necessary credit to finance strategic restructuring and invest in new technology. For a while, therefore, the difference between firms with dominant private ownership and state ownership, particularly over a short period of time, may not be significant.

Search and information costs take a central position in the transaction cost theory. For successful export activities, primarily, a systematic collection of information is required, since it can act as a catalyst to reduce the uncertainties of the international environment (Leonidou and Adams-Florou 1999). Due to their resource constraints, SMEs appear to be more dependent than large firms on services, information and contacts generated through business associations (Bennett 1998). Hence, we investigate the impact of membership of business associations on SME export behaviour.

Many country-level studies have demonstrated the importance of the development of financial markets for firms’ export activities (Beck 2002, 2003; Manova 2006; etc.). SMEs have even greater need for credit relative to large firms due to their limited capital resources. Moreover, SMEs face greater difficulties in obtaining external finance (due to information asymmetries and/or institutional factors), which may be reflected in their overall performance, including international activities (Beck et al. 2006, 2008; Hutchinson and Xavier 2006). Hence, we investigate the export effect of the availability of external finance.

Two additional firm-related factors that we investigate are the level of capacity utilisation (facilities and manpower) and market share. Conventionally, capacity utilisation is treated as a simple proxy for pressure of demand. However, more recent literature suggests a more subtle supply-side interpretation. Drawing on Bansak et al. (2007), capacity utilisation may reflect the possibility that new and relatively cheap technologies available to firms allow them to hold excess capacities to respond to increases in demand. In other words, when there is a boost in export demand then SMEs use their excess capacities to respond. They explain as follows (p. 633): “…new technologies may make it easier to ramp production up and down. Combined with falling prices of high-tech equipment, this may encourage firms to install a broader margin of excess capacity—operating at lower average utilization—to be able to handle upswings in demand”. These alternative approaches imply different signs on the estimated effect of capacity utilisation: a positive relationship with export activity in the case of the demand-side interpretation, a negative one in the case of the supply-side interpretation, and an insignificant effect in the event either that neither effect is present in the data or that both are and offset one another. The demand-side explanation raises a further issue; namely, because greater exports may increase demand pressure and thus capacity utilisation, there may be simultaneity and thus potential endogeneity. However, as with the technology variables, the potential endogeneity of capacity utilisation is addressed by using the question on the level of capacity utilisation that refers to a period before the current one to which export intensity refers.Footnote 14 Second, we assume that firms with a greater share of the domestic market would have an incentive to try to expand their activity across borders to take advantage of additional demand in foreign markets. Accordingly, we anticipate that the likelihood of exporting would be higher for firms that have a larger share of the domestic market.

Finally, we control for differences in the exporting behaviour of firms for each individual country and over time. Regarding the former, we introduce country dummies to take into account that there are significant institutional, cultural, and other differences among TCs. Moreover, the country dummies play an important role in our estimation strategy by controlling for influences that otherwise would be difficult to address.Footnote 15 On the other hand, the three rounds of the survey have taken place in three different periods with different external environments, covering almost a decade of varying conditions affecting the export behaviour of firms across the sample (such as rates of growth in the EU). Whether and how these changes were reflected in the firm’s export propensity and intensity we aim to capture through period dummies.

3 Empirical strategy

3.1 Methodology

As the BEEPS database contains information on exporters and non-exporters, the dependent variable (y)—percentage share of export sales in total sales—is zero in a significant number of cases (i.e. for non-exporters), and the observations for exporters are roughly continuous over the positive range of values. This type of data is addressed by the generalised tobit model (Wooldridge 2006, p. 587). In tobit estimation, zero-value observations are incorporated into the model as the outcome of a decision-making process. In effect, tobit estimation models a dual decision making process: in our case, firms’ decisions as to whether or not to export, and, if so, how much to export. In this way, tobit estimation addresses the potential endogeneity of our independent variables that would arise if the self-selection of firms into the export market were to be omitted from the model. This differs from two-step models that first estimate the determinants of the decision as to whether or not to export, and then, in a second-stage regression, the determinants of firms’ exports conditional upon the probability of exporting at all. Which approach is the more appropriate depends upon whether or not the two parts of firms’ export decisions have the same or different determinants. The theory informing our empirical specification suggests that the same factors affect both the firm’s propensity to export and, if it exports at all, the firm’s export intensity (see Section 2, above). Moreover, diagnostic assessment of these alternatives—i.e. the same or different determinants—also endorses the tobit model as a valid estimator for our data (reported below). Finally, an additional complication is suggested by the dependent variable being defined as a proportion, therefore bounded by zero and one. However, this should not be a problem in our samples, because few firms export all or even nearly all of their output. Moreover, robustness checks using estimators capable of directly addressing this problem yield estimates fully consistent with those reported below.Footnote 16

The model for cross-section data has the following form:

$$ \begin{aligned} & y_{i} = \left\{ {\begin{array}{ll} {x_{i} \beta + \varepsilon_{i} > 0} & {{\text{for}}\,{\text{exporters}}} \\ {0} & {{\text{for}}\,{\text{non-exporters}}} \\ \end{array} } \right. \hfill \\ \varepsilon_{i} \sim N(0,\sigma^{2} ) \hfill \\ & \end{aligned} $$

where i = 1, 2,…, N firms, x i is a 1 × (k + 1) vector containing the k variables of interest discussed in Section 2 together with an intercept term, and β is the corresponding (k + 1) × 1 vector of coefficients to be estimated, and ε i are independently and individually distributed (iid) over the whole sample with mean zero and variance σ 2. And, for panel data:

$$ \begin{aligned} & y_{\text{it}} = \left\{ {\begin{array}{ll} {x_{\text{it}} \beta + \alpha_{i} + \varepsilon_{it} > 0} & {{\text{for}}\,{\text{exporters}}} \\ 0 & {{\text{for}}\,{\text{non-exporters}}} \\ \end{array} } \right. \hfill \\ & \alpha_{i} \sim N(0,\sigma_{\alpha }^{2} ) \hfill \\ & \varepsilon_{it} \sim N(0,\sigma_{\varepsilon }^{2} ) \hfill \\ \end{aligned} $$

where, in addition, t = 2002, 2005 and 2008/2009, x it is a 1 × (k + 1) vector containing the k variables of interest together with an intercept term, the a i are time invariant (fixed) effects distributed iid over the firms with mean zero and variance σ 2 a , and the ε it are iid over the whole sample with variance σ 2 ε .

Although a 2-year or 3-year panel sample may not be sufficient to identify any dynamics in the data, it is sufficient to estimate a tobit model, which accounts for unobserved effects that are constant (fixed) over time but vary between firms by means of the firm-specific error term α i .Footnote 17

Further, we follow Wooldridge (2002, pp. 521–524) who distinguishes between two types of marginal effects: the ‘conditional’ marginal effects, which account for changes in the expected (or predicted) value of exports (y) for the subpopulation of firms for which exporting activity is observed (y > 0), and the ‘unconditional’ marginal effects that account, in addition, for the effect of changing values of the independent variables on the probability that exporting will take place at all (i.e. will change from zero to positive and, hence, be observed). For dummy variables, both conditional and unconditional marginal effects are calculated as the discrete change in the expected value of the dependent variable as the dummy variable changes from zero to one.

3.2 The dataFootnote 18

The data used in this investigation are from BEEPS, an extensive survey targeting the business environment and the performance of enterprises in TCs (see Appendix Table 5 for a two-way table showing the countries and number of observations for each country in each round of the survey). We employ data for 2002 including 5,330 firms, 2005 with 5,385 firms, and 2008/2009 with 7,247.Footnote 19 As our study concentrates on the SME sector, we utilise these datasets by estimating six specifications of the econometric model on five samples of SMEs: namely, three for the individual years 2002, 2005 and 2008/2009, a pooled dataset, a panel of companies surveyed in 2005 and 2008/2009, and, a panel of companies surveyed in all three rounds.Footnote 20 , Footnote 21 The definition of our variables and their summary statistics are provided in Table 1 and in the Appendix, Table 6.Footnote 22

Table 1 Description of variables used in the econometric specifications
Full size table

The summary statistics (Appendix, Table 5) show a fairly low involvement of SMEs in export activities: data from the sample specifications show that only around 20 % of SMEs in the TCs engage in export activities. In addition, as expected, the calculations also show that export markets are not the domain of SMEs, with the average share of export sales in total sales at around 7 %. However, SMEs in TCs seem fairly well endowed with human capital, in terms of the education of their workforce and the skills that they possess. Moreover, sample companies have continuously invested in new infrastructure and technology. With regards to the latter, over 65 % of companies in the pooled sample have introduced new product lines (as well as upgrading old ones) and new technologies during the last decade.

Other firm-specific data show that, on average, sample firms have just over 40 employees, and are overwhelmingly privately (around 95 %) and domestically owned (on average around 90 %). Less than one-third are located in the capital city, generally engaging in trade and services. Data show that imports are a significant source of inputs and other supplies, while large domestic firms are an important market for SMEs, more than are foreign MNEs. Finally, over one-third of sample companies are members of business associations.

3.3 Robustness checking—handling missing data

The proportion of missing data for most variables in the samples is fairly low (see Appendix, Table 6).Footnote 23 However, both the gross investment and spending on R&D variables have a high percentage of missing responses. Given the prominence of these variables in our discussion (Section 2 above), we do not want to drop them entirely from our estimated models: first, because their particular effects are of interest, and second, because we want to demonstrate that the estimates of our other variables of interest are not greatly affected by the inclusion or omission of these two variables. Moreover, we argue that it is not an entirely satisfactory option to drop observations with missing values of these variables. This is for two reasons: first, because we would then be unable to utilize fully the data available for other variables, and second, because dropping observations with missing values (“listwise deletion”) is justified only on the strong assumption that these values are ‘missing completely at random’ (MCAR); otherwise, the consequence is again inefficient and biased coefficient estimates (Schafer and Graham 2002), arising from differences between the distribution of the missing observations and the distribution of the observed items. We conclude that it is not satisfactory to drop either variables or observations. Consequently, we imputed the missing values (the percentage of imputed values for each variable corresponds to the percentage of missing values detailed in Appendix, Table 6). To this end, we applied multiple imputation (MI) as the technique most favoured in the statistical literature on analysing survey data with missing values.

Rubin (1987, p. 2) defines MI as a “technique that replaces each missing or deficient value with two or more acceptable values representing a distribution of possibilities”. The MI procedure first estimates an imputation model on the observed data m times to create m complete datasets (i.e. datasets in which the observed data is completed by the imputed values). To preserve the features of the joint distribution in the imputed values, the imputation model includes both factors known to be associated with missingness and factors that explain the variation of the target variables (Schafer and Graham 2002; Van Buuren et al. 1999). Then, each dataset is analysed by an estimation method specified by the researcher. Finally, the results obtained from the m analyses—one for each dataset—are combined to produce parameter estimates and standard errors that fully reflect missing-data uncertainty. The particular feature of MI is that it allows for both within-imputation and between-imputations dimensions of randomness—hence, errors—in estimating parameters, which are reflected in the accompanying standard errors. This procedure yields valid estimates from imputed datasets of the standard errors in addition to approximately unbiased estimates of all parameters. This is the most significant advantage of the MI technique.Footnote 24 Moreover, MI rests on the assumption that values are “missing at random” (MAR), which is a weaker assumption than the MCAR assumption entailed by analysing a dataset with missing values.Footnote 25

Accordingly, in all six datasets used for estimation—i.e. the 2- and 3-year panels, the pooled, and the cross-section samples for 2008/2009, 2005 and 2002—all the missing values are imputed, regardless of the number of missing values for individual variables. Consequently, the sample sizes have increased substantially in relation to the non-imputed samples: the longitudinal datasets by 14 and140 %, respectively, the pooled dataset by 7 %, and the 2008/2009, 2005 and 2002 datasets by, respectively, 13.5, 25 and 330 %.Footnote 26 This large increase in the size of the datasets is reflected in the results. Although the signs on the estimated effects in the imputed and non-imputed samples are remarkably consistent, some estimates differ in their statistical significance and in their magnitude. For survey datasets with the typical characteristics displayed by BEEPS, we propose that results from imputed datasets should be taken to complement, rather than substitute for, the results from non-imputed datasets. Accordingly, the results from MI serve as a robustness check of the non-imputed results.

4 Results and discussion

We analyse a range of datasets and apply different specifications (reflecting nonconformities between the surveys). The consistency of estimates across the different datasets, both imputed and non-imputed, and the correspondingly different specifications, suggests robustness of the reported results. The results are overwhelmingly consistent in terms of the direction of the estimated effects. Moreover, most of the estimated coefficients are consistent across different specifications in terms of statistical significance. There are slight differences in the magnitude of the coefficients, albeit not worth dwelling upon as they do not imply any change in our conclusions.

Tables 2 and 3 report the unconditional marginal effects and their respective standard errors from estimating tobit models on the six samples; in each case, both non-imputed and imputed. We comment in detail only on the unconditional marginal effects, because these refer to the whole population of firms (i.e. both potential exporters and those that are exporting), and are therefore the effects most relevant for our discussion.Footnote 27 To adopt a conservative approach to inference, we report cluster-robust standard errors for the cross-section estimates (clustered on country). For the panel estimates, the standard maximum likelihood estimates are compared to bootstrap estimates, with no noteworthy difference in the corresponding levels of statistical significance.Footnote 28

Table 2 Unconditional marginal effects: panel tobit estimation of the determinants of SME export behaviour in transition countries
Full size table
Table 3 Unconditional marginal effects: tobit estimation of the determinants of SME export behaviour in transition countries
Full size table

To assess the validity of tobit estimation, there is a diagnostic check suggested by Greene (2003b, p. 768) and Wooldridge (2002, p. 534), which evaluates the appropriateness of the tobit model. As this check requires, we find that the probit and tobit coefficient estimates are consistent after appropriate transformations.Footnote 29 , Footnote 30 This diagnostic check suggests that, for each of the six regressions on the non-imputed samples reported in Tables 2 and 3, the determinants are similar for both the propensity of firms to export in a particular period and the intensity of exports by those firms that do export in a particular period. This finding is consistent with the theory informing our empirical specification and has useful policy implications, which are developed in the conclusion to this paper. Here, the import of these checks is to suggest that, in each case, the tobit models provide consistent and unbiased estimates.

For ease of interpretation, Table 4 provides a summary of the outcomes for each econometric model estimated for the TCs. To provide a robustness check, Table 4 also summarises the effects of MI, by providing information on the results from the imputed datasets as well as highlighting differences with the estimates arising from the non-imputed datasets. These results take the empirical analysis as far as is permitted by the available data, which is restricted to cross-section samples together with the 2- and 3-year panels. Nonetheless, our results do indicate the importance of various factors influencing the export behaviour of firms.

Table 4 Summary of estimation results for SME export behaviour in TCs
Full size table

First, the human capital measures affect positively firms’ export behaviour. The hypothesised positive relationship between human resources and exporting—Hypothesis 1—has been confirmed by the estimated effects of the education of the workforce (ftwor_edu), which shows that the greater the percentage of employees with higher education, the higher the expected percentage share of exports in a firm’s sales. The unconditional marginal effects for the samples indicate, ceteris paribus, that a 1 % increase in the pool of employees with higher education will increase the percentage share of exports in a firm’s turnover, within a range of between 0.030 and 0.064 %. This indicates that better quality of human resources at the firm level may produce lower unit costs and/or higher quality of production, enabling firms to trade higher quality products at lower prices in international markets.

However, association between export behaviour and the other human-related variables is weak at best. The estimates of on-the-job training (training) are statistically insignificant. The explanation for this result can be viewed from the resource-based perspective. The literature on human resources supports the view that large firms commit greater resources to training than do small firms (Bryan 2006; Colombano and Krkoska 2006; etc.). Bryan (2006, p. 635) explains that “small firms are less likely to train employees than larger firms, because they suffer higher labour turnover and higher failure rates, and they tend to have shallow hierarchies that limit long-term career prospects”. Flexibility of SMEs with regards to organisational structure (org_str) also has no significant effect on export behaviour. As we pointed out, the literature hails SMEs for their ability to change and adapt, treating this as source of their competitiveness. Especially in the transition context, with many countries with unstable business environments, we assumed that this factor would come into play. However, the resulting insignificant coefficient should not be viewed as a complete story. The lack of data made it impossible to investigate any other form of flexibility—such as price, product or technology—in relation to the export behaviour of firms. Finally, education of the CEO (ceo_edu) produces a positive relationship with the export behaviour of firms. In the imputed dataset for 2002, the coefficient is significant at the 5 % level. As pointed out, the evidence on the education of managers and its impact on firms’ behaviour is scarce. Gottesman and Morey (2006) believe that managers with higher educational attainment will be more adaptive and innovative, and more likely to possess other characteristics that may improve firm’s export propensity and intensity. Following the discussion of firm flexibility, we can argue that being adaptive and innovative is crucial for a successful manager in the highly uncertain transition environment.

We argued that the accumulation of technology (through investments in capital goods and innovation) will lower costs, enhance the quality characteristics of products and determine the emergence of new products. These are routes for increasing firms’ productivity and, thereby, increasing competitiveness in the market. The hypothesised positive relationship between technological and innovation capabilities and exporting—Hypothesis 2—has been confirmed by the consistently positive and significant coefficients on the dummy variable, capturing whether or not the firm has “established new, upgraded a product line or introduced a new technology” in the recent past (prli_tech). The estimated unconditional marginal effect ranges from around 1 to 3, meaning that firms reporting recent product or process innovation export up to 3 % more of their output than firms that do not. The other three technology-related variables, namely, the investment-sales ratio (gross_inv), R&D intensity (inv_rd), and the assessment of the firm’s technology relative to its competitors (tech) generally yield insignificant results, especially the R&D intensity variable. Gross investments and firm’s technology relative to its competitors appear significant for the imputed 2002 sample. For the 2002 sample, gross investment is reported as percentage of average annual sales in a previous period: hence, the statistically significant estimated coefficient for gross investment in the imputed sample means that a 1 percentage point increase in the share of gross investment in sales leads to an increase in the export share in sales of 0.075 of a percentage point. With regards to the firm’s assessment of its technology level, a discrete change from 0 to 1—signifying better technology relative to firm’s competitors—increases the export share in total sales by up to 0.8 of the percentage point.

Among the variables included to capture productivity-enhancing spillover effects, we find a suggestive contrast between the capital city effect (entres) estimated from those samples in which the three variables measuring industry—especially vertical—linkages (sales to MNEs—mne_sal, sales to domestic large firms—large_sal, and import intensity—impint) are absent and the capital city effect estimated from those samples in which the vertical linkages effects are present:

  • two of the samples in which the industry (vertical) linkages variables are absent (the 3- and 2-year panel samples) yield a statistically significant capital city effect; yet

  • both samples in which the industry (vertical) linkages variables are present (the 2002 and 2005 samples) yield insignificant capital city effects.

This contrast suggests that, when the capital city indicator is the only variable able to capture productivity externalities, whether arising from agglomeration or industry (vertical) linkages, then it may capture all such effects. This is informative about such effects in general but cannot tell us anything about the origins of such effects. However, the 2002 and 2005 samples, which contain both the capital city effect and the three variables measuring industry (vertical) linkage effects, yield highly significant linkage effects of the anticipated sign but utterly insignificant capital city effects. Together, these results are consistent with Hypothesis 3b (that industry—especially vertical—linkages positively influence export behavior in TCs) but tend to refute Hypothesis 3a (that agglomeration effects associated with capital cities positively influence export behavior in TCs).

On this interpretation, no capital city effect on the exporting of SMEs in TCs has been identified (location), although one would anticipate such a relationship from the prominence given to informal networks and communication in studies of firms’ performance in TCs as productivity-enhancing mechanisms (see Smallbone and Welter 2001). In contrast, statistically significant results are found for the effects of sales to MNEs (mne_sal) and large domestic companies (large_sal) as well as import intensity (impint). These mirror the productivity spillovers derived from industry linkages. With regards to the interactions with MNEs and large domestic companies, our findings are in line with the outcomes in Konings (2001), Yudaeva et al. (2003), Javorcik (2004) and Gorodnichenko et al. (2010), among others, linking knowledge transfers from MNEs to domestic firms. The unconditional marginal effects for non-imputed samples in 2002 and 2005 indicate, ceteris paribus, that a 1 % increase in the share of sales to multinationals will increase the percentage share of exports in a firm’s turnover by 0.075 and 0.080 %, respectively. A novelty of our investigation is the indication that the same effect is produced by interaction with large domestic firms. Here, too, the unconditional marginal effects for non-imputed samples indicate, ceteris paribus, a 1 % increase in firms’ sales to large domestic companies will increase the percentage share of exports in total sales by 0.031 and 0.073 %, respectively. It is logical to assume that within industries there is no single source of spillovers, but rather all gain from interactions with each other. With regards to the final industry-linkage variable, the higher the share of imported inputs, all things being equal, the greater the propensity to engage in exporting as well as to export more. Damijan et al. (2004) argue that firms can exhibit significant productivity gains especially from serving advanced markets. Following this logic, we believe that not only exporting to advanced countries but also importing (or any other form of international interactions) will induce productivity increases in SMEs in TCs. This is what the import intensity variable (impint) indicates in our case: a 1 % increase in the share of imported material input relative to the total leads to an increase in the percentage share of exports in total sales by 0.019 and 0.028 %, respectively. In this context, we should point out the reorientation of the international trade of TCs towards developed countries after the breakdown of socialism; this pattern still persists, with the European Union being the main trading partner for the majority of TCs.

The other firm-specific variables included in our specifications are in line with what SME literature generally predicts, including empirical investigations for TCs. With regards to the controlling variables, as the unconditional marginal effects show, size (size) is an important factor in the exporting behaviour of SME sector in TCs. This result is consistent with other studies pointing out the significance of size in foreign markets. The results indicate that, as firms grow in size, there is a higher propensity for non-exporters to export as well as for exporters to export more. However, the size variable does not reveal non-linear effects on the degree of export involvement. In addition, the estimated coefficient on the age variable is generally statistically insignificant, indicating no obvious role of experience in firms’ export behaviour (age).

As far as the ownership variables (foreign vs. domestic owned and state vs. private owned) are concerned, the results are generally as expected. The effect of foreign ownership on the export behaviour of SMEs is positive and consistent across the various datasets (foreign). The unconditional marginal effects estimated across the various samples indicate that a 1 % increase in the foreign ownership of SMEs increases the percentage of export sales in total sales by between 0.03 and 0.07 % (rounded).Footnote 31 Our results confirm what has been already found in many studies regarding TCs that highlight the superior performance of foreign firms vis-à-vis domestic firms on the grounds of the productivity gap between these two categories of firms.Footnote 32 Although it was expected that state ownership would have a negative and significant effect on firm performance in general, and on export behaviour in particular, the empirical evidence is rather weak. The results for state-owned companies (soe) indicate statistically insignificant effects on export behaviour throughout the samples, except for the pooled estimates which is negative and statistically significant at 10 %. This is consistent with some of the earlier studies of the impact of privatisation, which also failed to show the negative relationship between state ownership and firms’ performance (see Bevan et al. 1999). Furthermore, this result can be explained by the fact that the sample contained only a small number of state-owned firms (ranging from a maximum of 10.73 % in 2002 to 1.41 % in 2008/2009), and that a large number of firms (almost 20 % of the total in 2008/2009 sample), were privatised in the 2002–2008 period, and therefore did not have enough time to adjust to market-based conditions.

The sector of activity variable (entact) indicates, ceteris paribus, that companies involved in production activities engage more in exporting relative to trade and service companies (the result appears in all samples, apart from the imputed 2002 sample). The difference in the magnitude of the marginal effect for the panel samples relative to other samples is a product of the difference in the definition of the variable. For the panel sample, a discrete change from 0 to 1—signifying production activities in relation to other activities—increases the export share in total sales by up to 13 %. For other samples, a 1 % increase in the share of production in total sales increases the export share in total sales in a range from 0.015 to 0.065 %. Although the service sector is rapidly gaining in importance in many emerging markets (Kandilov and Grennes 2010), it seems that production activities of the SME sector in TCs have an edge compared to service activities when it comes to exporting.

Other important outcomes of our estimations relate to access to external finance, market share, membership of business organisations, and capacity utilisation. Our results demonstrate that access to finance (credit) is indeed an important factor that enhances an SME’s competitive edge. The results overwhelmingly confirm the positive relationship between availability of finance and export behaviour. In addition, in line with other studies on TCs (see Hobdari et al. 2009), the estimated effect of market share (mark_shar) indicates that SMEs with a greater share of the domestic market tend to take advantage of greater demand in foreign markets. Furthermore, the results are consistent with the suggestion that business associations (buss_assoc) are an invaluable source of export information for SMEs in TCs and provide a route for SMEs to engage in international business networks. Finally, the results for capacity utilisation are generally too weak to be conclusive. However, the estimates for the 2002 sample—imputed and non-imputed—yield a negative rather than a positive relationship with exporting, which is in line with a supply-side rather than the conventional demand-side interpretation. Should this supply-side interpretation be supported by future studies, then researchers may need to reconsider the routine use of capacity utilisation as a simple proxy for demand pressure. For, in this case, capacity utilisation may reflect a more subtle supply-side strategy, whereby firms carry excess capacity in order to increase their flexibility to respond to the uncertain opportunities of export markets.

Two final groups of variables attempt to identify country and period disparities in the export behaviour of firms in TCs. In the cross-section tobit estimates, the country dummies are generally significant (of the 201 estimated country fixed effects, only 3 are not highly significant). This is consistent with our assumption that the former communist legacy, the pace of reforms, institutions, culture, and varying macroeconomic conditions may have had an impact on the export behaviour of firms. In contrast, in the panel estimates, of the 87 estimated country fixed effects only 2 are significant. Together, these results suggest that the firm-level time invariant (fixed) effects in our panel models and the country fixed effects in our cross-section models play a similar role with respect to controlling for otherwise unmodelled systematic influences on the dependent variable. In addition, the year dummies control for otherwise unobserved period-specific but group-invariant influences. We see higher export intensity and propensity of firms surveyed in 2005 than in 2008/2009 (no significant difference is found for 2002). We should note that TCs have suffered greatly during the 2008/2009 financial crisis, especially primary commodity-producing countries in the CIS. In addition, highly financially integrated countries in CEB and SEE also felt heavily the brunt of the crisis. Hence, the results for the period variables partially confirm these claims.Footnote 33

5 Conclusions

This paper investigates the determinants of the export behaviour of SMEs in transition countries (TCs), using cross-sectional and panel SME samples from the World Bank/EBRD Business Environment and Enterprise Performance Survey (BEEPS) carried out in 2002, 2005, and 2008/2009. We were concerned, in particular, with the impact of human capital- and technology-related factors, as productivity-enhancing factors, on the export behaviour of firms. In the same light, we investigated two potential channels of productivity-enhancing spillovers, namely, beneficial externalities arising, respectively, from location in the capital city and from industry—especially vertical—linkages. We also investigated the effects of firm-size, experience-related factors, location, type of ownership, sector of activity, financial constraints, and membership of business associations. Tobit models were employed to analyse the relationship between firms’ export behaviour (measured by the share of total sales generated by exports) and these potential factors. This econometric approach enabled us to differentiate between exporters and non-exporters, while including both in our investigation. Hence, we have analysed firms’ export behaviour by taking into account the likelihood that firms will export at all as well as the level of export activity.

A recent strand of international trade literature argues that a firm’s export entry and exit decisions are determined by the interplay of firm-level variation in productivity and sunk costs. We draw largely on Melitz’s dynamic model of export participation for the core of our empirical model. We augment this with a number of additional strands of thought, which either substantiate or complement Melitz’s approach, to specify our empirical model of the export behaviour of SMEs in TCs. Although confined to broadly supply-side factors, the scope of our model specification—the large number of factors included—contributes to a greater understanding of SME behaviour in international markets.

Missing values are endemic to survey analysis, and thus a common problem in empirical SME research. Missing values entail the possibility of sample bias induced by non-random “missingness”. Accordingly, we re-estimated each model on samples completed by imputed values. We propose this is a robustness check on our estimates in addition to estimating variant model specifications on six different samples (2- and 3-year panels, and the three waves individually and pooled). The following results are robust in that they appear not only with the same sign but also with similar size and statistical significance across the majority of estimates—i.e. over variant model specifications estimated on different samples using both non-imputed and imputed datasets.

Under the ceteris paribus condition, we find that both human capital- and technology-related factors are important sources of international competitiveness for small firms in TCs. Companies with a larger share of educated workers have higher export propensity and intensity. Similarly positive effects are identified for the introduction or upgrading of products and technologies (i.e. product and process innovation). Three variables included in the analysis indicate the importance of industry—especially vertical—linkages for the export behaviour of SMEs: sales to multinationals, sales to large domestic companies, and import intensity. Moreover, we find the following positive influences on the propensity and intensity of SMEs to export: firm size, foreign capital share, sector of activity (i.e. production), the availability of external finance, and, membership in a business association.

The policy implications are three-fold: export promotion may best be targeted at potential rather than existing exporters, the same policies may both encourage potential exporters to export and existing exporters to export more, and a wide-range of complementary supply-side measures are necessary to make a substantial impact on firms’ export behaviour.

Following Melitz, the premise of our discussion is that high-productivity firms self-select into export markets. The implication is that policy makers should target potential exporters rather than actual exporters as, by definition, the latter do not require policy intervention to enter foreign markets. As López (2004, 2005) argues, notably in the context of developing countries, firms “consciously” increase their productivity levels in order to serve export markets. In other words, firms make productivity-enhancing investments with the aim of penetrating foreign markets. Policy makers can support this process by developing appropriate policy instruments to facilitate productivity enhancement: in particular, support for innovation, technology improvement, R&D spending, and development of human capital. As Greenaway and Kneller (2007, p. 157) explain, intervention may “stimulate more conscious self-selection and deliver a productivity boost”.

It has already been mentioned that the probit estimations (conducted as a diagnostic check) show the factors that influence the firms’ propensity to export. A comparision of the probit and tobit estimates indicate that the same set of variables influence—with the same sign—both decisions as to whether or not to export (propensity), and, if so, how much to export (intensity). These results suggest that the same policies can increase both the probability that potential exporters will decide to export and the export efforts of existing exporters. In other words, policy prescriptions can be focused on the same areas for both potential exporters and actual exporters.

The results presented in this paper show that the marginal effects are generally rather small. This suggests that no single policy can transform the export activity of SMEs. Rather, a wide range of well-designed and consistently implemented policies will be required to boost SME exporting and, indeed, to promote the sector more generally. According to this study, such policies could usefully embrace: education, support for SME investment in technology and R&D, foreign participation in SME ownership, support for integration into the supply chains of larger firms, especially MNEs, measures to increase banking sector competitiveness, which improve the availability of finance for SMEs, and measures to promote membership of business associations (and, perhaps, by implication, SME networking in general).