A5 - Fisnik Recica Full Paper

*I would like to express my special gratitude to Professor Iraj Hashi and J ean Mangan for helpful comments and
support with regards to the theoretical and econometric work of this study.

Working Paper
IMPACT OF INNOVATION ON EXPORT PERFORMANCE
EVIDENCE FROM TRANSITION ECONOMIES
Fisnik REICA*
PhD Candidate, Staffordshire University Business School, Stoke on Trent, U.K.,
Lecturer at the Faculty of Economics, AABRiinvest University & Researcher at the Riinvest Institute, Prishtine, Kosova
[email protected]
Abstract
Innovation has been widely studied for its impact on firm performance and is considered a
driving force behind international trade. The aim of this paper is to investigate if there is a
significant impact of innovation on export intensity in Central and South Eastern Europe and
possible differences between the regions of Western Balkan, South Eastern candidate countries
for European Union, Central Eastern Europe and Baltic countries controlling for the various
factors which may affect firms export behaviour. We use the datasets from Business
Environment and Enterprise Performance Surveys of 2002, 2005 and 2008 jointly conducted by
the World Bank and European Bank for Reconstruction and Development. One major issue in
the literature is the endogeneity between innovation and export because innovation affects
exporting but improved export performance affects innovation as well. Therefore, export
intensity of firms is analysed using Tobit estimation method with instrumental variable to avoid
for any possible bias of estimates.
The results indicate that innovation is a significant and positive determinant of export intensity.
Other variables indicating positive and significant effect on export intensity are foreign
ownership, affiliation of firms in business association, finance of investments by commercial
banks and proportion of staff with the University education. Firm size indicates that as larger the
firm is the effect is stronger. Western Balkan countries indicate higher degree of export intensity
than SEE candidate countries but lower than others, while innovative firms of Western Balkan
indicate higher degree of export intensity in comparison to other three regions.
Impact of Innovation on Export Performance: Evidence from Transition Economies
2

1. INTRODUCTION
An important determinant of the success of firms in the modern era is their ability to
develop new products and processes in order to increase their efficiency and gain competitive
advantage. As more innovative they are firms are supposed to be more competitive, have better
performance and have greater chances to approach international markets. Based on the theories
of Vernon (1966) and Krugman (1979) innovation is considered to be a driving force behind the
international trade.
This study aims to estimate the impact of innovation on export intensity of the firms in
transition economies of Central Eastern Europe (CEE) and South Easter Europe (SEE) and the
differences between Western Balkan countries, European Union candidate countries of SEE,
CEE and Baltic countries. All countries of our sample belong to the transition economies, but our
focus is in the Western Balkan countries which have gone through a similar development phase
in last two decades associated with economical and political problems.
Previous studies which investigated innovation impact on export performance of the
firms have generally found positive and significant effect of innovation (Wakelin 1998,
Sterlachini 1999, etc). Their results may have been biased as they did not control the possibility
of endogeneity since innovation affects performance but improved performance affects
innovation as well. This issue is predicted by the global-economy models of endogenous
innovation and growth (Grossman and Helpman, 1994). To avoid this problem, recent studies
(Damijan and Kostevc 2008, Anh et al 2009, etc.) have additionally controlled for the
endogeneity issue and the results of innovation impact on export performance are similar with
the previous ones.
Using large datasets from the Business Environment and Performance Surveys (BEEPS)
conducted by the World Bank and European Bank for Reconstruction and Development (EBRD)
on 2002, 2005 and 2008 we employ a Tobit estimation method with instrumental variable to
control for the endogeneity problem. Tobit estimation method enables us to incorporate in one
model both exporting and non-exporting firms by censoring the dependent variable. Since panel
estimation is not possible due to the slight changes in the 2008 questionnaire, we estimate our
3

export intensity model with pooled data of 2002 and 2005 and separately for 2002, 2005 and
2008 for the robustness of the results. The dependent variable, export intensity, is expressed as
the share of exports on total sales, while innovation activities are indicated by innovation output
captured by the survey questions on new products and processes introduced in previous 36
months. We also control for the age of the firm, size, sector, foreign ownership, business
association membership, finance of investments by credits from commercial banks, quality of
products, changes in organisational structure and human capital factors such as proportion of
skilled and trained workers and proportion of workers with university education. Furthermore
we investigate whether there is any difference between the regions of interest in export intensity
in general and for innovative firms in particular.
The paper is structured as follows: the following section critically reviews the literature
on innovation and its impact on export performance performance. Section three presents
empirical analysis starting with the theoretical basis, methodology used for analysis, description
of data and interpretation of our empirical findings. The last section presents concluding
remarks.

2. LITERATURE REVIEW
This section will aim to review the literature on innovation activities and the impact of
innovation on export performance as one of the significant determinants for the success and
growth of companies.

2.1. Innovation activities
The process in which firm decides to innovate, the determinants of that decision and
impacts of the innovation activities on firm performance have been examined by different
authors using different methodologies and different innovation indicators.
Schumpeters work of 1934 defines innovation as an introduction of a new product, a
new process or a new organizational arrangement, opening or identifying of a new market and as
4

the conquest of a new source of supply of raw materials or half manufactured goods.
1
Innovation
is considered an important element of a firms strategy to gain competitive advantage and
maintain its dominant position in the market. The now prevalent view is that firms become
industry leaders by conducting research and development (R&D) leading to innovation in their
production technologies or the products they provide (Pepall, et al., 2008, p.572). This has led to
a more particular focus on the role played by innovation in relation to the productive processes
and the growth of firms. As it is emphasized in the DTI (2003) economics paper 7, innovation
can lead to productivity growth through the development of more valuable products and services
or new processes that increase efficiency. Positive impact of innovative activities on firms
performance gives an advantage to innovative firms compared to those that do not innovate.
However, there is no single measure of innovations, and a range of innovation indicators are
required to show innovation activities from input to innovation output.
In the innovative processes, as argued by Vermeulen, et al. (2003) there is a relationship
between innovation inputs, throughputs and outputs, as an indicator that firms which invest more
in first two phases of the process will achieve higher levels of innovative output. The most
emphasized indicators of innovation input is R&D effort (Acs and Audretsch, 1990; Hitt et al.,
1991) and the number of R&D employees (Scherer, 1965; Schmookler, 1966).) These input
indicators, through a transformation process will lead to innovation output such as new or
significantly improved products, processes, services, etc. Vermeulen, et al. (2003) points out that
among various innovation output indicators, three of them have received the most attention:
number of patents, new product announcements and the degree of newness of new products.
Studying the relationship between innovative activities and export performance in non-
R&D intensive industries, Sterlacchini (1999) argues that traditional R&D indicators are not
adequate for studying innovation in non-R&D intensive industries and small firms, giving
priority to other measures of innovation input such as design work and engineering development.
Moreover, Bleaney and Wakelin (2002) suggest that an innovation output measure is likely to be

1
Similar to the Schumpeters definition of innovation is given by the Oslo Manual (2005, p.46) which defines
innovation as the implementation of a new or significantly improved product (good or service) or process, a new
marketing method or a new organizational method in business practice, workplace organizations or external
relations.
5

preferable to an input measure such as R&D intensity, because as Acs and Audrestch (1987)
argue, R&D measures suffer from indicating only the budgeted resources allocated towards
trying to produce innovative activity, but not the actual amount of resulting innovations.

2.2. The Impact of innovation on export performance
International trade is driven by many factors but innovation is considered to be a driving
force behind it in the process of firm development and its efforts to penetrate into the
international markets based on the models of Vernon (1966) and Krugman (1979). Innovation is
seen as an important segment of firm strategic strengths and given its excepted impact on export
performance it gives firms an incentive to invest more in new product and process developments,
in order to enter new markets worldwide and be more competitive. According to Vernon (1966)
new products are introduced by firms to meet national needs and are first exported to similar
countries
2
. This is important for companies to reduce risks and costs, increase revenue and
market share through markets outside the national boundaries.
On the other hand, Krugman (1979) developed a model of international trade in which
patterns of trade are determined by a continuing process of innovation and technology transfer.
3

In his model, Krugman postulated a world of two countries: innovating North and non-
innovating South. Innovation first takes place in the industrialized North with the introduction of
new products that will lead to a later production of those products in South after a lag in which
the South will adopt new technology and imitate innovative products of the North. In this process
North must continually innovate to maintain its relative position and maintain its real income, by
exporting new products and importing old products imitated by South. This implies that each
good is first exported by North, but eventually it will become an export of South, suggesting a
causation effect between innovation and exports and a rise to product cycle in international trade.
As Grossman and Helpman (1994) predicted, in the economy global- models of innovation and
growth, the increase of exports driven by innovation will in reverse increase firm investments in
innovative activities, thus raising the endogeneity issue between innovation and exports.

2
Countries with similar needs, preferences and incomes.
3
Process of transformation of new products into old products
6

In support of the models which indicated that innovation should have a positive impact in
the exporting activities of the firm, the study by Hirsch and Biaouji (1985) who examined the
impact of R&D intensity on export performance of Israeli firms with a sample of 111 firms,
found positive impact of innovation on export performance. Moreover, the number of R&D
employees is found to have a positive and significant effect on the export growth.
Kumar and Siddharthan (1994) have studied the impact of technology and size on export
performance in developing countries, using the data on 640 Indian firms for the period 1987 to
1990. In their study, technology factor is captured by the R&D intensity as a measure of
innovative activities. They also controlled for the firm size and technology imports, advertising
and capital intensity of operations (automation), affiliation with multinational enterprises, and
government policy factors related to export incentives and concessions. Since a large numbers of
Indian enterprises do not export, the dependent variable has a zero value in many cases, so they
used a Tobit model for empirical analysis as an appropriate one, and the results suggest that
innovation activities captured by R&D of the firms favourably influences export behaviour only
in the low and medium technology industries. Indications that there is no positive impact of
R&D in high technology industries is related to the fact that Indian firms do not have the ability
to engage in export activities if they rely on their own technological activities. Moreover, the
relationship of firm size and export performance suggests an inverted U-shaped form meaning
that firms export more when they start to grow, but after a certain point of growth the effect on
exports starts to decline. This might imply that large enterprises act as oligopolist over the
protected domestic market which generally makes them less prone to export than the other firms.
Wakelin (1998) examined the role of innovation in determining export behaviour for a
sample of UK firms, and treated it in two ways, as a probability of a firm exporting and the
propensity to the export of the exporting firms, including innovating and non-innovating firms.
As a proxy for innovation activities this study used innovation history of the firms. Outlining
characteristics of innovation at the firm level, the author suggests that firms enjoy benefits of the
innovation in terms of cost reductions, new markets and potential monopoly rents. This makes
the firm a suitable unit of analysis in the context of the role of innovation in export behaviour.
Data used for this analysis covers 5 years from 1988 to 1992, however, the number of total
7

observation was lowered as data are missing for some firms in some years. The main variables
used in the model included propensity to export measured by the proportion of exports on total
sales, average capital intensity
4
, average wages, unit labour costs, and the number of produced
innovations. Size variable was set according to the number of employees.
In the first specification, Wakelin estimated a single censored Tobit model, including the
decision of whether or not to export and the level of exports, in one model. In the second
specification, a two stage model was used separating the decision of whether or not to export
from the decision of how much to export. In the first stage, the decision whether or not to export
is considered using a Probit model, while in the second stage only the subset of firms which
export have been considered using a truncated estimation procedure as the dependent variable is
observed only if it is greater than zero. Double specification was tested against the Tobit model
as the restricted model, and Tobit model was rejected, which might indicate model
misspecification.
The findings of the Wakelin study suggest that innovating and non-innovating firms
behave differently both in terms of probability of exporting and the level of exports, implying
that the capacity to innovate changes the behaviour of the firm. In terms of size, small innovating
firms are less likely to enter export markets than non-innovating firms, but large innovative firms
are likely to export, and the more innovations they have had, the higher is the probability to enter
export markets. Indications are that smaller innovative firms are less likely to export, since the
cost of entering export markets is higher for smaller firms, therefore, they would prefer to stay in
domestic markets. Another result of the study is that the production of innovations at the sector
level improves probability of all firms exporting, no matter if they are innovative or non-
innovative, which implies that the innovative environment is important and would encourage
firms to export, even though the same relationship was not supported by the results for the export

4
Capital intensity is defined relative to sales and not to labour as the capital to labour ratio was found to be
collinear with the average remuneration variable, making it difficult to separate the effect of the two variables
(Wakelin, 1998).
8

propensity.
5
This might indicate that positive spill-over effects are significant for the increase of
probability for first time exporters, but not for the increase of export propensity.
Taking into account that R&D statistics can be misleading in the case of smaller firms,
which often do not have formal R&D units
6
, Sterlacchini (1999) takes into account other
innovation inputs such as the innovative content of capital stock
7
, the ratio of expenditure on
design engineering and trial production to sales, and the shares of costs for acquiring innovative
capital goods on sales. Study was based on the questionnaire administered to firms at the end of
1997 by means of direct interviews, investigating the period between 1994-1996, using a sample
of 143 firms with fewer than 200 employees classified as supplier dominated and specialised
suppliers.
The estimation method used in this analysis follows the method of Wakelin (1998), but
when the whole sample is considered, the likelihood ratio test suggested that the single censored
Tobit model should be preferred to the two-stage model, indicating that the parameters of both
probability to export and export intensity could be measured in one model. This might be a result
of the small sample used for analysis. On the other hand, when only the subset of innovating
firms is considered, the author does not clearly indicate which one is the best model
specification. Tobit estimates are justified as correct estimates for the empirical analysis since the
proportion of export to sales variable has a large number of observations in its lower limit or
better saying when firms do not export at all, and a few observations are in the upper limit when
firms sales derive from exports. Therefore, OLS estimates would be downward biased
8
, and by
using censored Tobit estimates that bias would be avoided.
Except innovation indicators Sterlacchini used a dummy variable for innovating firms,
taking a value of 1 if the firm introduced at least one innovation during the period 1994-1996.

5
Production of innovation at the sector level is measured by the number of total innovations produced in the
sector from 1979-1983, from the survey, excluding each firms individual innovations, scaled by the number of
enterprises in the sector (Wakelin, 1998).
6
See Santarelli ad Sterlacchini, 1990; and Kleinknecht and Reijnen, 1991).
7
To assess the innovative content of a firms capital stock, the main criterion used was the distinction, for each
production phase, between machines and logistical devices based on micro-electronics and those which relied on
electro-mechanical technologies. (Sterlacchini, 1999)
8
Downward bias of the OLS estimates may predict values of the dependent variable outside its possible range (See
Sterlacchini, 1999 and Wooldridge, 2006, pp. 98-99).
9

Furthermore, as control variable for firm characteristics he uses firms size measured by the level
of sales in 1996 and the share of sales in 1996 due to the productive decentralisation of other
firms it has been assumed that if this is greater than 60 percent, the firm is classified as a sub-
contractor. The other firm characteristics variable is a dummy variable identifying the affiliation
of a firm with an industrial business group. The only significant variables were firm size and the
innovation/technological level of capital stock. Results found from the empirical analysis of the
Tobit estimates indicated that export performance has an inverted U-shaped relationship with
firm size. The innovation variable measuring intensity of expenditures on design, engineering
and trial production was positive and particularly significant. Moreover, the innovation dummy
has a less significant but a positive impact on export performance. On the other hand, in the two
stage model, apart from the effect of firm size on export performance, the probability of being an
exporter is affected positively but with a low significance level by all three indicators of
innovation. Furthermore, in the second stage, the propensity to export does not seem to be
affected by firm size but is significantly impacted by the innovation dummy and the intensity of
expenditure on design, engineering and trial production.
As the implication of the analysis by Sterlacchini (1999), it might be suggested that given
the positive and significant impact of the innovation input indicators used in the study, the
product and process innovations development which are shown to be important for the
development of the small firms, do not necessarily require R&D efforts.
Using the data on Slovenian firms for the period 1996-2002, and combining firm level
information of foreign trade flows and innovation activity for the same period from Community
Innovation Surveys (CIS1, CIS2, CIS3), Damijan and Kostevc (2008) examined the causal
relationship of exports and innovation. Slovenian domestic market is small, therefore, 85 percent
of the Slovenian manufacturing firms are exporters, and most of their exports are oriented
towards European Union markets. Employing bivariate Probit estimation, and including firm
innovation outcome as the variable of interest measured by actual product or process innovations
undertaken by firms, they test the prediction that positive effects of exporting will manifest
themselves in improved firm ability to innovate, while improved ability to innovate would
increase the probability of becoming an exporter. Based on their findings, they could not clearly
10

establish the direction of the causal relationship between exporting and innovation. They found
that innovating status increases the probability of exporting, but it does not increase the
probability of becoming first time exporter, and exporting status increases probability of
innovating, but it does not increase the probability of becoming a first time innovator. Even
though their findings do not clearly establish the causal relationship, they emphasize the
importance of the endogeneity problem when measuring the impact of innovation on exports
which might lead to biased estimates if not considered.
Another study which has considered the endogenour relationship between innovation and
exports is done by Anh, et al. (2009) for the developing country of Vietnam, based on the
knowledge that the export sector is a major driving force in Vietnams economic growth. They
based their initiative for the study on the assumption that innovation can affect a firms
competitiveness and hence export status by increasing productivity and developing new goods
for the international market. They captured innovation activities as a new product innovation, a
new production process and a modification of existing products. The result suggests a significant
impact of all three innovation measures on export performance, amongst which product
innovation showed the strongest effect.
Generally, the literature reviewed suggests a positive and significant effect of innovation
indicators on export performance, no matter if input or output indicators have been used as
innovation measure. As noted in the literature review, innovation affects export performance
positively, but increased performance affects innovative activities as well, thus suggesting the
causal relationship. Most of the studies have not taken into account the endogeneity issue, so
their results might have been biased and therefore questionable, but some studies have indeed
dealt with the issue by controlling for endogeneity.

3. EMPIRICAL ANALYSIS
Empirical analysis in this paper aims to assess the impact of innovation on export
intensity, controlling also for other variables which are important in determining the firms
export behaviour. The first part covers theoretical basis of the model while the second one
11

describes methodology used. Then, we describe data sources and its limitations, following with
the variable specification and empirical findings.

3.1. Theoretical basis
As discussed in the previous section, trade models first developed by Vernon (1966) and
Krugman (1979) highlighted the importance of the role of innovation as the main determinant of
the international trade and an important tool to achieve competitive advantage. According to
Lefebvre and Lefebvre (2001) competition is increasingly technology based, therefore, it is
expected that the role of technological capabilities from which new product and processes
emerge would be very important in determining the firms ability to export. However, this raises
the endogeneity issue as predicted by global-economy models of endogenous innovation and
growth (Grossman and Helpman, 1994) because with the increase of exports driven by
innovative activities firms are expected to invest more in future innovation, thus creating the
reverse relationship.
From the literature review it is known that a variety of factors affect export performance,
both the firms probability of exporting and the export intensity (Wakelin, 1998; Sterlacchini,
1999; Kumar and Siddharthan, 1994; Anh et al., 2009; etc.). Following studies which estimated
both the probability of exporting and the export intensity, we may assume that generally the
same variables affect both outcomes even though they may differ in the direction and magnitude
of the impact on the two separate decisions.
Considering the availability of data and the control variables used by other authors we
control for innovative activities as the main variable of interest and also other variables which
are expected to affect export behaviour of the firms. Therefore, Export intensity is expressed as a
function of innovative activities, human capital factors, and firm characteristics (Higon and
Driffield, 2007; Gashi, 2008).
Export intensity = f (Innovation activities, human capital factors, firm characteristics).

12

3.2. Methodology
Most of the firms observed included in the Business Environment and Enterprise
Performance Survey (BEEPS) dataset which we use in our analysis are not exporters, having a
zero value for the share of exports in total sales. Therefore, to examine export intensity of firms,
only the subset of exporting firms might be used, but this may produce some degree of sample
selection bias since most of the firms in the whole sample are not exporters, so the Ordinary
Least Square (OLS) estimation method is not suitable. OLS can give estimates which imply
predictions of the propensity to export outside its possible range, i.e., higher than one and lower
than zero. OLS estimates of a sample considering only one subset of observations while omitting
others creates sample selection bias, thus OLS estimates will be biased and inconsistent.
(Wakelin, 1998; and Gujarati, 2003).
Following Wakelin (1998) and Sterlacchini (1999), a single censored Tobit model
9
will
be employed, because it includes all the available information from the explanatory variables
where the decision on whether the firm exports at all and the decision on the level of exports are
incorporated into one model. The Tobit model is referred to as the corner solution response
where the dependent variable is zero for a nontrivial fraction of the population but is roughly
continuously distributed over positive values (Wooldridge, 2006). Information on the regressand
is available only for some observations, and the sample is known as a censored sample which
makes a Tobit model a censored regression model (Gujarati, 2003).
Following Wooldridge (2006) and Greene (2003) we investigate the appropriateness of
the Tobit model as it incorporates in one model both the probability to export and export
intensity of the firms. Wooldridge (2006) refers to it as an informal way to evaluate the
appropriateness of the Tobit model. This test is carried out by dividing the Tobit estimates with
the overall standard error sigma from the Tobit regression and comparing them with Probit
coefficients. If they are not significantly different from Probit coefficients, then the Tobit model
can be considered as appropriate. As Wooldridge notes, sign changes and differences in
magnitude for explanatory variables that are insignificant in both can be ignored.

9
Tobit model is originally developed by James Tobin, the Nobel laureate economist. (Gujarati, 2003).
13

To control for the endogeneity issue of innovation and export intensity we employ Tobit
model with instrumental variable.
The Tobit model of export intensity function can be expressed as:
) 0 (
otherwise exporters - non for 0
0 if exporters for
2
, ~N
x x
y
i
i i i i
i

Independent variables in the model are expressed as
i
x , represents the coefficients of the
variables and the intercept and
i
is the error term. The latent variable
i
y satisfies the classical
linear model assumptions; in particular, it has a normal, homoskedastic distribution with a linear
conditional mean (Wooldridge, 2006).
3.3. Data
This study uses the BEEPS data which has been conducted by the World Bank and
EBRD in 28 transition economies as well as Turkey in years 1999, 2002, 2005 and 2008. In these
Surveys the sectoral composition is determined relative to the contribution to GDP. The sample
was designed to be as representative as possible to the population of firms which are selected
randomly from manufacturing and services (including trade) and to be distributed between at
least two major industrial regions within each country. Moreover, at least ten percent of the
sample in each country was designed to be in small, large, foreign owned and exporting firms.
10

Because the 2008 questionnaire has slight changes it was not possible to combine all of
them in the pooled or panel estimation, therefore only 2002 and 2005 surveys are used in pooled
regression for this analysis. Data from 15 SEE
11
and CEE
12
countries in 2002 and 2005 surveys
are pooled together to give a larger number of observations. This pooling technique is supported
by Wooldridge (2006, p. 449) who argues that by pooling random samples drawn from the same
population but at different points in time, we can get more precise estimators and test statistics
with more power. Our pooled dataset has approximately 7715 observations (firms) in 15

10
See BEEPS dataset available at http://www.ebrd.com/country/sector/econo/surveys
11
SEE countries included are Albania, BiH, Bulgaria, Croatia, Macedonia, Rumania and FRY
12
CEE countries included are Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia.
14

transition economies of CEE and SEE. To control for any changes taking place between 2002
and 2005 a year dummy is included. Observations in 2005 belonging to the same firms in 2002
have been excluded from the dataset as they would be correlated. Panel estimation would be
desired as it controls for dynamic effects but because of only two years and insufficient number
of observations it might not give appropriate results, therefore it is not part of this analysis.
One limitation of the BEEPS dataset is the quality of responses since the answers are
provided by the managers of the firms and may be subjective. Another is the large number of
missing observations for some variables such as Research and development as an innovation
input indicator and International cooperation for product development as a knowledge spillover
effect. As it was not possible to establish if the data is missing randomly, these variables could
not be included in the model as this might have produced biased estimates. The dataset of 2008
survey has slight changes as it is missing some important variables that we use in our model such
as process innovation, membership in business associations and change in organisational
structure and it has large number of missing observations for finance from commercial banks and
reinvested profits. Therefore, being unable to incorporate it in a pooled regression or run panel
estimation, 2008 regression is estimated separately for the robustness of results.
3.4. Variable specification
13

The dependent variable, Export intensity, is expressed as proportion of exports on total
sales and the variable is constructed by adding together the proportion of the sales exported
directly and indirectly.
Innovation is an important determinant of the export performance which affects firms
productivity and ability to compete on international markets. From the BEEPS survey,
innovation activities can be indicated by the innovation output captured by three survey
questions: if a firm introduced a new product or service, if it upgraded an existing product or
service or if it has introduced a new technology as a process innovation in the previous 36
months. Following Sterlacchini (1999), a dummy variable taking a value of 1 if the firm
introduced at least one innovation is constructed, which will measure the impact of innovation on

13
See Appendix A.1. for the detailed description of the variables
15

export intensity. The use of R&D investment as the innovation input indicator is limited due to
the large number of missing observations in the dataset. Using this variable would have resulted
in the loss of many observations, which cannot be treated as random, and therefore it had to be
excluded from the model.
14
On the basis of the theory and the empirical findings of previous
studies innovation is expected to have a positive impact on the export intensity.
Although the possible endogeneity of exports and innovation may have caused an
estimation problem, the wording of the question on innovation avoids this problem. The
questions on innovation indicators refer to the innovation activities which took place in previous
36 months while the export intensity refers to the current period. As argued by Gashi (2008), the
effect of current export performance could not have influenced the past innovation activities, and
this can imply that the causality effect between the two variables is minimized in this case.
However, considering the limitations of the dataset, the potential endogeneity issue might be
investigated using instrumental variables which are assumed to be exogenous to exports.
15
In our
case only the variable reinvested profits in the previous year is available as an instrument that is
expected to be correlated with the innovative activities but not with export intensity.
16

Quality accreditation is expected to have positive impact on export behaviour because
higher quality products might be better accepted in foreign markets. As Kesidou and Szirmai
(2007) argue, quality accreditation is a necessary step for many firms in developing countries
that try to enter foreign markets and gain the trust of demanding consumers. So, the quality
variable controls if the accreditation with ISO 9000, 9002 or 14000 and AGCCP has an impact
on export intensity. The relevant question in the 2002 survey asks only for ISO 9000
accreditation, while in the 2005 survey the question refers to ISO 9000 as well as a number of

14
34 percent of observations for R&D investment in 2002 and 83.5 percent in 2005 are missing.
15
Lachenmaier and Woessmann (2006) treated innovation impulses and obstacles such as suggestions from the
firms production and resource management and on innovation obstacles such as lack of equity capital as the proxy
variable for innovation arguing that these variables are credibly exogenous to firms export performance.
16
Gashi (2008) used percentage of profits reinvested in the previous year of the survey as the proxy variable for
innovation. In 2002 the answers are given in interval values of the percentage, while in 2005 are given as
percentage. In pooled dataset the 2002 interval values are taken as mean value of percentage intervals.
Atanassov, et al. (2007) argue that banks being unable to evaluate novel technologies will tend to discourage
investing in innovative projects and be more prone to shut down ones that are ongoing, which suggests that
reinvested profit is an important financial source for innovation.
16

other kite marks. Therefore, it will be used in the pooled regression and is expected to have
positive impact.
We investigate the impact of the size of the firm to see whether there is a positive effect
frombigger size firms as they have more resources to enter foreign markets. The relationship
between export propensity and firm size in general has been found positive but non-linear (Roper
and Love, 2002). However, given that the size of the firms is reported as a dummy variable in
this data set, the non-linearity and expected inverse U-shaped relationship cannot be investigated.
An inverse U-shaped relationship is also found by Wakelin (1998) and Sterlacchini (1999).
The age of the firm is expected to play an important role in exporting behaviour since
more experienced firms may be expected to have more advantages to penetrate in foreign
markets though there is no consensus on this in the literature. As argued by Higgon and Driffield
(2008), years of accumulated experience may capture learning by doing effects, but the
opposite is expected if younger firms may behave more proactive, flexible and aggressively.
Therefore, the expected sign of the age coefficient is ambiguous and a U-shaped relationship
may also be possible.
Another useful control variable is Foreign Ownership as an indicator that firm might
have better export performance. If the business group is international, firm could take advantage
of the access to resources such as finance, physical or human capital, branding, marketing, and
distribution (Roper, et al. 2006) which may prove useful to an exporting firm, therefore we
expect positive impact on export intensity.
Dummies for sectors are used to check for the differences in exporting activity in three
distinct sectors of production, services and trade because some of the sectors are more inclined to
exports than others and generally the sector of production is expected to be more prominent in
exporting.
To investigate differences in export behaviour of the SEE and CEE countries, four
dummy variables are included. Two regions are divided into four sub-regions, so we control for
differences between west Balkan countries (Albania, BiH, Macedonia, Serbia and Montenegro),
countries which at the time of the 2002 and 2005 surveys were EU candidate members in SEE
17

(Croatia, Bulgaria and Romania), Central Eastern Europe (Poland, Czech Republic, Slovakia,
Slovenia and Hungary) and the Baltic countries (Estonia, Latvia, Lithuania). Western Balkan
countries of the SEE region are expected to have lower exports intensity compared to more
developed CEE countries due to the economic and political problems they faced during the
1990s.
Higgon and Driffield (2008) argue that information constraints limit the exporting
behaviour of firms, thus highlighting the importance of networking activities. Therefore, we
control for the membership in business associations as an indicator of networking connections of
the firm potentially enabling it to create new cooperation and go beyond the local market. The
impact of this indicator on export intensity is expected to be positive
17
.
Organizational structure might also influence a firms export performance; therefore we
investigate if the change in organisational arrangements in the previous three years has a positive
effect. Dosoglu-Guner (2001) points out that since organizational culture affects corporate
behavior and strategy it can be argued that firms that have an entrepreneur, adaptable, risk-
taking, and future-oriented organizational culture might perceive international expansion as part
of their corporate culture.
According to Grossman and Helpman (1994), human capital factors indicated by the
knowledge capacity are the engine of economic growth. At the micro level these factors
positively affect productivity because the improved knowledge will increase working efficiency
and utilisation of the capital, so we can assume that these factors positively affect export
performance of the firms. Following Higon and Driffield (2007) and Gashi (2008), we control
for human capital factors using proxy variables such as education level of the workers,
proportion of skilled workers and training of the skilled workers which are expected to have a
positive relationship with exporting activities.
Access to finance is an important factor affecting the development of new products and
processes and also the level of investments, both of which are essential for maintaining
competitiveness in international markets. Beck (2002) found that economies with more
developed financial sectors export more. The percentage of a firms working capital and fixed

17
Positive impact of the affiliation of a firm with an business is group is found by Gashi 2008.
18

investment financed by credit from banks is used to measure the extent of external financing of
firms and is expected to have a positive impact on export intensity.
We have also included a dummy variable to control for the year difference from 2002 to
2005.
3.5. Descriptive statistics
Descriptive statistics of all variables used in the pooled data model of export behaviour
are given in table 3.1.
Table 3.1. Descriptive Statistics of the pooled dataset of Export Behaviour Variables in
CEE and SEE Region
Variable Mean Std. Dev. Min Max
Expint 12.25 26.09 0 100
Expprob 0.32 0.46 0 1
Age 16.16 18.80 3 202
Agesq 614.80 1965.45 9 40804
W_balk 0.20 0.40 0 1
SEE_EU 0.21 0.41 0 1
CEE 0.45 0.50 0 1
Baltic 0.14 0.34 0 1
y05 0.56 0.50 0 1
Prodsect 42.81 46.98 0 100
Sersect 24.43 41.60 0 100
Trsect 32.76 43.97 0 100
Forown 0.08 0.27 0 1
Buss_Ass 0.48 0.50 0 1
Finbank 10.23 22.66 0 100
Innov 0.64 0.48 0 1
Quality 0.15 0.36 0 1
Org_Str 0.19 0.40 0 1
Skilled 51.58 32.00 0 100
Uni_Edu 23.23 27.44 0 100
Trskill 0.44 0.50 0 1
L_size 0.11 0.32 0 1
M_size 0.18 0.39 0 1
S_size 0.70 0.46 0 1
Reinvpr 38.58 39.73 0 100
19

Most of the variables in the model are dummies or are given as proportions ranging from
zero to one hundred. Only age is a continuous variable, used together with age square to
investigate the U-shaped relationship between the firms experience and exports. Descriptive
statistics for the 2008 sample are presented in the Appendix A.9.
Statistical figures from table 3.2. indicate that CEE countries are proportionally more
engaged in exporting activities (though only marginally above the Western Balkan countries) but
they have the lowest rate of innovation. On the other hand, SEE EU candidate countries have the
largest number of innovation activities but the lowest proportion of exports compared to other
regions, while the SEE countries of Western Balkans and Baltic countries have a similar level of
engagement in innovation and exports which is in between of other two regions.
The findings of the empirical analysis and the appropriateness of the model used will be explored
in the next section.
Table 3.2. Proportion of innovation and exports.
Region W_balk SEE_EU CEE Baltic
Introduced Product or
Process Innovation 66% 73.03% 57.75% 63.58%
Proportion of firms
which have exported 32.29% 27.33% 32.47% 30.84%
Proportion of
Exporting firms which
have Introduced a
product or a process
Innovation 74.14% 83.48% 71.78% 74.92%

3.6. Empirical findings
This section of empirical findings consists of section on diagnostics, interpretation of
results and a short presentation of 2008 sample estimation results.

20

3.6.1. Diagnostics
The available econometric software does not provide straight forward information on
estimation diagnostics, therefore we have undertaken couple of diagnostic checks.
Table 3.3. Tobit model estimation with endogenous regressor using pooled dataset
Tobit model with endogenous regressors Number of obs = 5737
Wald chi2(19) = 814.05
Log likelihood =-15111.267 Prob >chi2 = 0.0000
Coef. Std. Err. Z P>z

Innov 65.8673*** 23.6916 2.7800 0.0050
y05 1.3590 2.1165 0.6400 0.5210
SEE_EU -7.5232** 3.3786 -2.2300 0.0260
CEE 10.2489*** 3.8802 2.6400 0.0080
Baltic 6.6944* 3.7662 1.7800 0.0750
Age 0.1199 0.1271 0.9400 0.3460
Agesq -0.0001 0.0011 -0.0900 0.9250
Prodsect 0.2390*** 0.0436 5.4800 0.0000
Sersect 0.1717*** 0.0309 5.5600 0.0000
Forown 31.0335*** 3.5136 8.8300 0.0000
Buss_Ass 14.3562*** 2.1493 6.6800 0.0000
Finbank 0.1541*** 0.0509 3.0300 0.0020
Quality 3.3477 3.7756 0.8900 0.3750
Org_Str 0.0293 3.7851 0.0100 0.9940
Skilled 0.0121 0.0393 0.3100 0.7590
Uni_Edu 0.1710*** 0.0520 3.2900 0.0010
Trskill -6.2520 4.1084 -1.5200 0.1280
M_size -6.1045* 3.3837 -1.8000 0.0710
S_size -28.4019*** 3.4975 -8.1200 0.0000
_cons -87.1532*** 13.9524 -6.2500 0.0000

Sigma 55.1014 1.0321
Instrumented: Innov
Instrument: Reinvpr
Wald test of exogeneity (/alpha = 0): chi2(1) = 5.83 Prob > chi2 = 0.0157
Obs. summary: 3912 left-censored observations at expint<=0
1825 uncensored observations
0 right-censored observations
*** Significant at 1 percent level of significance.
** Significant at 5 percent level of significance.
*Significant at 10 percent level of significance.
To control for the possibility of endogeneity, we used Ivtobit model which is a Tobit
model estimation with endogenous regressor, using instrumental variables. The result indicated
by the Wald test for exogeneity presented in table 3.3. indicates that there is insufficient evidence
to reject the hypothesis of no endogeneity at 1 percent level of significance which suggests that
21

the estimation with endogenous regressor is preferred, suggesting that the point estimates from
Ivtobit are consistent
18
. Since our endogenous variable is given as a dummy variable indicating if
the firm carried out any product or process innovation, it was not possible to test for the possible
weakness of the instrument used because there is no statistical test or a rule given for Wald test
values in order to define an instrument as weak or strong.
Additionally, a Tobit regression without considering the endogeneity issue is run for the
purpose of comparison. The statistically significant estimates of Ivtobit regression are more
significant and two of them which were previously insignificant (quality and orgstr) are now
significant
19
. The slight change in estimates of the two regressions might support the relaxation
of the endogeneity assumption.
Table. 3.4. Comparison of IVTobit estimates divided by sigma with IVProbit estimates
20

Var Tobit Coef. IVTobit Estimate/sigma IVProbit estimates

Innov 65.8673 1.195383*** 1.177720***
y05 1.3590 0.024663 -0.014073
SEE_EU -7.5232 -0.136534** -0.255211***
CEE 10.2489 0.186001*** 0.202572***
Baltic 6.6944 0.121492* 0.061602
Age 0.1199 0.002175 0.004944**
Agesq -0.0001 -0.000002 -0.000002
Prodsect 0.2390 0.004337*** 0.003101***
Sersect 0.1717 0.003116*** 0.001596***
Forown 31.0335 0.563207*** 0.531962***
Buss_Ass 14.3562 0.260541*** 0.253954***
Finbank 0.1541 0.002796*** 0.003201***
Quality 3.3477 0.060755 0.126054
Org_Str 0.0293 0.000531 0.019026
Skilled 0.0121 0.000219 -0.000189
Uni_Edu 0.1710 0.003103*** 0.004455***
Trskill -6.2520 -0.113464 -0.062404
M_size -6.1045 -0.110787* -0.105297
S_size -28.4019 -0.515447*** -0.464204***
_cons -87.1532 -1.581686*** -1.487750***
Sigma 55.1014
*** Significant at 1 percent level of significance.
** Significant at 5 percent level of significance.
*Significant at 10 percent level of significance.

18
Information for this test is provided by Stata Base Reference Manual Volume 2 I-P release 10
19
See Appendix A.2. for results of tobit model estimation
20
See Appendix A.3. for more details on IVProbit regression of pooled data.
22

However, because the Wald test of exogeneity suggests that estimation should control for
endogeneity, the results of Ivtobit will be reported even though that the standard errors of Tobit
estimates are likely to be smaller.
To test the appropriateness of the model, a Probit estimation with endogenous regressors
is undertaken in order to compare the estimates of Ivtobit model divided by the estimated
standard error of the regression sigma with the Ivprobit estimates. Comparison of the estimates is
presented in table 3.4. indicating that significant coefficients of both estimations have no large
differences between them indicating that the Ivtobit model is appropriate for estimating export
intensity
21
. This suggests that truncation of the sample to only exporting firms does not give
significantly different results, so, it can be estimated as a one model with all firms included.
Table 3.5. Estimates for separate years 2002, 2005, and pooled data
22
.
2002 2005 Pooled
y05 1.3590
Innov 70.0463* 60.4201** 65.8672***
SEE_EU -5.5830 -8.7256* -7.5232**
CEE 12.5671** 9.3136 10.2489***
Baltic 7.2461 8.5653 6.6943*
Age 0.1491 0.1148 0.1198
Agesq -0.0004 -0.0001 -0.0001
Prodsect 0.1705** 0.3104*** 0.2389***
Sersect 0.1133** 0.2366*** 0.1717***
Forown 25.0515*** 38.4733*** 31.0334***
Buss_Ass 17.8232*** 11.0206*** 14.3561***
Finbank 0.0657 0.2193*** 0.1540***
Quality 7.0022 0.9104 3.3476
Org_Str -3.5758 4.2323 0.0292
Skilled -0.0212 0.0468 0.01205
Uni_Edu 0.1242* 0.2292*** 0.1709***
Trskill -6.1983 -6.1695 -6.2520
M_size -3.8367 -8.7088* -6.1045*
S_size -24.1051*** -33.2918*** -28.4018***
_cons -85.6715*** -88.0523*** -87.1531***
*** Significant at 1 percent level of significance
** Significant at 5 percent level of significance
*Significant at 10 percent level of significance
As another robustness check of the pooled estimation, we have carried out individual regressions
for 2002 and 2005 and the comparative estimates are given in table 3.5. The estimates in all three

21
Wooldridge (2006) suggests that only significant variables is important to have close estimates.
22
See appendices A.4. and A.5. for IVTobit regression results of 2002 and 2005.
23

regressions are consistent as regards to the sign of coefficients, while the significance of the
estimates is generally similar.
When we look at the statistical significance of the coefficients, the impact of innovation
on export intensity seems to be stronger in 2005 relative to 2002. This is the case also with
sectors, finances from commercial banks and university education. However, relying on the
consistency of statistically significant estimates and because pooled data estimates are stronger
and more precise we will interpret only pooled data estimates. The 2008 estimation results are
presented in a separate sub-section.
3.6.2. Interpretation of empirical results
Innovation, foreign ownership, business association, financing from commercial banks,
university education of staff and size of the firms indicate a significant impact on export
intensity. To measure the actual effects of each variable, the marginal effects are presented in
table 3.6.
Here, both conditional marginal effects (when the sub-sample of exporting are
considered) and the unconditional marginal effects (when the whole sample of the firms is
considered) are presented. Even though unconditional marginal effects are slightly larger, we
will report only them to avoid possible biasness of the estimates as the result of sample selection
bias when only the subset of exporting firms is considered.
Innovation as expected indicates to be a positive determinant of export intensity and is
highly significant (significant at 1, 5 and 10 percent level of significance). Unconditional
marginal effects suggest that ceteris paribus, on average, holding other variables at their mean
values, if a firm introduces an innovation will have 17.4 percent higher export intensity than
firms which did not. This supports the theory that innovation is a driving force behind
international trade.

24

Table 3.6. Unconditional and Conditional Marginal effects for pooled estimation
23

variable
Unconditional Marginal Effects
for Pooled estimation dy/dx
Conditional Marginal Effects
for Pooled estimation dy/dx
Innov* 17.38177*** 16.34038***
y05* 0.41900 0.36472
SEE_EU* -2.22865** -1.97727**
CEE* 3.18111*** 2.76065***
Baltic* 2.16635* 1.84414*
Age 0.03701 0.03219
Agesq -0.00003 -0.00003
Prodsect 0.07378*** 0.06418***
Sersect 0.05302*** 0.04612***
Forown* 12.10510*** 9.54626***
Buss_Ass* 4.45498*** 3.86691***
Finbank 0.04757*** 0.04138***
Quality* 1.05584 0.90941
Org_Str* 0.00904 0.00787
Skilled 0.00372 0.00324
Uni_Edu 0.05279*** 0.04592***
Trskill* -1.91695 -1.67304
M_size* -1.81941* -1.60946*
S_size* -9.72543*** -8.08266***
(*) on variable name - dy/dx is for discrete change of dummy variable from 0 to 1
When adding control interaction dummies of innovations and regions in the model, the
innovation estimate is still significant while the interaction dummies suggest that innovative
firms in western Balkans are more export intensive than innovative firms from the other three
regions
24
. This may indicate that firms with export intentions in this region are oriented to
innovative products more than the ones in more advanced economies where they may be in a
stable state of development or their innovative products may be targeting more significantly local
markets. However, caution is required when interpreting the results because as Hashi and
Krasniqi (2008) argue the measures used for innovation activities in the BEEPS surveys are

23
See Appendices A.7. and A.8. for estimation results of unconditional and conditional marginal effects for the
pooled regression.
24
See Appendix A.6. for the IVTobit regression including dummy interactions of innovation and regions.
25

rather crude (whether certain changes have taken place or implemented over the three years prior
of the survey) and lack any information on the quality of those changes.
Estimates for control variables of regions are statistically significant for all three regions,
with CEE countries being highly significant; SEE candidate countries are significant at 5 percent,
while Baltic countries are significant only at 10 percent level of statistical significance. Ceteris
Paribus, at the sample mean, comparing to western Balkan countries, CEE region export
intensity is 3.18 percent higher while in the Baltic countries around 2.16 percent higher. An
interesting result is indicated for SEE candidate countries which suggest that they have lower
export intensity than the western Balkan countries for about 2.3 percent which is contrary to
what we expected. An explanation for this might be that the proportion of exporting firms in the
sample for this group of countries is lower compared to the western Balkan countries.
Year dummy indicates that export performance was better in 2005 but it is insignificant.
Age estimates are statistically insignificant, but given that the age coefficient has a
positive sign while age square coefficient has a negative sign, it indicates an inverse U-shaped
relationship with export intensity. This indication is consistent with the theory and suggests that
after a certain point of experience firms export intensity starts to decline but the indicated effect
seems to be insignificant.
Production and Service sector are both positive and highly significant on export intensity
as it was expected. Unconditional marginal effects suggest that in comparison with the Trade
sector, being part of the Production sector, ceteris paribus, at the sample mean, firm has 0.07
percent higher export intensity, while being part of Service sector firm has 0.05 percent higher
export intensity.
Foreign ownership is statistically highly significant and positive, and consistent with the
expectations. Ceteris paribus, at the sample mean, the unconditional marginal effects suggests
that if a firm belongs to a foreign company its export intensity is 12 percent higher in comparison
to other forms of ownerships. The magnitude of the coefficient indicates a vital importance of
foreign companies in transition economies as the generators of exports, which would be an
26

indicator for companies in these regions to look forward for joint ventures with foreign
companies.
Membership of business associations is also highly significant and positive, raising the
importance of networking as a determinant of entrance and expansion in foreign markets. Ceteris
paribus, at the sample mean, unconditional marginal effects suggest that being a member of
business association on average it increases the export sales as share of total sales by 4.5 percent
relative to non members.
Financing by commercial banks as expected has a highly significant and positive impact
on export intensity. Ceteris paribus, at the sample mean, if a firm increases its investments in
working capital and new fixed investments financed from commercial banks by 100 percent, on
average it will increase export sales as share of total sales by 4.7 percent.
Quality accreditation of products as predicted indicates a positive impact on export sales
but it is insignificant. We may say that because the foreign products are expected to have higher
quality even in the absence of accreditations, the effect of quality accreditations is marginally
insignificant.
Human capital related variables such as changes in organizational structure of the firm,
training of employees and proportion of skilled workers in total number of workers have
insignificant effect on export intensity. On the other hand, the proportion of the employees with
some university degree and higher is highly significant and positive. Ceteris paribus, at the
sample mean, unconditional marginal effects suggest that on average if the proportion of workers
with some university degree or higher increases by 100 percent it will impact an increase of
exports sales as the share on total sales by 5 percent, which is not so strong.
In terms of the size, the dummy for small firms is highly significant while for medium
firms it is significant only at 10 percent level of statistical significance and both are negative.
Ceteris paribus, at the sample mean, unconditional marginal effects suggest that on average a
firm of medium size has 1.81 percent lower export intensity than a large firm, while a small firm
has 9.72 percent lower export intensity in comparison to large firms. As predicted by the theory
the larger firms are expected to have higher export intensity. This again might lead to the
27

conclusion that large firms have more resources to enter foreign markets, and they may also be
more efficient and therefore more competitive on foreign markets. One limitation was that we
could not investigate for U-shaped relationship of the size due to missing data on actual number
of employees in the pooled dataset.
3.6.3. Interpretation of the 2008 sample estimation results
Due to the important missing variables in the 2008 dataset such as process innovation,
membership of business associations, etc., and large number of missing observations on
reinvested profit as our instrumental variable for innovation, it was not possible to include this
data in a pooled estimation or undertake panel estimation. Therefore, we estimated it separately
in order to get another robustness specification for our results and conclusions. Descriptive
statistics and results of 2008 estimation are presented in Appendices A.9. and A.10. Differently
from previous estimation of 2002 and 2005 samples, Bulgaria and Romania were EU members at
the time of the 2008 survey, while Kosova is added in the Western Balkan group of countries in
the 2008 regression since data were available.
In terms of sign and significance 2008 regression gives different results compared to
pooled and separate regressions for 2002 and 2005 only for Quality accreditation which is now
significant at 1 percent level of significance and University education is now insignificant. In
terms of size, as expected, the variable size has a positive value and size square a negative value
indicating an inverse U-shaped relationship. Even though Bulgaria and Romania became EU
members, their group with Croatia as an EU candidate still indicates lower export intensity
compared to Western Balkan group of countries. Innovation as our main variable of interest is
captured by two questions of the survey: if the firms introduced a new product or service, or if it
has upgraded a new product or service. The innovation coefficient is positive and significant at 1
percent level of significance. In general, other factors indicate similar significance level and sign
as in the pooled regression which may be considered as another fact for the robustness of results.
However, since we could not control for endogeneity, these results should be considered with
caution.
28

To sum up, as suggested by the literature, innovation seems to have a strong impact on
export intensity both in terms of magnitude and significance. This indication further supports the
importance of innovation as a key segment for the firm growth and expansion in foreign markets.
Foreign companies also indicate significant impact of a large magnitude. Other factors have the
expected impact on export intensity but most of the effects are small, hence, policy makers who
deal with the international trade and firm exports should consider all significant determinants of
export performance in general, but innovation and foreign ownership in particular.

4. CONCLUSION
Innovation has always been important for the firm performance but in the era of
globalization and international competition it becomes more important through its effects on
export performance. Innovation is by many authors referred to as a driving force behind the
international trade based on theoretical models of Vernon (1966) and Krugman (1979). Using the
BEEPS datasets of 2002, 2005 and 2008 which are jointly conducted by the World Bank and
EBRD this study examined impact of innovation on export intensity in transition countries of
CEE and SEE. Part of the SEE is the Western Balkan which countries have gone through a
difficult phase of political and economical problems during the 1990s.
Theoretical work review identified variables that affect export performance of the firm
with the special focus on innovation as main variable of interest examining their direction and
magnitude of impact. Based on the global economy models of endogenous innovation and
growth (Grossman and Helpman, 1994) the reverse relationship of innovation and exports is
expected since innovation affects export performance but export performance affects innovation
as well. This causes the possibility of endogeneity problem which is controlled for by the recent
studies (Damijan and Kostevc 2008, Anh et al 2009, etc.), an issue we control as well.
The methodology used is a Tobit estimation with instrumental variable. Tobit model is
employed following Wakelin (1998) and Sterlacchini (1999) as it includes all the information
from the explanatory variables where the decision on whether the firm exports at all and the
decision on the level of exports are incorporated into one model. We estimated the model of
29

2002 and 2005 pooled data as it gives more powerful and precise estimates but also separate
regressions for the robustness of results which are consistent in terms of sign and significance.
The appropriateness of the Tobit model is supported by comparing the Tobit estimates divided
by overall standard error sigma with the probit estimates which are similar.
Empirical findings of this study indicate that innovation as it is suggested by the literature
has positive and statistically significant effect on export intensity. Under the ceteris paribus
conditions, on average, the unconditional marginal effects of the pooled data estimation suggest
that if a firm introduces an innovation it will have 17.4 percent higher export intensity than firms
which did not. This result supports the argument that innovation is a driving force behind the
international trade, therefore it must be considered seriously by companies which target
international markets. Age as expected is shown to have an inverse U-shaped relationship
suggesting marginal diminishing returns as firms grow older but the indicated effect is
insignificant to export intensity. The size of the firm indicates that as bigger the firm is it will
have higher rate of export intensity. Companies with foreign ownership indicate positive and
significant impact on export intensity which supports foreign investments in transition
economies, while similar result is suggested if the firm is a member of business associations
giving importance to networking activities as a way towards foreign markets. Credit from
commercial banks is as expected indicates significant and positive effect on export intensity.
From human capital factors, the proportion of employees with some university degree or higher
is indicated to be significant and positive but the magnitude of the effect is not so high which is
the case also with some other variables such as business association membership and finance of
investments from commercial banks. Moreover, quality certification of products, changes in
organisation structure and the proportion of skilled workers and their training do not indicate
significant effects.
Furthermore, firms in production sector and service sector indicate positive and
significant impact on export intensity relative to the trade sector, with production sector
indicating the strongest effect. By controlling for regional differences we found that Balkan
countries have higher degree of export intensity than SEE candidate countries but lower than
30

CEE and Baltic countries. When controlling for innovative firms of the regions, Western Balkan
countries indicate higher level of export intensity compared to the other three regions.
A number of limitations in our study should be mentioned. One limitation is the quality
of data since the answers are provided by the managers of the firms and may be subjective.
Another is the data availability because variables which we wanted to use such as R&D as an
innovation input indicator and the knowledge spillover indicators could not be used due to large
number of missing observations. Furthermore, the instrumental variable used in the model could
not be investigated for its strength. Moreover, the 2008 dataset which would enable us to
undertake panel estimation could be used only as a separate regression because of the slight
changes in the 2008 questionnaire.
To conclude, innovation indicates to be a very important determinant of export behaviour
and policies in Western Balkan countries as our region of interest should be designed to help
firms innovate and invest more in relevant innovative activities in order to help them grow and
improve the international trade balance. Being part of the same region and because of the similar
trend of challenges and developments, indications from the conclusions drawn for Western
Balkan countries can be referred to Kosova as well, which had a period of stagnation and war
between 1990 and 1999. All in all, as most of the estimated effects are relatively small, policies
that would give rise to exports should give more support to factors such as innovation and
foreign ownership which are indicated to have stronger impact on export intensity.

31

REFERENCES:
Acs, Z.J., Audretsch D., 1987. Innovation, market structure, and firm size. The Review of
Economics and Statistics, Vol LXIX No 4, 567-574.
Acs Z.J ., Audretsch D., 1990. Innovation and Small Firms. MIT Press, Cambridge.
Anh, N.N., Jones, N., Nhat, N.D., Chuc, N.D., 2009. Capitalising on Innovation for Exports by
the SME sector. Working paper series No: 2009/15, DEPOCEN Development and Policies
Research Center.
Anon Higon, D., Driffield, N. (2007). Exporting and Innovation Performance: Analysis of the
Annual Small Business Survey in UK. Aston Business School, mimeo.
Atanassov, J., Nanda, V., Seru, A., 2007. Finance and innovation: the case of publicly traded
firms, Working papers series, SSRN eLibrary.
Beck, Th., 2002. Financial Development and International Trade. Is There a Link?, Journal of
International Economics, LVII(1), 107-131.

Bleaney, M., Wakelin, K., 2002. Efficiency, Innovation and Exports, Oxford Bulletin of
Economics and Statistics, v64, 3-15.

Damijan, J ., Kostevc, C., 2008. Causal Link between Exporting and Innovation Activity.
Evidence from Slovenian firms. Working Papers, DYNREG Dynamic Regions in a Knowledge
Driven Global Economy.
Dosoglu Gunner, B. (2001) Can organizational behavior explain the export intention of firms?
The effects of organizational culture and ownership type, International Business Review 10
(2001) 7189.
DTI, 2003 (2). Competing in the Global Economy The innovation Challenge, Economics
Paper No.7.
32

Gashi, P., 2008. Export Behaviour and the Small and Medium Enterprise Sector in South Eastern
Europe, with Special Reference to Kosova, PhD Thesis, Staffordshire University.
Greene, W. (2003), Econometric Analysis, 5
th
edition. New J ersey: Prentice Hall.
Grossman, G. M. and Helpman, E. (1994) Endogenous Innovation in the Theory of Growth,
Journal of Economic Perspectives 8 (1): 23 44.
Gujarati, N.G (2003): Basic Econometrics, 4
th
Edition, McGraw Hill, USA.
Hashi, I., Krasniqi, B., 2008. Enterpreneurship and SME Growth: Evidence from Advanced and
Laggard Transition Economies, SSRN Working Paper, submitted to Research on Emerging
Economies.
Hirsch, S., I. Bijaoui., 1985. R&D intensity in export performance: a micro view.
Weltwirtschaftliches Archiv. 121, 138-251.
Hitt M.A., Hoskisson R.E., Ireland R.D., Harrison J., 1991. Effects of acquisition on R&D inputs
and outputs. Academy of Management Journal 34(3), 693-706.
Kleinknecht, A., Reijnen, J .O.N., 1991. More evidence on the undercounting of small firm R&D.
Research Policy 20, 579-587.
Kumar, N., Siddharthan, N.S., 1994. Technology, firm size and export behaviour in developing
countries: the case of Indian enterprise. Journal of Development Studies 32 (2). 288-309
Krugman, P., 1979. A Model of Innovation, Technology Transfer, and the World Distribution of
Income. Journal of Political Economy, 87 (2), 253-266.
Lachenmaier, S. and Woessmann, L. (2006) Does Innnovation Cause Exporrts? Evidence from
Exogenous Innovation Impulses and Obstacles using German Micro Data, Oxford Economics
Papers, 582, 317 350.
Lefebvre, E. Lefebvre, L.A. (2001) Innovative Capabilities as Determinants of Export
Performance and Behaviour: A Longitudinal Study of Manufacturing SMEs, published in
33

Innovation and Firm Performance, London and Basingstoke: Palgrave (Edited by Kleinknecht,
A. and Mohnen, P. 2002).
Loof, H., Heshmati, A., Asplaund, R., Naas, S., 2001. Innovation and performance in
manufacturing industries: a Comparison of the Nordic Countries, SSE/EFI Working Paper Series
in Economics and Finance No. 457
OECD Oslo Manual., 2005. Guidelines for collecting and Interpreting Innovation Data, Third
Edition. OECD, Paris.
Pepall, L., Richards, D., Norman, G., 2008. Industrial Organization: Contemporary Theory and
Empirical Applications, 4
th
Edition, Blackwell Publishing Ltd.
Roper, S., J.H. Love., 2002. Innovation and Export Performance: Evidence form UK and
German Manufacturing Plants, Research Policy 32: 1087-1102
Santareeli, E., Sterlacchini, A., 1990. Innovation, formal vs. informal R&D, and firm size: some
evidence from Italian manufacturing firms, Small Business Economics 2, 223-228.
Scherer F.M.., 1965. Firm Size, Market Structure, Opportunity, and the Output of Patented
Inventions. American Economic Review 55(5), 1097-1125
Schmookler, J., 1996. Invention and Economic Growth. Harvard University Press, Cambridge.
StataCorp. 2007. Stata Statistical Software: Release 10. College Station, TX: StataCorp LP.
Sterlacchini A., 1999. Do innovative activities matter to small firms in non-R&D intensive
industries? An application to export performance. Research Policy 28(8), 819-832.
Vermeulen, P., OShaughnessy, KC., Jong, J ., 2003. Innovation in SMEs: An empirical
Investigation of the Input-Throughput-Output-Performance, The SCALES-paper series.
Vernon, R., 1966. International Investment and International Trade in the Product Cycle.
Quarterly Journal of Economics, 80 (2), 190-207.
Wakelin, K., 1998. Innovation and export behaviour at firm level. Research Policy 26, 829-841.
34

Wooldridge, J. M. (2003), Introductory Econometrics: A Modern Approach, 2
nd
edition, Mason:
South-Western.

APPENDICES
A.1. Description of the variables used in the model
Dependent Variable
Expint
Sum of direct and indirect export sales of the firm as the share of total sales expressed
in terms of percentage
Expprob
Dummy for exporting firms (Firms with positive value of export sales as the share of
total sales)
Independent Variables
Innov
Dummy for innovation - if the firm has developed a major product line/service,
upgraded an existing product line/service or has acquired new production technology
over the last 36 months
y05 Dummy for 2005
SEE_EU Dummy for SEE EU candidate countries (Croatia, Bulgaria, Romania)
CEE Dummy for CEE countries (Czech Republic, Hungary, Poland, Slovakia and Slovenia)
Baltic Dummy for Baltic countries (Estonia, Latvia and Lithuania)
Age Business Experince of the firm - since establishment
Prodsect Production sector - share of sales generated by production sector
Sersect Service sector - share of sales generated by service sector
Forown Dummy for firms with foreign ownership
Buss_Ass Dummy for firms which are member of business associations
Finbank
Percentage of firms working capital and new fixed investments financed by
commercial banks
Quality Dummy for firms which have obtained any quality accreditation
Org_Str
Dummy for firms which have completely new organizational structure or a major
reallocations of responsibility and resources between departments
Skilled Percentage of skilled workers as the share of total workforce
Uni_Edu Percentage of the workforce with some University Education or higher
Trskill Dummy for firms which have offered formal training to skilled workers
M_size Dummy for medium size firms
S_size Dummy for small size firms

35

A.2. Tobit Regression
Tobi t r egr essi on
Number of obs = 5950

LR chi 2( 19) = 1119. 29

Pr ob > chi 2 = 0. 0000
Log l i kel i hood = - 12072. 898
Pseudo R2 = 0. 0443
expint Coef. Std. Err. T P>t [95% Conf. Interval]

y05 0. 815178 1. 909856 0. 43 0. 67 - 2. 928834 4. 559191
SEE_EU - 5. 01712 2. 901701 - 1. 73 0. 084 - 10. 70551 0. 671268
CEE 2. 857965 2. 515893 1. 14 0. 256 - 2. 074102 7. 790032
Baltic 3. 95874 3. 342365 1. 18 0. 236 - 2. 593513 10. 51099
Age 0. 003271 0. 110393 0. 03 0. 976 - 0. 2131401 0. 219682
Agesq 0. 000652 0. 000968 0. 67 0. 501 - 0. 0012465 0. 00255
Prodsect 0. 323114 0. 025646 12. 6 0 0. 2728385 0. 37339
Sersect 0. 183697 0. 02831 6. 49 0 0. 1281998 0. 239194
Forown 31. 6923 3. 088718 10. 26 0 25. 63729 37. 74732
Buss_Ass 14. 97247 1. 965162 7. 62 0 11. 12004 18. 8249
Finbank 0. 216587 0. 037677 5. 75 0 0. 142726 0. 290448
Innov 8. 625281 2. 086327 4. 13 0 4. 535321 12. 71524
Quality 9. 86398 2. 397043 4. 12 0 5. 164903 14. 56306
Org_Str 6. 810254 2. 243677 3. 04 0. 002 2. 41183 11. 20868
Skilled 0. 017505 0. 036253 0. 48 0. 629 - 0. 0535637 0. 088573
Uni_Edu 0. 238216 0. 042124 5. 66 0 0. 1556373 0. 320795
Trskill 2. 241131 1. 952803 1. 15 0. 251 - 1. 587073 6. 069335
M_size - 5. 17759 3. 039413 - 1. 7 0. 089 - 11. 13595 0. 780767
S_size - 29. 8148 3. 093376 - 9. 64 0 - 35. 87893 - 23. 7507
_cons - 56. 7658 5. 402839 - 10. 51 0 - 67. 3573 - 46. 1742

/ si gma 55. 02033 1. 008739 53. 04283 56. 99782

Obs. summar y: 4046 l ef t - censor ed obser vat i ons at expi nt <=0
36

1904 uncensor ed obser vat i ons
0 r i ght - censor ed
obser vat i ons

A.3. IVProbit Regression
Pr obi t model wi t h endogenous r egr essor s Number of obs = 5753

Wal d chi 2( 19) = 1321. 04
Log l i kel i hood = - 6589. 619 Pr ob
> chi 2 = 0. 0000
Coef. Std. Err. z P>z [95% Conf. Interval]

Innov 1. 17772 0. 326737 3. 6 0 0. 5373275 1. 818112
y05 - 0. 01407 0. 038151 - 0. 37 0. 712 - 0. 088848 0. 0607021
SEE_EU - 0. 25521 0. 056785 - 4. 49 0 - 0. 3665084 - 0. 143914
CEE 0. 202572 0. 058884 3. 44 0. 001 0. 0871624 0. 3179816
Baltic 0. 061602 0. 066116 0. 93 0. 351 - 0. 0679837 0. 1911868
Age 0. 004944 0. 002355 2. 1 0. 036 0. 0003287 0. 0095583
Agesq - 1. 91E- 06 2. 23E- 05 - 0. 09 0. 932 - 0. 0000457 0. 0000419
Prodsect 0. 003101 0. 000989 3. 14 0. 002 0. 0011632 0. 0050389
Sersect 0. 001596 0. 000568 2. 81 0. 005 0. 0004815 0. 0027098
Forown 0. 531962 0. 088102 6. 04 0 0. 3592858 0. 7046374
Buss_Ass 0. 253954 0. 04522 5. 62 0 0. 1653241 0. 3425838
Finbank 0. 003201 0. 001121 2. 86 0. 004 0. 0010051 0. 0053976
Quality 0. 126054 0. 076812 1. 64 0. 101 - 0. 024494 0. 2766016
Org_Str 0. 019026 0. 069341 0. 27 0. 784 - 0. 1168797 0. 1549308
Skilled - 0. 00019 0. 000698 - 0. 27 0. 786 - 0. 0015567 0. 0011782
Uni_Edu 0. 004455 0. 001144 3. 9 0 0. 0022135 0. 0066961
Trskill - 0. 0624 0. 069362 - 0. 9 0. 368 - 0. 1983508 0. 073542
M_size - 0. 1053 0. 064602 - 1. 63 0. 103 - 0. 2319158 0. 0213211
S_size - 0. 4642 0. 081761 - 5. 68 0 - 0. 6244521 - 0. 303956
_cons - 1. 48775 0. 143976 - 10. 33 0 - 1. 769938 - 1. 205561

/ at hr ho - 0. 47232 0. 191136 - 2. 47 0. 013 - 0. 8469402 - 0. 097702
/ l nsi gma - 0. 80291 0. 009323 - 86. 13 0 - 0. 8211832 - 0. 784639
Rho - 0. 44007 0. 15412 - 0. 6894676 - 0. 097392
si gma 0. 448023 0. 004177 0. 4399108 0. 4562843

37

I nst r ument ed:
i nnov

I nst r ument s: y05 see_eu cee bal t i c age agesq pr odsect ser sect f or own
bussass f i nbank qual i t y ski l l ed uni edu t r ski l l m_si ze s_si ze
or gst r r ei nvpr
Wal d t est of exogenei t y ( / at hr ho = 0) : chi 2( 1) = 6. 11 Pr ob > chi 2 = 0. 0135

A.4. IVTobit regression of 2005
Tobit model with endogenous regressors Number of
obs = 3183

Wal d chi 2( 18) = 463. 25
Log l i kel i hood = - 8240. 2123
Pr ob > chi 2 = 0. 0000
Innov 60. 4202 27. 2669 2. 2200 0. 0270 6. 9781 113. 8622
SEE_EU - 8. 7256 4. 5970 - 1. 9000 0. 0580 - 17. 7356 0. 2844
CEE 9. 3136 6. 0373 1. 5400 0. 1230 - 2. 5193 21. 1466
Baltic 8. 5654 5. 2625 1. 6300 0. 1040 - 1. 7490 18. 8797
Age 0. 1149 0. 0890 1. 2900 0. 1970 - 0. 0596 0. 2894
Agesq - 0. 0001 0. 0003 - 0. 3600 0. 7200 - 0. 0007 0. 0005
Prodsect 0. 3104 0. 0572 5. 4300 0. 0000 0. 1983 0. 4225
Sersect 0. 2367 0. 0466 5. 0700 0. 0000 0. 1452 0. 3281
Forown 38. 4734 5. 2284 7. 3600 0. 0000 28. 2259 48. 7209
Buss_Ass 11. 0207 2. 9557 3. 7300 0. 0000 5. 2275 16. 8138
Finbank 0. 2193 0. 0652 3. 3600 0. 0010 0. 0915 0. 3471
Quality 0. 9105 4. 8158 0. 1900 0. 8500 - 8. 5284 10. 3493
Org_Str 4. 2323 4. 3689 0. 9700 0. 3330 - 4. 3306 12. 7953
Skilled 0. 0468 0. 0581 0. 8100 0. 4200 - 0. 0670 0. 1607
Uni_Edu 0. 2292 0. 0751 3. 0500 0. 0020 0. 0821 0. 3763
Trskill - 6. 1696 5. 4219 - 1. 1400 0. 2550 - 16. 7964 4. 4572
M_size - 8. 7089 4. 8477 - 1. 8000 0. 0720 - 18. 2103 0. 7925
S_size - 33. 2919 4. 9396 - 6. 7400 0. 0000 - 42. 9733 - 23. 6105
_cons - 88. 0524 17. 8397 - 4. 9400 0. 0000 - 123. 0175 - 53. 0872
/ al pha - 49. 83074 27. 37485 - 1. 82 0. 069 - 103. 4845 3. 822978
/ l ns 4. 036203 0. 0256162 157. 56 0 3. 985996 4. 08641
/ l nv - 0. 805248 0. 0125333 - 64. 25 0 - 0. 8298133 - 0. 7806835

S 56. 61096 1. 450158 53. 83888 59. 52578
V 0. 4469769 0. 0056021 0. 4361307 0. 4580928
I nst r ument ed: i nnov
38

I nst r ument s: see_eu cee bal t i c age agesq pr odsect ser sect f or own bussass
f i nbank qual i t y or gst r ski l l ed uni edu t r ski l l m_si ze s_si ze
Rei nvpr
0 r i ght - censor ed obser vat i ons
A.5. IVTobit regression of 2002
Tobi t model wi t h endogenous r egr essor s

Wal d chi 2( 18) = 366. 09
Log l i kel i hood = - 6841. 1749
Pr ob > chi 2 = 0. 0000
Innov 70. 04637 40. 12375 1. 75 0. 081 - 8. 594743 148. 6875
SEE_EU - 5. 583015 5. 230645 - 1. 07 0. 286 - 15. 83489 4. 66886
CEE 12. 56713 5. 235289 2. 4 0. 016 2. 306153 22. 82811
Baltic 7. 246123 5. 595853 1. 29 0. 195 - 3. 721547 18. 21379
Age 0. 1491309 0. 1887148 0. 79 0. 429 - 0. 2207432 0. 5190051
Agesq - 0. 0004672 0. 0016251 - 0. 29 0. 774 - 0. 0036524 0. 0027179
Prodsect 0. 1705586 0. 0676503 2. 52 0. 012 0. 0379665 0. 3031508
Sersect 0. 1133099 0. 0439867 2. 58 0. 01 0. 0270976 0. 1995221
Forown 25. 05152 4. 747508 5. 28 0 15. 74658 34. 35647
Buss_Ass 17. 82327 3. 138973 5. 68 0 11. 67099 23. 97554
Finbank 0. 0657078 0. 0794394 0. 83 0. 408 - 0. 0899906 0. 2214061
Quality 7. 002285 5. 782013 1. 21 0. 226 - 4. 330252 18. 33482
Org_Str - 3. 575853 6. 866738 - 0. 52 0. 603 - 17. 03441 9. 882707
Skilled - 0. 0212951 0. 0561528 - 0. 38 0. 705 - 0. 1313525 0. 0887622
Uni_Edu 0. 1242603 0. 0745312 1. 67 0. 095 - 0. 0218183 0. 2703388
Trskill - 6. 198314 5. 601014 - 1. 11 0. 268 - 17. 1761 4. 779472
M_size - 3. 836799 4. 742711 - 0. 81 0. 419 - 13. 13234 5. 458743
S_size - 24. 10515 5. 145635 - 4. 68 0 - 34. 19041 - 14. 01989
_cons - 85. 67151 22. 57177 - 3. 8 0 - 129. 9114 - 41. 43166
/ al pha - 64. 47893 40. 20676 - 1. 6 0. 109 - 143. 2827 14. 32487
/ l ns 3. 969851 0. 0274394 144. 68 0 3. 916071 4. 023632
/ l nv - 0. 8077339 0. 0139836 - 57. 76 0 - 0. 8351413 - 0. 7803265
S 52. 97666 1. 453647 50. 20282 55. 90376
V 0. 4458673 0. 0062348 0. 4338132 0. 4582564
I nst r ument s: see_eu cee bal t i c age agesq pr odsect ser sect f or own bussass
39

f i nbank qual i t y or gst r ski l l ed uni edu t r ski l l m_si ze s_si ze
r ei nvpr

A.6. IVTobit model with endogenous regressor including dummy interactions for
innovation in the regions.
Tobi t model wi t h endogenous r egr essor Number of obs = 5737

Wal d chi 2( 22) = 758. 05
Log l i kel i hood = - 10572. 924 Pr ob >
chi 2 = 0. 0000
Innov 170. 2502 70. 69823 2. 41 0. 016 31. 68418 308. 8161
Inn_SEE_EU - 159. 4097 69. 7298 - 2. 29 0. 022 - 296. 0776 - 22. 74176
Inn_CEE - 153. 6528 69. 36068 - 2. 22 0. 027 - 289. 5972 - 17. 70835
Inn_Baltic - 162. 3809 69. 81256 - 2. 33 0. 02 - 299. 211 - 25. 55079
y05 0. 0507282 2. 203044 0. 02 0. 982 - 4. 267158 4. 368615
SEE_EU 103. 8194 47. 73855 2. 17 0. 03 10. 25356 197. 3852
CEE 109. 0342 47. 69911 2. 29 0. 022 15. 54564 202. 5227
Baltic 114. 9204 47. 67155 2. 41 0. 016 21. 48592 208. 355
Age 0. 1831348 0. 1441593 1. 27 0. 204 - 0. 0994122 0. 4656819
Agesq - 0. 000728 0. 0012426 - 0. 59 0. 558 - 0. 0031635 0. 0017075
Prodsect 0. 2421002 0. 0436039 5. 55 0 0. 1566381 0. 3275623
Sersect 0. 1533551 0. 0338913 4. 52 0 0. 0869295 0. 2197808
Forown 35. 55039 3. 855571 9. 22 0 27. 99361 43. 10717
Buss_Ass 14. 00613 2. 276257 6. 15 0 9. 544745 18. 46751
Finbank 0. 1777211 0. 0480346 3. 7 0 0. 083575 0. 2718672
Quality 5. 379719 3. 361818 1. 6 0. 11 - 1. 209323 11. 96876
Org_Str 1. 952284 3. 344538 0. 58 0. 559 - 4. 602891 8. 507458
Skilled - 0. 0083511 0. 0418585 - 0. 2 0. 842 - 0. 0903923 0. 0736901
Uni_Edu 0. 2194427 0. 0481933 4. 55 0 0. 1249855 0. 3138998
Trskill - 2. 904201 3. 055422 - 0. 95 0. 342 - 8. 892718 3. 084316
M_size - 8. 878455 3. 793636 - 2. 34 0. 019 - 16. 31385 - 1. 443065
S_size - 32. 11797 3. 711493 - 8. 65 0 - 39. 39236 - 24. 84358
_cons - 155. 9203 44. 19182 - 3. 53 0 - 242. 5346 - 69. 30588
/ al pha - 170. 6511 70. 84397 - 2. 41 0. 016 - 309. 5028 - 31. 79952
40

/ l ns 4. 008133 0. 0187277 214 0 3. 971428 4. 044839
/ l nv - 1. 593768 0. 0093356 - 171 0 - 1. 612065 - 1. 57547
S 55. 04402 1. 030849 53. 06022 57. 10198
V 0. 2031587 0. 0018966 0. 1994752 0. 2069102
I nst r ument s: I nnsee_eu I nncee I nnbal t i c y05 see_eu cee bal t i c age agesq
pr odsect ser sect f or own bussass f i nbank qual i t y
or gst r
ski l l ed uni edu t r ski l l m_si ze s_si ze
r ei nvpr

Obs. summar y: 3912 l ef t - censor ed obser vat i ons at
expi nt <=0
1825 uncensor ed
obser vat i ons
0 r i ght - censor ed
obser vat i ons

A.7. Unconditional Marginal Effects of IVTobit pooled regression.
Mar gi nal ef f ect s af t er i vt obi t
y = E( expi nt *| expi nt >0) ( pr edi ct , yst ar ( 0, . ) )
= 12. 036382

variable dy/dx
Std.
Err. z P>z [ 95% C.I. ] X

Innov* 17. 38177 5. 50147 3. 16 0. 002 6. 59908 28. 1645 0. 65034
y05* 0. 419002 0. 6517 0. 64 0. 52 - 0. 85831 1. 69632 0. 554297
SEE_EU* - 2. 228654 0. 95771
-
2. 33 0. 02 - 4. 10573 - 0. 35158 0. 2076
CEE* 3. 181114 1. 21033 2. 63 0. 009 0. 808901 5. 55333 0. 475858
Baltic* 2. 166345 1. 27555 1. 7 0. 089 - 0. 33368 4. 66637 0. 122712
Age 0. 0370083 0. 03924 0. 94 0. 346 - 0. 0399 0. 113912 16. 7772
Agesq - 0. 000032 0. 00034
-
0. 09 0. 925 - 0. 0007 0. 000637 649. 609
Prodsect 0. 0737767 0. 01344 5. 49 0 0. 047428 0. 100125 46. 9308
Sersect 0. 0530177 0. 00952 5. 57 0 0. 034351 0. 071685 22. 5356
Forown* 12. 1051 1. 65224 7. 33 0 8. 86677 15. 3434 0. 074429
Buss_Ass* 4. 454981 0. 66751 6. 67 0 3. 14669 5. 76327 0. 485968
Finbank 0. 0475668 0. 01572 3. 03 0. 002 0. 016755 0. 078378 10. 5745
Quality* 1. 055837 1. 21638 0. 87 0. 385 - 1. 32822 3. 4399 0. 15984
Org_Str* 0. 0090428 1. 16903 0. 01 0. 994 - 2. 28222 2. 3003 0. 194875
Skilled 0. 0037224 0. 01212 0. 31 0. 759 - 0. 02004 0. 027482 57. 8431
41

Uni_Edu 0. 0527869 0. 01599 3. 3 0. 001 0. 021449 0. 084125 20. 2815
Trskill* - 1. 916949 1. 25133
-
1. 53 0. 126 - 4. 36952 0. 53562 0. 438905
M_size* - 1. 819414 0. 97389
-
1. 87 0. 062 - 3. 72821 0. 089385 0. 192784
S_size* - 9. 725426 1. 32133
-
7. 36 0 - 12. 3152 - 7. 13568 0. 692348
(*) dy/dx is for discrete change of dummy variable from 0 to 1

A.8. Conditional Marginal effects of IVTobit Pooled regression
Mar gi nal ef f ect s af t er i vt obi t
y = E( expi nt | expi nt >0) ( pr edi ct , e( 0, . ) )
= 38. 984941

variable dy/dx
Std.
Err. z P>z [ 95% C.I. ] X

Innov* 16. 34038 5. 60815 2. 91 0. 004 5. 34862 27. 3321 0. 65034
y05* 0. 3647243 0. 56771 0. 64 0. 521 - 0. 74797 1. 47742 0. 554297
SEE_EU* - 1. 977266 0. 87137
-
2. 27 0. 023 - 3. 68513 - 0. 2694 0. 2076
CEE* 2. 760654 1. 05847 2. 61 0. 009 0. 686089 4. 83522 0. 475858
Baltic* 1. 844141 1. 06664 1. 73 0. 084 - 0. 24643 3. 93471 0. 122712
Age 0. 0321934 0. 03419 0. 94 0. 346 - 0. 03481 0. 099197 16. 7772
Agesq - 0. 0000278 0. 0003
-
0. 09 0. 925 - 0. 00061 0. 000554 649. 609
Prodsect 0. 0641782 0. 0114 5. 63 0 0. 041837 0. 08652 46. 9308
Sersect 0. 04612 0. 00826 5. 59 0 0. 029939 0. 062301 22. 5356
Forown* 9. 546258 1. 21905 7. 83 0 7. 15697 11. 9355 0. 074429
Buss_Ass* 3. 86691 0. 57673 6. 7 0 2. 73655 4. 99727 0. 485968
Finbank 0. 0413783 0. 01354 3. 06 0. 002 0. 014839 0. 067917 10. 5745
Quality* 0. 9094133 1. 03404 0. 88 0. 379 - 1. 11726 2. 93609 0. 15984
Org_Str* 0. 0078657 1. 01678 0. 01 0. 994 - 1. 98498 2. 00071 0. 194875
Skilled 0. 0032381 0. 01055 0. 31 0. 759 - 0. 01743 0. 023906 57. 8431
Uni_Edu 0. 0459192 0. 01381 3. 33 0. 001 0. 018858 0. 07298 20. 2815
Trskill* - 1. 673044 1. 10317
-
1. 52 0. 129 - 3. 83521 0. 489121 0. 438905
M_size* - 1. 609459 0. 87619
-
1. 84 0. 066 - 3. 32677 0. 107848 0. 192784
S_size* - 8. 082657 1. 04628
-
7. 73 0 - 10. 1333 - 6. 03198 0. 692348
(*) dy/dx is for discrete change of dummy variable from 0 to 1

42

A.9. Descriptive Statistics of Variables used in the Tobit Estimation of BEEPS 2008 sample
Variable Obs Mean
Std.
Dev. Min Max

W_Balkan 5031 0. 333135 0. 471381 0 1
SEE_EU 5031 0. 196382 0. 397301 0 1
CEE 5031 0. 307494 0. 461502 0 1
Baltic 5031 0. 16299 0. 369393 0 1
ProdSect 5031 0. 354005 0. 478258 0 1
ExpInt 5008 14. 19808 28. 54956 0 100
Innov 5031 0. 799642 0. 400308 0 1
Age 4938 16. 30174 16. 10265 0 183
Agesq 4938 524. 9895 1678. 988 0 33489
Size 4970 105. 7135 429. 5515 0 18208
Sizesq 4970 195652. 7 5355167 0 3. 32E+08
Forown 4964 10. 20165 28. 57173 0 100
Quality 4952 0. 321082 0. 46694 0 1
Uni_Edu 4780 15. 14498 20. 64983 0 100

43

A.10. Tobit Regression of BEEPS 2008 sample
Tobi t r egr essi on

LR chi 2( 12) = 1014. 22

Pr ob > chi 2 = 0. 0000
Log l i kel i hood = - 9661. 0967 Pseudo
R2 = 0. 0499
ExpInt Coef. Std. Err. t P>t [95% Conf. Interval]

SEE_EU - 16. 77992*** 3. 196523 - 5. 3 0 - 23. 04665 - 10. 51319
CEE 3. 269605 2. 518731 1. 3 0. 194 - 1. 668324 8. 207534
Baltic 8. 212801*** 3. 046078 2. 7 0. 007 2. 241018 14. 18458
ProdSect 46. 79662*** 2. 160043 21. 7 0 42. 56189 51. 03134
Innov 11. 26127*** 2. 893899 3. 89 0 5. 587832 16. 93471
Age 0. 4457482*** 0. 1437442 3. 1 0. 002 0. 1639402 0. 727556
Agesq
-
0. 0037302*** 0. 0013371 - 2. 8 0. 005 - 0. 0063516 - 0. 001109
Size 0. 0353211*** 0. 0060306 5. 86 0 0. 0234983 0. 047144
Sizesq - 4. 76E- 06*** 1. 02E- 06 - 4. 7 0 - 6. 76E- 06 - 2. 75E- 06
Forown 0. 3313549*** 0. 0340416 9. 73 0 0. 264617 0. 398093
Quality 19. 58053*** 2. 261062 8. 66 0 15. 14776 24. 01331
Uni_Edu 0. 0665013 0. 0508708 1. 31 0. 191 - 0. 03323 0. 166233
_cons - 70. 58554*** 4. 018363 - 18 0 - 78. 46347 - 62. 70761
/ si gma 55. 00428 1. 11592 52. 81654 57. 19202
Obs. summar y: 3051 l ef t - censor ed obser vat i ons at ExpI nt <=0



A5 - Fisnik Recica Full Paper

Uploaded by

Copyright:

Available Formats

A5 - Fisnik Recica Full Paper

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A5 - Fisnik Recica Full Paper

Uploaded by

Copyright:

Available Formats

*I would like to express my special gratitude to Professor Iraj Hashi and J ean Mangan for helpful comments and

You might also like