Firm Behaviour and the Introduction of New exports: Evidence from
Brazil
Xavier Cirera 1, Anabel Marin 2 and Ricardo Markwald 3
Uncompleted draft- 31st January 2011
1. Introduction
One of the main objectives of economic development policies is to achieve export
diversification. A widely accepted empirical result establishes that at least until relatively
high levels of per capita income are reached, economic development is associated with
the evolving diversification of production into a progressively wider array of new types of
industries and exported products (Imbs and Wacziarg, 2003). Diversification is crucial for
achieving economic development for several reasons. It reduces vulnerability with
respect to external shocks (Haddad, Lim and Saborowski 2009), decreases the
incidence of trade shocks (Ghosh and Ostry, 1994), creates learning opportunities and,
more importantly, it is clearly correlated with high rates of growth (Al-Marhubi 2000;
Herzer and Nowak-Lehnmann 2006; Hesse 2009).
While the benefits of diversification are clear, it is less evident how to achieve it; and
there are indications that breaking into new export markets is becoming increasingly
difficult. One of the major realities of the growth paths of many developing and emerging
economies is the difficulties in reaching a higher stage of diversification that would
enable them to sustain growth and development. Their composition of production and
exports still involves relatively high levels of concentration on „natural‟ resource-based
activities, with slow rates of diversification away from this concentration.
Diversification, however, occurs at the firm level, the main unit of production.
Consequently, understanding diversification requires understanding the process and
capabilities by which firms introduce new products for production and export. Several
approaches have directly or indirectly analysed possible determinants of the process of
diversification. The objective of this paper is to test empirically firm determinants of
diversification by drawing on contributions coming from the economics, innovation and
business literature; as well as understanding firm dynamics around the diversification
process.
In order to analyse the determinants of firm diversification we use a unique micro dataset
that links production, trade and innovation data at the firm level in Brazil. This dataset
allows analysing the impact of innovation, trade and production dynamics on
diversification behaviour.
Despite being one of the most diversified economies in Latin America, Brazil still lags
well behind advanced economies and other emerging economies like China and Mexico
regarding diversification (Hummels and Klenow, 2005). Its export basket is still heavily
1
Institute of Development Studies, University of Sussex, U.K
SPRU, University of Sussex
3
FUNCEX, Brazil
2
1
dependent on natural resources. At the same time, however, Brazil has developed a few
competitive manufacturing sectors, which makes the country an interesting case study.
The paper is organised as follows. Section 2 surveys the literature analysing firm
diversification. Section 3 describes the dataset and methodology used. Section 4
characterises firm export diversification in Brazil. Section 5 analyses the main
determinants of firm export diversification. The last section concludes.
2. Explaining firm export diversification. A survey of the literature.
Firms‟ dynamics around export processes have been analysed by different stands of the
economics and management literature. This section tries to understand these firm level
mechanisms necessary for firm export diversification by surveying the relevant literature
and summarising the main testable predictions associated to the different theoretical
frameworks.
The underlying main determinant for firm diversification across the literature is the
expectation of higher profitability. Firms diversify in order to obtain higher profits. As a
result, the important question to answer is how this diversification occurs and how the
diversification path looks like. Different stands of the literature have approached these
questions in different ways, mainly focusing on their own core research question. Some
of the literature on innovation or management has focused on the introduction of new
products for production in general, which may be exported or not. The insights of this
literature are critical for export diversification. Another example is the trade and
heterogeneous firms‟ literature, based on (Melitz 2003) workhorse model, which has
focused on describing firm trade dynamics regarding trade flows, number of firms
engaged in trade, quantities and prices. Despite this last literature is agnostic regarding
the endogenous processes of firm selection to export, it still offers important predictions
on exogenous factors, such as productivity levels and trade costs, which matter for
exporting. A final example is the “discovery” literature (Hausmann and Rodrik 2003),
which focuses only on new products, “discoveries”, for the country, and gives important
predictions about the constraints for firm diversification in general.
As a result of this heterogeneity of predictions arising from the different frameworks,
finding a sequential way of synthesising this literature is difficult, since different angles
look at different research questions and include different internal and external
determinants. For this reason, our approach is to summarise each stand of the literature
separately, but doing so by focusing only on the predictions relevant to the micro
determinants of export diversification.
2.1 Heterogeneous Firms Models and the Extensive Margin of Trade
In recent years, a large number of empirical studies have emerged based on (Melitz
2003) model of heterogeneous firms. The key element of this framework is the issue of
self-selection of firms for exporting. One important stylised fact of this literature is the
fact that firms involved in international trade tend to be larger and more productive
(Bernard, Jensen, Redding and Schott 2007; Mayer and Ottaviano 2008). This model is
able to replicate these stylized facts. Nevertheless, its static version treats productivity
as exogenous to the firm, and does not address how firms achieve the productivity
thresholds that allow them to export. Accordingly, the key elements shaping the decision
2
for exporting are productivity levels and trade costs, which will determine what firm will
access export markets and to what markets will be able to export.
Several papers have analysed empirically the predictions of the model. Export growth
can be decomposed by growth at the intensive margin, growth in value of existing
products; or growth at the extensive margin, growth in value of new products exported or
new markets. A key finding of the literature is the fact that most export growth tends to
occur at the intensive margin, on existing products. Breaking into new products and
markets tends to be more difficult than increasing existing shares. A collection of papers
have analysed export growth at the extensive margin and linked this type of growth to
the predictions of the Melitz‟s (2003) model. The focus of these papers has been how
tariffs and trade costs, including the business environment, affect growth at the
extensive margin. For example, (Amurgo-Pacheco and Pierola 2008; Dennis and
Shepherd 2007) find that reducing trade costs via improving the business environment,
reduction in transport costs or reducing tariffs increase extensive margin growth
(diversification).
In recent years, the Melitz‟s framework has been modified in order to accommodate
other important stylised facts. For example, (Baldwin and Harrigan 2007) modify the
model in order to accommodate quality differentiation and the fact that more productive
firms also earn higher (rather than lower) export prices. (Ruhl and Willis 2009) modify
the model to incorporate entry costs that change with changes in relative prices,
productivity and demand shocks. This allows the model to be able to replicate gradual
growth in exporters‟ market share. Finally, (Arkolakis and Muendler 2009) modify the
model to incorporate multiproduct firms.
A more relevant modification of this framework for the question of this paper is
(Costantini and Melitz 2007). The authors modified the Melitz (2003) model in order to
include dynamic elements, where entry, exit and innovation are jointly determined
depending on entry and trade costs. In addition, (Aw, Roberts and Xu 2009) also
develop a dynamic industry model with firm heterogeneity where investment on
innovation decisions depends on profitability and entry sunk costs. These models
represent a substantial improvement in trying to make productivity dynamics
endogenous.
Despite these new
framework, the focus
explaining innovation
is that diversification
costs.
advancements to include dynamic issues within the Melitz
of this literature relies mainly on entry and trade costs, rather than
or productivity changes. Thus, the key prediction for new products
will occur mainly via reductions on trade costs, tariffs and entry
2.2 Multiproduct Firms, Product Churning and Survival
Traditional trade theory has considered firms as producers of one product. The reality,
however, is the fact that firms are multi-product and export multiple products to multiple
destinations (Bernard, Redding and Schott 2010). This implies that firms look at the
decision to export a new product in conjunction with the optimal production and export
mix.
(Broda and Weinstein 2010) for example analyse product creation and destruction for a
large sample of firms in the US. The authors document large product turnover and the
3
importance of aggregate demand for creation of products. Creation is largely pro-cyclical
and does not occur until high potential demand is identified.
The large extent of product churning and adjustment of multi-product firms to productivity
and demand shocks via adding and dropping new products is also evident when
analysing export flows. Due to uncertainty regarding trade relationships, most trade
flows tend to be small and short-lived. (Besedes and Prusa 2006; Besedes and Prusa
2006) using trade data for the US at the product level document the fact that most trade
transactions are small and do not survive more than three years. Survival rates are
larger for differentiated products since the trade relationships required to sustain this
type of trade need to be solid. (Martincus and carballo 2009) find a similar pattern for
Peru and suggest that export market diversification is more important than product
diversification in increasing survival. (Volker 2009) also finds low survival rates for
German imports, and the fact that survival depends on how larger and closer exporters
are, the demand elasticity of substitution, and exporters market power.
There are several important implications from this literature regarding firm export
diversification. First, the role of demand shocks as determinant for adding new products.
Second, the need for considering product churning as a multi-product decision rather
than a single product decision in isolation from other products. Third, the fact that
survival of new products may be short-lived. Forth, the fact that more differentiated
goods require more stable relationships and, therefore, are less difficult to introduce but
more stable. Given these stylised facts, an additional important question for firm
diversification is how previous experience, both positive from successful exports and
negative from exits, affects the current decision for introducing a new product.
2.3 Preparation for Exporting and Sequential Exporting
The issue of uncertainty is a key element when establishing new export relationships.
While there is a large literature on learning-by-exporting related to changes in
productivity, there is less evidence on the dynamics preceding the introduction of new
products for exporting. One type of studies, innovation studies, has looked at innovation
efforts (see section 2.5). A new angle of the literature has started to look at preparation
efforts or sequential exporting.
(Iacovone and Javorcik 2010) using a firm level dataset from Mexico document firm
market preparation for exporting. Firms first introduce varieties domestically and there is
evidence of increases in quality proxied by increases in prices preceding exports. Once
domestic varieties have matured, then they are exported. Given the asymmetry and
uncertainty of information, multi-product firms start exporting a small number of varieties
in small volumes. They also find that most varieties do not survive for more than a year
and that export discoveries are rare.
(López 2009) focus on a different type of preparation for exporting, firm investments.
Analysing the relationship between exporting and productivity, the author shows using
Chilean firm-level data significant increases in investment and productivity prior to firms
start exporting. This supports the view of self-selection of exporters, in this case using
investments to increase productivity.
(Albornoz, Calvo-Pardo, Corcos and Emanue 2010) focus on a different type of
uncertainty: expected profitability. In their model, while there are high sunk costs for
4
exporting, profitability can only be determined once the firm has started to export. This
implies a sequential process where firms first decide whether to export based on
expected profitability, and then adjust quantities, prices and markets once the real
profitability can be estimated. This gradual and sequential export expansion is also
affected by distance and trade costs, since expansion to other markets depends on
similarity and distance.
The importance of this literature is that it gives very relevant predictions for
understanding the determinants of exporting new varieties. In particular, the following
elements need to be considered in preparation for exports:
 Maturity of product for domestic market
 Process of quality upgrading
 Increases in investment
 Increases in productivity
 Familiarity with export markets
2.4 Export “Discovery”
Following the seminal work of (Hausmann and Rodrik 2003) the attention on export
diversification at the country level has focused on understanding the processes through
which a firms start exporting a product new to the country. The original model focused
on the role of market failures constraining export “discoveries” when initial entrepreneurs
are unable to capture all the positive externalities generated with the product discovery
due to rapid imitation by other firms. A large number of empirical work and case studies
has focused in describing these discovery episodes.4
(Freund and Pierola 2009) document new export “discoveries” for Peru, and compare
them with new exports to the firm. The authors find that new exports to the firm have
less chance of survival than “discoveries”, and that only large experienced exporters
engage in “discoveries”, since the later require larger sunk costs. (Klinger and Lederman
2004) find that “discoveries” are not restricted to ‟dynamic‟ industries but also to
traditional sectors such as agriculture. More importantly, the authors find that the
determinants of entrepreneurship are not correlated with the frequency of discovery,
which implies the existence of market failures.
(Klinger 2007) focuses on the different types of uncertainties related to export
“discoveries”. The author analyses eight case studies of different sectors in different
countries and suggests two types of uncertainty: productivity, costs and quality, and
demand, market segment characteristics and price. The author finds that when
uncertainty is high, two main alternative strategies emerge: “discovery” in similar
products or the use of FDI investors to obtain technology and knowledge.
While the “discovery” literature focuses on a narrow type of firm diversification, new
export products for the country, it stresses relevant elements for firm diversification in
general. For example, one may characterise both processes of firm diversification by a
similar process where only the magnitude of sunk costs for exporting is different, much
larger for “discoveries” and the capacity for imitating also differs, much lower for
discoveries. If this is the case, then the role of productivity and demand uncertainties
4
See for example the collection of country case studies commissioned by the IADB for the
project “The Emergence of New Successful Export Activities in Latin America”
5
and the strategy to overcome these uncertainties, product relatedness and FDI
investment, are extremely relevant for firm diversification in general.
2.5 Innovation Studies
Another stand of the literature that has concentrated on the issue of preparation for
exporting is the innovation literature. Although this is a very large and broad literature,
some papers analyse the impact of innovation on the propensity to export. One problem
of this literature is the proxy for innovation, since often only broad measures such as
R&D expenditure are used as proxy.
(Aw, Roberts and Xu 2009) for Taiwan and (Cassiman and Martinez-Ros 2007) for
Spain find that R&D investments increase the probability of exporting. (Becker and
Egger 2007) look at the types of innovation efforts in more detail. Concretely, they
compare the impact of product vis-a-vis process innovation on exporting. One problem
when looking at the impact of innovation on exports is the endogeneity of innovation
expenditure, since exporters tend to invest more on innovation. The authors using a
sample of German firms and PSM techniques to control for endogeneity find that product
innovation is the key factor for exporting. In a similar paper using a sample of Belgium
firms and also controlling for the endogeneity of innovation, (Beveren and
Vandenbussche 2009) find that it is the mixture of both product and process innovation
which increases the propensity to export. The authors also stress the importance of
anticipation, which implies the need for controlling for past export activity and the fact
that innovation efforts occur before exporting. In another related paper, (Damijan,
Kostevc and Polanec 2010) look at the relationship between innovation and exporting in
Slovenia. Using also PSM methods to correct for endogeneity they find no impact of
product or process innovation on exports. On the contrary, the authors find that
exporting tends to increase innovation efforts.
This last finding is consistent with some of the management literature on global
engagement and innovation (Lederman 2009). Firms engaged in international trade and
MNEs are more likely to innovate because they have better access to technology
diffusion. (Lederman 2009) also suggests an additional hypothesis related to innovation,
market structure. Concretely, based on existing models one can think of inverted-U
relationship between the degree of competition and innovation. Firms may require large
monopoly rents in order to innovate when they are further away from the technological
frontier, while once they are closer to the frontier they can compete with little innovation
investment.
Despite the lack of stylised facts of some of this literature, it stresses clearly the
importance of considering product and process innovation when explaining export
diversification. One important difference of this literature with the question of this survey
is the focus on whether firms are engaged in export activities, rather than on the
introduction of new products. Likely as the result of the data they use, they cannot
explain firm diversification. In addition, due to the questionnaires used the innovation
efforts are only aggregated into product vs process, rather than concrete innovation
measures. Despite these caveats, the studies are methodologically very sound and
stress the importance of focusing on innovation efforts and on its endogeneity for
exporting.
6
On the other hand, more clear predictions arise from the business literature of innovation
determinants. Concretely, market structure and global engagement or international
exposure may explain cross-sector differences in innovation, and, therefore, product
adding.
2.6 Firm diversification and relatedness
A very relevant link with the dynamics of firm diversification is provided by the business
literature on firm diversification. This literature has analysed the determinants of firm
diversification to other activities and focused on the type of new activities that the firm is
more likely to produce.
(Lien and Klein 2010) summarise the main determinants in two main types. On the one
hand, the industrial organization literature emphasises the role of profitability in
explaining firms‟ decisions to entry in new sectors. However, the fact that certain sectors
are more profitable than others, imply also higher barriers to entry. This implies that
entry to a new product/sector should be positively correlated to product/industry growth
and negatively correlated with sector/product specific barriers to entry. On the other
hand, the resource-based approach stresses the importance of excess capabilities and
resources explaining diversification. Accordingly, firms will diversify only to those sectors
which are similar and can be handled by existing resources and capabilities. This raises
the question of relatedness or similarity between products or activities. Diversification is,
therefore, not random and follows “feasible” paths along the “product space”.
A large number of measures have been proposed ranging from categorical measures to
SIC classification distances, input ratios, commodity flows or survival measures (see
(Lien and Klein 2009)). Changes in SIC classification at 3 or 4 digits measure whether
pairs of products are within the same classification category. This type of measures,
while simple to calculate fail to capture that certain activities outside a core product are
easier to produce with existing capabilities. Other measures are based on correlations
between sector input uses or commodity flows across sectors. These measures provide
a proxy of similarity in the production process across sectors. A final measure is based
on co-occurrences at country or plant level. (Hidalgo and Hausmann 2009) build a
network representing the product space based on co-ocurrences of countries exporting
the same product. They show that diversification occurs in countries by moving to similar
and closer products in the product space (Hidalgo, Klinger, Barabasi and Hausmann
2007). (Neffke and Henning 2008) measure co-ocurrence at the plant level using firm
production portfolios. The assumption is that more plants produce the same pair of
products the more similar the capabilities required to produce them.
While this literature does not focus on how firms introduce new products and when
would that happen, it gives very strong predictions on where diversification is likely to
occur. Accordingly, diversification should be strongly conditioned by product/sector
profitability and existing capabilities and resources.
2.7 Multinational Enterprises, FDI and Value Chains
One way of overcoming the uncertainties related to firm-diversification is via foreign
investors or participating in foreign value chains. As suggested above, Klinger (2007)
finds FDI an instrument to overcome productivity and demand uncertainties for export
activities. However, the empirical literature on technology transfer for firm export
7
diversification is rather thin, likely the result of the difficulties to measure
information/technology transfer. (Swenson 2008) studies the relationship between MNEs
and export relationships in China and find that MNEs enhance export capabilities mainly
via information spillovers. (Greenaway, Sousa and Wakelin 2004) find for a panel of UK
firms that export propensity is larger in sectors with larger MNE presence. They find this
result as evidence of positive spillovers. (Bekes, Kleinert and Toubal 2009) analyse the
impact of MNEs in Hungary. The authors find evidence of positive spillovers on more
productive firms but not on exporters. (Aitken, Hanson and Harrison 1997) look at the
impact of MNE‟s spillovers using data from Mexico. They find positive spillovers from
MNEs on the propensity to export, although the export intensity of MNEs does not affect
the propensity to export by domestic firms. Finally, (Javorcik and Spatareanu 2009)
analyse the spillover effects of MNEs on domestic firms for a sample of Czech firms. The
authors find evidence that high productivity firms have a higher probability of supplying
MNEs, but also that suppliers learn from their relationships with multinationals; although
they do not focus on the role played by MNEs on firm diversification.
Regarding value chain relationships, (Egan and Mody 1992) study buyer-seller
relationships for the bicycle and garment imports to the US. They find that these
relationships help exporters in developing countries to lower entry costs and act as
instruments for information, technology diffusion and access to industrial networks.
(Hobday and Rush 2007) analyse the role of MNEs in electronics in building export
capabilities for exporting on domestic subsidiaries in Thailand. They find that some
subsidiaries upgraded capabilities while others remained as assembling plants. A key
element in determining the outcome on capabilities seemed to be the degree of
centralisation of technology decisions within the value chain. More decentralised
networks allowed for plants to upgrade, while value chains where technology decisions
and processes were tightly controlled within the parent headquarter remained as
assembly plants. Thus, while in some cases MNEs can be an important vehicle for
technology transmission to domestic subsidiaries, they can also become an important
constraint in highly centralised value chains.
While the studies above suggest that there is evidence of MNEs spillover effects within
industries on export propensity, there is little evidence on the specific role of MNEs, FDI
or value chain relationship on firm export diversification. Specifically, little is known about
the mechanisms through which MNEs, FDI or value chain relationships may facilitate
domestic firms export diversification. In addition, different degrees of value chain
governance can play different roles in fostering or constraining firm diversification.
2.8 A summary of the main hypothesis
Firms introduce new products for exporting because of the expectation of higher profits.
This is the underlying main determinant of firm diversification for all the frameworks
surveyed above. The literature, however, differs on the answers to two key questions
required for understanding firm export diversification:
 How firms introduce new products for exporting
 Where or in what products diversification is more likely to occur
Table 1 below attempts to summarise the main mechanisms suggested by the literature
to explain these two questions. The last column of the table suggests some initial
proxies that can be used to test the different predictions.
8
Table 1 The determinants of firm diversification for exporting
Question
How to introduce new
products
Literature
Heterogeneous
Melitz
firms
Multiproduct firms
Preparation for exporting
Sequential exporting
Innovation
Global engagement
Market structure
MNEs
Value chains
Where or in what
products diversification
is more likely to occur
Profitability
Resource view
Export discovery
Mechanisms
Responses to positive productivity shocks
and reduction in trade costs that make export
of existing produced products profitable
Product churning and adjustment to demand
and supply shocks, and trade policy
Explicit preparation to export via:
- Product maturity in domestic market
- Quality upgrading before exporting
- Increased investment and productivity prior
to exporting
Due to uncertainty on profitability, sequential
expansion of markets and volumes
Investment on process, and especially
product innovation, in order to export.
Access to Technology diffusion via exports,
imports and FDI
Increased innovation with competition until
close to technological frontier
Information spillovers in sectors with MNE
presence
Capabilities adoption via value chain
relationships
Diversification likely to occur in sectors of
more profitability and lower barriers to entry
Diversification likely to occur in related
products where firm has enough capabilities
and resources
Easier where productivity and demand
uncertainty
are
lower
and
positive
externalities are more likely to be captured by
entrepreneur
9
Potential proxies
Productivity index
Tariffs
Product specific transport and trade costs
size
Number of products
Sector/product demand growth
Previous introduction of product for
domestic market
Unit value changes prior to exporting
Investment changes prior to exporting
Prior experience
Prior number of exporting markets
Different Innovation measures
PINTEC
Product/sector highly traded
Product sector with high FDI
Concentration ratios
Barriers to entry proxies
Sector output MNE share
From
Type of relationship with suppliers via PIA
and PINTEC
Type of value chain governance
Product/sector growth
Relatedness
measures
(literature)
between existing a new products
Relatedness measures
Barriers to entry proxies
Regarding the first question, how this process of firm diversification happens, different
frameworks focus on several external and internal (to the firm) factors. The
heterogeneous firms‟ framework assumes self-selection and, therefore, focuses on
exogenous positive changes in productivity levels or trade costs that will allow firms
already producing domestically to achieve the required productivity threshold for
exporting. The multi-product literature emphasises the role of adding products as
response to demand and supply adjustments. The global engagement, MNEs and value
chains literature emphasise the role of engagement with external actors in order to
obtain the capabilities to export a new product via information spillovers or technology
diffusion. Finally, some studies, included under preparation for exporting, sequential
exporting and innovation, focus on specific processes through which firms introduce new
products such as prior domestic market preparation, investment or sequential
expansion.
In relation to the second question, where diversification is more likely to occur, a key
element is the role of proximity or relatedness. Diversification will occur more likely in
products where there are capabilities and resources. This is also emphasised by the
discovery literature, and more concretely by the product space literature (Hidalgo,
Klinger, Barabasi and Hausmann 2007), which suggests that diversification to other
sectors is more likely to arise when countries already produce in connected sectors. One
way, however, to jump to more distant activities when uncertainty is large is via FDI or
integration in value chains, which facilitate the acquisition of the relevant know-how.
Understanding the determinants of firm export diversification, therefore, requires
considering both, the internal processes of diversification, such as increases in
innovation efforts or prior preparation via domestic markets, and the external factors that
facilitate exports, such as changes in demand or in trade costs. It also involves
understanding what activities are more likely to enter diversification strategies and under
what circumstances. The relative importance of these determinants is likely to depend
on firm characteristics (size, global engagement,..), type of product, sector, demand, and
the proximity to existing production patterns and capabilities. This more comprehensive
view of firm export diversification is missing from the existing literature, which tends to
focus on individual sets of determinants, while ignoring other important factors.
3. Data and methodology
In order to analyse firm export diversification we build a unique dataset that links
production, trade and innovation data for Brazilian firms. Concretely, we use the
following databases:
PIA (Pesquisa Industrial Anual)
PIA is a firm survey for manufacturing and mining sectors conducted annually by IBGE
(Instituto Brasileiro de Geografia e Estatistica). PIA has two different modules, PIA
empresa, which focus on firm characteristics, and PIA produto, which describes the
production and sales portfolio for each firm.
It surveys firms in the formal sector, with tax identification number, and with a core
activity in manufacturing or mining. Firms with 30 or more employees are included in the
sample, while smaller firms up to 29 workers are included randomly in the sample. In
10
total PIA covers more than 40,000 firms. PIA produto is based on the PIA empresa
sample. However, before 2004 only the largest firms from PIA empresa were included.
PINTEC (Pesquisa de Inovação Tecnológica)
PINTEC is a survey in innovation based on the CIS-4 surveys of the European
Community. It provides detailed information on R&D expenditure and innovation
processes for a sample of firms. Firms with more than 500 workers are automatically
included in the sample, while firms from 5 to 499 workers are included randomly.
PINTEC is available for 2000, although with a different questionnaire, 2003 and 2005.
SECEX (Secretaria Comercio Exterior)
SECEX provides the universe of registered trade flows at the firm level, by HS-8 product
and market destination. The dataset used aggregates export fob values per year,
product and destination.
Due to its most restrictive sampling methodology, estimations are based on the sample
of firms surveyed in PINTEC. However, the overall dataset includes all the data
available. When merging PIA and PINTEC, 73% of observations from PINTEC are
matched with PIA data, while the remaining 27% are not matched. Interestingly, all
exporters from SECEX have been surveyed by PIA or PINTEC, and they represent 17%
of the overall sample.
In order to avoid simultaneity problems we use lagged variables from PIA and PINTEC
to explain new exports in t. Since PINTEC has only three years available, we effectively
use production, firm and innovation data for each firm i (vector Xkt-1) in 2000, 2003 and
2005 to explain the probability that firm i introduces a new exported product in
2001,2004 and 2006 (Yit).
Yit   0  k  k X kit 1  uit
(1)
Based on table 1 we calculate different variables to proxy the predictions of the literature
regarding firm diversification. The main variables used are summarised in Table 2. We
use sector dummies at CNAE two digits in order to control for sector specific elements
such as trade costs, profitability and changes in foreign demand. In line with the
heterogeneous firm‟s framework we use firm size and productivity. We also control for
the number of firms that firm produce and export. Regarding, preparation for exports we
look at the evolution of unit values and markets shares of the firm previous to the
introduction of the new product. We also use information on the type of innovation, the
size of the innovation effort and firm ownership. Finally, we look at the degree of
relatedness between new exports, existing exports and the production bundle. To
measure relatedness we use different measures based on crude classification code
distances and correlations based on input usage across sectors and proximity in the
product space. Section 4 describes in more detail the construction of some of these
variables.
11
Table 2 proxies used in empirical analysis
Literature
Heterogeneous firms
Variables used
lnTFP - Total factor productivity -TFP using Levinsohn and Petrin
and value added
Multiproduct firms
Preparation
exporting
for
Innovation
Global engagement
Market structure
MNEs
Value chains
Profitability
Resource view
export discovery
and
ln(L) Size- log of employment
Sector dummies
Herf - Herfindahl index of concentration for both the production and
export basket
Number of products produced and exported
Average ratio firm unit value vis-a-vis product unit value average
Change in average unit value ratio
Average firm share in domestic production for each product
Change in average product share
Dummy- inova -either in product or process
Dummy – inoprod - only product innovation
Dummy –radinov –if radical innovation to the world
Log innovation expenditure
Skills_proxy - ratio firm to average sector wage
skills_rd – the sum of technical workers
Sector dummies
Ownership dummy
Dummy
ISO certification
Sector dummies
Ownership/foreign capital
client_dep - high information from foreign client
group_dep- high information from foreign group
Sector dummies
Relatedness measures:
- distance_3d 3 digits is distance min and max CNAE 3 digits
multiplied by 2 if change CNAE1 production basket
- distance_2d 2 digits is distance min and max CNAE 3 digits
multiplied by 2 if change_cnae1 – production basket
- hs2 distance – simple difference between hs2 sectors – export
basket
- Correlation in input use across national accounts sectors (see section
4)
- Hidalgo and Hausmann SITC4 correlations (See section 4)
The challenge in estimating equation (1) is the definition of new product Yit. We use
alternative definitions described in section 4. Furthermore, the definition of the control
group once we identify episodes of firm diversification is critical. Two main alternatives
for control group exist. First, if diversification processes for existing exporters are similar
to the processes required by firms breaking into export markets, then new exporters
should include both diversifying firms and new exporters vis-a vis non diversifying
exporters. Alternatively, new exporters should be considered separately or part of nondiversifying exporters.
In order to overcome these problems we follow a sensitivity analysis approach and
compare different results from different specifications. First, we use two alternative
measures of new export. The first measure, Y1it, is one when the firm diversifying is a
new exporter, and zero if the firm is an existing exporter not diversifying or a new
12
exporter. The second measure Y2it, is one when the firm is diversifying or it is a new
exporter, and zero when it is an existing exporter that does not diversify.
Equation (1) can be estimated as a Probit random effects model for both alternative
indexes. However, in doing so, we omit from the sample non-exporter firms. It is possible
that unexplained factors that determine whether a firm is a exporter are also correlated
with the probability of firm diversification, and therefore, posing a risk of sample selection
bias. In order to correct for this potential problem, we also perform a Heckman selection
and use a Heckprobit estimator, where the calculated inverse mills ratio (IMR) from a
first selection model for the probability of exporting are introduced as regressor of the
level equation; the probability of a firm diversifying. The estimates are summarised in
section 5.
4.
Firm export diversification in Brazil
The key variable in this paper is to identify episodes where new products are introduced
for exporting. There are two challenges in doing so. First, we only observe exports flows
for the period of our sample, so we cannot determine whether a product was introduced
before this period. Second, as suggested by the literature (Besedes and Prusa 2006;
Martincus and carballo 2009) most export flows tend to be short lived. Table 3 shows the
number of years that each variety (product/firm) is exported for all sectors. Only 2.28%
of varieties are observed during the whole period and only 10% of varieties are exported
from the year that they are introduced to 2008.This is an indication of extremely large
short duration of export flows, with 57% of varieties exported only one year.
There are two main implications of low survival rates for new exports. Low survival rates
make more uncertain to know whether a firm has exported a product in the period prior
to our sample. Second, low survival rates imply that not considering issues of
sustainability when determining new exports would be translated into an extremely large
number of products considered.
Table 3 Duration of exports - product by firm- all sectors
Number of years exported
Number of varieties
1 year
360,300
2 year
105,877
3 year
55,351
4 year
30,327
5 year
21,181
6 year
15,968
7 year
12,371
8 year
9,808
9 year
6,959
10 year
14,443
Total
632,585
63,187
continuous until 2008
Source: Authors‟ own elaboration from SECEX
13
% share
56.96
16.74
8.75
4.79
3.35
2.52
1.96
1.55
1.1
2.28
100
9.99
In order to address the issue of firm export diversification and sustainability, we compute
three alternative measures with different degrees of rigidity regarding survival after being
exported for first time. The three classifications are summarised in Table 4. Classification
1 is clearly the most rigid methodology since requires continuously exporting the
product, once it is introduced until 2008. All new products identified in classification 1 are
part of the other two methodologies. Classification 2 allows for some degree of
intermittence during the period, while the last methodology is the most flexible.
Table 4 Methodologies to identify new products
Classification
Description
Classification 1 New product not exported in 2000, and once introduced is exported
continuously until 2008; or in 2007, 2008 and 2009.
Classification 2 New product not exported in 2000, and once introduced exported at
least 5 years; or 2006, 2007 and 2008; or 2007, 2008 and 2009
Classification 3 New product not exported before 2002 and exported at least three
years after
The different methodologies are applied to the SECEX dataset ignoring the destination
market dimension. One could define variety as firm, product and destination. However,
survival rates at such level of disaggregation are even lower. Besides, the main interest
of the paper is on product diversification, rather than market diversification.
The SECEX dataset includes the universe of exporters and, therefore, it has firms
belonging to all the sectors of the economy. One element that arises when looking at the
sector composition of exports at the firm level (Table 5) is the extremely large number of
exported products for the construction sector. This sector is largely dominated by a small
number of MNEs with large construction contracts in Latin America and Africa, which
export a very large number of products in order to supply their operations abroad. This
explains that on average firms in that sector export 60 different products. The number of
products exported in the trading and services sector is also larger than manufacturing
and agriculture. Since the objective of the paper is to focus on how firms produce and
introduce new products for exporting, in the remaining of the paper we focus on the
manufacturing sector, for which innovation and production data is available on the other
surveys.
Table 5 Number of exported products by firm – Sector decomposition
Standard
Sector
mean median
p99
deviation
Agriculture
2.10
1
13
3.27
Trading
9.92
2
143
30.75
Construction 60.39
2
2105
289.05
Industry
6.71
2
71
19.38
Services
7.32
1
106
29.20
Other
5.75
1
63
19.36
Source: Authors‟ own elaboration from SECEX
N
3,859
50,621
623
124,327
6,052
1,221
sum
8,098
501,926
37,620
834,279
44,294
7,023
range
106
1013
2577
603
746
460
max
107
1014
2578
604
747
461
The different methodologies described in Table 4 are implemented to firms classified in
the manufacturing sector according to whether their core activity in terms of sales lies in
this sector. In addition to the three different classification methods, we differentiate
14
among three different types of new exported product. Concretely, the three following
typologies of products are identified:



New products – these are new products introduced by existing exporters
New exporters – these are new products introduced by firms that were not
exporting previously
Discovery – these are new products that were never exported in sample, and,
therefore, are new products to the country.
Table 6 shows the summary of new products, exporters and discoveries according to
each classification for the sample period. In general, new exports are not such a rare
event. Around 20% of firms introduce a new product under one of the different
classifications in our sample, and when considering the period 2001-2007, an average of
28% of firms a year introduces a new product for exporting. This rate falls to 17% when
considering the more restrictive classification 1.
Classification 1 identifies from 816 to 1,901 firms introducing new products. It clearly
becomes less rigid after 2006, due to the fact that the likelihood to be exported only
three years increases substantially. On the other hand classification 3, which requires
less sustainability of the export flow, shows a larger number of firms introducing new
products, ranging from 3,212 to 2,218 firms, and new exporters, ranging from 1,358 to
524 firms.
While more than 50% of firms introduce only one new product each time they diversify
according to each classification, some firms introduce more than one product at the
same time, with an average that oscillates between 2 or 3 products. In a few cases, a
very small number of firms introduce in the same year more than one hundred new
products. While these firms are likely to be very large firms, it is also possible that are
firms involved in trading activities at the same time than production.
An important element to analyse is the sector in which new exports are more likely to
happen. When looking at the distribution of new products and exporters by sector of
activity, we need to consider the number of existing product lines at each sector, since
some sectors have a larger number of products that can be exporter. Appendix 1 shows
the number of new products and exporters as a share of total flows per each HS-2
chapter. The sectors with larger share of new exports are 60 fabrics, 41 skins and
leather, 30 pharmaceutical, 31 fertilizers and 81 other base metals. 5
5
We also observes some new exports related to agricultural exports that correspond to firms that
also produce agricultural products, despite their core activity is considered manufacturing.
15
Table 6 New products in the manufacturing sector
year
2001
2002
2003
2004
2005
2006
2007
obs
mean
sd
p99
max
obs
mean
sd
p99
max
obs
mean
Sd
P99
max
Obs
mean
sd
p99
max
Obs
mean
Sd
P99
max
Obs
mean
Sd
P99
max
Obs
mean
Sd
P99
max
New products
Class 1 Class 2 Class 3
816
1931
1.92
2.81
2.39
4.74
12
26
26
74
1074
2053
2.09
2.61
3.01
4.15
17
20
44
69
1335
2096
3212
2.13
2.53
2.98
3.11
4.05
4.86
19
25
26
37
53
69
1474
1857
3112
2.19
2.41
2.86
5.32
5.50
6.11
16
19
21
170
188
252
1649
1777
2757
2.12
2.19
2.61
3.69
3.80
4.55
13
14
19
115
117
148
1901
1901
2218
2.16
2.16
2.36
3.57
3.57
3.96
14
14
17
95
95
99
3784
2692
2692
2.64
2.29
2.29
4.11
3.31
3.31
21
17
17
68
47
47
New exporters
Class 1 Class 2 Class 3
329
827
1.66
1.74
1.91
2.13
11
11
18
25
334
680
1.71
1.84
2.27
2.81
12
17
22
36
501
785
1358
1.76
1.94
2.06
3.00
4.00
3.94
9
12
13
45
87
109
438
544
1000
1.86
1.83
1.97
4.37
4.27
4.53
12
12
14
81
88
123
358
371
593
1.80
1.83
1.90
2.68
2.75
2.80
13
16
13
32
34
43
460
460
524
2.35
2.35
2.30
15.24
15.24
14.91
9
9
9
327
327
341
651
405
0
2.17
2.13
5.29
5.53
19
17
114
97
Class 1
1
1
1
1
3
1
0
1
1
2
1
0
1
1
1
1
1
1
3
1
0
1
1
2
1
0
1
1
7
1.14
0.38
2
2
Discovery
Class 2 Class 3
4
1
0
1
1
8
1.25
0.46
2
2
3
8
1
1
0
0
1
1
1
1
1
2
1
1
1
1
3
1
0
1
1
2
1
0
1
1
7
1.14
0.38
2
2
1
1
5
1
0
1
1
2
1
0
1
1
7
1.14
0.38
2
2
Obs- number of firms that introduce new products; mean- average number of new products per
firm that introduces new products; sd – standard deviation; p99 – number of new products at the
th
99 percentile; max- maximum number of products introduced by a single firm.
Source: Authors‟ own elaboration from SECEX
The last three columns in Table 6 measure the number of new products for the country,
“discoveries”. As expected, “discoveries” are very rare, 39 new products in total, ranging
from one to a maximum number of eight discoveries in the same year. By sector, most
discoveries are concentrated in the chemicals (HS2 28, 29, 38) and pharmaceutical
products (30), which account for 77% of “discoveries”
16
Table 7 Product “discoveries” by HS2 sector
HS2
2
28
29
30
32
38
39
48
50
51
55
84
91
Description
MEAT AND EDIBLE MEAT
INORGANIC CHEMICALS
ORGANIC CHEMICALS
PHARMACEUTICAL PRODUCTS
TANNING OR DYEING EXTRACTS; TANNINS
MISCELLANEOUS CHEMICAL PRODUCTS
PLASTICS AND ARTICLES THEREOF
PAPER AND PAPERBOARD; ARTICLES OF PAPER
SILK
WOOL, FINE OR COARSE ANIMAL HAIR; HORSE
MAN-MADE STAPLE FIBRES
NUCLEAR REACTORS, BOILERS, MACHINERY
CLOCKS AND WATCHES AND PARTS THEREOF
Number
Discoveries
1
6
15
5
1
4
1
1
1
1
1
1
1
%
2.56
15.38
38.46
12.82
2.56
10.26
2.56
2.56
2.56
2.56
2.56
2.56
2.56
Source: Authors‟ own elaboration from SECEX
Relatedness in exports
Related to the question on how firms diversify, is the question to what new activities are
firms more likely to diversify. This question can be approached by analysing the degree
of relatedness between new products introduced and the existing production or export
portfolio. The resource based approach to firm diversification (Lien and Klein 2009)
suggest that firms can diversify only to those activities where there are existing
capabilities within the firm. Thus, we should expect that diversification occurs to similar
products.
In the context of multi-product firms, defining relatedness is problematic since firms are
able to produce and export products classified in different sectors. That implies that firms
are also able to introduce more than one product for exporting belonging to different
sectors.
This is reflected in our dataset where in a significant number of cases firms introduce
more than one new product for exporting. 6 If we identify for each firm/year those new
products according to any of the methodologies, and compare the variety of new
products, there is a significant degree of sector variation. We compute the maximum and
minimum HS code for all identified new products and new exporters under alternative
classifications. For only 8.4% of the cases the difference between products is within the
same HS-4 digits sector and 23% within the same HS-2 chapter. In addition, we
calculate a simple difference in HS-2 chapter code. Figure 1 shows the probability
distribution function of the calculated distances. Interestingly, there is a second peak on
the right of the distribution between a distance of 40 and 60 HS2 chapters. Interestingly,
the large majority of distances in this second cluster is explained by firms that introduce
6
Since we look across methodologies for new products and new exporters, we identify more than
one product in a large number of cases, 86% of firms that diversify or enter export markets.
17
a product from the plastic and rubber sectors, as well as products from the machinery
and transportation sectors. In general, the export basket tends to be diversified along the
value chain, with firms being able to export inputs and processed inputs.
Figure 1 Probability distribution function HS-2 distances – New products basket
.02
0
.01
Density
.03
.04
Kernel density estimate
0
20
40
60
80
100
distance
kernel = epanechnikov, bandwidth = 3.0628
While comparing the set of new products introduced for exports gives a clear idea of the
level of firm diversification, the issue of relatedness should be considered in relation to
existing exports. In order to do so, we concentrate the analysis to those exporting firms
that introduce one new product for exporting under the more stringent classification. We
have more than 12,000 firm/years where new products are introduced, from which in
45.38% of case firms introduce one product and 20% two new products. The remaining
firms introduce more than two products to a maximum of 252. It is possible that some of
these firms perform trading activities at the same time than manufacturing.
In order to simplify the analysis we focus on the sample of 45% of firms that introduce
one new product for exporting in a given year. We compare the new product introduced
with the export basket of the previous year and calculate the following set of relatedness
measures:



Correlation based on the input use of the input-output matrix in 2005. We
calculate the correlation in terms of input use between the 55 national account
sectors, and then map the correlations from sectors to activities (CNAE 1.0) and
then to the HS-8 product level of the Mercosur nomenclature (NCM). For each
firm and year, we calculate the correlation between each product exported in t
and the new product exported in t+1. Then, we take the maximum correlation as
the measure of relatedness. If one of the products exported in t is in the same
HS-4 digits sector than the new product introduced the correlation is one and the
new product is highly related to existing exports.
Correlation based on the input use of the input-output Leontieff matrix in 2005.
Same as above, but using the Leontieff input requirements matrix.
Correlation based on (Hidalgo, Klinger, Barabasi and Hausmann 2007) product
space. The authors develop a methodology where SITC4 sectors are related in
18


terms of co-occurrences defined by the conditional probability that any given pair
of SITC4 products are exported by countries in the world. 7 We map these
correlations to HS-4 sectors and replace the correlation to unity when two
products belong to the same HS-4. Again, we use the maximum correlation
between all the products exported in t and the new product in t+1.
Minimum difference between the existing and new product at HS-4 level. Thus, a
difference of zero when one of the products is exported in t is in the same HS-4
sector than then new product introduced in t+1.
Minimum difference between the existing and new product at HS-2 chapter. As
above but looking at the HS-2 chapter for each product.
Figure 2 shows the probability distribution function for most of these relatedness
measures. The upper panels (a) and (b) plot the distribution function for all firm/year
where there was an introduction of a new product defined by classification
methodology 1. Panel (a) shows the different correlation measures and (b) the HS-2
difference. As expected most new products in t+1 are very related to existing exports
in t. Concretely, 75% of new exports are in the same HS-2 chapter than existing
exports in t, and 51% in the same HS-4 sector. Alternatively, for those cases where
we have available correlation measures, only 16% of cases show correlations of
input-output measures different from one (24% for the hidalgo-Hausmann measure).
Clearly most export diversification occurs in closely related products.
Panels (c) and (d) focus on those cases that are not completely related and do not
belong to same or very similar HS-4 sectors. Regarding the correlation measures,
the results indicate that once the firm introduce a new product for exporting in a
different sector from their core exports, the new product has a very low level of
relatedness. On the other hand, the Hidalgo-Hausmann correlation index, suggests
that most of these new products are mildly related with mean correlation coefficients
below 0.5.
These results tend to confirm the resource based approach to firm diversification,
where new exported products are similar related to existing core activities. The
resulting interesting question is what are the processes through which diversification
to more unrelated products occur.
A final characterisation of relatedness is to analyse whether diversification to related
products is more or less likely in specific type of products. Appendix 2 tabulates the
HS-2 chapter of each new product for each relatedness measure. We differentiate
between the number of new products at HS-2 that are totally related to the existing
export basket (correlations equal to one or HS-2 difference equal to zero) and those
that are somehow more distant. We have some agriculture new products due to the
fact that some firms are multi-sector. Focusing on those sectors with at least 10 new
products, we identify less related new HS2 chapters, as those sectors with lower
correlation than average and higher distance than average. We focus only on the
correlation of the usage I-O matrix, the Hidalgo-Hausmann correlation and HS-2
distance, since the other two measures are highly correlated.
7
A country is considered to export a given product if it has a revealed comparative advantage
larger than one.
19
Figure 2 Distribution of relatedness measures
(a) Relatedness measures distributions –
all firms with one new product
(b) Distribution HS-2 distances - all firms
with one new product
Probability distribution function - Relatedness measures
.15
Kernel density estimate
0
2
.05
4
Density
.1
6
8
Firms that introduce one new product for exporting
.5
1
0
0
Correlation Input usage matrix
Correlation Hidalgo-Hausmann
Correlation leontieff matrix
0
5
10
15
(mean) hs2dif
20
25
kernel = epanechnikov, bandwidth = 1.0010
Source: Author own elaboration from SECEX and IBGE
(c) Relatedness measures distributions –
products with different HS-4 only
(d) Distribution HS-2 distances - products
with different HS-4 only
Probability distribution function - Relatedness measures
.04
Kernel density estimate
0
.5
Correlation Input usage matrix
Correlation Hidalgo-Hausmann
.02
1
Correlation leontieff matrix
0
0
.01
2
4
Density
6
8
.03
10
Firms that introduce one new product for exporting with different HS4
0
10
20
30
(mean) hs2dif
kernel = epanechnikov, bandwidth = 2.7105
Source: Author own elaboration from SECEX and IBGE
20
40
50
Concretely the more distant HS-2 chapters are: 10 Cereals, 16 Preparations of meat
or Fish, 18 Cocoa and cocoa preparations, 25 Salt, sulphur, earth cement.., 40
Rubber and rubber articles, 50 Silk, yarns,.., 53 Vegetable textile fibbers, 65
Headgear and parts, 99 Other products services. Clearly, these most distant HS-2
chapters tend to be concentrated in non-manufacturing sectors.
Sophistication
One additional question regarding firm diversification is whether the path of
diversification leads to exporting more sophisticated products. While firms will
prioritise profitability of new activities for the given set of capabilities they have, it is
important to analyse whether these diversification paths are conducive to products
with larger value added or technological content.
The definition of sophistication is clearly problematic, since it can be defined along
several dimensions: quality, value added, technological content or conducive to
higher country growth. We focus in two dimensions of sophistication:
 PRODY – We use the measure of sophistication introduced by Hausmann et
al., 2007 and Lall et al. (2006). Using BACI HS-6 data from CEPII we
calculate for each product and year from 2001 to 2007, the measure defined
in (2). This is to give each product the GDP per capita weighted by the export
share of each country i for each product k in relation to the sum of exports
shares for that product and year.
PRODYkt  i
xikt / X kt
GDPcapit
i xikt / X kt
(2)
Once PRODY is calculated we re-scale the measure as the ratio with the mean
PRODY on that specific year. Then we use the ratio to compare existing exports
in t with the sophistication measure of the new exported product in t+1. WE
calculate the change in sophistication ratio from the most sophisticated product
in the export basket to the new product. This sophistication change is zero when
the new product introduced has the same sophistication level or it is within the
same HS-6 code.
 OECD classification – we use the technological content sophistication index
from the OECD. This classification, groups products according to the
following rankings: 1 – not industrial products, 2-low technological intensity,
3-low/medium intensity, 4-medium/high intensity and 5- high technological
intensity.
Once we have grouped existing exports in these groups, we use the existing
highest technological group and calculate the difference with the technological
group of the new exported product. We set the difference to zero, when the new
product is within the same HS-6 digit group than an existing product in t.
The results indicate that 64% of diversification cases occur in the existing
technological intensity level, 31% of new exported products imply diversification
towards a lower technological intensity, and only 5% of cases imply some
21
technological upgrading. When using the Hausmann et al. (2007) methodology, the
results suggest that in 19% of cases diversification is carried out in the same level of
sophistication, in 61% diversification is to lower sophistication products, and only in
20% of cases there is diversification upgrading.
Figure 3 Distribution of change in sophistication in product diversification
0
.5
Density
1
1.5
Kernel density estimate
0
5
(mean) sophis_diver
10
kernel = epanechnikov, bandwidth = 0.0503
Summing up, prima facie, firm export diversification is not such a rare event. Firms
tend to diversify exports with relative frequency, and this diversification is quite even
across HS-2 chapters. In addition, a significant number of firms introduce more than
one new product at a time. When looking at the path of diversification, the analysis
suggests that most new products are related or within the boundaries of existing
capabilities of the firm. Furthermore, more diversification occurs in products of
similar or lower sophistication or technological intensity.
5. The determinants of firm diversification in Brazil
In order to determine the main elements explaining firm export diversification we
estimate equation (1). We use sector dummies based on the CNAE classification of
national economic activities to control for sector specific unobserved factors and firm
random effects. Table 8 shows the Probit random effects and Table 9 shows a Probit
with Heckman sample selection for the probability of exporting.
22
The results show that larger and more productive firms are more likely to diversify.
Firms with larger unit values wis-a-vis other firms in the sector diversify less, maybe
as a result of higher existing profitability. Firms more concentrated in a few core
activities tend to diversify less, while firms with a larger number of products in their
production bundle diversify more. Regarding innovation, firms that engage in product
innovation are more likely to diversify, but there is no statistical significant level
regarding process innovation or the level of skills. Firms that do not engage in any
innovation process are less likely to diversify. Finally the variables that proxy the
origin of relevant flows for innovation or links with MNES and value chains are not
statistically significant.
Table 8 Probit random effects estimates
Ln(L)
new1_1
new2_1
new1_2
new2_2
new1_3
new2_3
0.0285
0.0579**
0.034
0.0418*
0.0656**
0.0775***
0.0241
0.0259
0.0247
0.0252
0.0266
0.0247
0.0695**
0.0587*
0.0679**
0.0601*
0.0309
0.0325
0.0297
0.0314
-0.0351*
-0.0458**
-0.0474**
-0.0346*
0.0188
0.0207
0.0188
0.0188
0.0205
0.0204
0.1401
0.2604*
0.1857
0.1752
0.3355**
0.3706**
0.1421
0.1475
0.138
0.1435
0.1429
0.1438
-0.2088***
-0.2119***
-0.2127***
-0.2195***
0.0658
0.0689
0.0643
0.0672
0.0055
0.0069
0.0056
0.0067
0.0048
0.005
0.0047
0.0049
n_prod
0.0209***
0.0223***
0.021***
0.0208***
0.0013
0.0014
0.0014
0.0013
prod_inno
0.1769***
0.2111***
0.1618***
0.1949***
0.0508
0.0533
0.0499
0.0521
0.043
0.0278
0.0199
0.0007
lnTFPLP
uv_ratio
prod_share
herf2 (production)
distance_2d
process_inno
-0.0554***
0.0225***
0.0222***
0.0014
0.0014
0.0492
0.0515
0.048
0.0502
-1.6875
-1.605
-1.9111*
-1.8422
1.0588
1.1044
1.0778
1.1234
group_dep
0.0779
0.0722
0.1051
0.0877
0.0902
client_dep
-0.0072
0.0415
0.1075
0.1105
0.0339
skills_rd
foreign_cap
iso
-0.055***
0.0853
0.0878
0.0941
0.086
0.0881
0.0884
0.0885
-0.0037
-0.0042
0.0501
0.051
0.1052
0.1072
0.1079
0.1074
0.1283*
0.039
0.0424
0.1298*
0.1302*
0.0689
0.0715
0.0673
0.0706
0.0698
0.071
0.0121
0.0261
0.0176
0.0176
0.0218
0.0271
0.0492
0.0514
0.0482
0.0491
0.0503
0.05
0.0739**
0.0564*
lnTFPFE
herf
23
0.0332
0.0333
-0.3264***
-0.3679***
0.0916
0.0933
distance_3d
0.0004
no_rd
Skills_proxy
_cons
0.0005
0.0005
0.0005
-0.1399**
-0.1134**
0.0526
0.0536
-0.0103
0.0019
0.0235
0.0219
-9.0127
-9.4613
-0.9603
-8.2068
-1.2298
-0.5352
3.60E+03
4.10E+03
1.2141
3.50E+03
1.2424
1.2241
6766
6766
6766
6766
6766
6766
*** significant at 1% confidence level, ** significant at 5% confidence level and * significant at
10% confidence. Year and CNAE 2 digits sector dummies coefficients not reported
As suggested above, using only the sample of exporters may lead to sample
selection bias, since the determinants of a firm exporting or not may be correlated
with the decision of diversifying. Table 9 shows the results correcting for the potential
sample selection bias. We first estimate a selection equation for exporters, where the
probability of exporting depends on size, productivity, unit value ratio, product share
and firm wages, and add the inverse mills ration in the second stage. The inverse
mills ratio is statistically significant, indicating that uncorrelated factors may be
explaining selection to export and diversification. Interestingly, the results are very
similar to Table 8, with the main exception than size and productivity are no longer
statistically significant determinants of export diversification.
6. Conclusions
24
Table 9 Probit with heckman selection
Variable
Ln(L)
lnTFPLP
uv_ratio
prod_share
herf2
distance_2d
n_prod
prod_inno
process_inno
skills_rd
PROBI~1
PROBI~2
PROBI~3
PROBI~4
PROBI~5
PROBI~6
PROBI~7
PROBI~9
PROB~10
PROB~11
PROB~12
-0.0376
-0.0398
-0.0247
-0.0275
-0.0463
-0.0515
-0.0331
-0.0425
-0.0187
-0.0288
-0.0423
-0.0601
0.0322
0.0333
0.0309
0.032
0.035
0.0362
0.0347
0.0358
0.0332
0.0343
0.0379
0.0393
-0.0131
-0.0513
-0.0105
-0.0495
-0.0113
-0.0593
-0.0055
-0.059
0.0058
-0.0482
0.0244
-0.0443
0.0389
0.0402
0.0375
0.0386
0.041
0.042
0.0396
0.0408
0.0383
0.0393
0.0418
0.0429
-0.0304**
-0.0393**
-0.0421***
-0.0487***
-0.0347**
-0.0326**
-0.03*
-0.0392**
-0.0413***
-0.0482***
-0.0323**
-0.0308*
0.0151
0.0163
0.0152
0.0162
0.0154
0.0158
0.0151
0.0163
0.0152
0.0161
0.0155
0.0158
-0.0468
-0.0114
0.0021
0.0491
-0.0027
-0.0261
-0.0024
0.0242
0.0562
0.0932
0.0701
0.0293
0.1262
0.1288
0.1225
0.125
0.1293
0.1325
0.1259
0.1284
0.1224
0.1247
0.1293
0.1323
-0.1785***
-0.1775***
-0.1841***
-0.1835***
-0.1627***
-0.1538***
-0.2815***
-0.2869***
-0.3074***
-0.3113***
-0.276***
-0.267***
0.0517
0.053
0.0504
0.0517
0.0511
0.0522
0.0714
0.073
0.0699
0.0714
0.0714
0.0727
0.0045
0.0057
0.0047
0.0056
0.001
0.0026
0.0003
0.0004
0.0003
0.0004
-0.0001
0.0001
0.0038
0.0038
0.0037
0.0038
0.0038
0.0039
0.0004
0.0004
0.0004
0.0004
0.0004
0.0004
0.0181***
0.01924***
0.01854***
0.01994***
0.0324***
0.03454***
0.01814***
0.01924***
0.01834***
0.01974***
0.0324***
0.03464***
0.0009
0.0009
0.0009
0.001
0.0013
0.0013
0.0009
0.0009
0.0009
0.0009
0.0013
0.0013
0.1514***
0.1757****
0.13544***
0.15884***
0.09294***
0.11524***
0.0407
0.0418
0.0398
0.0408
0.0403
0.0412
0.0418
0.026
0.0189
0.0009
0.0446
0.0203
0.0397
0.0409
0.0388
0.0399
0.0391
0.04
-1.229
-1.068
-1.281
-1.11
0.4311
0.7414
0.8547
0.8727
0.8544
0.8702
0.7542
0.7513
-0.12634***
-0.12164***
-0.10064**
-0.09494**
-0.14684***
-0.13254***
0.0422
0.0435
0.0409
0.0422
0.0408
0.0419
-0.0062
0.0079
-0.0106
0.0046
-0.0282
-0.0127
0.0182
0.0176
0.0184
0.0176
0.0195
0.0192
no_rd
Skills_proxy
group_dep
PROBI~8
0.0756
0.0751
0.1056
0.099
0.0589
0.0422
0.08
0.0798
0.1089
0.1024
0.0545
0.0382
0.0705
0.0711
0.0693
0.0699
0.0751
0.0757
0.0708
0.0714
0.0695
0.0702
0.0755
0.0761
25
client_dep
0.0028
0.0471
-0.0033
0.0443
-0.0613
-0.0093
0.0048
0.0521
-0.0046
0.0453
-0.068
-0.0105
0.0872
0.088
0.0854
0.0862
0.0898
0.0906
0.087
0.0878
0.0852
0.086
0.0896
0.0904
0.0145
0.0824
0.0215
0.0846
0.0281
0.1086**
0.0177
0.075
0.028
0.0802
0.0524
0.121***
0.0538
0.0545
0.0525
0.0532
0.0548
0.0553
0.055
0.0556
0.0537
0.0543
0.0563
0.0568
0.0057
0.0181
0.0086
0.012
0.0297
0.0325
0.0113
0.0256
0.0127
0.0176
0.0322
0.0375
0.0398
0.0408
0.0389
0.0399
0.0396
0.0404
0.0396
0.0406
0.0388
0.0395
0.0403
-0.3718**
-0.52484***
-0.32344**
-0.49144***
-0.34024*
-0.56854***
-0.34634**
-0.52294***
-0.27684*
0.0397
0.46734***
-0.2492
-0.5206***
0.1606
0.1682
0.1513
0.1584
0.1796
0.1886
0.1562
0.1636
0.1474
0.1543
0.1748
0.1844
-4.8384
-4.1923
1.0039
1.7307
-8.6043
-7.8987
-4.9115
-4.576
0.8979
1.3008
-8.7889
-8.4747
0
0
1.2573
1.2797
0
0
0
0
1.0762
1.0887
0
0
N
6,800
6,700
6,800
6,700
6,800
6,800
6,800
6,700
6,800
6,700
6,800
6,800
r2_p
0.164
0.1861
0.1616
0.184
0.2436
0.267
0.1635
0.1852
0.1615
0.1834
0.2447
0.2673
foreign_cap
iso
imr
_cons
*** significant at 1% confidence level, ** significant at 5% confidence level and * significant at 10% confidence level
26
7. References
Aitken, B., G. H. Hanson, and A. E. Harrison, 1997. Spillovers, foreign investment,
and export behavior, Journal of International Economics 43, 103-132.
Al-Marhubi, F., 2000. Export Diversification and Growth: An Empirical Investigation,
Applied Economics Letters 7, 559-562.
Albornoz, F., H. Calvo-Pardo, G. Corcos, and O. Emanue, 2010. Sequential
Exporting(Centre for Economic Performance, LSE).
Amurgo-Pacheco, A., and M. D. Pierola, 2008. Patterns of export diversification in
developing countries : intensive and extensive margins(The World Bank).
Arkolakis, C., and M.-A. Muendler, 2009. The Extensive Margin of Exporting Goods:
A Firm-level Analysis.
Aw, B. Y., M. J. Roberts, and D. Y. Xu, 2009. R&D Investment, Exporting, and
Productivity Dynamics(National Bureau of Economic Research, Inc).
Baldwin, R., and J. Harrigan, 2007. Zeros, Quality and Space: Trade Theory and
Trade Evidence(National Bureau of Economic Research, Inc).
Becker, S. O., and P. Egger, 2007. Endogenous Product Versus Process Innovation
and a Firm's Propensity to Export, SSRN eLibrary.
Bekes, G., J. Kleinert, and F. Toubal, 2009. Spillovers from Multinationals to
Heterogeneous Domestic Firms: Evidence from Hungary, World Economy,
Vol. 32, Issue 10, pp. 1408-1433, October 2009.
Bernard, A. B., J. B. Jensen, S. J. Redding, and P. K. Schott, 2007. Firms in
International Trade, Journal of Economic Perspectives 21, 105-130.
Bernard, A. B., S. J. Redding, and P. K. Schott, 2010. Multiple-Product Firms and
Product Switching, American Economic Review 100, 70-97.
Besedes, T., and T. Prusa, 2006. Ins, outs, and the duration of trade, Canadian
Journal of Economics 39, 266-295.
Besedes, T., and T. J. Prusa, 2006. Product differentiation and duration of US import
trade, Journal of International Economics 70, 339-358.
Beveren, I. V., and H. Vandenbussche, 2009. Product and Process Innovation and
the Decision to Export: Firm-level Evidence for Belgium(LICOS - Centre for
Institutions and Economic Performance, K.U.Leuven).
Broda, C., and D. E. Weinstein, 2010. Product Creation and Destruction: Evidence
and Price Implications, American Economic Review 100, 691–723.
Cassiman, B., and E. Martinez-Ros, 2007. Product Innovation and Exports. Evidence
from Spanish Manufacturing(Madrid).
Costantini, J., and M. Melitz, 2007. The Dynamics of Firm-Level Adjustment to Trade
Liberalization.
Damijan, J. P., C. Kostevc, and S. Polanec, 2010. From Innovation to Exporting or
Vice Versa?, The World Economy 33, 374-398.
Dennis, A., and B. Shepherd, 2007. Trade costs, barriers to entry, and export
diversification in developing countries(The World Bank).
Egan, M. L., and A. Mody, 1992. Buyer-seller links in export development, World
Development 20, 321-334.
Freund, C., and M. D. Pierola, 2009. Export Entrepreneurs: Evidence from Peru.
Greenaway, D., N. Sousa, and K. Wakelin, 2004. Do domestic firms learn to export
from multinationals?, European Journal of Political Economy 20, 1027-1043.
Haddad, M. E., J. J. Lim, and C. Saborowski, 2009. Trade Openness Reduces
Growth Volatility When Countries are Well Diversied.
27
Hausmann, R., and D. Rodrik, 2003. Economic development as self-discovery,
Journal of Development Economics 72, 603-633.
Herzer, D., and F. Nowak-Lehnmann, 2006. What does export diversification do for
growth? An econometric analysis, Applied Economics 38, 1825-1838.
Hesse, H., 2009. Export Diversication and Economic Growth(The World bank,
Washington, DC).
Hidalgo, C. A., and R. Hausmann, 2009. The Building Blocks of Economic
Complexity(arXiv.org).
Hidalgo, C. A., B. Klinger, A. L. Barabasi, and R. Hausmann, 2007. The Product
Space Conditions the Development of Nations, Science 317.
Hobday, M., and H. Rush, 2007. Upgrading the technological capabilities of foreign
transnational subsidiaries in developing countries: The case of electronics in
Thailand, Research Policy 36, 1335-1356.
Iacovone, L., and B. S. Javorcik, 2010. Multi-Product Exporters: Product Churning,
Uncertainty and Export Discoveries, The Economic Journal, Vol. 120, No.
544, pp. 481-499, May 2010 120, 481-499.
Javorcik, B. S., and M. Spatareanu, 2009. Tough Love: Do Czech Suppliers Learn
from their Relationships with Multinationals?, Scandinavian Journal of
Economics 111, 811-833.
Klinger, B., 2007. Uncertainty in the search for new exports.
Klinger, B., and D. Lederman, 2004. Discovery and Development: An Empirical
Exploration of New Products, SSRN eLibrary.
Lederman, D., 2009. The business of product innovation : international empirical
evidence(The World Bank).
Lien, L. B., and P. G. Klein, 2009. Using Competition to Measure Relatedness,
Journal of Management 35, 1078-1107.
Lien, L. B., and P. G. Klein, 2010. Why There? Decomposing the Choice of Target
Industry, SSRN eLibrary.
López, R. A., 2009. Do Firms Increase Productivity in Order to Become Exporters?,
Oxford Bulletin of Economics and Statistics 71, 621-642.
Martincus, C. V., and J. carballo, 2009. Survival of new exporters in developing
countries: does it matter how they diversify?
Mayer, T., and G. Ottaviano, 2008. The Happy Few: The Internationalisation of
European Firms, Intereconomics 43, 135-148.
Melitz, M. J., 2003. The Impact of Trade on Intra-Industry Reallocations and
Aggregate Industry Productivity, Econometrica 71, 1695-1725.
Neffke, F., and M. S. Henning, 2008. Revealed Relatedness: Mapping Industry
Space(Utrecht University, Section of Economic Geography).
Ruhl, K. J., and J. L. Willis, 2009. New Exporter Dynamics.
Swenson, D. L., 2008. Multinationals and the Creation of Chinese Trade Linkages,
Canadian Journal of Economics/Revue canadienne d'économique, Vol. 41,
Issue 2, pp. 596-618, May/mai 2008.
Volker, N., 2009. Die another day: duration in German import trade, Review of World
Economics (Weltwirtschaftliches Archiv) 145, 133-154.
28
Appendix 1 Distribution of new products and exporters by HS2 chapter – percentage of total product lines for each sector
10.00%
9.00%
8.00%
7.00%
6.00%
5.00%
4.00%
3.00%
2.00%
1.00%
0.00%
1 2 3 4 5 6 7 8 9 10111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576787980818283848586878889909192939495969799
29
Appendix 2 Distribution of relatedness measures across HS-2 sectors
correlation use
matrix
hs2
02
03
04
05
07
08
09
10
11
12
13
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
46
47
48
49
50
51
52
53
54
55
New
products
43
21
20
34
6
19
11
19
9
14
23
37
20
54
17
41
36
39
53
39
4
85
10
41
56
111
84
29
94
45
62
36
3
7
131
476
166
103
43
6
488
2
2
149
83
2
4
63
2
50
21
cor<1
1
1
1
1
1
3
3
27
6
6
1
5
8
17
3
1
20
11
13
9
1
1
26
75
12
12
13
22
27
1
2
cor==1
43
19
16
30
5
18
6
15
9
5
20
31
18
48
16
40
35
34
23
23
4
45
7
25
42
83
75
27
63
33
42
20
6
88
331
119
88
23
5
379
2
1
103
35
2
3
63
2
45
17
correlation
Leontieff matrix
cor<1
1
1
1
1
1
3
3
27
6
6
1
5
8
17
3
1
20
11
13
9
1
1
26
75
12
12
13
22
27
1
2
cor==1
43
19
16
30
5
18
6
15
9
5
20
31
18
48
16
40
35
34
23
23
4
45
7
25
42
83
75
27
63
33
42
20
6
88
331
119
88
23
5
379
2
1
103
35
2
3
63
2
45
17
30
correlation
hidalgoHausmann
cor<1
2
3
6
2
5
1
4
5
4
7
4
3
4
7
10
2
3
4
3
4
37
3
10
18
4
13
10
2
82
2
9
5
54
1
35
17
2
4
11
5
cor==1
19
17
6
19
2
6
4
10
1
2
18
9
5
32
6
27
21
20
12
8
1
34
2
25
18
39
63
14
40
22
33
14
5
66
197
57
63
11
2
328
2
78
18
1
42
1
28
9
HS2 sector
distance
different
10
10
8
2
10
6
8
4
9
5
18
15
17
11
7
6
14
32
26
39
5
11
17
33
12
4
30
13
23
21
3
1
39
135
72
5
28
1
27
2
46
49
1
4
12
1
18
6
same
33
21
10
26
4
9
5
11
5
5
18
19
5
37
6
34
30
25
21
13
4
46
5
30
39
78
72
25
64
32
39
15
6
92
341
94
98
15
5
461
2
103
34
1
51
1
32
15
HS4 sector
distance
different
24
4
14
15
4
13
7
9
8
12
5
28
15
22
11
14
15
19
41
31
3
51
8
16
38
72
21
15
54
23
29
22
3
2
65
279
109
40
32
4
160
2
71
65
1
4
21
1
22
12
same
19
17
6
19
2
6
4
10
1
2
18
9
5
32
6
27
21
20
12
8
1
34
2
25
18
39
63
14
40
22
33
14
5
66
197
57
63
11
2
328
2
78
18
1
42
1
28
9
56
57
58
59
60
61
62
63
64
65
66
68
69
70
71
72
73
74
75
76
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
99
Total
44
7
22
32
29
153
109
39
192
12
1
142
38
61
81
69
355
42
1
72
3
4
6
6
90
77
982
459
13
191
2
2
237
9
12
4
235
24
26
1
14
6911
2
3
3
5
4
17
7
18
16
10
51
10
16
2
1
4
13
18
98
71
8
40
1
2
44
3
1
1
21
3
11
834
37
7
19
17
26
143
104
26
176
7
1
118
30
29
55
44
263
26
1
47
1
4
5
2
65
44
821
348
5
151
1
184
6
11
3
194
19
15
5
5292
2
3
3
5
4
17
7
18
16
10
51
10
16
2
1
4
13
18
98
71
8
40
1
2
44
3
1
1
21
3
11
834
37
7
19
17
26
143
104
26
176
7
1
118
30
29
55
44
263
26
1
47
1
4
5
2
65
44
821
348
5
151
1
184
6
11
3
194
19
15
5
5292
31
12
2
4
7
2
18
13
7
16
4
9
6
12
77
44
17
145
1
11
5
19
4
9
64
3
99
13
14
35
21
175
14
9
2
21
1
4
1
1
6
25
135
65
7
24
10
872
1
37
22
577
237
5
129
1
128
2
5
2
153
18
7
5
3514
19
1
11
24
4
22
18
19
12
6
1
31
13
32
19
26
106
19
33
2
5
5
31
46
108
68
8
45
1
2
51
4
1
1
39
4
16
1
9
1739
25
6
11
8
25
131
91
20
180
6
111
25
29
62
43
249
23
1
39
1
4
1
1
59
31
874
391
5
146
1
186
5
11
3
196
20
10
5
5172
28
3
13
26
17
76
65
22
47
11
1
43
25
47
46
48
180
28
1
51
2
6
5
53
55
405
222
8
62
1
2
109
7
7
2
82
6
19
1
9
3397
16
4
9
6
12
77
44
17
145
1
99
13
14
35
21
175
14
21
1
4
1
37
22
577
237
5
129
1
128
2
5
2
153
18
7
5
3514