The Region as a Platform to the World.DOC

More precisely, do increases in market share within the regional market allow exporters to improve their export performance in the rest of the world through information spillovers?. The

The Region as a Platform to the World

Abstract: An argument often advanced, when discussing the benefits of regional

integration in Latin America, is that the formation of a larger market may serve as aplatform for (potential) exporters to world markets Not only economies of scale andlearning by doing, when facing a larger demand may allow firms to become morecompetitive, but information about customs procedures, required design of exportproducts and firm reputation is generated through exports to the regional market Theknowledge acquired in the regional market can then be used to penetrate more distantmarkets outside the regional agreement The evidence for the Southern Cone CommonMarket (Mercosur) is mixed Results suggest that the region was an important “class-room” for Argentina and Brazil’s exporters, but not for the small members of Mercosur(Uruguay and Paraguay)

JEL: F12, F13, F14, F15

Keywords: Exports, Regional Integration, Information Spillovers, Mercosur.

 We are grateful to Kala Krishna, Jaime de Melo and participants at the “Leitner Conference on Political and Economic Aspects of Regional Integration” at Yale University, April 7-8, 2000 for very helpful comments The views expressed here are those of the authors and not necessarily those of the institutions to which they are affiliated.

 Development Research Group (DECRG), The World Bank, 1818 H Street, NW, Washington DC 20433, e-mail: anicita@worldbank.org.

 DECRG, The World Bank, and CEPR, London, e-mail: molarreaga@worldbank.org.

 DECRG, The World Bank and University of Maryland, e-mail: isoloaga@worldbank.org.

“We cannot continue alone Soon we will have to open up to the whole Americas and our traders will not have gained the necessary experience to

A recent literature looks at the contribution of social and ethnic networks to overcomethese informational barriers by helping to match buyers and sellers across internationalborders and serving as a deterrent for opportunistic behaviour (Rauch and Casella, 1998,Rauch, 1999 or Rauch and Trinidade, 1999)

In addition, repeated interactions with foreign importers will help exporters overcomethese barriers More importantly, a successful business relationship may generateimportant information spillovers beyond the two trading partners On the demand side,importers may use other importers, who have had direct experiences with the potentialsuppliers, as a source of information on the performance of potential exporters (WorldBank, 1989) On the supply side, a successful exporter may generate demonstrationeffects for other firms, which become aware of potential opportunities in foreign markets

These information spillovers may not be limited to national borders Potential importersmay look at the performance of a potential future supplier in a third country to decidefuture transactions Also export activities with one particular country generate a betterunderstanding of consumer tastes and customs administrations in similar countries (Nicitaand Olarreaga, 2000) International social or ethnic networks may also help transmitinformation about successful business oportunities to third countries (Rauch, 1999)

One of the repeated justifications for regional integration efforts, in the WesternHemisphere in particular, is based on these types of arguments Exporters may obtainsufficient experience in the regional market, which could then be used as a platform forinternational export expansion (Devlin and French-Davis, 1999) The region can serve as

a “class-room” for potential exporters, where they learn “how to export” and createreputation as reliable suppliers.1 From a policy perspective, in the presence of exportinformation spillovers, regional trade agreements may create sufficiently large gains tocompensate for possible trade-diverting effects

The objective of this paper is to try to find some evidence of the presence of information spillovers” in the case of Mercosur member countries (Argentina, Brazil,

“export-Paraguay and Uruguay) More precisely, do increases in market share within the regional market allow exporters to improve their export performance in the rest of the world through information spillovers?

In the empirical section, to capture information flows among countries, we use the value(in real terms) of total trade in newspapers and periodicals (SITC 8922) between

countries These includes trade oriented papers such as the Journal of Commerce (US), Export Channel (US), Made for export (Europe), Asian Channel (Hong Kong), Gazeta Mercantil Latinoamericana (in Mercosur countries) but also journals, magazines and

1 Another important branch of the literature has been interested on whether export activities allow firms to become more productive Contrary to popular presumptions, little evidence has been found for developed (Bernard and Jensen, 1999) and developing (Clerides, Lach and Tybout, 1998) countries This literature concludes that highly productive firms self-select into export activities, which explains that exporting firms are more productive than non-exporting firms Note that these do not rule out the presence of spillover

effects into other firms (or other markets), as found by Clerides et al (1998)

periodicals It has several advantages over the more traditionally used measure ofdistance.2 First, it allows for size effects The amount of information between Mercosurcountries and Mexico is larger than the amount of information between Mercosurcountries and Honduras, though the distance between them is very similar Second, itallows to control for cultural factors such as language and colonial links Informationflows between Mercosur countries and Spain are much larger than between Mercosurcountries and South Africa, though the latter is closer to Mercosur than Spain To capturethe export-information flows, we weigh bilateral newspaper trade by the market share atthe tariff line that Mercosur exporters have in the regional market, which serves as anindicator of their performance in a particular product in the regional market

Figure 1 illustrates how export information spillovers at period t-1 (the kink and dash

arrows in figure 1) affect export of a Mercosur country to third markets (the partner) at

period t (the large horizontal arrow) The export performance of a Mercosur country (call

it “0”) in other Mercosur members markets or the ROW at period t-1 will be transmitted

to a third market through information flows (e.g., trade in newspapers) and consumers inthird markets will learn about the export performance of country 0 exporters At the sametime exporters in country “0” will learn more about consumer tastes in the third marketwhen exporting to countries with which the third market has significant informationflows (the dash arrows) These information flows (the kink and dash arrows) will in turn

positively affect export flows of country 0 to the third market at period t

To anticipate results, we found that Argentina and Brazil’s (the large members ofMercosur) exporters benefit from information spillovers to penetrate third markets bothfrom within Mercosur and from outside Mercosur export performance However, exportinformation spillovers appeared to be larger when the original export market is Mercosur.Both spillovers from Mercosur and from the rest-of-the-world (ROW) increase after

2 In another context, Portes and Rey (1999) measure information flows using the amount of bilateral telephone calls between two countries This also corrects for the drawbacks of distance However, in our context, we believe that trade in newspapers is more likely to generate spillovers than private phone calls Rauch and Trindade (1999) use the share of common ethnic population to capture these information flows, but focus only on chinese ethnic networks as a source of information flows Given our problem we would need a matrix of all bilateral ethnic networks in the world, which is not available to our knowledge Moreover, trade flows of newspapers are likely to be correlated with bilateral shares of foreign population.

1991 On the other hand, for Paraguay and Uruguay (the small members of Mercosur)exporters, information spillovers were larger when originating in ROW countries, and thecreation of Mercosur in 1991 tend to reduce rather than increase the effect of exportinformation spillovers on the export performance of Paraguay and Uruguay to ROWcountries Thus, we can only identify a “platform” effect for the larger members ofMercosur

The rest of the paper is structured as follows Section 2 describes the evolution of bothintra and extra-Mercosur exports from 1980 to 1998 Section 3 focuses on the empiricalmodel, whereas section 4 presents the results Section 5 concludes

2 Evolution of intra and extra-Mercosur exports in the 1990s

The Treaty of Asunción led to the creation of a regional market among Argentina, Brazil,Paraguay and Uruguay, called Mercosur (The Southern Cone Common Market) in 1991

By January 1995 a quasi-customs union was in place, with 75 percent of tariff linessubject to a common external tariff and 97 percent of tariff lines set at zero among tradingpartners.3

Intra-Mercosur exports have been growing much more quickly than extra-Mercosurexports (26 and 7 percent) Note that this is not inconsistent with our story as informationspillovers are certainly going to be stronger within the region, and is probably just thereflection of tariff preferences On the other hand, we observed an acceleration ofMercosur exports to the ROW in the late 90s The average growth of Mercosur exports tothe rest of the world in the early 1990s was 0.9 percent (1990-1993) to be compared with

a 9.1 percent annual increase during the period 1994-1998 This suggests an acceleration

of the rate of integration of Mercosur exporters into world markets consistent with the

“region as a platform” hypothesis However, before rushing into conclusions, one needs

3 The exempted sectors being computer products, telecommunication equipment, automobiles and sugar for which special regimes have been negotiated and convergence to a full customs union should be achieve by 2005.

to control for determinants of export flows, such as size of bilateral markets, demandconditions, cultural links, etc A recent study of the effects of regionalism on aggregateexports to the ROW using a gravity-type approach concludes that the creation ofMercosur had no effects on exports to third countries (Soloaga and Winters, 1999)

If instead of using aggregate exports, we focus on exports of industrial products thepicture is also different Moreover, keeping in mind the objective of the paper,information spillovers are likely to be more important for products in which Mercosurmembers do not have a “natural” comparative advantage and have not already acquired asignificant reputation and know-how in world markets (World Bank 1989).4 Table 1provides the average export growth of industrial products of all Mercosur members toother Mercosur members and to the rest of the world for two periods: 1980-1991 (pre-Mercosur) and 1981-1998 (post-Mercosur) Intra-Mercosur export growth of industrialproducts has significantly increased and reaches an impressive 26 percent annual increasefor the period 1991-1998 (compared to an annual –3 percent for the period 1980-1991).The contrast with the evolution of industrial exports to the ROW is striking The growth

of Mercosur exports of industrial products to the ROW has declined from 5 percent in1980-1990 to 2 percent in 1991-1998 This evolution is also consistent across Mercosurcountries

The decline in export growth of industrial products to the ROW is consistent with thetariff preferences that are granted for within Mercosur trade, but puts some doubts on the

“Mercosur as a platform” hypothesis However, Mercosur exports to the ROW havecontinued to increase (with the exception of Uruguay) as shown in figure 2 Moreover, to

be able to capture “platform”-type effects it is necessary to go beyond aggregate data Inthe next section we develop an empirical model to try to identify at the product levelplatform-type effects linked to information spillovers

4 This is somewhat consistent with Rauch (1999) proposition that asymmetric information in international trade is more costly for differentiated products

The basic model draws on Nicita and Olarreaga (2000) They show within a simplemodel with linear demands, constant marginal transport and production costs, andinternationally segmented markets that in the presence of cross-country spillovers on the

demand or supply side, market equilibrium implies that exports of product p to country c

at time t are given by:

1 , 1 , 1

, ,

 is the “potential” size of country c market for product p at time t (this is measured

as total imports of country c of product p at time t); d is distance from country O to c

country c;  0is the constant marginal transport cost; s c,t 1is country O market share

of product p in total imports of country c in the previous period and 0captures “own

market” effects The idea is that the larger the market share of country O in country c total imports in period t-1, the larger are going to be exports of country O to country c in period t Thus, as in Farrell (1986), or Froot and Kemplerer (1989), past market shares

determine future demand (and in equilibrium supply) Own-market effects partly includeexport-information spillovers within the two bilateral trading partners, but may alsocapture hysteresis effects as in Baldwin (1988) or Baldwin and Krugman (1989)

The last term in (1) is the one we are interested in this paper It captures informationspillover effects across countries: N c, t 1 is a vector of newspaper trade between country c

and all other export markets of country O; S p, t 1 is a vector of country O market share in

all of its trading partners at period t-1, and  captures the importance of export

information spillover across countries on country O export performance

To capture the potential platform effect of intra-Mercosur trade on exports to the ROW,

we first take country O as being one of Mercosur members and country c as ROW

countries We further decompose the third term in (1) into two A first element that

captures the information spillovers that intra-Mercosur trade creates on exports to theROW and a second element that captures information spillovers among ROW markets.Equation (1) becomes:

Row t p

Row t Row M

t p

M t

M t c c

t c t

where S p M, t 1is a vector containing the market share of country O (i.e., a Mercosur

member) exports on other Mercosur members total imports; N t- M1 is a vector of other

Mercosur countries share in total trade of newspapers of country c;5 Row

t

N1 is a vector of

the share of newspaper trade between country c and ROW countries and S p Row, t 1is a vector

of country O export market share in each ROW countries.6

ROW through information spillovers regarding export performance within the Mercosur

countries on their exports to third countries

The estimation of information spillovers ( ) through equation (2) may be biased by theabsence of some other important variables, or that are related to comparative advantageaspects or other types of externalities that are absent in (2) To control for these, we addthree variables to the right hand side of (2) First, bilateral trade preference and cultural

links may affect our estimates and to capture this we introduce gravity, which is the total exports of the source country to country c (purged of product p exports to country c) It

can also be interpreted as capturing all the explanatory variables of a gravity equation foreach Mercosur member (including regional trade agreements with ROW countries) Itmay also be seen as across products own market effects Second comparative advantageaspects may also affect our measure of information spillovers (in some products our

5 The normalization in terms of shares allow us to interpret the information flow, as the probability of observing information flows among countries.

6 To proxy for information flows, we also used distance and total trade between partners instead of

newspaper trade in the empirical section Results were robust to the use of different information proxies.

source country may be a “natural exporter” and in others not) and to control for this we

introduce ca , which is defined as total exports to the rest of the world., denoted ca

(again purged) Third, we also control for possible within sector (and country)externalities by taking the share of bilateral exports at the industry level on total imports

of country c at the industry level and denote it industry

Finally, regarding the Mercosur “platform” effect, given that Mercosur was created only

in 1991 and that data is available from 1980 we introduced two “dummies” to measurethe marginal effect of the creation of Mercosur in 1991 on the information spillovereffects from Mercosur and from ROW countries These variables are constructed as

M t

Row t

p,1 T1,1991N1 S ,1

value 1 if t1991 and 0 otherwise.7

3.1 Data

Trade data is obtained from United Nations Comtrade Data Base for the period

1980-1998 at the 3-digit level of the SITC classification For each of the Mercosur exportingcountries we only use data on products that have been exported at least once during theperiod 1980-1998 to at least one market The ROW is composed of 54 countries thatrepresent on average over the period more than 85 percent of rest-of-the- world trade Alltrade data is adjusted to 1997 US dollars (units are thousands of 1997 US dollars) Thebilateral distance matrix is calculated using distance between capitals of differentcountries, provided by Volcano software For variable construction see the appendix

3.2 Econometric specification

The equation to be estimated is a stochastic version of equation (2) which includes thecontrol variables described above The non-existence of Mercosur exports in manyproducts across trading partners leads to a large presence of zeroes in the endogenousvariables To correct for this bias introduced by censoring we use a Two-Stage Tobittechnique (Maddala, 1983, p.221-222):

7 On a study of the dynamic effects of trade integration, Freund and McLaren show that the actual trade flows in Mercosur started changing 2 to 3 years before the agreement was signed These anticipation effects will not be captured by our variables.

t c

Row t c p

M t c t

c t

p t

c

Row t p

Row t Row M

t p

M t

M t c p c t c t c

t c p

t c t

c

t

c

Time

dig ca

gravity

S N S

N s

d x

x

x x

x

,

1 , 5 1 , 4 , 3 , 2 , 1

1 , 1 1

, 1 1

, ,

* ,

* ,

* ,

*

,

,

2

and

0 if ,

0 0 if ,

















































































where Time is a time dummy introduced to control for general trends in Mercosur

members exports to the ROW;  c,t c c,t is an error term containing a random country effect to control for other export market countries characteristics.8

The first step of the estimation procedure consist in estimating the probability density

function (pdf) of each observation and its probability of occurrence (i.e., being non-zero)

using a Probit estimator where the endogenous variable takes the value 1 if an export flow is observed an zero otherwise In the second stage, we weight each observation by

its probability of occurrence and include the pdf as an explanatory variable to obtain the

unconditional equation for x p c,t.9 An interesting feature of the two-stage procedure is that

in the first step of the estimation we will observe the effect of information flows spillovers across countries on the probability of observing a trade flow in the next period

Finally, given that each regression contains a large number of observations (more than 65 thousand), it will be difficult to perform a detailed analysis of influential observations To make it tractable, we exclude from the estimations any observations that when run over the whole sample had a standardised error above 2.5 This procedure never eliminated more that 5 percent of the observations on each tail of the error distribution

8 Fixed effect may yield inconsistent estimates when using the two-stage tobit method or in the presence of spatial effects (i.e., spillovers across countries) as discussed by Anselin (1988).

9 As suggested by Maddala (1983) and unlike Heckman (1976) we include the non-zero observations in the second stage.

4 Results

Table 2 reports results of the estimation of equation (3) for Argentina and Brazil using aTwo-Stage Tobit technique Table 3 reports results for the smaller members ofMercosur (Paraguay and Uruguay) In the case of small members results of thesecond-stage are obtained using a simple tobit The reason for this is that given thelarge presence of censored data for small members (Paraguay 97 percent andUruguay 89 percent), the two-stage approach yields inefficient results.10

In the case of Argentina, it appears that export information spillovers originating in eitherMercosur or ROW do not seem to significantly affect the probability of exporting to athird market Moreover, the creation of Mercosur has not led to a change in theseregimes Similarly, export information spillovers do not seem to affect the level ofexports to third countries in the case of Argentina until 1991 However, the creation ofMercosur has led to an increase in export-information spillovers originating within themarket The explanation of this change in regime could be due to the general interest thatthe signature of the Mercosur agreement created in the region

In the case of Brazil, export information spillovers originating in Mercosur increase theprobability of exporting to a third market, but tend to decrease it when they originate inROW countries The creation of Mercosur has led to an increase in both Mercosur andROW based information spillovers The same pattern can be observed for the level ofexports to third countries Again, it appears that the creation of Mercosur has led to achange in regime, which could be explained by a larger interest in the ROW on Mercosuractivities from 1991

In the case of Paraguay, exports information spillover originating in Mercosur seem toincrease the probability of exporting to the ROW, but information spillovers originating

10 Results using two-part models also yield consistent results, but efficiency measured in terms of R 2 or statistics Leung and Yu (1996) argued using Monte-Carlo evidence that in data sets with high degree of censoring the two-step estimator may be inefficient when compared to two-part models In our data set, this does not seem to be the case (note that our two-step procedure includes the non-zero observations as suggested with Maddala (1983), whereas Leung and Yi (1996) use Heckman’s procedure

t-in the ROW have a statistically t-insignficant effect on exports to third countries Thecreation of Mercosur in 1991 increase the effect that information spillovers originating inMercosur countries have on the probability of exporting to third countries, but have noeffect on exports information spillovers originating in ROW The level of exports ispositively affected by information spillovers originating in both Mercosur and ROW andthe creation of Mercosur in 1991 had no effect on the information spillovers originating

in Mercosur, but decrease information spillovers originating in ROW The explanation forthis last phenomenon may be linked to the general re-orientation of Paraguay’s exports tothe Mercosur market, as the large members of Mercosur can absorbed an significant share

of Paraguay’s exports at preferential tariffs

For Uruguay, the estimates describe a very similar picture to Paraguay’s experience.Information spillovers originating in Mercosur increase the probability of exporting andthe level of exports to third markets Spillovers originating in the ROW also increase thelevel of exports to third markets The creation of Mercosur in 1991 had no effect on eitherthe probability or the change in the level of exports to third markets due to informationspillovers

To summarise, it seems from the estimates reported in tables 2 and 3 that the creation ofMercosur had a positive and statistical significant effect on the changes in the level ofexports attributed to information spillovers only for the large members of Mercosur(Argentina and Brazil) This change in regime can be partly explained by a generalincrease in ROW interest in the region following the creation of Mercosur in 1991 In thecase of the small members (Paraguay and Uruguay), there is evidence of the presence ofexport information spillovers both from within the Mercosur market and from the ROW,but the creation of Mercosur in 1991 has apparently not yield a platform effect.11

To capture the size of the information spillovers effect we will need to calculate theimplicit elasticities, given that the regressions are not log-linear These is done in the nextsection

11 If anthing, the creation of Mercosur in 1991, has reduced the effect that export information spillovers from the ROW had on Paraguay’s level of exports to third countries.