ECON

Journal of International Economics 47 (1999) 225–244

Intra-industry trade and strategic interaction: Theory and evidence

*Daniel M. Bernhofen Department of Economics, Clark University, Worcester, MA 01610, USA

Accepted 14 March 1997

Abstract

We conduct a theory-based empirical study of intraindustry trade in homogeneous products. We derive an oligopolistic model of intra-industry trade, which is an extension of the segmented market model of trade, initially proposed by [Brander, J. A., 1981, Intra-industry trade in identical commodities, Journal of International Economics 11, 1–14]. The empirical implementations of the model are investigated in the context of the petrochemical industry. Our analysis employs a unique data set containing detailed product- and location-specific data on the petrochemical industries in Germany and the United States. Allowing for different empirical specifications, we find that cross-product variations in bilateral intra-industry trade of petrochemicals are well explained by the variables suggested by the theoretical model.  1999 Elsevier Science B.V. All rights reserved.

Keywords: Intra-industry trade; Strategic trade; Petrochemical industry

JEL classification: F12; F14; L65

1. Introduction

Explaining trade patterns lies at the heart of international microeconomics. Since a substantial amount of world trade occurs in similar products, a phenom- enon termed intra-industry trade (IIT), a vast theoretical and empirical literature has emerged on this subject. The first studies on IIT by Balassa (1966); Grubel (1967) and Grubel and Lloyd (1975) were of empirical nature. Since Grubel and

*Tel.: 11 508 7937185; fax: 11 508 7938849; e-mail: [email protected]

0022-1996 / 99 / $ – see front matter  1999 Elsevier Science B.V. All rights reserved. P I I : S 0 0 2 2 - 1 9 9 6 ( 9 7 ) 0 0 0 2 7 - 5

226 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

Lloyd’s documentation of an extensive amount of IIT among industrialized (and similarly endowed) countries seemed at odds with the traditional theories of comparative advantage, they provided motivation for the development of the ‘‘new

1trade theory’’ under imperfect competition. The ‘‘new trade theory’’ has supplied us with two types of models, commonly

2categorized into ‘‘large numbers’’ and ‘‘small numbers’’ explanations of IIT. The ‘‘large numbers’’ explanation refers to the fairly well developed theory of international trade under monopolistic competition. In the monopolistically competitive model, and its variations, intra-industry trade in differentiated products

3results from the interaction between preferences for variety and scale economies. The ‘‘small numbers’’ model, initially proposed by Brander (1981), has provided us with a homogeneous-product explanation of intra-industry trade in segmented, duopolistic markets. In Brander’s model, intra-industry trade arises from firms’ incentives to capture some of the foreign monopoly rents. Although the essence of the model, that firms’ exports are the result of their profit motives and that, consequently, trade increases competition, seems to capture an important aspect of international trade, the Brander model has been given relatively little attention in

4the discussions on intra-industry trade. We believe that this is partly due to the ‘‘common perception’’ that the theory of intra-industry trade under oligopolistic competition has little empirical content.

The empirical literature on IIT has evolved independently of the theoretical 5literature. Most empirical intra-industry trade studies have emphasized correlates

rather than determinants. Consequently, the theoretical underpinnings of the empirical studies are often unclear. An exception is the study by Helpman (1987), who conducts an empirical study of intra-industry trade based on the model of monopolistic competition. Recently, Hummels and Levinsohn (1995) have revi- sited Helpman’s ‘‘supportive evidence’’ for the model of monopolistic competi- tion.

Applying different data and econometric methods, they cast some doubt on the

1Although it is ‘‘conventional wisdom’’ that intra-industry trade results from imperfect competition, Chipman (1992); Davis (1995) have demonstrated that intra-industry trade can be explained in the context of the Heckscher-Ohlin-Samuelson theory. For a survey of the theoretical literature of IIT under constant returns to scale, see Bhagwati and Davis (1994).

2The terms ‘‘large’’ and ‘‘small’’ refer to the number of firms under the respective market structures. 3The pioneering theoretical articles of IIT in differentiated products go back to Krugman (1979),

Krugman (1980); Lancaster (1980); Helpman (1981). For a good synthesized discussion see Helpman and Krugman (1985).

4Due to the extension of the model by Brander and Krugman (1983), which lead to the well-known ‘‘reciprocal dumping’’ model, the model has had a much bigger influence on the dumping literature than on the intra-industry trade literature. For recent applications of this model in a dumping context, see Weinstein (1992); Bernhofen (1995).

5For good general surveys see Deardorff (1984), Greenaway and Milner (1987), Leamer (1992) and Leamer and Levinsohn (1995). Grant et al. (1992) provide a nice policy-oriented assessment of the empirical literature on IIT.

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 227

empirical support of the country characteristic hypotheses which are derived from this model. One of their major findings is that a majority of intra-industry trade seems to be specific to country-pairs.

In this paper, we conduct a theory-based empirical analysis of intra-industry trade in oligopolistic markets. We derive a model of IIT, which is an extension of the Brander model. Allowing for country-specific differences in market structures, we demonstrate that the theory implies testable hypotheses on the effects of national industry characteristics on the Grubel–Lloyd index of intra-industry trade. In particular, we show that country-specific differences in cost, market size, and firm concentration each have a negative effect on the intensity of bilateral intra-industry trade.

The empirical implications of the model are investigated in the context of bilateral intra-industry trade in petrochemicals between the United States and

6Germany. Our approach is in the spirit of the ‘‘new empirical industrial organization literature’’, which has focused its attention on the study of single

7industries rather than on a cross-section of many industries. Recognizing the importance of industry-specific idiosyncracies, this literature emphasizes the importance of the link between theory and econometric specification. This approach is becoming more widely endorsed by trade economists, exemplified by Leamer and Levinsohn (1995, p. 1342), who write:

‘‘The proper function of empirical work is not to test the validity of the theory but to determine if the theory is working adequately in its limited domain. International trade in lumber might well be characterized by a factor endowments-based model, while an endogenous growth model might better explain trade patterns in computer memory chips, and a model of monopolis- tic competition might best characterize international trade in varieties of furniture.’’

We believe the petrochemical industry to be the appropriate ‘‘domain’’ for a homogeneous-product theory of intra-industry trade under oligopolistic competi- tion. Petrochemicals are homogeneous bulk chemicals, characterized by varying degrees of bilateral intra-industry trade between the US and Germany. The two countries are the world’s main producers and are also the two largest exporters and importers of petrochemicals. The relatively concentrated industry structures in the US and Germany seem compatible with a partial-equilibrium model of imperfect competition.

Our empirical analysis employs a unique data set containing detailed, and internationally comparable, product- and location-specific data on the petrochemi- cal industries in the United States and Germany. Allowing for different empirical

6Due to the limited availability of data, we had to restrict ourselves to a single country-pair. 7For a good survey of this literature see Bresnahan (1989).

228 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

specifications, we find that cross-product variations in bilateral intra-industry trade of petrochemicals are well explained by the variables suggested by the theory. In particular, country-specific differences in cost efficiency, market size and market concentration tend to reduce the intensity of intra-industry trade, as measured by the Grubel–Lloyd index.

The paper is organized as follows. In Section 2, we develop the theoretical framework. Section 3 discusses the empirical implementation and describes the data set. The estimation results are contained in Section 4. Concluding remarks are given in Section 5.

2. The theory

We consider a two-country, single-industry model with n firms located in theh home country and n firms located in the foreign country. The number of firms inf each country is fixed and all firms are assumed to produce the same homogeneous

8product. Each firm makes decisions about how much to produce for the domestic market and how much to export, regarding each country as a separate market. Recent empirical work on firms’ pricing-to-market behavior provides support for

9the segmented market specification. iHome and foreign shipments to the home market are denoted by x (i 5 1,...,hh

j n ) and x ( j 5 1,..., n ), respectively, while home and foreign shipments to theh fh f

i jforeign market are denoted by x (i 5 1,..., n ) and x ( j 5 1,..., n ), respectively.hf h ff f Firms’ production costs are assumed to be the same for firms within a country, but may differ across countries. Country-specific cost differences can be interpreted as the result of national differences in endowments or efficiency. Throughout this paper we assume firms’ marginal costs to be constant. The marginal costs of the representative home and foreign firm are denoted by c and c , respectively.h f

Per customer demand is assumed to be identical in each country and is given by D( p), with D9( p) , 0. Total industry demand at home is s D( p) and total industryh demand in the foreign market is s D( p), where s and s denote the parameters thatf h f

10capture the market size in each country. Firms’ profit functions can then be written down as:

h i 21 i 21 i i p 5 x D (z /s ) 1 x D (z /s ) 2 c (x 1 x ), for (i 5 1,..., n ) (1)i hh h h hf f f h hh hf h

8Although scale economies are not explicitly modelled, it is the presence of fixed costs that limits the number of firms in the industry. It is an implicit assumption that firms which are actually ‘‘in the market’’ have borne the sunk entry costs necessary to operate in the industry.

9A survey of this literature can be found in Goldberg and Knetter (1996). 10Our assumption embodies the notion that the demand behavior of an individual customer is the

same in each country. However, the countries may differ in the number of customers demanding the good.

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 229

f i 21 j 21 j j p 5 x D (z /s ) 1 x D (z /s ) 2 c (x 1 x ), for ( j 5 1,..., n ) (2)j fh h h ff f f f fh ff f

where

n n n nh f h f i j i j

z 5Ox 1Ox and z 5Ox 1Oxh hh fh f hf ff i 51 j 51 i 51 j 51

Each firm is assumed to play a Cournot–Nash game by selecting profit maximizing output levels, given its rivals’ outputs. Differentiating Eqs. (1) and (2) with respect to these choice variables, we obtain the firms’ first-order conditions:

i p(z /s ) 1 (x /s )p9(z /s ) 2 c 5 0, (i 5 1,..., n ) (3)h h hh h h h h h

i p(z /s ) 1 (x /s )p9(z /s ) 2 c 5 0, (i 5 1,..., n ) (4)h f hf f f f h h

j p(z /s ) 1 (x /s )p9(z /s ) 2 c 5 0, ( j 5 1,..., n ) (5)h h fh h h h f f

j p(z /s ) 1 (x /s )p9(z /s ) 2 c 5 0, ( j 5 1,..., n ) (6)f f ff f f f f f

21 i i j jwhere p 5D . The Cournot equilibrium is the vector (x , x , x , x ) whichhh hf fh ff satisfies (3)–(6) simultaneously. Eqs. (3)–(6) are sufficient conditions for a unique equilibrium to exist if industry demand satisfies p91zp0,0, guaranteeing that each

11firm’s reaction curve is downward sloping. Our assumption that home and foreign firms are identical allows us to focus on the output supplies of a

irepresentative home and foreign firm. Consequently, we can write: x 5x ,hh hh i j j

x 5x , x 5x and x 5x .hf hf fh fh ff ff We are interested in how country-specific cost and market conditions affect the

bilateral trade flows within an industry. Let us denote total exports from home to foreign by X and total exports from foreign to home by X . Our symmetryhf fh assumption implies X 5n x and X 5n x . When we apply the bilateralhf h hf fh f fh intra-industry trade flows to the standard measure of IIT, the Grubel–Lloyd index, we get:

un x 2 n x uh hf f fh ]]]]r 5 1 2 (7) n x 1 n xh hf f fh

The index varies between 0 and 1 with higher values of r indicating a higher intensity of intra-industry trade. From Eq. (7), one can see that r varies with the number of firms in each country and the equilibrium output levels of the representative home and foreign firm. Let us first consider the benchmark case where both countries are identical.

11For a discussion of this condition see Dixit (1986).

230 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

Proposition 1. If the home and foreign industry are identical with respect to market size, i.e. s 5 s , market concentration, i.e. n 5 n , and cost, i.e c 5 c , thenh f h f h f there is 100% intra-industry trade, i.e. r51.

Symmetry of cost and market size imply that x 5x . Since n 5n , it followshf fh h f immediately from Eq. (7) that there is 100% intra-industry trade. In what follows, we investigate how country-specific differences in the number of firms, marginal cost, and market size affect the intensity of bilateral intra-industry trade, as measured by the Grubel–Lloyd index. We deviate from the benchmark case by varying one factor at a time. Let us first consider the case where cost and market size are identical across countries, i.e. c 5c and s 5s , but where the degrees ofh f h f market concentration differ, i.e. n ±n . Since this implies that x 5x , Eq. (7)h f hf fh reduces to

r 5 1 2 un 2 n u /(n 1 n )h f h f

Proposition 2. Under symmetric cost and market size, i.e. c 5 c and s 5 s , theh f h f level of intra-industry trade decreases with the asymmetry in market concen- tration.

The less concentrated an industry is relative to its foreign counterpart, the larger the number of firms that penetrate into the foreign economy; consequently, the higher the industry’s total export flows relative to its import flows. On the other hand, a country with a relatively concentrated industry imports more than it exports. Reciprocal market penetration is identical, and the IIT level is at its highest, when both industries are equally concentrated.

Let us now consider the case where the market size and firm concentration are identical, i.e. s 5s and n 5n , but where the country-specific costs are different.h f h f From Eqs. (4) and (5) we obtain ( p 2c ) /( p 2c )5x /x , which implies thath f hf fh x .x if and only if c ,c . Substitution into Eq. (7) yields r 512c 2c /(2p 2hf fh h f h f uc 2c u). A characteristic feature of a Cournot equilibrium is that the equilibriumh f price is an increasing function of the sum of the firms’ marginal cost: p 5f(c 1c )h f

12where f 9.0. Assuming a demand structure which yields p 5k(c 1c ), where k ish f a factor of proportionality with k .0.5, we obtain: r 512uc 2c u / [(2k 21)(c 1h f h c )].f

Proposition 3. Under identical market size and firm concentration, i.e. s 5 s andh f n 5 n , the level of intra-industry trade decreases with the asymmetry in country-h f specific cost structures.

12For a good discussion of the properties of the Cournot model see Shapiro (1989).

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 231

Under Cournot competition, a firm’s profit and market share decrease in its own cost and increase in its rival’s cost. Since intra-industry trade results from the firms’ incentive to capture some of the foreign rents, reciprocal market penetration will be highest if the firms’ cost structures are identical. If, on the other hand, the cost structures become more dissimilar, the country with the low-cost producers will have a competitive advantage and will be a net exporter of this good; while the country with the high-cost producers will be a net importer.

Finally, let us consider the case where market concentrations and costs are the same, n 5n and c 5c , but where the market sizes are different. From Eqs. (4)h f h f and (5) it follows that x /s 5x /s , which implies that x .x if and only ifhf f fh h hf fh s .s . From Eq. (7) we obtain r 512us 2s u /(s 1s ), which gives us proposi-f h h f h f tion 4.

Proposition 4. Under identical market concentration and cost, i.e. n 5 n andh f c 5 c , the level of intra-industry trade decreases with the asymmetry in marketh f size.

Proposition 4 provides us with a demand side explanation for changes in the Grubel–Lloyd measure of intra-industry trade. The more similar the market size at home and abroad, the higher the degree of reciprocal penetration.

3. Empirical implementation

The preceding section contains a strategic trade model where trade results from firms’ incentives to capture some of the foreign monopoly rents. The centerpiece of our theoretical model is Eq. (7), which links the Grubel–Lloyd measure of intra-industry trade to industry concentrations and equilibrium output supplies of the representative firms. The theory suggests three partial determinants of the level of IIT: country-specific relative differences in cost, market size and firm concentration.

uc 2 c u us 2 s u un 2 n uh f h f h f ]]] ]]] ]]]r 5 r , , (8)S Dc 1 c s 1 s n 1 nh f h f h f

In propositions 2 through 4 we have shown that, starting from a point of complete symmetry, the Grubel–Lloyd index is inversely related to each of these three market structure variables.

From an empirical standpoint, it is a limitation of the theory that the partial relationships—≠r / ≠[uc 2c u /(c 1c )],0, ≠r / ≠(us 2s u /(s 1s ),0 and ≠r /h f h f h f h f ≠(n 2n /(un 1n u),0—have been formally shown to hold only under symmetryh f h f of the other parameters. We proceed by specifying an empirical model based on Eq. (8) in the conventional way—i.e. assuming that the partial derivatives hold for

232 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

all market structure configurations—and then examine potential biases of this specification in the context of our data. Furthermore, we perceive the ‘‘specific’’ demand structure, which yields p 5k(c 1c ), as a maintained hypothesis and treath f k as unobservable.

Our empirical implementation of the theory is in the form of an industry analysis of bilateral intra-industry trade, using detailed product- and location- specific data on the petrochemical industry in Germany and the United States during 1988–1992.

In what follows, we provide a short profile of the petrochemical industry and discuss the compatibility of our empirical analysis and data with the underlying theory.

3.1. Industry characteristics

In contrast to other chemical products—in particular specialty chemicals— petrochemicals are standardized products whose chemical properties are well- defined. Since there is practically no product differentiation at the individual product level, these products can be classified as homogeneous. As shown in Table 1, there exists a varying degree of bilateral US–German intra-industry trade at the individual product level. Consequently, the existence of intra-industry trade is

13certainly not a ‘‘figment of the product grouping’’. Our sample consists of 38 traded petrochemicals for which industry data was available for the time period

141988–1992. There is a consensus among industry experts that the commercial structure of

15the petrochemical industry is best characterized as being oligopolistic. This stems mainly from the fact that the industry is very capital-intensive and that petrochemi- cal production requires substantial initial investment costs. Hence, scale economies in the form of sunk costs constitute the main barrier to entry in this industry.

The United States and Germany are the two largest producers and exporters of petrochemicals in the world. Due to the fact that both countries experienced similar historical developments in this industry, the petrochemical products in our

16sample have become standardized products in both countries. Business case studies of individual products suggest that oligopolistic profits in petrochemicals

13Finger (1975) has argued that the occurrence of IIT at aggregated product levels (e.g. at the 3-digit level or higher) can be explained by variations of factor proportions within a group. He asserts that IIT is due to aggregation and is, therefore, not inconsistent with the factor-abundance explanation for trade.

14Our classification of ‘‘traded’’ refers to petrochemicals for which some trade took place between the US and Germany during 1988–1992. The products listed in Table 1 fall into the Harmonized System (HS) Codes 28, 29, 39 and 40.

15See for instance Chapman (Chapman, 1991, pp. 22–34) and Stobaugh (Stobaugh, 1988, pp. 67–78).

16For a study of the historical development of the international petrochemical industry, see Spitz (1988).

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 233

Table 1 Intra-industry trade between Germany and the US: 1988–1992

Product Trade volume Average IIT min max no. obs.

Plastics Low-density Polyethylene 297 379 0.2667 0.2069 0.3057 5 High-density Polyethylene 281 987 0.1127 0.0838 0.1567 5 Ehylene-vinyl acetate copolymers 437 204 0.0632 0.0378 0.0884 5 Polypropylene 191 903 0.9262 0.8157 0.9981 5 Propylen, copolymer 71 084 0.6237 0.4769 0.7338 5 Polystyrene 453 618 0.4021 0.3508 0.4576 5 ABS (Acrylnotrile-butadiene- 84 897 0.1867 0.1866 0.1866 1

styrene copolymer) Polyvinylchloride 507 954 0.5303 0.4449 0.6370 5 Vinylchloride / acetate copolymer 50 603 0.1439 0.0929 0.2038 5 Vinylidenechloride copolymer 18 304 0 0 0 5 Polymer of Vinylacetate 211 425 0.1595 0 0.2550 5 Polyvinyl alcohol 63 354 0.5490 0.5490 0.5490 1 Polymethylmethacrylate 20 843 0.5024 0.3598 0.5728 5 Polyacetal 767 220 0.4404 0.3441 0.610 5 Polycarbonate 899 786 0.7409 0.6980 0.8038 5 Polyethyleneterephtalate 242 604 0.0317 0 0.084 5 Urea resin 35 596 0.4259 0.3534 0.4954 5 Melamine resin 82 308 0.3694 0.3171 0.4008 5 Polyurethane in primary form 248 737 0.4136 0.3244 0.4655 5 Silicon, in primary form 528 596 0.3097 0.2667 0.3546 5 Petroleum resin 98 541 0.5962 0.5172 0.6647 5

Synthetic Rubbers Styrene-Butadiene-Rubber (SBR) 452 288 0.02677 0 0.0541 5 Polybutadiene 839 243 0.9222 0.8752 0.9905 5 Ethylene-Propylene-Rubber 476 511 0.8792 0.8389 0.9159 5

Basic Petrochemicals Carbon Black 190 065 0.0790 0.0510 0.1092 5 Butadiene 65 163 0.0076 0 0.0305 4 Benzene 897 039 0 0 0 5

Industrial Chemicals Paraxylene 104 123 0 0 0 4 Styrene 301 628 0.1352 0 0.3219 5 Ethylbenzene 707 651 0 0 0 5 Cumene 605 512 0 0 0 5 Ethylene glycol 366 330 0.1894 0 0.5188 5 Propylene glycol 13 836 0.0761 0 0.3807 5 Oxirane (ethylene oxide) 256 0 0 0 5 Methyloxirane (propylene oxide) 14 143 0 0 0 5 Acetone (propanol) 736 436 0.6664 0.4783 0.8548 5 Formic acid 368 285 0 0 0 4 Acrylonitrile 36 339 0.2103 0 0.4207 5

‘‘Trade volume’’ refers to total bilateral trade (in 100 kg) during 1988–92; ‘‘min’’ is the lowest of the annual IIT values; ‘‘max’’ is the largest of the annual IIT values; ‘‘average’’ is the average of the annual IIT values; ‘‘no. obs.’’ are the number of years for which complete trade data were available.

234 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

17are determined by the number of producers and by estimated production costs. Hence, it seems that the Cournot–Nash model is a reasonable framework for the study of the petrochemical industry. We hold the assumption that firms perceive countries as separate markets as a ‘‘maintained hypothesis’’ in our empirical study.

Eq. (8) is based on a representative firm model where firms’ marginal costs are constant and differ only by the location of firms. We believe that the petrochemical industry provides an excellent focus for our empirical study in this respect. With special permission, we were able to obtain engineering cost data on petrochemi- cals, created by the Stanford Research Institute International (SRI International). A unique feature of this data is the use of a consistent methodology for calculating ‘‘representative’’ production costs in the two main regions of petrochemical

18production. Furthermore, the cost calculations indicate that unit variable costs— for a given plant size—are independent of the scale of production.

The theoretical model in Section 2 specifies that the product is sold to home or foreign customers who behave as price takers. Given that petrochemicals are intermediate goods, one might wonder whether the relatively high intensities of bilateral intra-industry trade (in Table 1) do not simply stem from intra-firm trade among multinationals. Since systematic evidence on intra-versus interfirm trade is only available at a relatively high degree of aggregation, we refer again to

19anecdotal evidence from the industry literature. Vertical integration has been, indisputably, an important feature of the petro-

chemical industry. However, vertical integration (within and across borders) has primarily taken place between oil refining and petrochemical production. Conse- quently, oil companies have had a significant influence on the development of the

20petrochemical industry. On an industry-scale, there has been very little vertical integration between the petrochemical sector and the sectors which use

21petrochemicals. Also, case study evidence on foreign direct investment of German petrochemical producers in the US indicate that cross-border acquisitions

22were not anticipated to change the trade patterns of petrochemicals. Given the many downstream uses of petrochemicals, we believe the price-taking assumption by customers to be justified in the context of this industry.

17See for instance Stobaugh’s (1988, p. 171ff) study of styrene. 18For a detailed description of the cost data, see Section 3.2. 19The benchmark surveys on US Multinationals—issued by the Department of Commerce—contain

only intra-firm trade information for an industry classification that lies between the 2 and 3 digit SIC aggregation level.

20For instance, in 1989, Exxon was the largest producer of petrochemicals in the US. 21According to Chapman (1991, pp. 33–34), the scattered diversification efforts between petrochemi-

cals and process industries that took place during the 1960s and 70s were dissolved after the second oil shock in the early 1980s. General Motors, which kept its stake in the petrochemical sector, is an exception.

22See for instance Harvard Business School (1991) case studies N9-391-140 / 1 on the acquisition of Celanese by Hoechst.

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 235

In our theoretical model, we have abstracted from barriers to trade, which can occur either in the form of transportation costs or government-imposed restric- tions. It is easy to show that the existence of barriers to trade do not have any effect on our measure of intra-industry trade, as long as the barriers are symmetric across countries. Since we are going to focus on bilateral intra-industry trade

23between Germany and the United States, transportation costs are symmetric. Furthermore, there are no significant trade restrictions in petrochemicals between these two countries.

3.2. The data

3.2.1. Dependent variable Our dependent variable is the Grubel–Lloyd measure of intra-industry trade,

given in Eq. (9):

uX 2 M uit it ]]]IIT 5 1 2 (9)it X 1 Mit it

where X denotes German shipments to the US and M denotes US shipments toit it Germany of product i at time t. For each product, the index has been constructed from annual bilateral trade flows obtained from annual issues of Aussenhandels- statisk, published by the Statistisches Bundesamt, annual issues, Wiesbaden, Germany. Since all of our products are homogeneous bulk chemicals, we were

24able to use quantity measures of trade flows. Table 1 contains the total trade volumes (exports plus imports) and the averages of the annual bilateral Grubel–

25Lloyd indices between Germany and the United States during 1988–1992. The last column in Table 1 gives for each product the number of years for which complete trade data was available.

3.2.2. Independent variables The scope of the empirical study has been determined by the availability of

26engineering data on chemical production costs. Data on production conditions are

23As pointed out by a referee, this reasoning assumes that shipping products from Germany to the US is as expensive as shipping them from the US to Germany. After talking to transportation companies, we found that shipping rates are about 10% higher westbound than eastbound. Given that shipping costs constitute, on average, at most 10% of per unit production costs (see Stobaugh, 1988, p. 162), the possible ‘‘cost bias’’ from ignoring transportation costs will be at most 1%.

24The use of quantity measures is consistent with what the theory suggests. In traditional cross- industry studies, trade flows are usually measured in terms of import and export values. Variations in trade flows can then result from variations in quantities or prices.

25All German data in this paper refer to the former West Germany. 26Our sample consists of all petrochemicals for which some trade took place between the US and

Germany during 1988–1992 and for which we were able to obtain the engineering cost data.

236 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

usually not available beyond the 4-digit aggregation, and internationally compar- able data only at higher levels. However, due to the standardization of production processes and the homogeneous nature of the products in the petrochemical industry, we were able to use product- and location-specific cost estimates for this study. With special permission, we obtained access to engineering cost data on

27petrochemicals created by the Stanford Research Institute International (SRI). A unique feature of the data is the use of a consistent methodology for calculating production costs across the two main regions of petrochemical production—the US Golf Coast and Germany. The international branches of SRI annually collect location-specific information on raw material costs, by-product credits (i.e. value of created by-products) utility costs (cooling water, steam, process water, electricity), etc., which determine the marginal production costs of a measuring unit (e.g. cents per pound of a chemical). An important characteristic of petrochemical production is that the marginal production costs are—for a given plant—independent of the scale of production. Since SRI estimates cost data for different production processes, we chose—for each product—the process with the

28lowest total marginal production costs. Total annual product sales in a country is used as a proxy for market size. For

the US, annual data on total product sales have been compiled from annual issues of Synthetic Organic Chemicals, published by the United States International Trade Commission. For Germany, product sales data were obtained from annual issues of Produzierendes Gewerbe, Fachserie 4, Reihe 3.1, published by the Statistisches Bundesamt in Wiesbaden, Germany.

For the US, the number of producers for each product has been compiled from Synthetic Organic Chemicals. For Germany the number of producers for each product has been collected from annual issues of the Worldwide Petrochemical

29DirectoriesPennwell Directories, annual issues. Ideally, we would have liked to obtain data on market shares of firms that are actually engaged in exporting. However, this data is not available.

4. Estimation

Before we proceed with estimating Eq. (8), we need to investigate the ‘‘potential biases’’ of this specification, given that the theoretical relationships

27SRI is an industry consulting firm, located in Menlo Park, California. The engineering cost data is compiled in the Process Economics Program Yearbooks, which SRI sells to its industry clients.

28For most products, the production process with the lowest marginal costs was the same in both regions. Ideally, we would have liked, for a given product, to use the cost associated with the process most widely used in a country; however, this information is not readily available.

29The fact that only very few foreign producers were listed in our product sample, which we included in our counting, indicates that direct investment considerations are negligible in the context of our study.

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 237

derived in propositions 2–4 have been only established under symmetry of the other parameters. The potential biases stem from the fact that the explanatory variables enter Eq. (8) in the form of absolute values, ignoring information about the direction of trade flows.

For instance, c 2c .0 suggests that home is a net importer; whereas n 2n .0h f h f suggests that home is a net exporter, ceteris paribus. If in our product sample, (c 2c ) /(c 1c ) and (n 2n ) /(n 1n ) have a high positive correlation, theh f h f h f h f interaction between these two independent variables will have an impact on the intra-industry trade index that would be ignored in our empirical specification. Under such a ‘‘bad’’ configuration, Eq. (8) would lead to an underestimation of the intra-industry trade index. Similarly, we need to investigate the potential of ‘‘bad’’ configurations with respect to the interaction of each of these two independent variables with the ‘‘market size’’ variable. Recognizing that, ceteris paribus, s 2s .0 implies that home is a net importer, ‘‘bad configurations’’ are equivalenth f to high (pos.) values of corrh(c 2c ) /(c 1c ), (s 2s ) /(s 1s )j or corrh(n 2n ) /h f h f f h h f h f (n 1n ), (s 2s ) /(s 1s )j.h f h f h f

Table 2 contains the correlation coefficients corresponding to the three interactions among the explanatory variables. A ‘‘bad’’ configuration is associated with a large positive correlation coefficient (possibly close to 1). Since our data consists of 5 years, we obtain a total of 15 correlation coefficients. The data indicates, in general, weak interaction terms; 8 negative and 7 positive, with the median of the positive coefficients being 0.056. Hence, our independent variables seem to measure ‘‘distinct influences’’ on the level of intra-industry trade.

Table A1 in Appendix A contains descriptive statistics for each variable. The explanatory variables from Eq. (8) are denoted by COSTDIF, SIZEDIF and FIRMDIF. Due to missing observations in the dependent and independent variables, our complete sample size, for a given year, has been reduced to between 20 and 28 observations. Since our sample contains a significant number of products for which the IIT index is zero, we provide estimates for both the ‘‘full sample’’ (i.e. including the zero IIT values) and a ‘‘restricted sample’’ excluding observations with zero IIT values. It should be noted that products with a zero IIT value are traded, but only in one direction (i.e. X 50 or M 50, but not both). Weit it provide estimates for the ‘‘full sample’’ and the ‘‘restricted sample’’ to check for the robustness of our results with regard to the ‘‘zero value outliers’’. Table 1 reveals that most basic and industrial petrochemicals are characterized by

Table 2 Likelihood of ‘‘bad’’ industry configurations

1992 1991 1990 1989 1988

corr((c 2c ) /(c 1c ), (n 2n ) /(n 1n )) 0.056 0.107 0.059 20.044 20.029h f h f h f h f corr((c 2c ) /(c 1c ), (s 2s ) /(s 1s )) 20.24 20.13 20.19 20.129 0.057h f h f f h h f corr((n 2n ) /(n 1n ), (s 2s ) /(s 1s )) 20.032 0.259 20.059 0.04 0.024h f h f h f h f

238 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

relatively little or no intra-industry trade. In order to capture the effect of 30intermediate petrochemicals, we introduce a dummy variable INTERME—taking

the value 1 if the product is either a basic or an industrial petrochemical and 0 31otherwise—when estimating the full sample.

The theoretical analysis in Section 2 has provided us with ‘‘sign relationships’’ between the GL measure and the three industry-structure variables. However, it doesn’t supply the functional form for the relationship between the IIT index and

32the explanatory variables. In what follows, we estimate Eq. (8) under alternative empirical specifications and investigate the robustness of the results.

We first consider a linear regression specification. In the context of the classical linear regression model, we use ordinary least squares (OLS) both, on the ‘‘raw Grubel–Lloyd index’’ and on the log-linear adjusted index. The logistic trans- formation (log(x /(12x))) of the IIT index is performed since the index varies

33between 0 and 1. In addition to the linear regression model, we consider a nonlinear regression specification. In the latter case, we use the logit probability

34function as a functional form :

IIT 5 1 /(1 1 exp(2b 9z )) 1 e (10)it it it

where z pertains to the vector of explanatory variables and [ to the disturbanceit it term. Eq. (10) is estimated using nonlinear least squares (Nonlinear LS). Assuming a normally distributed disturbance term, the nonlinear least squares

35estimator corresponds to the Maximum Likelihood Estimator. Consequently, it has all the desirable properties of the Maximum Likelihood Estimator, including asymptotic efficiency.

Table 4 contains the estimation results pertaining to individual years. Due to the relatively small number of observations in a given year, we also pooled our data across years and introduced year dummies. The estimation results from the pooled data set are given in Table 3. In general, Tables 3 and 4 indicate that the model fits the data fairly well and the estimates are fairly robust with regard to the different empirical specifications.

We first consider the estimates from pooling the data across years. An adjusted 2

R that ranges between 50% and 70% indicates a good overall fit of the model.

30Basic and industrial petrochemicals are used as intermediates in the production of synthetics or other chemical products.

31Since the majority of the basic / industrial petrochemicals have zero IIT values and are, therefore, already excluded in the restricted sample, there is no need to include the INTERME variable when running regressions on the restricted sample.

32It can be shown that even under a ‘‘maintained hypothesis’’ of linear demand, the theory doesn’t seem to predict a functional form which could be exploited empirically.

33The rationale behind this specification is that this function maps the interval (0,1) onto (2`, `). 34See also Balassa (1986) who applies the nonlinear regression estimation to a multi-country,

multi-industry analysis of IIT. 35See Greene (1993), pp. 314–315.

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 239

Table 3 Estimates for pooled data

Nonlinear LS OLS on IIT Nonlinear LS OLS on IIT OLS on

log(IIT /(12IIT))

Full sample Excluding zero observations

CONSTANT 3.065*** 0.801*** 2.745*** 0.954*** 2.968***

(5.79) (8.93) (5.63) (12.21) (5.97)

COSTDIF 236.12*** 20.788* 231.079*** 21.194*** 28.994***

(27.19) (21.86) (26.60) (23.55) (23.76)

SIZEDIF 21.063** 20.668*** 21.055* 20.678*** 24.494***

(21.97) (27.85) (21.89) (28.21) (28.29)

FIRMDIF 21.170** 20.061 20.995** 20.172** 20.850*

(22.54) (20.64) (22.20) (22.06) (21.64)

INTERME 24.744*** 20.345*** 2 2 2

(23.82) (26.16)

1991 20.26 0.0003 20.235 20.042 20.111

(0.34) (0.004) (20.70) (20.64) (20.24)

1990 0.050 20.009 20.243 20.089 20.618

(0.31) (20.13) (20.78) (21.32) (21.63)

1989 0.017 20.003 20.104 20.038 0.035

(0.34) (20.05) (20.31) (20.56) (0.07)

1988 20.283 20.004 20.215 20.055 20.304

(20.83) (20.06) (20.64) (21.02) (21.04) 2

R adj. 0.70 0.50 0.63 0.53 0.55

t values are given in parentheses and are based on White’s consistent estimators. ***,(**),(*) indicates significance at the 1%, (5%), (10%) level.

2The adjusted R from the pooled data indicates a relatively small change in the goodness of fit when the zero observations are excluded. The signs and significance of the explanatory variables are fairly consistent across the different specifications. The insignificance of all the year dummies indicate no systematic year effects. INTERME has a highly significant negative effect for each spe- cification. This result is not surprising, given the relatively little two-way trade in intermediate petrochemicals, as can be seen in Table 1.

In the nonlinear regression specification, COSTDIF, SIZEDIF and FIRMDIF always have the expected negative sign and are all significant (at the 5% level for the full sample and at the 10% level for the restricted sample). Similarly, for the linear regression model, the industry-structure variables all have the expected negative sign and almost all are statistically significant. The exception is FIRMDIF, which turns out to be insignificant when running ordinary least squares on the full sample.

2Looking at the adjusted R for the yearly estimates, Table 4 reveals that the independent variables explain between 39% and 87% of the variation in the dependent variable. The model seems to fit the data better when the zero observations are excluded than when they are included. Considering the entire sample, SIZEDIF always has the expected sign and is significant at the 1% level in

240 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

Table 4 Estimates for annual German–US bilateral intra-industry trade

1992 1991 1990 1989 1988

OLS estimation on IIT (including zero observations)

CONSTANT 0.885*** 0.634*** 0.763*** 1.105*** 0.864*** (5.51) (4.22) (4.99) (8.75) (10.43)

COSTDIF 21.354 20.259 20.663 25.333*** 20.093 (21.17) (21.35) (20.51) (23.07) (21.10)

SIZEDIF 20.661*** 20.731*** 20.742*** 20.181 20.852*** (23.97) (24.89) (23.77) (20.77) (27.29)

FIRMDIF 20.031 0.220 0.127 20.414** 20.098 (20.134) (1.00) (0.61) (22.32) (20.939

INTERME 20.452*** 20.318*** 20.434*** 20.668*** 20.487*** (24.12) (22.91) (24.65) (26.16) (25.52)

2 R adj. 0.42 0.46 0.54 0.77 0.80 N 28 27 26 20 20

OLS estimation on IIT (excluding zero observations) CONSTANT 1.178*** 0.907*** 0.759*** 1.179*** 0.966***

(8.88) (9.21) (4.70) (9.72) (18.74) COSTDIF 21.447* 20.895*** 22.473 26.87*** 23.074***

(21.61) (27.36) (21.53) (23.80) (22.94) SIZEDIF 20.870*** 20.821*** 20.413* 20.030 20.615***

(25.99) (26.78) (21.81) (20.13) (23.70) FIRMDIF 20.328 20.058 0.029 20.475*** 20.125

(21.49) (20.33) (0.20) (22.73) (21.22) 2

R adj. 0.58 0.62 0.39 0.75 0.87 N 18 17 18 14 14

OLS estimation on log(IIT /(12IIT)) (excluding zero observations) CONSTANT 4.384*** 3.024** 1.646** 5.249*** 2.353***

(5.78) (2.24) (1.96) (4.16) (6.94) COSTDIF 210.722** 27.205*** 218.003* 255.836*** 214.026**

(21.99) (24.57) (21.76) (22.99) (22.41) SIZEDIF 25.650*** 25.429*** 22.89** 0.543 23.759***

(26.02) (24.66) (22.02) (0.23) (23.48) FIRMDIF 21.907 20.466 0.612 23.248** 20.250

(21.63) (20.33) (0.83) (22.22) (20.38) 2

R adj. 0.66 0.50 0.54 0.69 0.85 N 18 17 18 14 14

t values are given in parentheses and are based on White’s consistent estimators. ***, (**), (*) indicates significance at the 1%, (5%), (10%) level.

all years except for 1989. COSTDIF and FIRMDIF appear to be significant in only a single year (1989); the year in which SIZEDIF is insignificant. Similar to the pooled estimates, INTERME has a highly significant negative impact each year.

Considering the restricted sample, one can see that with regard to the significance of the independent variables, the estimates are invariant to the

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 241

transformation of the dependent variable. SIZEDIF has the expected sign, is significant at the 1% level for three years, at the 10% level for one year and is again insignificant in 1989. FIRMDIF is again only significant in 1989, but COSTDIF is now significant four of the five years and always has the expected sign. We believe that cost similarities (across the two regions) of petrochemicals with no intra-industry trade might be responsible for this change of significance when moving from the full sample to the restricted sample. The yearly estimates indicate that for products for which the IIT index is positive, country-specific differences in cost efficiency and market size have a significant negative impact on the intensity of bilateral intra-industry trade.

Overall, we can conclude that the cross-product variations in the intensity of bilateral intra-industry trade in petrochemicals are well explained by our industry structure variables. In particular, country-specific differences in cost efficiency, market size and market concentration tend each to have a negative impact on the IIT index.

5. Concluding remarks

The phenomenon of intra-industry trade has spawned a voluminous theoretical and empirical literature. However, empirical studies that go beyond ‘‘descriptive empiricism’’ and test hypotheses based on economic theories of intra-industry trade are in short supply. The few existing theory-based studies have investigated the empirical implications of the intra-industry trade theory under monopolistic competition. The major focus of these studies has been on the testing of country-characteristic hypotheses—using aggregated (4-digit) SITC data.

This paper suggests a new approach to the empirical analysis of intra-industry trade, motivated by the new empirical industrial organization literature. Our ‘‘industrial organization approach’’ has focused on the empirical analysis of bilateral intra-industry trade in the context of a single industry. We believe the main contributions of this paper to be the following:

(1) From a mere descriptive viewpoint, our trade data on petrochemicals reveal the existence and prevalence of intra-industry trade at the individual product level. Since the choice of aggregation scheme has been recognized to be crucial in the measurement of intra-industry trade, our data shows that intra-industry trade in petrochemicals cannot be dismissed as a ‘‘figment of the product grouping’’. Furthermore, there is substantial variation in the intensity of bilateral intra-industry trade in products that can be classified as being homogeneous.

(2) A simple theoretical extension of the Brander model has shown to yield testable hypotheses about the effects of country-specific differences in industry characteristics on the Grubel–Lloyd index of intra-industry trade.

(3) Our ‘‘industrial organization approach’’ has enabled us to conduct an empirical study in the context of an industry that seems compatible with the

242 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

underlying assumptions of the theory. The availability of internationally compar- able, product-specific data on petrochemicals allowed us to generate reasonable empirical proxies for the variables suggested by the theory. Our empirical results indicate that the effects of the industry-structure variables on the intensity of bilateral intra-industry trade are in accord with what the theory suggests.

The few empirical studies based on recent insights of the ‘‘strategic approach’’ to international microeconomics have focused on the normative side of the literature; in particular, on the role of (national) welfare improving strategic trade

36policies. This paper constitutes one of the first empirical steps towards the ‘‘positive origins’’ of the new trade theory: the explanation of trade patterns associated with intra-industry trade. Given that a huge amount of intra-industry trade takes place in industries that are most accurately characterized as being oligopolistic, it seems that there is a fair amount of new ‘‘empirical territory’’ which still needs to be investigated by trade economists.

Acknowledgements

This project has been supported by a Clark University faculty research grant. My gratitude goes to Dr. Wang of SRI International and to Frau Spindler of the Verband der Chemischen Industrie for their hospitality during my data collection visits to Menlo Park, California and Frankfurt, Germany. I am grateful to John Brown, David Hummels, Amy Glass, Arthur Lewbel, Taiji Furusawa, David Weinstein, two anonymous referees and participants at the Midwest International Economics Meetings in Minnesota and at department seminars at Boston, Brandeis, Clark and Connecticut for helpful comments. Thanks to J. David Richardson for encouragement and comments during the early stages of this project. Zahid Hafeez and Aylin Sertkaya provided helpful research assistance.

Appendix A

aTable A1. Means and standard deviations

1992 1991 1990 1989 1988

Full sample (including zero IIT values) IIT 0.290 0.253 0.281 0.322 0.304

(0.331) (0.327) (0.310) (0.350) (0.312) COSTDIF 0.074 0.098 0.063 0.062 0.132

36See for instance the collection of papers in Krugman and Smith (1994).

D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244 243

(0.062) (0.124) (0.050) (0.041) (0.274) SIZEDIF 0.484 0.450 0.430 0.437 0.444

(0.255) (0.257) (0.251) (0.259) (0.256) FIRMDIF 0.451 0.466 0.492 0.503 0.499

(0.246) (0.235) (0.216) (0.206) (0.232) N 28 27 26 20 20

Restricted sample (excluding zero IIT values) IIT 0.451 0.401 0.405 0.460 0.434

(0.312) (0.332) (0.296) (0.333) (0.285) COSTDIF 0.079 0.105 0.074 0.066 0.064

(0.066) (0.149) (0.048) (0.041) (0.039) SIZEDIF 0.514 0.463 0.451 0.420 0.444

(0.269) (0.278) (0.280) (0.261) (0.262) FIRMDIF 0.507 0.540 0.536 0.538 0.409

(0.259) (0.229) (0.227) (0.223) (0.259) N 18 17 18 14 14 a Standard deviations are given in parentheses.

References

Balassa, B., 1966. Tariff reductions and trade in manufactures among the industrial countries. American Economic Review 56, 466–473.

Balassa, B., 1986. The determinants of intra-industry specialization in United States trade. Oxford Economic Papers 38, 220–233.

Bernhofen, D.M., 1995. Price dumping in intermediate good markets. Journal of International Economics 39, 159–173.

Brander, J.A., 1981. Intra-industry trade in identical commodities. Journal of International Economics 11, 1–14.

Brander, J.A., Krugman, P., 1983. A ‘‘reciprocal dumping’’ model of international trade. Journal of International Economics 15, 313–321.

Bresnahan, T.F., 1989. Empirical studies of industries with market power. In: Schmalensee, R., Willig, R. (Eds.), Handbook of Industrial Organization, vol. 2. North-Holland, Amsterdam, pp. 1011–1057.

Bhagwati, J., Davis, D., 1994. Intra-industry trade: issues and theory. Harvard University Discussion Paper No. 1695.

Chapman, K., 1991. The International Petrochemical Industry: Evolution and Location. Blackwell, Cambridge, MA.

Chipman, J.S., 1992. Intra-industry trade in a loglinear model. Mimeo, University of Minnesota. Davis, D.R., 1995. Intra-industry trade: A Heckscher–Ohlin–Ricardo approach. Journal of International

Economics 39, 201–226. Deardorff, A.V., 1984. Testing trade theories and predicting trade flows. In: Jones, R., Kenen, P. (Eds.),

Handbook of International Economics, vol. 1. North–Holland, Amsterdam, pp. 467–517. Dixit, A., 1986. Comparative statics for oligopoly. International Economic Review 27, 107–122. Finger, M., 1975. Trade overlap and intra-industry trade. Economic Inquiry 13, 581–589.

244 D.M. Bernhofen / Journal of International Economics 47 (1999) 225 –244

Goldberg, P., Knetter, M., 1996. Goods prices and exchange rates: what we have learned? NBER Working Paper No. 5862.

Grant, R., Papadakis, M., Richardson, J.D., 1992. Global trade flows: old structures, new issues, empirical evidence. Mimeo, Syracuse University.

Greene, W.H., 1993 Econometric Analysis. Prentice Hall, Englewood Cliffs, NJ. Greenaway, D., Milner, C., 1987. Intra-industry trade: current perspectives and unresolved issues.

Weltwirtschaftliches Archiv 123, 39–57. Grubel, H.G., 1967. Intra-industry specialization and the pattern of trade. Canadian Journal of

Economics and Political Science 33, 374–388. Grubel, H.G., Lloyd, P.J., 1975. Intra-industry Trade: The Theory and Measurement of International

Trade in Differentiated Products. Wiley, New York. Harvard Business School, 1991, Hoechst in the United States (A and B), case study N9-391-140 / 1,

Harvard Business School, Boston, MA. Helpman, E., 1981. International trade in the presence of product differentiation, economies of scale

and monopolistic competition: a Chamberlin–Heckscher–Ohlin approach. Journal of International Economics 11, 305–340.

Helpman, E., 1987. Imperfect competition and international trade: evidence from fourteen industrial countries. Journal of the Japanese and International Economies 1, 62–81.

Helpman, E., Krugman, P., 1985. Market Structure and Foreign Trade. MIT Press, Cambridge, MA. Hummels, D., Levinsohn, J., 1995. Monopolistic competition and international trade: reconsidering the

evidence. Quarterly Journal of Economics 110, 799–836. Krugman, P., 1979. Increasing returns, monopolistic competition, and international trade. Journal of

International Economics 9, 469–479. Krugman, P., 1980. Scale economies, product differentiation, and the pattern of trade. American

Economic Review 70, 950–959. Krugman, P., Smith, A., 1994. Empirical Studies of Strategic Trade Policy. University of Chicago

Press, Chicago, IL. Lancaster, K., 1980. Intra-industry trade under perfect monopolistic competition. Journal of Internation-

al Economics 10, 151–175. Leamer, E., 1992. Testing trade theory, NBER working paper No. 3957. Leamer, E., Levinsohn, J., 1995. International trade theory: the evidence. In: Grossman, G., Rogoff, K.

(Eds.), Handbook of International Economics, vol. 3. North–Holland, Amsterdam, pp. 1339–1394. Pennwell Directories, annual issues, Worldwide Petrochemical Directories. Pennwell Publisher, Tulsa,

Oklahoma. Shapiro, C., 1989. Theories of oligopoly theory. In: Schmalensee, R., Willig, R. (Eds.), Handbook of

Industrial Organization, vol. 1. North–Holland, Amsterdam, pp. 329–414. Spitz, P., 1988. Petrochemicals: The Rise of an Industry. Wiley, New York. Statistisches Bundesamt, annual issues. Aussenhandelsstatistik, Wiesbaden, Germany. Stobaugh, R., 1988. Innovation and Competition: The Global Management of Petrochemical Products.

Harvard Business School Press, Boston, MA. United States International Trade Commission, annual issues, Synthetic Organic Chemicals, US

Production and Sales, GPO, Washington. Weinstein, D.E., 1992. Competition and unilateral dumping. Journal of International Economics 32,

379–388.