Essays in International Trade and Financial Economics

BY

Huiwen Lai

A thcsii submitted in conformity with the requirements for the degree of Doetor of Phi)osophy

Graduade Department Department of Economics University of Toronto

O Copyright by H u k n Lai (2000)

ubitioris and "9 Bb i iogrephi Services

Acquisitions et services bibliographiques

The author bas granted a nonexclusive licence dowing the National Library of Canada to reproduce, loan, distribute or seii copies of this thesis in rnicrofonn, paper or electronic fonnats.

L'auteur a accordé une licence non exclusive permettant à la Bibliothèque nationale du Canada de reproduire, prêter, distriiuer ou vendre des copies de cette thèse sous la forme de microfiche/^ de reproduction sur papier ou sur format électronique.

The author retains ownership of the copyright in this thesis. Neither the thesis nor substantial extracts fiom it may be printed or otherwise reproduced without the authoc's permission.

L'auteur conserve la propri6té du droit d'auteur qui protège cette thèse. Ni la these ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.

Abstract Essay in international Trade and Financial Economics Ph.D. Thesis (2000) by Huiwen Lai Department of Economics University of Toronto

This thesis includes three essays related to international economics and financial economics. The first essay presents a mode1 of trade in the presence of multinationals, asymmetric

trade barriers, and international differences in production costs. The first part of the essay presents the rnodelrs implications for bilateral trade. The estimation reveals more reasonable parameters for elasticity of substitution and trade costs than that suggested by previous research. The simulation indicates that tariff liberalization will shift trade f rom rich countries to poor countries and from preferential trading areas to inter-continental trading partners. The second p a r t of the essay derives the multinational production and export

equations Fmplied by the mode1 and estimates these equations simultaneously by recognizing the cross-equation restrictions on parameters

and

error terms.

It

suggests

that

the

elimination of tariffs would substantially increase U. S. exports, but would not affect U . S .

production abroad.

The

second essay models general equilibrium product price effects using the CES monopolistic competition model in international trade. We then estimate the model and, mimicking computable general equilibrium (CGE) models, use the model to estimate the

compensating

variation

associated

with

trade

liberalization. We find gains from trade liberalization that

are much larger than those usually reported.

In addition,

extensive specification testing is conducted to evaluate the performances of this model and its alternatives. The results point to the types of model specifications needed before the model can usefully be applied to policy questions. The third essay

studies the properties of Canadian

interest rates

relative to those of U.S. In sharp contrast to the U.S. evidence, the conditional variances of Canadian macroeconomics variables are found to be insignificant predictors of term

premia in the Canadian T-bill term structure. However, the conditional variances of

U.S.

macroeconomic variables are

found to be important deteminants of Canadian premia.

Contents 1 The DetertiaUiants of Bilateral 'Ikade and Multinational Production

.................................. The Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 The Consumer's Problem . . . . . . . . . . . . . . . . . . . . . . 1.2.2 The Producer's Problem . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 BilateralTradeEquation . . . . . . . . . . . . . . . . . . . . . . . TheData . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bilateral 'lkade Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 TheFixedMects . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Influentid observations . . . . . . . . . . . . . . . . . . . . . . . 1.4.3 Economic Implications . . . . . . . . . . . . . . . . . . . . . . . . nade and Multinational Production . . . . . . . . . . . . . . . . . . . . 1.5.1 Bilateral ?tade and Multinational Production Equations . . . . . 1.5.2 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.3 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . .

1.1 Introduction

1.2

1.3 1.4

1.5

1.6 Conclusions

..................................

...................... Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8.1 Table 1.1. The34 Coutries inthe BilateralThdeEquation . . .

1.7 Appendix Panel Data on TaxifFk 1.8

1.8.2 Table 1.2. Estimates of the Bilateral Trade Equation

.......

1.8.5

. ... .. . . . . Table 1.4. Residual Analysis of the Fked Effects Model . . . . . . Table 1.5. Senitivity R d t s of the Bilaterai Dade Equation . . .

1.8.6

Table 1.6. Hypothetical T d and Distance Enects, 1992 (with

1.8.3 Table 1.3. Residual Analpis of the Basic Mode1

34

1.8.4

35 36

Fixed Mects Model). Grouped by OECD and non-OECD Countries 37 1.8.7 Table 1.7. Hypothetical Tariff and Distance Effects, 1992 (with Fixed Effects Model), Grouped by Regions

. . ..... ... . . .

38

1.8.8 Table 1.8. CountriesintheU.S. ExpottandtheU.S. Multinational

. . . . . . . . . . ,. . . . . . . . . . . . . . . . . . . Table 1.9. Estimates of CroeEquation Restrictions . . . . . . . . Table 1.10. Sensitivity Results of CroskEquation Restrictions . .

Production

1.8.9 1.8.10

39 40

41

1.8.11 Table 1.11. Tarin and Distance Effects on the U.S. Exports and the U.S. Multinational Production

........... .... . ,

1.9 Fig u r e s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

43

1.9.1 Figure 1.1. Mode1 Fit of the Bilateral Trade Equation . 1.9.2

42

......

43

Figure 1.2. Model Fit of the Cross-Equation Restrictions: the U.S. Export Equation

. .. . . .... .... . . . . . .. .. . . . ..

44

i .9.3 Figure 1.3. Model Fit of the CroeEquation Restrictions: the U.S.

Multinational Production Equation

1.9.4

.. . .. ....... . ....

45

Figure 1.4. Effectsof Hypothetical Reductionon the U.S.Exports and Multinational Production (1992) . . . . . . . . .

. ......

46

1.9.5 Figure 1.5. Effects of Hypothetical Distance-Related nade Cost Reductions on the US. Exports and Multinational Production (1992) 47

The Gains M m made: Standard Errors with the CES Monopohtic Competition Mode1 48 2.1 Introduction.

. .. . . .. . . . ... . . . . . . . . .. . . . . .. . . . .

2.2 The0ry.....................................

2.3 TheData

. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

51 53

2.4 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Identification

.................................

............................... 2.7 Thde Liberalization and Compensating Variation . . . . . . . . . . . . . 2.8 A Cntical Etoadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 By hdustry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.10 Mis-Specification of the Income, Price. and Data Identity Terms . . . . .

2.6 Empirical R d t s

2.11 Interpreting the Fixed Effects

........................

2.12 Specification Tests and Model Selection . . . . . . . . . . . . . . . . . . .

.................................. 2.14Appendixs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.14.1 AppendOc A.Deflators . . . . . . . . . . . . . . . . . . . . . . . . 2.14.2 Appendix B . Likelihood Fùnction . . . . . . . . . . . . . . . . . . 2.15 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.15.1 Table 2.1. List of Importers and Exporters . . . . . . . . . . . . . 2.13 Conclusiom

2.15.2 Table 2.2. 1992 Bilateral Tariff Rates for Aggregate Manuf'turing

2.15.3 Table 2.3. Estimation for Aggregate Manufacturing

........

2.15.4 Table 2.4. Estimation for Aggregate Manufacturing

........

2.155 Table 2.5. Tariff Reductions Be-

1972 and 1992 . . . . . . . .

2.15.6 Table 2.6. Compensating Variation for the 14 hporting Countries

2.15.7 Table 2.7. Compensating Variation for the 36 Exporthg Countries 2.15.8 Table 2.8. Estimatee by hdustry

..................

2.15.9 Table 2.9. Evidence of Mis-Specifi~8tionfrom hh&jt = fl, ln si:

+

......................... .. . . . . . . . . . . . . . . 2.15.10TabIe 2.10. The Fixed Effects . . . . &in@,

,+&inQjt

2.15.11Table 2.11. Correlates of the Fixed MectS (ki = a +Pxii) 2.l!Ll2 Table 2.12. Formal Model Specincation Testing 2.16Figures

....

..........

.....................................

.........................

89

2.16.2 Figure 2.2. Compensating Veriation . . . . . . . . . . . . . . . . .

90

2.16.1 Figure 2.1. Model Fit

2.16.3 Figure 2.3. Just How Good is the Fit?

2.16.4 Figure 2.4. Fied Effects and Distance

............... ...............

91

92

3 The Expectation Hypothesis. Term Premia and the Canadian Term

Structure of Interest Rates 3.1 Introduction

93

..................................

................. ...............................

96

.....................

97

3.2 Expectation Theory of the Term Structure 3.2.1

Literature

3.2.2 The Expectation Hypothesis

........................... Second-Moment Determinants of Term Premia . . . . . . . . . . . . . . 3.3.1 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 An ARCH-M Model of Term Premia . . . . . . . . . . . . . . . . Conciusio~ls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..................................... a 3.5.1 Table 3.1. Descriptive Statistics . . . . . . . . . . . . . . . . . . .

3.2.3 Empirical results 3.3

3.4

3.5

94

96

99 102 102 103

107 109

109

3.5.2 Table 3.2. OLS Regression of Excess Returns on theYieldSpread 110 3.5.3 Table 3.3. Estimation of ARCH-M Premia

.............

111

3.5.4 Table 3.4. Estimation of ARCH-M Premia on Canadian T-bib

.................. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

112

3.6.1 Figure 3.1. Excess Returns on Zmonth T-bills

...........

113

Figure 3.2. Excess Returns on %monthT-bills

...........

114

Using U.S. Conditional Variances

3.6

3.6.2

113

Chapter 1

The Determinants of Bilateral Trade and Multinational Production

1.1 Introduction New trade theory, as represented by the monopolistic competition model hm been very

succesgful in explaining intra-industry trade among OECD countries (Krugman 1979 and 1981, Lancaster 1980, and Helpman 1981). It has also been expanded to explain the rapid growth of multinational production over the last two decades (Helpman 1984, Brainard 1993a, Merkusen 1995). While the model has dominateci tke theoretical literature since the late 1970s, it hm had somewhat less impact on the empincal literature (see Leamer

and Levinsohn 1995).

A few empirical papers have stuàied the implications of the monopolistic competition model for the relationship between trade costs and bilateral trade. Some studies have found that large variations in trade are explaineci by tariff and non-tariff barriers to trade

(Harrigan 1993, Haveman et. al l996), while others have modelled trade costs indirectly as unexplainecl econometric fixeci efFects (Helpman 1987, Hummeis and LeWisohn 1995, Harrigan 1996, and Jensen 1996). Dacie costs have also been found to be important in

determining patterns of multinational production (Brainard 1997). Besidea the monope listic competition model, a variety of empirical atudies using gravity modeia have found that trade barriers explain a large portion of trade (e.g. Bergstrand 1985).

Trade costs operate primarily via prices, yet a problem with the existing literature is that it does not ngorously incorporate product prices into the analysis. In the context of the monopolistic competition model, the difliculty is crested by the complexity of the

constant elasticity of substitution (CES)price index Ui the presence of asymxnetric trade costs. For example, Harrigan (1993) and Haveman et. al (1996) estimateci only part of the full price index term. rii Chapter 2 we will estimate the complex CES price t e m ~ only for the limited case where the primary international asymmetry is bilateral t

e rates.

In the context of the gravity model, the focua has been more on income temis than price terms.

M h e r , lack of conseasus on the theoretical foundations of the gravity model

have made it &Ecult to rigorously derive rich and detded predictions about the price

term (Anderson 1979, Bergstrand 1985 and 1989, HeIpmrui and Knigman 1985, Hdprnan

1998, Evenett and K d e r 1998, and DeerdorfF 1998).

This paper adds to the iiterature by estimating a monopolistic competition model that incorporates multinationals and a richer set of international asymmetries. These include agymmetric trade barriers and international clifferences in production costs. In the first part of the paper we consider the model's implications for bilateral trade. While the model is in the monopolistic competition fmnework and thus yields an estimating equation related to those in Herrigan (l993), our next chapter, and gravity models, there are important Merences. Models used in previous studies are not about multinationals,

but about firms that produce and export fiom a single country. Approachuig bilateral import equations as the outcome of decisions by multinationals leads one to coneider tariffs, distance, and international differences in production costs in a novel wa). W e develop the insights obtained kom this approach by deriving a bilateral trade equation that is directly implied by the theory. Taris, distance, and production costs enter Enns' pricing and output decisions. These decisions, when set against the backdrop of

CES

prderences, yield a precise b i l a t d trade estimating equation. Indeed, the theory even

predicts the functiond f o m for the dependence of bilateral trade on tariffs, distance, and production costs.

In thinlcing about this problem, we borrow fiom the large literature on multinational

6irms. In perticular, we build on Brainard's insight about the importance of distance and production

C O B ~ ~ See .~

Hallward-Driemeier (l996),Chakrabarti (l997), and Soboleva

(1997)for detailed discussions of location decisions. We emphasize that our p a p a is not about the location decisions of multinationals. Our focus is instead on the trade and production decisions of &ting plants. In order to make this decision problem tractable

we take the location decision of multinationals as fixeci. We hope to persuade the reader 'The traditionalexplanation of multinational activity is the U t i r proportionshypothesirr» (Helpman 19û4, Markuaen 1984). A more recent argument is the "prowmity-concentration tradidf explanationn by Brainard (19938). Brainatd (l993b,1997) tests these two theoriea separately and favm the iatta. This means that traàe costz3, but not fector ciifferences (or, in our contact, the factor p h M i c e s refiected in production cost dinerences), are s i e c a n t in muitinatid production determination. In thir paper, we are not teating which one is the correct moc&l.

that our approach is a usehl heuristic.

We draw from a large database on trade, endowments, and wages describeci in Antweiler and Tlrder (1997). In addition, we take great d o r t to build a unique and comprehensive time series on b i l a t d tarifEs by industry and year (see Section 1.7 for a data appen-

dix). With this rich database, we estimate the highly non-linear model. A panel data approach is applied to address the importance of the country-pair fixeci effets (Hummels and Levinshon 1995). No other pape has incorporateci both t d s and a full set of country-pair fixeci dects. This is because no other study has had acceas to a detailed tarEpanel. Further, serious and obvious identification issues mise in modeis with b i l a t d distance and a full set of country-pair fixed &&S.

We show that the CES monopolistic

cornpetition implies restrictions on the estimating equation%functiond form that solve the identification problem.

Estimation of the non-iinear model reveals a number of interesthg results. When we take account of the country-pair k e d effects, we obtain estimates of the elasticity of

substitution across varieties and the impact of trade costs that are smaller than usually reporteci. In a world of already low tarins and distances among major trading partners, the smder estimates meet the pnor expectations for the parameters while stiU accounting for the Large missing trade between actual and predicted data? The estimates imply that the elimhation of t d s would create more trade for poor countries (8%) than for richer countries (3%). They also imply that tarifE elimination would divert trade away from continentel preférentid treding are85 (e.g., European Union and NAFTA) and torwards inter-continental trading partners. That is, tarB liberalization would shift trade from the rich to the poor and from the local to the global.

In the second part of the paper, we move from bilaterai t r d e predictions to predicting the short-nin trade and production decisions of multinationals. It is traditional in *~hereis a debata as to whether the elasticity of substitution (o)should be smdl as many ampirical studies estimate, or big as most of the theoreticai Litenrture assume (Goldsteh and Khan 1985). To reconile the large misabg trade observed in the data and the fact that tarifEo and àistances among major trading partners (OECD,Ewopean Union, and NAFTA) are Inr,a andlot trsnsportation mts have to be tmeammbly Luge. Taking care of the country-pair Gred eftects largely solves the puzzle.

the literature to debate about whether trade and multinational production are comple-

m a t s or substitutes, Le., whether multinational production is for the domestic market or whether it is for export. Clearly, both elements are at play. It depends on whether one is looking at foreign direct invatment (FDI) directecl at rich countries or poor countries. The wual method for considering FDI is to postdate a regression of trade on FDI and use a positive coefncient as evidence of complementarity- A negative CO&cients is used as evidence of substitutability. Alternatively,

FDI is regressed on traàe

and similar inferences are ciram. Occasionally, some authors recognize the problem of endogeneity and use instruments (e.g., Brainard 1997). We use a radicaily different a p proach. We rigorously derive the production and export equations implied by our model

of rnultinationals and monopolistic cornpetition. These equations share common parameters and, one suspects, share cornmon omitted factors that result fiom the possibility that even oiir augrnented monopolistic cornpetition model is not a complete picture of

all the deterxninants. We t hen estimate these two equations simultaneously, recognizing the cross-equation restrictions on parameters and error terms. The data set combines the database used in the trade equation with data fiom the Bureau of Economics Analpis (BEA) of the U.S. Department of Commerce. We only work with U.S. multinationals due to data availability. Estimation yields some important results. It suggests that the elimination of teriffs, especiaily in poorer countries, would substantidy increase U.S. exports to these countries

(12% on average with mbstantial regional variation), but would have only a very small negative &èct on U.S. production abroad. Etimation &O yields one result that at first

seems unexpected: we over-predict U.S. production in small countries. However, this prediction is consistent with Brainard (1993a), Hallward-Driemeier (1996), and Soboleva (1997) who find that such countries will have large country-specific h e d costs and none of the externalities associatecl with having rnany multinationals.

The next section develops the rudimentary theoretical framework. Section 1.3 d* scribes data sources. Section 1.4 presents estimates of the bilateral trade equation. Sec-

tion 1.5 extends the modd to obtain both biiateral trade and multinational production equations, and then estimates these equations simultaneously by exploring the crossequation restnctions. We conclude in Section 1.6.

1.2 The Theory 1.2.1

The Consumer's Problem

To emphasize the importance of multinational activities, we consider a mode1 of trade that includes mdtinationals. Since all firms are potentidy multinationals, for notational simplicity we refer to each firm as a multinational even if it only produces in one market.

Each multinationai sets up plants in different countries depending on its objectivee and the host countries' conditions. Products made in different coiuitfies by plants of the same mdtinational are treated as different varieties because of consumers perceptions

and quality control standards of dinerent countries. Consumers have Cobb-Douglas preferences over goods and CES preferences over mrieties. With CobbDouglas preferences we can look at one good at a t h e . Fix the good and mppress the goods index. In the first stage a consumer in country i allocates Y, to the good in question. In the second stage the country i representative consumer

r n d z e s the CES subutility function3 subject to the expenditure constraint:

3The CES subutility hction impiies consumers w i ü have demard for al1 varieties. It f a to d e c t the empiricai evidence that wnsumers actually purchase a d e r set of varietieis than what is available, Tbis &y be one cause of hiissing tradde'.

N is the number of countries; Cik is the set of varieties produced in country k; &t

is

country i's demand for a variety produced in country k by the j-th multinational; P$ is the price associateci with

q?'; and p

=1

- O, where a is the elasticity of substitution

between varieties (a> 1). A key parameter to be estimated is o.

Country 3s demand for the variety produced by the j-th multinational in country k is given by

Gi = O if j 4 Cik (Le., if the varie@ is not produced). Otherwise, if j E Rk then

for i = 1,...N,and 5 = 1,...N. When the number of varieties is large, in equilibrium a is both the elasticity of demand and the elasticity of substitution between varietied

1.2.2

The Producer9sProblem

Each multinational j produces differentiated products in dinerent countries. Following the conventionel setup, a firm incurs (a) h-specific nxed costs sitch as R&D, (b) plantlevel fixeci costs, and (c) market-specific muginai cost MC(&) where Xk is B vector of

Xi includes country k's wage rates, labor quaüty and 0thfeatures of the production environment. The firm first decides determines whether to set up a plant in country k. This decision depends on the cost fectors in (a), (b), and (c). We

cost variables in market k.

take this decision as fixecl and do not try to explain it. See Hdward-Driemeier (1996),

Chakrabarti (1997), and Soboleva (1997) for recent discussions of location decisions by multinationd finna. M e a d , we focus on multinationals' short-run production decisions. Once a plant is set up in country k, the h ' s fixed costs are irrdevant to price and quantity decisiow. The problem becomes that of choosing consumer price pFi and output &t

in each market so as to rnaxunize groes revenues over variable costs. Given consumer

price p$, the producer eams revenues tt d(Dt)

pf,

&t

where t t is 1 minus the ad

4For rigour, we shwld integrate over varieties in equations (1.1) and (1.2). Bowewr, when we to estimation, the mmmation wili maice %gs

cleare!r.

u a l o m tariff imposeci by country i on goods imported from country le, and d ( D t ) is the transport cost as a function of the bilateral distance D! between countries i and k. Note that

$5 1 with gk=

with d ( 0 ) = 1

1 when 2=k. Also, d

(0)

is a decreasing huiction of bilateral distance

.

Special attention should be given to d ( ~ 1 )Following . Brainard (lgMa), d(D,k) c a p

tiirer, the &ect of trade barriers that are not directly meastuable. It rdects the disadwtages of being distant from destination markets. Such disadvantages can include shipping costs, linguistic or cultural Merences, and slow responsiveness to changing consumer demanàs. Mathematically, it meam that for each unit of output reaching country

i, ( l / d ( D f )- 1) units is lost in the transaction pro ces^.^ Denoting nf as the gross profit of fum j s plant in country k, the b ' s problem is

The price that firm j charges on its country k product being sold to country i is solved ns k,

Pji

- (1 -t:$)-'hfC(xk) d(Df)

With no trade barriers and a constant marginal cost, this equation reduces to the pricing d e of the simple monopolistic competition mode1 (Helpman and Krugman 1985). In the present kamework, however, both trade barriers and a non-constant marginal cwrt

are incorporateci to determine the price of a product variety that hally reaches the importing country* Substituting this price back into (1.2) gives the quantily demandeci:

=Tor many pmducts, physicai shipping mts msy be d, but the cost associateci with gatheriog information about the demand amditions in düttant markets or dealhg with remote merchruits may be significantn (Hanson 1997).

Notice that output and price vary across plants. This is very different kom the usual monopolistic cornpetition results.

1.2.3

Bilateral made Equation

Ushg the price and quantity-dernanded equations, the value of country i's imports from

country k is derived as the sum across d e t i e s of country 2s demand for varieties p r e duced in country k

Where J(k) is the number of plants producing in country k. The number of plants includes those of both domestic and foreign h.

With Factor Price Equalization (FPE)and no trade barriers, equation (1.6) becornes

Mf = [J(k)/c$,, ~ ( g -K. ) ] That is,country 2's imports fiom country k are proportional to country 2's industriai expenditure with the proportion being country k's share of

plants in the world as a whole. This prediction is very similar to the symmetric model of Helpman and Kiugman (1985). Although central to the model, a does not show up in the symmetric version of the trade equation and thus is not identifid without international asymmetnes.

In equation (1.6), trade bartiers and marginal costs for dI corntries afFect imports for each country. The underlying parameters, including a,can be estimateci by relating the variation in trade patterns to variation in marginal costs, terins, and trade barriers. Specificdy, we have:

(1)captures the rolea of production cost; (2) and (3) capture the negative &ects of trade barriers. The derivatives have the expected signs. Variables related to countries other

than the trading partners &O enter the bilateral trade equation, but their &ts

are l e s

straightforward.

1.3 The Data Much of the data used in next section are the same as those used in nex chapter. Specificdy, the original data on trade flows is from Statistics Canada's World Thde Datebase. Gross industrial output is £rom UNIDO'S INDSTAT database. Income and wage rates

corne from the Penn World Table (see Summers and Heston 1991). Data on endowments, distance, language, country adjacency, and the c.i.f-f.0.b factor are from Antweiler (1996).

The expenditures on mmufacturing goods Yi are calculatecl as rnmuf~u:turuigoutput plus net imports. Bilaterd taxiff data by industry and year have been casefdly compileci from different sources and took 8 months of full-tirne work to construct. See Section 1.7 for details of a the data. The database represents a major hiprovernent on the types of tarB data previously used and allows us to work with a panel. To our knowIdge, no other paper has incorporateci both terins and a full set of country-pair fixeci eff'ts. This is because no other study has had access to a detailed tariff panel.

For the purpose of this paper, we draw data on the number of plants fiom the Yeorbook of Indwtnd Statistics and the International Yearbook of Industrial StatisticsCAU

estimation is at the level of aggregate rnanuf~cturing.Ih next chapter we will estimate

a based on both aggregate and more disaggregate data. Their r d t s are consistent with the interpretation that the a for aggregate manufacturing is an aggregate of the a fiom individual manuf8c-g

indushies. This means that the adpis of aggregate

mmufacturing industry is suffident to explore the key economic interpretations of our

model.

The y e m for which the trade equation was estirnated include 1980, 1984, 1988, and 1992. One reason for not using every year of data is that one would then need to

cmefully model serial correlation. In next chapter we will show that serial correlation is not a probkm with longer-spaced intervals such as ours provideci one use fixeci effects. Limitations on data for bilateral tariffs and the number of mmufacturing plants restrict the data to 34 countries, see table 1.1 for a List of countries. The countries

include most of the OECD,meny of the developing countries intensively engaged in trade, and some developing countries that axe not heavy traders.

1.4 Bilateral Trade Equation In this section we will estimate the structural parameters directly fiom equation (1.6). Taking logarithms, adding time subscripts, and an error t m , equation (1.6) becomes

The first two terms capture the effects of country size: the industrial supply of the exporter and the industrial demand of the importer. The third term captures intern* tional differences in production costs. Because the production cost dinerences can also

be vieweci as consequences of relative e n d m e n t ciifference among countries, the third term c m ais0 be interpreted as capturing the role of endowments for trade detemination.

The fourth and fiRh ternül capture the &ects of bilateral tari& and &tance barriers. The nnal term is the log murmation of some highly nonlinear tenns related to variables

of ali countries. This final term cornes from the denorninator of the CES price index. It couects factors that influence muitinationals' pricing decisions such as truiffs, production costs, and trade costs. Except our next chapter, no other study has exploreci this term

MC(*) = d(*) = 1, thus ignoring the role = d ( - ) = 1, of production costs and distance for explainhg trade. By assuming MC(*) M y . Fiutha, o u next chapter will asme

Chapter 2 will significantly aimpify equation (1.7). Equation (1.7) is a generalization of two types of apecifications used in the empirical literature. The first specification explains trade by country size variables and attributes all other trade determinants to the residual (Helpman 1987, Hummels and Levinsohn

1995, Hanigan 1996, and Jensen 1996). The second specification adds bilateral trade barriers (Hanigan 1993, Haveman et. a1 1996). Note that neither of these papers use

panel data. Equation (1.7) ad& the marginal cost of the exporthg country and a highly non-linear pricing term term. The notable feature of this equation is that it is directly derived fiom the theory of monopolistic cornpetition. Following Brainard (l993a), we set d ( D f ) = ë p o : , which has the required properties d ( 0 ) = 1 and d'< O. The parameter p is interpreted as the trade cost per unit of

distance. To measure marginal costs, we use wage rates adjusteci by labor productivity where labor productivity is measured by PPP-adjustecl GDP per worker. Specifically

wXk,~~= , d (w&)* (1-7), we have

~ h e r ey;,g,,=

wk

t

-

O D ~ , ~ & ~ , ~Substituthg

d ( = )and MC(*) into

We will first estimate parameters a = (1 - a)a,& = (1 - u)p, and ,& = o - 1. Then

we will infer the structural parameters a =

a - 1, p = -g, and

Q

= -a Bs-

No previous paper has estirnatecl all 3 parameters (c,p, and a). G r a m type r e gressions are norrnally used to estimate a parameter related to distance, which is simiiar to our &. In those regressions, o and p cannot be disentangleci to distinguish between

the trade cost &ect and the substitution efkct. Hamigan (1993) and our next chapter estimate o by introducing tarins into trade determination. However, they assume that p = O. By intmducing marginal costs and distancerelatecl trade costs in Bddition to

our specification allows for the estimation of a and p as well as a.

ta*,

We apply the SAS/IML Newton-Raphson algorithm to obtain maximum likeiihood

-

estimates based on the followixig alternative assumptions on the error tem: (2)

Busic model: et, = v$ and v t ,

N(O,6).

(ii) Fked effects model: et, = ~f + v:, and

-

N(0, b2).

Table 1.2 reports the results of both specifications. The top panel lists estimates of

Pi, &, and P3 and their standard m o n .

Both specificstions give the expected signs.

However, estimates from fixed effects model are smaller (in absolute value).

The nrSt

two parameters capture the dects of cost disadvantage on trade; they are a combination

of cost factors (a, p ) and substitutability (a).The third term

=o

- 1, is directly

related to the elasticity of substitution. The structural parameters a,p, and O are listed in the middle panel of table 1.2.

The estimates of a are 6.424 for the basic model and 3.989 for the fixeci effects model. These numbers M e r fiom those of many earlier studies, which are either very big or very smaii.

The esthated p, 0.060 and 0.048 fmm the two models respectively, both

imply significant distance effects. With distance scaled to 1000 miles per unit, a p of 0.048 irnpîies that 4.8% of a product's value is lost per thousand mile8 of shipping. The

estimates of a! dina signXcant1y across the two models: although both have the expected sign, the fixeci &ects estimate of or is insigniscant whereas the basic model estimate of a has t-statistics of 24.3,

The correlation between log(Mf) and its prediction, excluding the nxed effects, is between 0.750 and 0.781.

This is pleasantly high. Figure 1.1 plots log(^&) agaiast its

prediction using the fixed efEects model (The prediction excludes the &rd &wts term.) .

The high conelation of 0.75 is refiected in the plot. Overall, the predicted log(Met) is slightly biased upwards and more points lie above the 45-degree line than below. In particular, the model fits best for observations with larger bilateral trade flows.

1.4.1

The Fixed Effects

Many studies have emphasized the importance of country-pair &ceci effects that for c a p turing llllmeasured explanatory variables. Hummels and Levinshon (1995) provide one

of the most prominent examples in the trade literature. Although the specification in (1.8) has rigorously incorporated tariffs and distance in a general framework, a nurnber of potential explanatory variables are still excluded. Examples include n o n - t d barriers

(NTBs), language and adjacency diimmies, and c.i.f-f.0.b factors. The likelihood ratio test statistic (LRT)for the null of zero fixed effects is 7970, which indicates that including the fixed efFects is cruciai for understanding the sarnple variation. To further understand the fixed effects, we calculate the residuals fÎom both models

and seperately regress these residuals on a series of variables that may be attributed to the unexplained fixeci effects. We reelize that i t is only correct to do so when these variables are orthogonal to the variables already included in the model. As such, the approach is informal and the result needs to be interpreted with caution. These variables include trade barriers of different forms, variables used in gravie-me regressiom, the monopolistic cornpetition model, and the fador endowxnents model. Rom table 1.3, it is obvious that different sets of regressors have signiscant explanatory power for the Basic

model residuals. Table 1.4, however, shows that these same regressors have almost no explanatory power for the residuals of the fixed effects model. These results add tu our evidence that the fixecl effkcts model outperforms the Basic model. By choosing the h e d effects model as our pref'ed model, we obtain a smaller estimate of the elasticity of

substitution, a smailer per unit distance cost and an insimcant wage effectd 6TheiiiPigui6cant wage d e c t may be due to the paasiiility thst wia does not M y capture mginai

The smaller parameters estimated for the fixeci effects model have an interesthg explanation. The actual world trade volume is far less than predicted by the simple monopolistic cornpetition trade model (Helprnan and Krugman 1985) and the gravity type equation (bIcCallum 1995, W ei 1996). Given the slready low tarins among major

trading couutries and the short bilateral distances within major preferential trading areas

(EU and NAFTA), this gap c m be explaineci by either a large distance &ect (big p ) , a large marginal cost &et (big a),a large substitution effect (big O ) , andlor a large

unmeasured aects (fixeci efkcts). Without considering the fixed effects, a very large a, p, and a,or a combination of these three is needed to explain the large missing trade.

Accounting for the fixeci efïects allows these three parameters to be smaller, but stiil consistent with the real data. That is, although the mexplained fixecl effects account for a luge portion of the trade variation, the Basic mode1 explains another important aspect with much smaller set of parameters.

1.4.2

Influential Observations

Table 1.5 preaents sensitivity results for the futed &écts model. For each specification, the estimation of the three structural parameters and their standard mors are reported in one row. The 'Baseline' row c a d e s over results from table 1.2. The 'Domestic Shipments' row reports the results when observations with i = k are omitted. Clearly, these are not infiuential observations. The remaining rows omit all observations related to the indicated country. In generd, the estimation does not change signincantly when any single country is dropped. Two exceptions are the United States and Japan. When either is excludeci, p

shrinks to its minimum of egaentialiy zero and o rises to its maximum of up to 5.2. -

costa. ËmpiricalIy, it is âifIicuit to calculate an internationally comparable marpiaal oost (aee Cbalnabarti 1997). Our hope is that certain orthogonaiity conditions will ensure the consistmcy of other &ted parameters despite the patentid mecrsurement m o r s in marghd costs. In an eariier version, we used wage rates adjusted by total factor productivity (TFP),where TFP was calculateci in the spirit of Coe and Heipman (1995). However, there is no guarantee that this cboiœ îs bettes than the current specification due to the cWicuIty of caiculating internationally comparable TFP I d . In Coe and Helprnan (1995), the & r d rate of TFP is used.

When the United Kingdom is omitted p rises to its maximum and a falls to its minimum. By implication, o is not stable across corntries. Specificaily, Japan and United States have inelastic preferences (osmall) and high trade costs ( p large). In contrast, the United Kingdom has elastic preferences and low trade costs.

This may be due to these

countries' important role in international trade and the fact that Japan and United States

are located fat away from theh major trading partners. Deleting either of them throws out important sample variation relatiny; trade to distance.

1.4.3 Econornic Implications Focusing on the explainable part of the model, we c m quant*

the amount by the

trade barriers have restricted the volume of world trade. In the non-linear setting, the percentage change due to the hypothetical dimination of existing tarîfb in any time

period t is calcuiated as:

Tariffefect =

k k (C ~ [ ~ t ~= ; lO]t -&C E[MiVl lti,t > O])/ C E[Mi,tlti,, > O]. ik ik ik k

k

(1.9)

The distance effect and the cornbineci tariff and distance effects are defined similarly. Ftom the bottom of table 1.2, the tarin effect is 3.7%, which meam that the 1992

level of t

d reduced bilateral trade by 33%. The distance &ect is 23.3%, meaning

that if all distance between countries were eliminated bilateral trade would rise by 23.3%.

The combined effect of tariffs and distance is 25.7%. These numbers show that distancerelated transportation costs are a greater impedimeat to trade than are tariffs. As the distancerelateci transportation costs are iinlikely to be eliminated in the short run,these

numbers imply serious limits on the impact of trade policy on world trade. World-wide numbers disguise large regional variations in the &ect of trade policy.

Grouping countries by OECD and non-OECD status, most existing trade is among OECD countries and the highest tari& are among non-OECDcountries. As a result, the impact of tmifb on trade within OECD countries is likely less than 3.7% whereas the impact of

tari&s on trade between non-OECD countries likely exceeds 3.7%. This is boni out by table 1.6. The estimateci t

a &ect is only 3% for trade within OECD countries and

a much biggec 8% for trade among non-OECD countries. Indeed, tarins have reduced non-OECD imports from OECD countries by a very large 12%. See table 1.6 for more

deteils. Combining this with NTBs, which by omission appear in the fixed &ts,

the

potential benefits of increased trade iiberaiization by non-OECD coutries Likdy remain large.

Table 1.7 investigates huther. Thde volumes within the EU block and within the

NAFTA block actuaiiy demase with tarin liberalization. Ail other trade volumes rise. It even nses between Europe and North America despite the low level of 1992 t d . The distance aects on trade distribution are aven more ciramatic. See table 1.7. In summary, although the trade liberalization effect on overd world trade is not large, its distribution is skewed: it shifts trade hom rich countries to poor countries and from local preferential

trading areas to global inter-continental trade.

1.5 n a d e and Multinational Production This section is about the short-run trade and production of multinationals. We will extend the model in section 1.2 to obtain both bilateral tracle and multinational pro-

duction equations. Combinuig data on multinational activitiea and data used in section 1.4, we will then estimate these equations sirnultaneously by expIoMng the cross-equation

restrictions and discuss the implications of the results.

1.5.1

Bilateral made and Multinational Production Equations

The presence of multinational firms in section 1.2 d o m us to explore the connection between trade end multinational production. Our andysis will be based on U.S. multination& operating in the United States and abroad. Using eqyations (1.4) and (1.5), the aggregate production of U.S.multinationals in country k is

Where

S2r is defineci as the set of varieties produced by US. multinationals in country

k, J,(k)

is the number of U.S. multinational plants producing in country k, and 0 is

the vector of parameters. This equation irnpIies that production by U.S. multinationals alredy operating depends on (1) the host country's merginal cost MC(Xk), (2) the

industrial expenditure

Yi, and

(3) trade barriers tf and d ( ~ f ) .For each importer i,

the average trade barners it imposes on al1 countries can be treated as weight-average

[e](l-u) where J(~~')Mc(x~)('-@)

tanffs and distances in the form of @-, - ~(k')

MC(X i )

f . d ( ~ f)

is considered as the weight.

Rom equation (1.6) with k = US.and i = k, U.S. exports to country k me given by

Equations (1.10) and (1.11) fom the framework for evaluating the connection between

trade and multinational production. As trade and multinational production are inseparable components of international

commercial activities, we cannot discuss one without addressing the other. It is traditional to argue about whether trade and multinational production are complements or substitutes, Le., whether mdtinationais locate d o m ~ c a l l yin order to serve the d e

mestic market or in order to serve as an export platform. Clearly both elements are at play, with the extent depending on whether one is

looking at FDI into rich or poor

host countries. The usual method for addressing this begins with a regregsion of trade on FDI.A positive FDI coefficient ie seen as evidence of complementaxity between trade

and FDI. A negative codcient is seen as evidence of substutability betwm trade and investment (e.g., Goldberg and Klein 1997). Alternatively, FDI is regressed on trade and

simiiar conclusions drawn* Only Brainard (1997)recognizes the problem of endogeneity.

She instruments gross exports with next export in estimation containhg U.S. outward m a t e sales and exports equations.

A radically different approach is to recognize that both FDI and trade corne out of a single larger model such as the modd of section 1.2. Hence both are driven by a set of underlying or ultimate causes that are the exogenous variables of the larger section

1.2 model. Manipulating equations (1.10)and

by taking logarithms, adding t h e

subscripts, and error terms, we have

Where

Equations (1.12)and (1.13)are linked in two ways. First, they share a cornmon set of parameters O. Second, our augmentecl monopolistic cornpetition model does not completely incorporate ail the determinants of trade and multinational production. For example,it exdudes ttnme88u~edNTBs, language and adjacency dummies, exchAnge rate vatiabïlity,

political risk and other omitted determinants of trade and FDI. These omitted deter-

minents are arguably common to both equations, and are implicitly attributed to the residuals q k , t and ezk,t The existence of common omitted variables implies that the two error terms will be correlated. Recognizing the cross-equation restrictions on parameters and error terms, we esti-

mate the two equations simultaneously. Clearly, the joint estimation will improve efficiency compareci to single equation estimations. M h e r , it will yield structural par*

meters related to behavioral relationship. In contrast, regressing trade on FDI yields no structural parameters, only ambiguous reduced-form parameters.

We use the maximum likelihood estimates. The log likelihood function c m be written as

where N represents the number of coutries and T represents number of time periods. Assiiming the same hinction forms of trade costs as in section 1.4, we will estimate a,p,

a,6f,62,and p using the SAS/IML Newton-Raphson algorithm.

1.5.2

The Data

Data on U.S. ovemeas multinational activitiea cornes hom the Bureau of Economics Analysis (BEA). Spedcdly, US. multinational production data are kindly provided by Raymond Matdoni at

BEA.Niunber of U.S. overseas plants cornes fÎom BEA pubLished

tables Selected Dota for Fowign Afiliates in A11 Countries in Which Investment Wa9 Re-

ported Mthough more disaggregate data exist,the consideration of protecting individuBi

tirms' confidential information allows the BEA to pubüsh data only at a more aggregate level. While the availability of firm Ievel data allows one to address problems such as the firms' dynamic investment and trade decisions (a-g., Hallward-Driemeier 1996), only

published data are available to us? We use aggregate mandàcturing data and combine them with the variabla describeci in section 1.2. The years used are 1984, 1986, 1988, 1990, and 1992. The availability of BEA data d o m us to focus only on this time period.

The reasom for using data on discrete years and totai manufwturing industry are similar to those describecl in section 1.3. Table 1.8 lists the 31 countries in the BEA sample.

1.5.3

Estimation Results

Table 1.9 praents the maximum likelihood estimates. A large negative a (-2.04) may at first seem unexpected. It implies that the United States will produce more in countries with higher wage rates. This is inconsistent with cost minimization. However, it is con-

sistent with Brainard (1993a, 1997), who emphasizes the transportation costs rather than fector price clifkence as a determinant of multinational production; it is also consistent with conventional wisdom that rich countries invest more in other rich countries than in

poor countries. However, we will not exclude the possibility of misgpecification in the cost huiction and the smder sample8 compared to the sample used in the bilateral trade

equation. With the sample being limiteci to the US. and its trading partners, both p and a as estimatecl are srnaller (0.024 and 2.85 respectively) compared to the reaults from

larger sample used in the bilateral trade equation. The smailer p and a cire consistent with the findings in table 1.2. In particder, the srnail p consistent with the fact that (1) the U.S. is fa-away fiom most of its major trading partners, (2) on average U.S.exports

more to its trading partners thnn bilateral trade conducted among all the 34 countries of the larger sample, and (3) the US. as an exporter may have above average technology in transportation. As the U.S. exports more to non-OECD countries than many of the OECD coutries do (EU countries mostly t r d e amongst themselves), the smder a

may reflect the fact that for each manufactturing product, the receiving countries of US. ?We were told that it is possible to work on the BEA h level data, but one must be a U.S. citizen to do m. *Thissample is much smaller thau the sample used in section 1.4 becauae it b restrictedto U.S. and its trading partmm-

exports on average have fewer varieties than those of average countries in the bilateral trade equations. As a result, these varieties are also less substitutable for this former group of countries. Table 1.10 presents the sensitivity results of the crosgequation constraints. It is shown to be robust for most cases when observations related to one of the countries

are deleted. The only exception is that when India is dropped, a becomes significantly smder (in absolute d u e ) and a becomes significantly larger. This rnay reflect the fact that India has relatively low wages, less multinational production, and much higher t

6

than those of other coutries in the sample. However, it may serve as a representative of countnes with these characteristics (e.g., less developed countries in Asia and Afiica) so it is important to include it in the regrecsion.

Figure 1.2 plots log(XWVt(k)) against predicted l~g(X,,~(k))and Figure 1.3 plots 10g(Q,,~ (k)) against the predicted log(Q,,l (k) ) . The high correlations 0.8 and 0.9 are fits remarkably well, rdected in the plots. While Figure 1.2 shows that 1og(XwVt(k)) Figure 1.3 shows that, except for the US. observations, larger l~g(Q,,~(k))is slightly under-predicted but smaller l~g(Q,,~(k)is slightly over-predicted. The fiRt guess is that the inclusion of the U.S. as a host country in the regression may bias our estimation. However , dropping the U.S . observations does not significantly change the estimation

and plot. This is consistent with the arguments related to the country-apecific fixed costs requirement (Brainard lW?),sunk cost and extemaliw factors (Hallward-Driemeier 1997,

Soboleva 1997) in plant choice and continuhg investment decisions. S m d countries are nomaily in an adverse position in these respects. Despite the bias, one may argue that visually Figure 1.3 still fits quite weli. However, it is worth mentionhg that the data have been scaled by logarithmn. S m d clifferences

from the logatithrnic plot may translate to much larger differences in level. This phenom-

enon may be related to a problern reporteci in in our next chapter, where the estimatecl o is unreasonably large and many 'outliers' are revealed when estimation is in levd.

Fbrther research on this aspect is warranteci.

The estimation dows us to evaluate the relative effects of trade poücy on trade and multinational production. As trade policy variables vary, US. overseas multinational production can either substitute, complement, or be independent of U.S. exports to its trading partners. The substitution scenaxio is consistent with Brainard (1997) and Grossman and Helpman (1994) that multinational production arises from incentives for

fums to jump o v e trade barriers; the complementary scenario implies that both trade

and multinational production increase when trade barriers fa& and the world becomes more integrated as trade becomes more liberalid.

For the predictable part of the model, table 1.9 shows that in 1992, as bilateral tarins between the

US. and its trading partners were eliminated, U.S. total exports increBsed

by 13%. On the other hand, total U.S. overseea multinational productions decreased by 1.2%. The hypothetical elimination of distance related trade costs had similar effects,

with the magnitude being much larger (38.6% and -6.7%). Table 1.11 reports the same effects at the host countries' level. With the elimination of the taxi&, the effects of

U.S.

exports on host countries are straightforward: the higher the pre-liberalization tarin, the higher the percentage increase in post-liberalization exports. Two extrerne cases are

India (222%) and Hong Kong (0%). On the other hand, the effects on U.S. multinational production in host c o u n t h are mixed: Hong Kong, Indonesia, Portugal actudy increase for 1%; Austria, Spain, Greece, Italy, Korea, and New Zealand unchanged; the rest of

the host countries receive 1% to 3% less of U.S. multinational productions during the post-liberahetion period relative to the pre-liberalization period. The distance d'ts

are similar, with the magnitude being much larger. Figure 1.4 and Figure 1.5 simulate the gradua1 reduction dects of 1992 tarif63 and distance related trade costs on aggregate U.S. exports and US. ovmeas muitinationai production. These figures show that as tarin and distance barriers reduce step by step,

US. overseas production decreases slowly but U.S.exports inme-

rapidly.

Effects of the unpredictable part of the model c m be partidy implied by the pars meter p. Its estimator is negative and insigniflcant (-0.074 with standard error of 0.088).

These estimators mean that the dects of excluded factors sffecting US. exports have weak negative correlation with those of excluded factors &ecting US. overseas muitinational productions. This is similar to the &ects of the predictable part of the model. Combining both e f k t s , this model predicts that factors affëcting multinational production are negativdy but weakly correlated with the factors &ecting trade. Focusing on the model part and reaIizing the orthogonal sssumption we imposeci earlier, we conclude that the tarifï liberalization effort introduced by WTO may increase US. exports sigmûcantly

and with only a s m d decrease, if any, on U.S.overseas multinational production. We want to ernphasize that the number of multinational plants is assumeci to be fixed.

The model

does not account for the multinationals' long-nin decision on reallocating

plants in response to trade policy change. Therefore the trade policy effects we obtained

in this section and the previous section are short-nin effects.

1.6 Conclusions In this paper, we dweloped a model of international trade that tekes into account multinationah, asymmetric trade barriers, and international dinerences in production costs.

The model implies highiy non-linear bilaterai trade and multinational production equ* t ions. The empirical resdts based on this stmctural model dowed us to estimate the un-

derlying parameters. Thia is a significant contribution to the literature: previous studies placed zero restrictions on at least some of theae parameters. T&g

account of country-

pair fixecl effects leads to smaller estimates of parameters related to trade costs and the eiasticity of substitution. These smder and more reaaonable parsmeters, combineci with country-pair fixed dects, are consistent with large missing trade in a world of low tan&9

and small distances among major trading pBStLler~. Estimates of the stmctural parameters dlowed us to assess the impact of the 1992 worldwide t

d structure. Elimuiation of these tari& wodd raise world trade by 3.7%.

W h e r , the d e c t is skewed. For example, non-OECD imports from OECD countries would rise by 12%. More generally, trade liberalkation would shift trade from rich countries to poor countnes and from local preferentiai trading areas to inter-continental trading partners. Within a d e d framework, we explored the interdependence between trade and multinational production in the form of cross-equation restrictions in parameters and error terms. The results specific to the United States and its trading partners show that short-nui trade Liberalization effects on US. exports and U.S. overseas multinational production are weakly opposite. The hypotheticd elimination of bilateral t& between the United States and its trading partners siflcantly increases U.S. exports but insignificantly reduces its overseas multinational production.

We must emphasize that we focused on h' short-run production decisions. We did not mode1 firms' long-nui plant docation decisions. As a result, the trade liberalkation effects we estimated are short-=

efFects. Our future research will complement the

findings of this papa by &O exarnining long-nui dects.

1.7 Appendix. Panel Data on Tariffs Data Sources: (A). llNCTAl? TRade Anolysis and Infornation System (TRAINS).

This is the most comprehensive inventory of bilaterel t a a s available. It includes bilateral tari& and trade at &digit to l W g i t Harmonized TariE System (HS) levels for many developed and developing countries. For 1992, it covers the 36 comtries used in

this paper.

(B). GATT Tanf Study for the 14 OECD Countries.

The GATT TarjfE Study data are adable for 1979,1983,and 1987. The data are fiom the ssme source as used in Tkder (1993). United Ncrtions Confmce o n Thde and Devefopment

29

(C). Indicutors of Turiff and Non-tariff Bade Bamkrs. Data fiom this OECD CD-Rom includes 1988 and 1992 average tarifEs for 6 OECD countnes by importer and %digit ISIC. These 6 countries are Sweden, Austria, Spain, Portugal, Australia, and New Zealand.

(D).The Michigan Mode1 of World Production and %de:

Themy and Applications

(Deanlu# and Stem 1986). This book contains 1972 and 1979 average tariff data for 17 OECD countries. It includes al1 those coutries in (B) except Greece. It ais0 includes Sweden, Austria, Austraüa, and New Zealand. It is organized by importer and $digit ISIC (Revision 2) industry. These data are originaily fiom an esriy version of the GATT tarifE study.

(E). UNCTAD Dirrctory of Import R e m e s (DIR).

This UNCTAD publication includes average tariff rates for 16 non-OECD countriedO. The data are available for various years between 1980 and 1993, depending on the country. It is organized by importer and an industry classification system that is close to Sdigit

ISIC (Revision 2). (F). Government Finance Statlistica Yearbook (GFS) and Intemational Finance Sta-

tbtics Yearbook (IFS). These publications contain data on import duties and total imports by country Erom 1972 to 1992, for all36 countries. We calculateci teriff rates as the ratio of import duties to total imports. There is no industry dimension.

Four Steps in the Construction of the Tarifï Panel Data Set:

(1). Constmct 1992 bilateral t d rates by importer, exporter, and 3-digit ISIC (Revision 2) industries for d 36 countnes: Using converters from Statistic Canada, we converted the TRALNS data £rom Wgit HS to Pdigit ISIC (Revision 3) and then to 3-digit ISIC (Revision 2). We then computed 1992 biiateral tariff rates by importer, ex-

porter, and a g i t ISIC (Revision 2) industry. This is consistent with the data stmcture l0The 16 ~ < m t r i e sare Argentins, B r d , Chile, Ecdadot, Hong Kong, hdonesia, kidia, Korea, Malaysia, Mexico, Morocco, Singepore, Sri Lada, TbaiIand, ninisia, Venmela

that rrlready existeci in Antweilar and Thder (1997) database. (2). Constnict the panel data set for the 14 OECD countries: combining data in (1)

and (B), we obtain bilateral tarifb in 1979, 1983, 1987, and 1992 for these 14 importem. Data between these pars are interpolated. With 1979 bilateral tari&, we extrapolate 1972 tariffs for 13 countries (excluding Greece) using a ratio based on DeardorfF and

Stem data set on 1972 and 1979.11 For Greece, we use GFS and IFS data between 1972

and 1979 to f o m this ratio. Then we interpolate the tariffs between 1972 and 1979. (3). Constmct the panel data set for Sweden, Austria, Spain, Portugal, Australia, and New Zealand as importers: for each importer and industry, ftst we compute a thne series variable representing an ratio of tariffs data of different years relative to 1992, then we use this variable and 1992 crosssectional data to extrapolate tariffs before 1992. The

ratio is defineci similady as in (2). This variable is obtained by first interpolating for the periods when OECD data are available, and then by extrapolating to earlier y e m using Deardorff and Stern as well as GFS and IFS data sets. (4). Construct the panel data set for the 16 poorer countries. for each country and induatry, we compute a t h e series variable representing an ratio of tarin data of different years relative to 1992, then we use this variable and 1992 cross-sectionel data to extrapolate tarifi More 1992. The ratio is defineci similarly as in (2). This variable is obtained by first interpolating for the periods when DIR data are available, and then by extrapolating to earlier years using GFS and IFS data sets. (5). Combining (2), (3), and (4), we obtain panel data set on t a d b by year, ISIC,

importer, and exporter.

Tabie 1.1. The 34 Countries in the BilateralTmde Equation Europe Belgium Gemany Denmark

Finland Greece Ireland

United Kingdom Austria

Spain

Norlh America USA Canada Mexico

Asia Hong Kong lndonesia lndia Japan Kotea Malaysia Singapore Sri Lanka

Other

Australia New Zealand Brazil Chile Ecuador Morocco Venezuela Tunisia

.

93 g-geg$ o q c c 5wo ooc

Tabte 1.3. ResMual Analysis of the Basic Model

Trade Barriers

Mode1

Model

Model

Model -5.379

i Intempt -5.118

-1.100

AD+

1.401 (0.135) O. 594 (0.081)

Tl

4.257 (O. 153) -15,345 (1.103) 0.634 (0.020) 4164

CIFI

DlST4

MBER OF OBSERVATION

1.353 (0.136) 0,545 (0.081) -0.749 (O.173) -11.486 (1.305) 0.625 (0.029) 4164

1,437 (O.136) 0.600 (0.081) -0.332 (0.153) -15.980 (1.116) 0.645 (0.029) 4164

(093) 9.805 (2,286) 1,431 (0.135) 0.656 (0.082) -0.151 (O. 153) -15.990 (1.13) 0.665 (0.031) 4164

the GDP share of a country in the world. (5), HU is the human capital. (6). K is capital. (7).ADJ and LANG are adjacency and ianguage dummies respectively. (8). CIÇ is the average CIF factor betwean trading prbiers.

Table 1.4, ResidualAnalysk of the F W Effects Model

(O. 106)

(O. 0230) 0.008

0,036 (O.125) 4.002 (0.950) 0.016 (0.056) 0.013

DISTy

MBER OF OBSERVATION

(0.034) 0,173

(0.034) 0.018

(O.034) O.183

(0.458) 0.008 (0.012) 4164

(O. 539) -0.007

(0.464) 0.007

(0.012) 4164

(0.092) 4164

(0.013) 4164

F-STATISTICS 1.727 12,545 1,615 3,900 Note: (l), The Dependent Vaaiable is q,,fmrn Basic Model. (2), Y is GDP. (3),YC is pet capita GD?. (4), S is the GD? share of a country in the world. (5), HK b the humen capital. (6),K 1s capital. (7), ADJ and LANG are adjaamy and language dumnies respectively. (8). CIF 1s the average CIF factor between trading partnem.

Table 1.5. Sensitivity Results of the Bilaterai Trade Equaüm observations excluded*

a

standard emt

CI

standard emr

standard emlr 0.38 1

Domestic Shipments Ausûalia Austria b i gium Bratil Canada Chile Germany Denmark Ecuador Spain Finland United Kingdom Greece Hong Kong 1ndonesia India lreland itasc Japan

Korea Sri Lanka Morocco Mexico Malaysia Neaierlands Nonivay New Zeaiand Portug8l Singapom &mien Thaiiand Tunisia

USA Venezuela -0.001 0.040 0.055 No-: (1) "Basdine"rafars to the estimateci m u b (mm k e d emctr, mode1of tabie 2. (2) "Domsstitk shipment?refers to the case whem obs8wations with k k are orniW. (3)Each of the other rows refem to the casa when obsewatbns related to the conesponding country are excluded.

Table 1.8. The 31 Countriccr in aie US. Export and the U.S. MuMnational Pmduc(bn Equations Europe Belgium Gemany Denmarù

Finland Greece lrebnd l t ' Nether)atids Norway Untted Kingdom

Austria Porîugal Spain Sweden

North America U.S.A Canada Mexico

Asia Hong Kong lndonesia lndia Japan Korea Malaysia Singapore Thailand

Other Australia New Zealand Brazil Chiie Ecuador Venezuela

Table 1.9. Estimates of the Cross-Equation Restrictions

Number of Observations Lag likelihood Function ~Ort(lwl(&(k)), M ( X m ( k ) ) Co~(~~g(~us l ~( k g )( )Q~W s ( k ) ) Total Tariff effect on US Exports Total Tariff e f k t on US MultinationalProduct Total Distance efbct on US Exports Total Distance eftect on US Multinational P d

(0.088) 155 -166.7 0.805 0.905 0.134 4.012 0.386 -0.067

Table 1.IO. Sensitivity Results of Cross-Equations Restrictions observations excluded BaseIlne Australfa Austria Belgium Brazil Canada Chile Gennany Denmarù Ecuador Spain Finland United Kingdom Greece Hong Kong lndonesia lndia lreland Italy Japan

Korea Mexico Malaysia Nettieriands Norway New Zealand Portugal Singapore Sweden Thaiiand USA

standard etfor -2.037 0.075 a

p

standard

o

0.025

0.007

standard

ertor

emr 2.854

0.618

standard error 4.074 0.088 p

Table 1.1 1. Tariff and Distance Effects on the U S Exports and the U.S. Multinational Production

Countries

Tanff Efbcts on: I

Exporb Australia Austria Belgium Brazil Canada Chile Germany Denmark Ecuador Spain Finland United Kingdom Greece Hong Kong lndonesia lndia lreland italy Japan Kotea Mexico Malaysia Netherfands Norway New Zealand Portugal Singapore Sweden Thailand Venezuela

Distance Efkcts on:

..-

In

10%

Multinational Production -1%

Exports 74%

Multinational Produdi 2%

31%

-3%

17%

-11%

Chapter 2

The Gains From Trade: Standard Errors with the CES Monopolistic

Competition Mode1

2.1

Introduction

In the mid 1970s,as negotiations for the Tokyo Round of t a r a reductions gathered momentum, a spate of papeni appeared offeRng assessments of the potential e$ects of such

an agreement. Weaknesses in the prevailing input-output methodology led Robert Stem to solicit new approaches from 2 promising junior facuity members. The firat proposed

a sound econometric strategy, but arriveci at results that would generoiisly be c d e d 'mixed'. The second proposed a lineuized computable general quilibrium model thst

yielded remarkably plausible results. The junior faculty in this apocryphal story are Ed Leamer and Alan DeardorfE. We relate this story because it marks a watershed. Deardorff's Michigan Mode1 dong with developments by John WhaJley, Rick Harris and others foretold the ascendancy of computable general equilibrium modeling over econometrics in discussions of international t rade policy.

One goal of this peper is to re-estoblish the role of standard econometric methods for estimating the welfare gains h m world-wide tar8liberalization. Our starting point is the

CES monopolistic competition model (Dixit and Stiglitz 1977; Heipman and Krugman 1985) with its closed-form predictions about compensating variation and bilateral trade.

The bilateral trade prediction was initially examineci by Lawrence (1987) and Saxonhouse (1989). Its popularity rose with the careful analysis of Hamgan (1993,1996). Most of the related econometric literature has focusseci on the prediction's gravity-style teims rather than its price t e m . Harrigan (1993),Haveman et. al (1999), and Anderson (1999) are exceptions. However, these authors all simplify the price term, thus making the analysis

less u s a for our aim of estimating welfare gains.

One cannot assess the impact of tariff iiberalization without knowing how economic agents respond to product prices. Yet when t d s and hence product prices cliffer aaoss countries the bilateral trade and welfme predictions of the CES monopolistic competition model are complicated eg., Brown (1991). Indeed, prevailing wisdom has it that they are far too complicated to be examineci econometridy. Computable general equilibrium

(CGE) techniques must be used (e.g., Harris 1984). While there may be many reasom

for wanting to use CGE models, this is not one of them. We show that estimation of the model with asymmetric price &ects is straightforward.' The second goal of this paper is to econometricdy assess the CES monopolistic cornpetition modd. We are interested in quantifmg the violence done by the modd to the data. In paxticular, we ask whether the model adequately captures price and income efFects. We also ask whether the model is well-specined compared to such alternatives as the gavity and Heckscher-Ohlin models. Here we extend the work of Harrigan (1993, 1996), Hiimmels and Levinsohn (1993, 1995), Jensen (1996), Evenett and Keller (1998)

and others by examining mis-specification of the income and price effets and correlates of the country-pair fuced effects. Findy, our work has implications for CGE models. The advantage of these models is that they ailow for more elaborate modelling exercises than we consider? A key

parameter for these models is the elasticiv of substitution between varieties. We show graphicdy that estimates of welfare gains from tariff reductions are sensitive to the choice of elasticity of substitution. We estimate this parameter using the type of model and type of sample vaziation that is relevant for CGE models. Our estimates of around 5 are lower than those typically (but not always) used in the literature. For example,

Brown and Stem (1989) use 15. Our estimates are more similar to those in Feenstra (1994), but often smaller than those implied by Feenstra and Levinsohn (1995).

In examining the welfare gains from tariff liberalization, it is important to emphasize that we will be exclusively interested in the effect of tarifhi in distorthg consumers' consumption decisions. Restateci, we will focus on the consumptzon effect to the exclu-

sion of the production e-

OUFmethods are easily extended to the production dect.

This requires additionai modelling assumptions in order to pin down what is otherwise

an indeterminate distribution of output intemationdy. While this is not conceptudy 'Hem our aims are aimilar to Berry, Levinsoh, and Pakes (1999). Our approach is more conventionai. 2Thedebate about econometrics versus CGE techniques is not about who can go over Niagara Fails with more style. Each has its own advantages. The main advantage of econametrics is its rich tool Lit for examining the rehtionship betwean the model and the data e.g., hypothesir testing, spdcation tests, and model selection criteria.

difEcult, in the interest of bringing the main contribution of the paper to the fore we have shied away firom the requisite additional modelling assumptions. We will have more

to Say about this below.

The outline of the paper is complicated by the fact that in many ways it is r e d y 2 papers. The k t part (sections 2.1-2.7) presents the model, estimates it, and draws out implications for the welfare gains f'rom future rounds of t

a cuts. The second part

(sections 2.82.12) point to various problems with the CES monopolistic competition specscation, including mis-specification of the price and income terms.

2.2

Theory

We are interested in the standard CES monopolistic competition trade prediction. There

is a single consumer in each country and preferences are internationdy identical. Consiuners have Cobb-Douglas preferences over goods and

CES preferences over varieties.

2-stage budgeting allows us to concentrate on the decision about varieties. Let i index

consumer coutries, let j index producer countries, and let g index goods. Let w index varieties and let

Nd be the number (measure) of varieties produced in country j. Let

q , j ( ~ ) be 2's consumption of variety w produced in country j. Let pd(w) be the producer

price and let TM be one plus the ad v a l o m tariff so that p g j ( w ) ~ , jis the price faced by consumers in country i. Given data constraints, there are no benefits fkom allowing Tgj to depend on W . Consumer utility

In equilibrium, a, La the elasticity of substitution between varieties. The 6,j are most r e d y thought of as Armington (1969) parameters. They have other interpretations, but we do not push on any particdar one since the 6#j play h o s t no role in our empirical

~ o r k Given . ~ Zstage budgeting, the country i consumer chooses the q*j ( w ) to mlwmize

U, subject to prices P

~ ~ ( W )and T ~ income

agY,where

Yi is national income.

Producers face constant marginal costs cg and fixed costs F'.4 Utility maximization, profit m m z a t i o n , and zero profits together imply that the equilibrium producer price pgj(w) is independent of j and W , the equiiibrium producer supply qgj(w) is independent

of j and w , and equilibrium consumer demands qM(w) are independent of w. Denote these by p,, qg, and q,j, respectively.

Moving fiom vacieties to ~ O O let ~ SQgj , let M'

qgNgjbe country j ' s otitpiit of good g and

q,jNgj be country i's consumption of j's good g. For i # j , &Igj is i's imports

from j. For i = j , M,j is i's consumption of goods produced in i. Tedious, but f d a r

algebra yields optimal consumer demmd~qgij5 and hence a9Qg1 Mpii= bgij -

Ps?

(T

. .6

gr1 g t J

Ck (rgikbgik)

y,. '-Og

Qgk

To eliminate the unobservable pg and to push as hard as possible on the theory, we introduce the data identity that what country j produces (Q,) eqiials what country j ships to the world, inciuding itself

(x,Adsij). This is consistent with stipply equalling

demand. Thus,

(2-3)

M M /CkMgkj is i's share of j ' s shipments. TOsee that quation (2.3) appears repeat3There are 3 interpretations of the 6*j. First, they may be thought of as home-bias preference parameters. Second, they may be thought of as capturing unabserd heterogeneity a u w country-pair trades. For exaxnpIe, the United States imports highquaiity shoes h m Italy and Low-quality shoes fiom Brazil, but because of aggregation the data record only quantity, not quaiity. In this example, QusAPrrAty > 6g,LISA,BRAZIL. Thkd, the b~jjentathe utiiity function in Bcactly the same way as Lancaster's compensation function (LanCBSter 1979, Helpman 1981) and so have an interpretation in tenas of i d 4 types. 4111an ear1ie.r version of the paper we allowed 4 and cg tu vary internati~nally~ This dowed us to examine international dinerences in c m as a source of comparative advantage. Whlle this provideci the predicted insights, the paper is aiready too long to indude it. We d d o p tbis more sophisticated amt side in a mdtinationab context in chapter 1. 5 = b~jajaoS @o~,i6fij)-ow / Cr@o~fi~6pi~)1-ug N+

edly in, for example, Helpman and Krugrnan (1985),note that in the absence of tarins M@j/

CkMgkj = si where si

is 2's shace of world income. Plugging equation (2.2) into

equation (2.3) and t a h g logs yields

where the price term

md 6, =

{6mj)vijïj.

(a,,6,) is defineci as

Setting 6,j

= 1, the denominator of

cDM is the 'red' income of j's

trading partners Le., the s, are deflated by the CES price index. This denominator is often cded the 'market potential' hinction for country j. See Hanson (1998). To summarize, imports Qgj

M,j

depend on income si, a price term ,P