The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical

Evidence
Author(s): Jeffrey H. Bergstrand
Reviewed work(s):
Source: The Review of Economics and Statistics, Vol. 67, No. 3 (Aug., 1985), pp. 474-481
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/1925976 .
Accessed: 12/10/2012 08:06

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of
Economics and Statistics.

http://www.jstor.org
 THE GRAVITY EQUATION IN INTERNATIONAL TRADE:
SOME MICROECONOMIC FOUNDATIONS AND
EMPIRICAL EVIDENCE
Jeffrey H. Bergstrand*
Abstract-Despite the gravity equation's empirical success in (1980).1 Table I presents results from estimating a
"explaining" trade flows, the model's predictive potential has
been inhibited by an absence of strong theoretical foundations. gravity equation similar to (1) for 15 OECD coun-
A general equilibrium world trade model is presented from tries' trade flows.2 Coefficient estimates are stable
which a gravity equation is derived by making certain assump- across years and are representative of trade gravity
tions, including perfect international product substitutability.
If, however, trade flows are differentiated by origin as evidence equations.
suggests, the typical gravity equation is misspecified, omitting Despite the model's consistently high statistical
certain price variables. The last section presents empirical explanatory power, its use for predictive purposes
evidence supporting the notion that the gravity equation is a
reduced form from a partial equilibrium subsystem of a general has been inhibited owing to an absence of strong
equilibrium model with nationally differentiated products. theoretical foundations. The most common justifi-
cation-used in Linnemann (1966), Aitken (1973),
THE "gravity equation" has been long recog- Geraci and Prewo (1977), Prewo (1978), Abrams
nized for its consistent empirical success in (1980), and Sapir (1981)-was developed by
explaining many different types of flows, such as Linnemann and asserts that the gravity model is a
migration, commuting, tourism, and commodity reduced form from a four-equation partial equi-
shipping. Typically, the log-linear equation librium model of export supply and import de-
specifies that a flow from origin i to destination j mand. Prices are always excluded since "they
can be explained by economic forces at the flow's merely adjust to equate supply and demand."3
origin, economic forces at the flow's destination, However, critics have argued that this approach is
and economic forces either aiding or resisting the "loose" and does not explain the multiplicative
flow's movement from origin to destination. functional form.4
In international trade, bilateral gross aggregate This study addresses these and other issues in
trade flows are explained commonly using the developing further the microeconomic foundations
following specification: of the gravity equation. The "looseness" critique is
addressed by systematically describing assump-
PXi, = fo(yi) (I?) 2(Dij )3(A 1)4u (1) tions necessary to generate a gravity equation simi-
lar to (1) from a general equilibrium framework.
where PXij is the U.S. dollar value of the flow Specific, yet intuitively plausible, functions for
from country i to country j, Yi (Y1) is the U.S. utility and production generate the equation's
dollar value of nominal GDP in i (j), Dij is the multiplicative form. Section I presents a general
distance from the economic center of i to that of equilibrium model of world trade derived from
j, Aij is any other factor(s) either aiding or resist- utility- and profit-maximizing agent behavior in N
ing trade between i and j, and u is a log-nor- countries assuming a single factor of production in
mally distributed error term with E(ln uij) = 0.
This specification was used in Tinbergen (1962),
Poyhonen (1963a, 1963b), Pulliainen (1963), Geraci 'Linnemann (1966), Aitken (1973), Sattinger (1978) and Sapir
(1981) used the same general specification, but also included
and Prewo (1977), Prewo (1978), and Abrams exporter and importer populations. Microeconomic founda-
tions of this alternative specification are discussed in Bergstrand
(1984).
Received for publication June 16, 1983. Revision accepted for 2The countries are Canada, United States, Japan, Belgium-
publication December 12, 1984. Luxembourg, Denmark, France, West Germany, Italy, Nether-
*Federal Reserve Bank of Boston. lands, United Kingdom, Austria, Norway, Spain, Sweden, and
The author is very grateful to J. David Richardson, Robert Switzerland. The adjacency, EEC, and EFTA dummies are
Baldwin, Rachel McCulloch, James Alm, Saul Schwartz and explained in the appendix.
two anonymous referees for helpful comments on earlier drafts, 3Linnemann (1966), p. 41; Leamer and Stern (1970), p. 146;
and Doug Cleveland for research assistance. All errors remain (Geraci and Prewo (1977), p. 68; Prewo (1978), p. 344; and
the author's responsibility. The views expressed do not neces- Sapir (1981), p. 341.
sarily reflect the views of the Federal Reserve Bank of Boston 4See, for example, Anderson (1979), p. 106 and Leamer and
or the Federal Reserve System. Stern (1970), p. 158.

[ 474 ]
 GRAVITY EQUATION IN INTERNATIONAL TRADE 475

TABLE 1.-TYPICAL GRAVITY EQUATION COEFFICIENT ESTIMATES FOR

AGGREGATE TRADE FLOWS

Variables 1965 1966 1975 1976

Country i's Income 0.80b 0.80b 0.83b 0.84b

(20.10) (20.70) (19.03) (18.95)
Country j's Income 0.65b 0.66b 0.69b 0.69b
(16.14) (17.11) (15.83) (15.45)
Distance - 0.72b - 069b0h71b - - 0.72h
(12.15) (12.11) (10.35) (10.21)
Adjacency Dummy 0.61b 0.63b 0.69b 0.74b
(4.35) (4.61) (5.13) (5.36)
EEC Dummy 0.35a 0.41b 0.24a 0.30a
(2.02) (2.46) (1.80) (2.23)
EFTA Dummy 0.69b 0.73b 0.66b 0.69b
(4.94) (5.40) (3.35) (3.46)
Constant 1.gob 1.54b 0.64 0.56
(2.82) (2.36) (0.89) (0.77)
Adjusted R2 0.80 0.81 0.80 0.79
Root Mean Square Error 0.639 0.616 0.607 0.615
Number of Observations 210 210 210 210
Sources of data: See appendix.
Notes: All variables except dummies are expressed in natural logarithms; estimation is by ordinarv least squares.
t-statistics are in parentheses.
aSignificant at the 5% level, one-tail test.
5Significant at the 1% level, one-tail test.

each. The reduced form from this system specifies stitution (CES) utility function:
the trade flow from i to j as a function of all
countries' resource availabilities for a given year as
well as trade barriers and transport-cost factors
among all pairs of countries. However, this is not a ~1k==1 jA }
"gravity equation." A bilateral trade flow equation
must include exporter and importer incomes as j= 1,... N (2)
exogenous variables to be a gravity model, by where Xkj (Xjj) is the amount of k's aggregate
definition. Section II demonstrates that a gravity good (j's domestically produced good) demanded
model similar to (1) can be explicitly derived from by j's consumers, 4ij = (j- l)/j where is
this system by making certain simplifying assump- the CES between domestic and importable goods
tions, including perfect substitutability of goods in j (O < ?j< x), and Oj= (aj - 1)/aj where a
across countries. Yet if aggregate trade flows are is the CES among importables in j (O _ aj _ o).
differentiated by national origin, (1) misspecifies This specification allows the elasticity of substitu-
the gravity model by omitting certain price vari- tion between domestic and importable goods and
ables. In light of strong evidence implying the that among importables to differ.5 Equation (2)
existence of nationally differentiated products, sec- simplifies to a standard CES function when yj and
tion III presents estimates of a gravity equation a, are constrained to be equal. Expenditures in j
that includes price variables. The results support are constrained by income:
the notion that the gravity equation is a reduced N
form from a partial equilibrium subsystem of a Yj PkjXkj, j=1...,N (3)
general equilibrium trade model with nationally k~l
differentiated products. and where Pk is the k-cur-
=k PkjTkJCkj/Ekj
rency price of k's product sold in the jth market,
I. A General EquilibriumModel of
World Trade
5Some have argued that this "two-level" form suggests con-
sumers first choose between domestic and importable goods
Demand according to relative aggregate (domestic/import) prices, and
then choose among various foreign suppliers according to rela-
In each country j in each year, consumers are tive (bilateral) prices among importables. See Hickman and
assumed to share the constant-elasticity-of-sub- Lau (1973) and Geraci and Prewo (1982).
476 THE REVIEW OF ECONOMICS AND STATISTICS

Tk1 is one plus j's tariff rate on k's product differ.6 Equation (7) simplifies to a standard CET
(Tjj = 1), Ckj is the transport-cost (c.i.f./f.o.b.) function when qj and y, are constrained to be
factor to ship k's product to j (Cjj = 1), and Ekj equal. Henceforth, E' will denote summation over
is the spot value of j's currency in terms of k's k = 1,..., N, k * i. Substituting (7) into (6) and
currency (Ejj = 1). Henceforth, E" will denote maximizing the resulting equation yields N2 first-
summation over k = 1,I ..., N, k * j. Maximizing order conditions that are solvable for N(N - 1)
(2) subject to (3) generates N(N + 1) first-order bilateral aggregate export supply equations:
conditions that are solvable for N(N - 1) bilateral
aggregate import demand equations: X,J~
= YiPi
XSJ; ~~17(1yF,Pk+Y
[(E ~~~111.i
P'q)

D - [( E tt-1 )I/ 1(-a) ay-ll

x ([(E, ~1+ '
?Pl+ 1/(1+-yi) 1+ni +
pii+ii}-

i,j=1,...,N (j*j) (8)

and N domestic supply equations:
i,j=1,...,N (i*j) (4)
17(1
and N domestic demand equations: XZS= YiPini([(E Pk ]

+Pl+'i+......... = 1, ,N (9)
- 1
+p J l-j = 1,, ... , N. (5) where, with one factor of production, national
income in i is constrained by
Derivations, for above and below, are available Y,= WiRi, i= 1, ... N. (10)
from the author upon request.
Equilibrium
Supply
Assume N2 equilibrium conditions:
In each country i in each year, firms maximize
the profit function: = x.D = x,S i, j = N
Xii (I11)
,...........
N where Xij is the actual trade flow volume from i
1li Y, PlkXk
- WRi, to j. Equations (3)-(5) and (7)-(11) produce a
i = 1,..., N (6)
k=1 general equilibrium model of world trade with
4N 2 + 3N equations and endogenous variables.
where Ri is the amount available of the single,
The reduced form for Xij from this system
internationally immobile resource in a given year
would be a function of every R1 (i = 1.., N),
in i (e.g., labor hours) to produce the various
and Cij (i, j = 1,..., N; i * j). Yet such a
outputs and W, is the i-currency value of a unit of Tij
function is not a gravity equation, since this re-
Ri. R in each country is allocated according to the
duced form necessarily excludes endogenous ex-
constant-elasticity-of-transformation (CET) joint
porter and importer incomes. The next section
production surface:
demonstrates that a gravity equation similar to (1),
including incomes as exogenous variables, can
nevertheless be derived from this system using

k=1 I certain additional assumptions.

i=1, ... ,N (7) II. Solving for the Gravity Equation:

A Partial EquilibriumApproach
where 68 = (1 + q1)/q1 where 71 is i's CET be-
tween production for home and foreign markets Assumption 1
(0 ? ?
i < ) and 4, = (1 + yi)/yi where yi is i's
CET for production among export markets (O ? yi The first assumption is that the market for the
? oo). This specification allows the elasticity of aggregate trade flow from i to j is small relative to
transformation of supply between home and for- "
6Some have argued that this form suggests a two-level"
eign markets and that among foreign markets to decision for producers analogous to that for consumers.
 GRAVITY EQUATION IN INTERNATIONAL TRADE 477

the other N2 - 1 markets. This is analogous to the Assumption 2

small open economy assumption frequently used
An assumption of identical utility and produc-
in international finance studies, which implies that
tion functions across countries ensures that param-
the foreign price level, the foreign interest rate,
eters in (12) and (13) are constant across all coun-
and foreign income can be treated as exogenous.
try pairings. This assumption is common to trade
The small market assumption implies that varia-
analyses, including the Heckscher-Ohlin-Samuel-
tions in Xij and Pij to equilibrate X' and X,'
son model of interindustry trade and recent formal
have negligible impacts on Y,, Yj, Pi, Pjj, 'P,k+Yj
models of intraindustry trade, cf., Dixit and Nor-
and i"Pklaj. The general equilibrium system of
man (1980).] Combining (12), (13), and this as-
4N2 + 3N equations can then be considered N2
sumption yields:
partial equilibrium subsystems of 4 equations each
in 4 endogenous variables (Xij, X,P, XV, Pij) and PXIj Yi /yaf(+)(yacayl/ya

3N constraints. Combining one each of (4) and (8) XTiy (y-+ l)/(-y+c)Ei(y+ )/( )/(+c
with one of (11) yields: (E ) g -)(y +1( +y)(-y+a)
X p.lk+'y

(y 1ka) )( - )/(1 a)( -y+ a

Pij Yi aT
CljgiJ ji
ij X E

+
X ' pilk+'y )(.yi-,q)/(l yj) [( E ik )

I-(-l)/(y+)
+P1

[(
k zPI+y,)(? + j)/(l?y) ? pl?m]
-(y+l)/(y+a)
-1l/y +,q)(j) + p+-, (14)

where PXij is the value of the trade flow from i to

j (PXij = P,jX,j). Equation (14) is termed the
P/ ) (12)
"generalized" gravity equation. Given data limita-
tions, (14) can be estimated by ordinary least
and squares (OLS) when a constant and log-normally
distributed error term are appended. The specifica-
tion is "general" because it treats exporter and
Xij (Yigj Yj'YCi-, yigiT.j YiajE sYcj
importer incomes as exogenous yet imposes no
-I ) restrictions on parameter values other than being
X,= (UpIlk+Y'CY'UT(IY' +y'U
identical across all country pairings.

Assumptions 3-6
A gravity model more similar to (1) that ex-
-
x [ -1 )y (1-yj A)/(1 j )
cludes all price terms can be obtained with four
additional assumptions. Assuming perfect sub-
stitutability of goods internationally in production
and consumption, perfect commodity arbitrage,
_yl, l/(y+, ? )
zero tariffs, and zero transport costs (and normal-
izing all exchange rates to unity) implies Cij = Tij
= 1 and Pij = P for all i, j = 1,..., N. Since
i, j=1..N ( j). (13)
uf = = -Y = r1 = oo, (14) simplifies to
The small market assumption yields a reduced-
PX,J= (1/2) v/2Y9'/2 (15)
form bilateral trade equation with Y,and Yj treated
exogenously. A consequence of this assumption is which is similar to the gravity equation in (1).
that certain price terms are also treated exoge-
7Likewise, Anderson, (1979) assumed identical expenditure
nously. functions across countries in his theoretical foundation.
478 THE REVIEW OF ECONOMICS AND STATISTICS

III. An Empirical Model of the Generalized can approximate the cross-country variation of
Gravity Equation Y)1/(1 + ) across
exportpricelevels, thatis, of (Y3'wP,k+
all i. Likewise, j's import price index can ap-
While each of the previous six assumptions
proximate (_,f - i 's GDP deflator can
should be tested, this paper focuses on only the -j
approximate
last four. Numerous studies over the past decade -
have revealed large and persistent deviations of [( , 1 + )( + q)/(1+y l + ]
national price levels from purchasing power parity
(PPP). Isard (1977) found that for even the most and j 's GDP deflator can approximate
disaggregated manufactured commodities for
[( !_ _(1-y )/(1-a c) _ ]
which U.S. and foreign prices could be matched,
"relative price behavior ... marks them as differen-
tiated products, rather than near-perfect sub- The exchange rate index will indicate changes in
stitutes" (p. 942). Richardson (1978) found that the i-currency value of a unit of j 's currency since
commodity arbitrage did occur, but neither signifi- the common base period. A rise in this index
cantly for every commodity group nor at all for implies an appreciation (depreciation) of the im-
many groups. When commodity arbitrage did oc- porter's (exporter's) currency from the base.'0
cur, it was imperfect. Kravis and Lipsey (1984) As before, the generalized gravity model is
recently concluded, " we are pretty sure that estimated for 1965, 1966, 1975 and 1976. This will
equality of price levels among countries (PPP) will help indicate the stability of parameter estimates
not turn out to be the norm, even in the long run" from year to year, from one decade to another,
(p. 5). Thus, substantive deviations from PPP seem and from "fixed" to "floating" exchange rates.
to persist even in the absence of tariffs and trans- Previous partial-equilibrium time-series studies
port costs; the existence of these barriers only have suggested that price changes have multiyear
complicates problems. effects on trade flows. Magee (1975) noted evi-
This evidence suggests that assumptions 3-6 are dence of significant effects of price changes on the
restrictive and the generalized gravity equation volume of traded goods five years after the price
would be more appropriate to estimate. This sec- change. Base years 1960 for 1965/1966 estimates
tion presents an econometric version of (14) and and 1970 for 1975/1976 estimates not only allow a
discusses the results of its estimation. generous lag length but also reflect years of "nor-
mal" economic activity-that is, neither years of
recession troughs or expansion peaks nor years of
Econometric Issues exchange rate regime changes."

The econometric analogue to (14) is dis- '3Export/import unit value indexes were used rather than
tinguished from (1) predominantly by the former export/import price indexes because the former were available
for all 15 countries whereas the latter were available for only 4.
including price and exchange rate variables. The 9Such considerations form the analytical basis for comparing
tariff variable in (14) can be proxied by dummy internationally unit labor cost, unit value and aggregate price
variables indicating the presence of preferential indexes in the IMF's International Financial Statistics (Cost
and Price Comparisons). Critics may argue that construction
trading arrangements as in the basic gravity equa- and commodity composition of national price indexes varies
tion. The transport-cost factor can be proxied by widely across countries. Yet the aggregation/composition proW
the distance between economic centers of i and j lem in cross-country index comparisons is no more severe than
in time-series import and export demand studies. Country j's
and a dummy for adjacency. wholesale price index will also be used as a proxy for the
Calculating the complex price terms in (14) is weighted average of j's domestic and import prices.
1{ Local-currency-denominated import unit value indexes and
beyond this paper's scope; however, cross-country
importer GDP deflators implicitly reflect variations in exchange
differences in aggregate price levels can be ap- rates and tariff rates influencing (E"P,-1)7' (-f.
proximated by cross-country variation in aggre- "Critics may argue that 1970 was not a year of "normal"
gate price (or unit value) indexes. If (local-cur- economic activity. However, real GDP growth among in-
dustrial countries in 1970 was only 1 percentage point below
rency-denominated) export price indexes are the average for 1961-80. Consumer prices rose in 1970 only 0.4
calculated similarly for several countries using a percentage points below the average for the same period. 1970
common base period-the latter "well chosen" to avoided the 9-year global economic boom of the 1960s, yet
preceded the 1971-73 transition to floating rates as well as the
avoid large divergences from PPP-then variation 1973-74 oil shock and its aftermath. Data are from the IMF's
across countries in these indexes in a given year International Financial Statistics Yearbook (1983).
 GRAVITY EQUATION IN INTERNATIONAL TRADE 479

TABLE 2.-GENERALIZED GRAVITYEQUATIONCOEFFICIENTESTIMATESFOR

AGGREGATETRADE FLOWS

Variables 1965 1966 1975 1976

Country i's Income 0.75) 0.76b 0.80b 0.84b

(16.95) (17.24) (16.04) (15.79)
Country j's Income 0.63 0.66b 0.54b 0.56b
(14.03) (14.63) (9.12) (9.34)
Distance -0.78b -0.75b - 0.76b -0.77b
(12.70) (12.47) (11.22) (10.92)
Adjacency Dummy 0.58 0.5 b 0.74b 0.76b
(4.23) (4.43) (5.54) (5.62)
EEC Dummy 0.32a 0.35a 0.18 0.18
(1.82) (2.07) (1.37) (1.35)
EFTA Dummy 0.67b 0.73b 0.62b 0.73b
(4.92) (5.45) (3.20) (3.67)
Exchange Rate (E,j) 0.58 0.30 0.74a 0.73
(0.40) (0.23) (1.76) (1.62)
i's Export Unit - 1.39b - 1.14b - 0.55 - 0.96
Value Index (3.03) (2.83) (0.90) (1.55)
j's Import Unit 0.78 0.61 2.32 1.85b
Value Index (0.94) (0.77) (4.15) (4.14)
_ 1.g1b - 0.79
i's GDP Deflator - 1.46a - 0.05
(2.37) (2.00) (1.17) (0.07)
j 's GDP Deflator -1.28 - 0.41 - 1.13 - 1.12a
(1.43) (0.50) (1.55) (1.67)
Constant 30.19b 19.62a 4.06 4.50
(3.11) (2.22) (0.68) (0.95)
Adjusted R2 0.81 0.82 0.81 0.81
Root Mean Square Error 0.619 0.603 0.583 0.591
Number of Observations 210 210 210 210
Sources of data: See appendix.
Notes: See table 1.

Empirical Results All coefficient estimate signs conform to those

suggested above in all four years, as shown in table
Some remarks are first in order regarding 2. Importer income, adjacency, and preferential
expected coefficient signs. Four variables have trading arrangements have positive coefficient signs
unambiguous expected impacts. A rise in j's in- as in the basic gravity model; distance has a
come, an appreciation of j's currency, adjacency, negative coefficient sign. An appreciation of the
and the presence of preferential trading arrange- importer's currency increases the trade flow from i
ments should increase the trade flow from i to j; to j. A rise in exporter's income increases the trade
greater distance between these countries should flow, implying that a exceeds unity. The negative
reduce this flow. Remaining variables have coefficient estimate for i's GDP deflator supports
ambiguous expected effects. If the elasticity of this conclusion about a. The negative coefficient
substitution among importables (a) exceeds unity, estimate for i's export unit value index is con-
i's income and GDP deflator will have positive sistent with a greater than one and y greater than
and negative coefficients, respectively. If, addition- r1.The positive coefficient estimate for j's import
ally, tle elasticity of transformation among expor- unit value index is consistent with a greater than
tables (y) exceeds that between production for the I. The negative coefficient estimate for j's GDP
domestic market and for abroad (n), i's export deflator suggests that y is less than one. All these
unit value index will have a negative coefficient. elasticity conclusions are intuitively plausible.
Country j's import unit value index will have a Most coefficient estimates for traditional gravity
positive coefficient if the elasticity of substitution equation variables-incomes, distance, and dum-
among importables exceeds that between domestic mies are statistically significant in one-tail t-tests
and imported products (it). Finally, j's GDP in all four years and are similar in value to
deflator coefficient will be negative or positive estimates shown in table 1. Among price and
depending upon whether ,u is less than or greater exchange rate variables, 40% have statistically
than unity, respectively. significant coefficient estimates in one-tail t-tests.
480 THE REVIEW OF ECONOMICS AND STATISTICS

Interestingly, the exchange rate index coefficient certain assumptions. Many of these-such as per-
estimate, which was insignificant under fixed ex- fect substitutability of goods internationally in
change, rates, became statistically significant in consumption and production and perfect com-
1975 under flexible exchange rates, as this index modity arbitrage-have been refuted by recent
becamnemore volatile. The stability of coefficient empirical evidence, however. Because these as-
estimates across years within decades suggests that sumptions were refutable, a "generalized" gravity
the generalized gravity equation is rather robust.'2 equation could be derived. This methodology for
F-tests were conducted upon restrictions implied understanding the trade gravity equation mirrors
by assumptions 3-6, i.e., to test equation (15) that used to augment understanding the gravity
versus (14). The restrictions imposed by these as- equation for migration flows. As Greenwood (1975)
sumptions were overwhelmingly rejected. noted:
Finally, what can be inferred about (1)? This
Empiricallybased studiesthat have examinedplace-
equation cannot be explicitly derived from the to-place migration within this frameworkhave al-
theoretical model. Yet with the coefficients in (1) most universallyadopted for estimationpurposesa
not constrained a priori, the specification implies modifiedgravity-typemodel of gross migration.The
that the elasticities of substitution and transforma- models are "gravity-type"in that migrationis hy-
tion are not expected to equal infinity. But if these pothesized to be directly related to size of the
elasticities do not equal infinity a priori, (14) is relevantorigin and destinationpopulations,and in-
clearly more appropriate to estimate. Given the versely related to distance. The models are "mod-
theoretical framework, (1) would not sensibly be ified" in the sense that the variablesof the basic
estimated over (14), except in the absence of data gravitymodel are given behavioralcontent, and ad-
for price and exchange rate variables. To be com- ditional variables that are expected to importantly
influencethe decision to migrateare includedin the
plete, F-tests rejected (1) relative to (14), i.e., that
estimated relationships.The additionalvariablesare
all price and exchange rate variables had zero typically suggestedas proxiesfor variousarguments
coefficient estimates, at the 5% significance level in of individualutility functions.(p. 398)
all four years and at the 1% level in three years.
Incomes make our generalized equation a "grav-
IV. Conclusion ity-type" model. Yet price terms, derived from
underlying utility and production functions, im-
For twenty years, the gravity equation applied portantly influence trade flows and lend behavioral
to international aggregate trade flows has been content to the gravity equation.
estimated in a specification represented by equa- Empirically, the price and exchange rate vari-
tion (1). The equation's empirical success induced ables have plausible and significant effects on ag-
curiosity about its underlying "behavior." Over gregate trade flows. Coefficient estimates suggest
the years, theories-with and without economic that products are differentiated by national origin
content-surfaced to explain this equation. and commodity arbitrage is imperfect. Moreover,
In this study, a general equilibrium model of within the context of the theoretical model, these
world trade was introduced from which a gravity results imply that the elasticity of substitution
equation similar to (1) could be derived under among importables exceeds unity, the elasticity of
substitution between domestic and imported prod-
12The generalized gravity equation was also estimated for all ucts is below unity, and the elasticity of transfor-
four years substituting j's wholesale price index for j's GDP
deflator. The results do not differ substantively from those in
mation among export markets exceeds that be-
table 2. The small markets assumption enables the error term in tween production for domestic and foreign
the generalized gravity equation to be assumed uncorrelated markets.
with incomes, export and import price terms, and domestic
price terms. If P1J or Xi significantly influences any of these
variables, a simultaneous equations bias arises. The generalized APPENDIX
gravity model was reestimated using (one-year) lagged values of
incomes, unit value indexes, domestic price indexes, and the Bilateraltrade flows are from OECD, Statisticsof Foreign
exchange rate index. The results do not differ substantively Trade, Series C, Trade by}Commodities, various issues (c.i.f.
from those in table 2. However, this was not a powerful test import values; U.S. and Canadianf.o.b. values modifiedto
since current and lagged values of these variables were highly conform). Incomes are GDPs in U.S. dollars from OECD
correlated. Both sets of results are available from the author NationalAccounts,variousissues. For distance,sea distances
upon request. are from U.S. Naval OceanographicOffice, DistacnceBetween
 GRAVITY EQUATION IN INTERNATIONAL TRADE 481

Ports, H.O. Publication No. 151, U.S. Government Printing Hickman, Bert G., and Lawrence J. Lau, "Elasticities of Sub-
Office, 1965, and land distances are from Rand McNally Road stitution and Export Demands in a World Trade Model,"
Atlas of Europe, Rand McNally and Co., 1974. Land distances European Economic Review 4 (1973), 347-380.
were multiplied by a factor of two following W. H. Gruber and Isard, Peter, "How Far Can We Push the 'Law of One Price'?"
R. Vernon, " The Technology Factor in a World Trade Matrix," American Economic Review 67 (Dec. 1977), 942-948.
in R. Vernon (ed.), The Technology Factor in International Kravis, Irving B., and Robert E. Lipsey, "The Study of Inter-
Trade (New York: National Bureau of Economic Research, national Price Levels," research summary for the NBER
1970). Countries' economic centers are specified in Linnemann Conference on Research on Recent and Prospective
(1966). Exchange rates and export/import unit value indexes U.S. Trade Policy (Mar. 1984).
are from IMF International Financial Statistics, various issues.
Leamer, Edward E., and Robert Stern, Quantitative Interna-
GDP deflators were calculated from nominal and real GDPs in tional Economics (Chicago: Aldine Publishing Co., 1970).
OECD National Accounts. Wholesale price indexes are from Linnemann, Hans, An EconometricStudy of International Trade
UN Statistical Yearbook, 1968, 1977. The adjacency dummy Flows (Amsterdam: North-Holland Publishing Co.,
equals 1 if both countries share a land border, 0 otherwise. The 1966).
EEC (EFTA) dummy equals 1 if both countries are members of Magee, Stephen P., "Prices, Incomes, and Foreign Trade," in
the European Economic Community (European Free Trade P. B. Kenen (ed.), International Trade and Finance
Association), 0 otherwise. (Cambridge: Cambridge University Press, 1975).
Poyhonen, Pentti, "A Tentative Model for the Volume of Trade
between Countries," Weltwirtscha]tlichesArchivi, Band
REFERENCES 90, Heft 1 (1963a), 93-100.
Abrams, Richard K., "International Trade Flows under Flexi- , "Toward a General Theory of International Trade,"
ble Exchange Rates," Economic Review, Federal Reserve Economiska Samfundets Tidskrift 16 (1963b), 69-77.
Bank of Kansas City (Mar. 1980), 3-10. Prewo, Wilfried, "Determinants of the Trade Pattern among
Aitken, Norman D., " The Effect of EEC and EFTA on OECD Countries from 1958 to 1974," Jahrbucherfur
European Trade: A Temporal Cross-Section Analysis," Nationalekonomie und Statistik 193 (Aug. 1978),
Americ-anEconomic Review 63 (Dec. 1973), 881-892. 341-358.
Anderson, James E., "A Theoretical Foundation for the Grav- Pulliainen, Kyosti, "A World Trade Study: An Econometnrc
ity Equation," American Economic Reviewv69 (Mar. Model of the Pattern of the Commodity Flows of Inter-
1979), 106-116. national Trade in 1948-60," Economiska Sanmfundets
Bergstrand, Jeffrey H., "The Gravity Equation and the Factor- Tidskrift 16 (1963), 78-91.
Proportions Theory of International Trade: A Theoreti- Richardson, J. David, "Some Evidence on Commodity Arbi-
cal and Empirical Synthesis," processed paper, Federal trage and the Law of One Price," Journal of Internia-
Reserve Bank of Boston, 1984. tional Economics 8 (May 1978), 341-352.
Dixit, Avinash K., and Victor Norman, Theoryvof Interncational Sapir, Andre, "Trade Benefits under the EEC Generalized
Trade (London: Nisbet and Company, Ltd., 1980). System of Preferences," European Economic Review 15
Geraci, Vincent J., and Wilfried Prewo, "Bilateral Trade Flows (1981), 339-355.
and Transport Costs," this REVIEW 59 (Feb. 1977), Sattinger, Michael, "Trade Flows and Differences between
67-74. Countries," Atlantic Economic Journal 6 (July 1978),
, An Empirical Demand and Supply Model of Multi- 22-29.
lateral Trade," this REVIEW 64 (Aug. 1982), 432-441. Tinbergen, Jan, Shaping the WorldEconomv: Suggestionsfor atn
Greenwood, Michael J., "Research on Internal Migration in Intern(ational Economic Policy (New York: The
the U.S.: A Survey," Journal of Economic Literature 13 Twentieth Century Fund, 1962).
(June 1975), 397-435.