INTERNATIONAL TRADE

The Gravity Model

Course Instructor:
Dr. Devasmita Jena
Why Gravity?
Gravity modelling and international trade studies are almost inseparable
Widely used model of applied trade and trade policy research
Mostly, it is used to study trade patterns across countries and their determinants
(As we will see)Structural Gravity models are theoretically consistent
In empirical trade studies, gravity results have been found to be quite robust
• Estimation and Counterfactuals
Wide range of questions can be approached using the gravity model
Gravity model of trade are increasingly being modified, in other fields of research,
to study bilateral flows:
• Migration
• Financial flows between countries
• Tourism flows
• Transfer of emissions from one state to others
Gravity Questions
Mostly, it is used to study trade patterns across countries and their
determinants
Can you take theory to data in a gravity framework? As long as you focus
on bilateral aspects
• Comparative advantage (RCA measures)
• Difference in endowment
• Productivity difference
• Technological gap
• Home bias( remoteness: dist/gdp)
• Intra-industry trade and its determinants
• Etc.
Gravity Questions
Whether a policy – socioeconomic, political, trade-oriented – affects
international trade flows? Again the focus is on bilateral aspect
• Tariff
• Trade agreements
• Trade facilitation measures
• Membership in WTO
• Economic sanction
• Political alliance
• Cultural differences (preferences) etc.
• Political sentiments between countries
What gravity cannot help answer:
• Trade-inequality (intra-country, international, wage, gender, sectoral etc) linkages
• Trade as determinant of informality
• What determines entry into a trade agreement? Etc.
The “bilateral focus” in Gravity Modelling
The important point: Focus of the questions should be on bilateral trade costs
and its impact on trade flows
Basic unit of observation for most gravity is exporter-importer pair
Three groups of factor influence bilateral trade
• Exporter-specific “push” factor – for example, technological advancement leading to
lowering cost of production
• Importer-specific “pull” factor – for example, market size
• Bilateral “drag” factor – for example, distance between countries, tariff, trade agreements
etc.
• Trade is the result of interplay of these factors
• Country-specific (the push-pull factors) do not allow much flexibility in terms of
observing, measuring and being theoretically consistent
What about country-specific factors affecting trade?
• Can it be bilateral, by taking difference or ratio? Example technological gap
The Intuitive Gravity Model
The gravity model has its origin in Newton’s law of gravitational attraction
between two objects
• The force of gravitational attraction between two objects is directly proportional to
the product of their masses and inversely proportional to the square distance between
them
Accordingly, the intuitive gravity model:

The equation has pull, push and drag factors

 Analysis: Taking log and doing and OLS estimation
• Inter alia, zero trade issue
Augmented gravity model included other variables – population (another
measure of mass), colony, ethnicity, border, tariff, agreement etc.
But the intuitive gravity model are largely ad hoc and a-theoretical
The Structural Gravity Model
Anderson (1979): derived structural gravity equation under the assumption of
product differentiation
Bergstrand (1985): derived gravity equation in a general equilibrium framework
under the assumptions of perfect international product substitutability
Deardorff (1998):derived structural gravity equation using Heckscher-Ohlin
theory
Eaton and Kortum (2002): derived structural gravity equation using Ricardian
Theory
 Helpman and Krugman (1985) derived structural gravity equation using new
international trade theory frameworks
Anderson and Wincoop (2001, 2003): derived an operational gravity model
utilizing CES expenditure system that can be empirically estimated
Yotov V. Yotov (…): Gravity methodologies
The Anderson and Wincoop (2001, 2003) model
--- (1)
---(2)
---(3)
Xij,t :exports from i to j at time t; Ej,t :importer j’s total expenditure; Yi,t :value of total production in
exporter i
Yt :world output
tij,t is the bilateral trade costs between trade partners i and j; σ > 1 is the elasticity of substitution
among goods from different countries
(1)=> exports is a function of market size, ( ) and total trade cost ( )
Structural terms, Pj,t and Πi,t :inward & outward multilateral resistance terms
Inward MRT: importer j’s ease of market access; outward MRT: exporter i’s ease of market access
Effects of Bilateral Costs?
The MRTs helps translate the initial, partial equilibrium effects of trade policy at the bilateral
level to country-specific effects on consumer and producer prices
• The direct effects gives the initial impact effects of trade costs on trade flows
• The general equilibrium trade costs also take into account the changes in prices, incomes and
expenditures induced by trade cost changes
 and
Chain reaction of change in bilateral trade costs, due to incorporation of MRT terms
Consistent with theory, if trade costs between two countries change, then the impacts on third
countries are non-zero
Structural Gravity: The Empirical Model
 Many theories, one structure : bilateral trade is a function of push, pull, and drag factors
Structural Gravity: The Dependent Variable
The dependent variable should be: exports in value terms (not quantity)
• Imports (again in value terms) can be considered
• Trade balance, or the average of bilateral flows, or any other variable, cannot be the LHS of
gravity equation
Why focus on bilateral exports in value terms?
• That is how the structural gravity models are derived from microeconomic fundamentals
• Theoretical derivation of gravity model include ways of deflating trade flows relative to
some economic aggregate
o This may have implications for econometric estimation
Importer’s expenditure is as an explanatory variable
• Expenditure is the sum of imports from all sources: international as well as domestic
• No reason to exclude intra-national trade from gravity model: important for counterfactuals
Structural Gravity: The Independent Variables
What about including GDP, GDP per capita and population like we did for intuitive
gravity?
Structural gravity model has expenditure, in gross shipments terms
• Gross shipments terms is not equal to GDP, which is measured in value added terms—i.e., net of
intermediate inputs
No theoretical rationale for using population as a measure of economic size
GDP per capita
• Assumption of homothetic preferences doesn’t allow inclusion of GDP per capita
• If you’re interested in income effects, use an appropriate theory (Fieler, 2011)
What about price index?
• It is problematic to use CPI/ GDP deflator as proxy: price data on traded goods is not calculated
under CES assumption
• price index—equivalently, MRT —is unobservable
Structural Gravity: Estimation Issues
Never use OLS!
• Heteroscedasticity
• Zero trade values
• Estimates are biased and inconsistent: doesn’t go away even by increasing the size of data
Other important challenges in estimating gravity models:
• MRTs – theoretically important, practically unobservable
o accounted for by exporter-time and importer-time fixed effects, which will absorb:
 size variables (Ej,t and Yi,t ); and other observable and unobservable country-specific characteristics such national
policies, institutions, exchange rates etc.

• Proper specification of bilateral trade costs

o Standard practice: proxy it by distance, contiguity, common colony and language, RTA, tariff etc..
o A note on tariff: Tariff acts as a price shifter
Structural Gravity: Estimation Issues
Other important challenges in estimating gravity models:
• Endogeneity of trade policy (Trefler, 1993)
o Trade policies may suffer from “reverse causality”: Ex. Endogeneity between tariff and RTA
o Trade policy may be correlated with unobservable cross-sectional trade costs
o Use country-pair fixed effects: accounts for the unobservable linkages between the endogenous trade
policy variable and the error term in gravity regressions (Egger & Nigai, 2015; Agnosteva et al., 2014)
o Pair fixed effects will absorb all bilateral observable (e.g. distance, language, etc.) and unobservable time-
invariant trade costs
o Pair fixed effects will not absorb the effects of time-varying trade policy variables (e.g., RTA)
• Inclusion of country specific policy variable, such as trade subsidies, disaster etc.
o Difficult to capture due to inclusion of exporter-time and importer-time fixed effects (to account for
MRTs)
o Use of intra-national and international trade in gravity equation
o A 2-stage estimation: the estimates of the MRTs from the 1st-stage regression are explained in an auxiliary
Structural Gravity: Estimation Issues
Other important challenges in estimating gravity models:
• Adjustment to trade policy – happens with lag
o Trefler (2004): trade estimations pooled over consecutive years might be an issue
o Olivero and Yotov (2012): obtain gravity estimates with 3- and 5-year interval trade data
Note: As far as possible, use a panel data
• Improved estimation efficiency
• Enables application of the pair-fixed-effects to address endogeneity of trade policy
variables
• Allows for a flexible and comprehensive treatment and estimation of the effects of
time-invariant bilateral trade costs with pair fixed effects
Sector-wise gravity models can be estimated, in case the research interest is to
understand trade patterns in a particular sector
The PPML Estimator
Poisson Pseudo-Maximum Likelihood (PPML) estimator (Santos Silva and Tenreyro, 2006,
2021)
Recast model as a generalized linear model (GLM): OLS is part of GLM models
Use GLM from the exponential family

Estimation of gravity using cross-section data? Again, PPML estimator should be used
• Gravity specification should include intra-national trade and importer- and exporter-fixed effects
• Interval data can’t be used
• Pair fixed effects are no longer applicable - include the standard set of gravity variables (e.g., bilateral
distance, contiguity etc.)
PPML better than OLS
• Deals with heteroscedasticity: conditional variance of 𝑋𝑖𝑗 is proportional to its conditional mean, then this
NLS estimator is numerically equivalent to the Poisson estimator
• Deals with zero trade values
Data
CEPII
ITPD-E : has data on intra-national trade as well
Data on domestic shipments? Not available expansively
Take advantage of the multi-region input-output tables (MRIOs):
• World Input-Output Database (WIOD)
• OECD-WTO Trade in Value Added Database (TiVA)
• Asian Development Bank’s MRIO