cbcam_d2s3
TRANSCRIPT
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 1/32
Gravity Models: Theoretical
Foundations and related estimationissues
ARTNet Capacity Building Workshop for Trade Research
Phnom Penh, Cambodia2-6 June 2008
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 2/32
Outline
1. Theoretical foundations• From Tinbergen (1962)….• … to Anderson and Van Wincoop (2003)
• The question of zeros in the trade matrix (Helpman,Melitz and Rubinstein, 2005)
2. Empirical equations and some related estimation issues
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 3/32
The Gravity Model: what it is?
Econometric model (ex-post analysis)Workhorse in a number of fields. It has been used toanalyze the impact of GATT/WTO membership, RTAs,currency unions, migration flows, FDI between countries,disasters ...Initially, not based on a theoretical model
What explain its popularity?• High explanatory power • Data easily available• There are established standard practices that facilitate the work of
researchers
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 4/32
The gravity model: the origins
Proposed by Tinbergen (1962) to explain internationalbilateral tradeCalled “gravity model” for its analogy with Newton’s
law of u niversal gravitation
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 5/32
The gravity model: the origins
Newton’s Law of Universal Gravitation
Fij = G M i M j
D ij2
F= attractive force; M= mass; D=distance; G = gravitational constant
Gravity Model specification similar to Newton’s Law
X ij = K Y iα
Y jβ
T ijθ
Xij= exports from i to j; or total trade (i.e X ij +X ji)Y= economic size (GDP, POP)
T =Trade costs
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 6/32
Standard proxies for trade costs in gravity equations
Distance
AdjacencyCommon language
Colonial linksCommon currencyIsland, LandlockedInstitutions, infrastructures, migration flows,...Surprisingly, bilateral tariff barriers often missing
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 7/32
The gravity model: the origins
Bilateral trade between any two countries is positively
related to their size and negatively related to thetrade cost between them
Which country-pair trade more?
Country B
Country A
Distance Country D
Country C
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 8/32
The gravity model: the originsEstimated gravity equation ...
Newton’s Law-based Normal Trade
ln (X ij) =
= C + a ln(Y i) + b ln(Y j) + c ln(dist ij) + u ij
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 9/32
Theoretical foundations for the gravity equation
Deardoff (1998) “I suspect that just about anyplausible model of trade would yieldsomething very like the gravity equation”
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 10/32
Theoretical foundations of gravity equation: historicalevolution
Anderson (1979)• Armington assumption (i.e. goods differentiated by
country of origin)Bergstrand (1990)
• Anderson and Monopolistic competition• But, he continue using existing price indexes instead
of those derived through the theoryVan Wincoop (2003)
• Monopolistic competition• Provide a practical way to estimate gravity
coefficients in a cross sectionHelpman et al. (2006)
• Heterogenous firms model of trade
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 11/32
A theoretical foundation of the gravity equation:Anderson and Van Wincoop (2003) ..
Step 1 : The expenditure share identity
p ijx ij=s ijEj
where p ij=import price
Step 2 : Share depends on relative pricesAssuming a CES demand function and that all goods
are traded
s ij=(p ij/P j)1-σ
where P j=Σ ini((p ij)1-σ
)1/(1- σ )
is the CES price index
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 12/32
… A theoretical foundation of the gravity equation:Anderson and Van Wincoop (2003)
Step 3:adding the pass-through equation
p ij=p oit ij
superscript o denotes producer pricet=bilateral trade costs
Step 4 : aggregating across varietiesXij=n is ijEj=n i (p o
it ij /P j)1-σ Ej
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 13/32
… A theoretical foundation of the gravity equation:Anderson and Van Wincoop (2003)
Step 5:using GE condition (summing over allmarkets, including country i’s ownmarket)
Yi= Σ iX ij
Solving for n ipi and substituting in eq. inStep 4 yields:
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 14/32
A theoretical foundation of the gravityequation: Anderson and Van Wincoop(2003)Step 6 : the reduced form of a intra-industry trade model
Y i Y j tij1-σ
X ij = Π i
Pj
where Πi=(Σ i tij
1-σ E j/P j1-σ)1/(1- σ)
P and Π = Multilateral Resistance Terms
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 15/32
Countries’ distance from the “Rest of the World”
matters for their bilateral tradeWhich country-pair trade more?
Country B Country A
Rest of the World
A theoretical foundationfor the gravity equation: the intuition
Country BCountry A
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 16/32
A theoretical foundation for the gravityequation: the intuition
The important contribution of Anderson andVan Wincoop’s paper has been to highlightthat bilateral trade is determined by relativetrade costs
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 17/32
Estimated gravity equation ...Theoretically Founded“Normal” Trade
Normal trade with resistances
ln (X ij) =
= K + a ln(Y i) + b ln(Y j) + c ln(bilateral tradebarriers ij) + d ln(MRT i) + e ln(MRT j) + u ij
MRT= Multilateral Resistance TermMRT are not observable.
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 18/32
Multilateral Trade Resistances and thegravity equation
3 ways to take MTR into account:1. Use an iterative method to solve MRT as function of
observable (see Anderson and Van Wincoop, 2003)2. Calculate Remoteness (trade/GDP weighted average distances
from the rest of the world) whereby Remoteness i=Σ j distance jj/(GDP j/GDP W)
3. Use country fixed effects for importers and for exporters
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 19/32
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 20/32
Estimated gravity equation ...using fixed effects:Cross Country Analysis
In cross country analysis MRT are fixed. Therefore,
using country fixed effects yields consistentestimations
ln (X ij) = K +c ln(bilateral trade barriers ij) ++Σ d i I i + Σ e i I j + u ij
Where I=country specific dummiesThere are 2n dummies.Total observations= n(n-1)
It is Impossible to estimate the coefficient for GDPand other country-specific variables
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 21/32
Estimated gravity equation ...usingfixed effects: Panel Data (1)
It is now POSSIBLE to estimate the coefficient forGDP and other country-specific variables
ln (X ij)t =K + a ln(Y i)t + b ln(Y j)t +c ln(bilateraltrade barriers ij)t + Σ d i I i + Σ e i I j + u ij
Where I=country specific dummiesThere are 2n such dummiesTotal observations= n(n-1)T
It is still not possible to estimate time-invariantcountry specific characteristics (eg. Island,landlockedness)
There may be a bias due to the variation overtime of MTRs
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 22/32
Estimated gravity equation ... using time-
varying fixed effects : Panel Data (2)In the case of a panel, MRTs may change over time
(variation in transport costs or composition of trade)
ln (X ij)
t =
=K + c ln(bilateral trade barriers ij)t + Σ d it I it+ + Σ e it I jt + Σ f τ Lτ +u ij
Where I=time-varying country specific dummies;L=time dummy (to take global inflation trends into
account)There are 2nT+T dummies, where T denotes the
time periodImpossible to estimate the coefficient of GDP
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 23/32
Estimated gravity equation ... usingcountry-pair fixed effects : Panel Data(3)address the bias due to the correlation between the bilateral
trade barriers and the MRTs
ln (X ij)t ==K + a ln(Y i)t + b ln(Y j)t + c ln(bilateral trade
barriers ij)t + Σ d ij I ij + u ij
Where I= country-pair dummies; There are n(n-1)/2 such
dummies
Disadvantage : coefficients of bilateral variables are estimated onthe time dimension of the panel AND cannot estimatecoefficient for distance, common border, common language
…A solution: Use random effects and the Hausman test to choosebetween random and fixed effect estimation
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 24/32
How to proceed?
Sensitivity analysis, test the robustness of the results toalternative specifications of the gravity equationsReport the results for the different equations estimated
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 25/32
Recent Theoretical Developments
Recently, Helpman et al. (2006) derived agravity equation from an heterogeneousfirms model of tradeThe importance of this derivation relates tothree issues that previous models of tradecould not explain:
• z e r o -trade observations
• Asymmetric trade flows• The extensive margin of trade: more
countries trade over time
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 26/32
The incidence of zero trade
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 27/32
How to handle zero-trade data?
Traditionally,When taking logs, zero observations aredropped from the sample. Then, the OLS
estimation is run on positive values.
Take the log(1+X ij), but then use Tobit
estimation as the OLS would providebiased results
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 28/32
OLS Bias if zeroobservations are not used
OLS Fit
True fit
X
Trade ij Recent studies (Felbermayr and Kohler, 2005;Helpman, 2006) that have included unrecorded trade
flows in gravity equations have tended to find thatWTO membership has a strong and significant effecton the formation of bilateral trading relationships.
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 29/32
How to handle zero-trade data?
More recently,
Helpman et al. (2006) claim: biased results using the standardapproachIn their model, differences in trade costs across countries andfirms heterogeneity account for both asymmetric trade flows and
zero trade. Zero trade occurs when the productivity of all firms incountry i is below the threshold that would make exporting to jprofitable.
Problem: the probability of having positive trade between 2countries is correlated with unobserved characteristics of thatcountry pair. These characteristics also affect the volume of their bilateral trade, given that they trade.
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 30/32
How to handle zero-trade data?
More recently,Helpman et al. (2006)provide the following solution:
Use 2 stage estimation1. Probit on the likelihood that 2 countries trade2. Then, estimate the gravity equation, introducing the
estimated Mills Ratio to control for sample selection bias (as
in the estimation of a selection model a la Heckman) and avariable controlling for firms heterogeneity. Selection biasand omitted variable problem.
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 31/32
References
Anderson J.E. (1979) “A Theoretical Foundation for the Gravity Equation”, AER,69(1):106-116
Anderson J. and E. Van Wincoop (2003) “Gravity with Gravitas: A solution to the Border Puzzle” AER, 93:170-192Bergstarnd J.H. (1990) “The Heckscher-Ohlin-Samuelson model, the Linder Hypothesis,and the Determinants of Bilateral Intra-industry Trade” Economic Journal, 100(4):1216-1229Baldwin R. and D. Taglioni (2006) “Gravity for Dummies and Dummies for GravityEquations”, NBER WP 12516Deardoff (1998)”Determinants of Bilateral Trade: Does Gravity Work in a NeoclassicalFramework?” in J.A. Frankel (ed.) The Regionalization of the World Economy, Chicago:University of Chicago PressEvenett S. and W. Keller (2002) “On Theories Explaining the Gravity Equation”, Journal
of Political Economy, 110:281-316Feenstra, R.(2004) Advanced International Trade, MIT PressHead K. (2003)”Gravity for Beginners”, mimeo, University British ColumbiaHelpman, Melitz and Rubinstain (2006) “Trading Partners and Trading Volumes” mimeo.Westerlund J. and F. Wilhelmsson (2006) “Estimating the gravity Model without Gravity
Using Panel Data”, at www.nek.lu.se
8/7/2019 cbcam_d2s3
http://slidepdf.com/reader/full/cbcamd2s3 32/32
Data sources
Feenstra bilateral trade data (1962-2000)from UN Comtradehttp://cid.econ.ucdavis.edu/data/undata/undata.htmlCEPII Distance and other geographyvariableshttp://www.cepii.fr/anglaisgraph/bdd/distances.html