Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Lecture 6: Technology, Geography and Trade Gregory Corcos
[email protected]
Isabelle Méjean
[email protected]
International Trade Université Paris-Saclay Master in Economics, 2nd year. 16 November 2016
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
1 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Introduction
Ricardo and HO theories (Lectures 2-4) : I I
I
Specialization according to comparative advantage Inter-industry trade between countries that differ in terms of technologies (Ricardo) or endowments (HO) Gains from trade due to a better allocation of resources
Limited empirical support (Lecture 5) I
I
Missing features in HOV can partially explain poor empirical performances In any case, the R 2 of such regressions is small
No (explicit) role for geography I
Hardly reconcilable with the gravity equation.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
2 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
The gravity equation
Robust empirical model of bilateral trade in which size and distance effects enter multiplicatively : Xij = G × Si × Mj × dij Workhorse econometric model of bilateral trade flows since Tinbergen (1962) Rationalized in mainstream modeling frameworks under some -widely used- parametric restrictions (See Lecture 10)
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
3 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Trade and the size of countries Figure 1: Trade is proportional to size
Japanese exports inEU, the2006EU (a) Japan’s exports to
Japan imports the EU (b) Japan’s importsfrom from EU, 2006 DEU
NLD ESP
FRA ITA
SWE HUN CZE
GRC IRLAUT FIN POL
SVK
PRT DNK
slope = 1.001 fit = .85
CYP EST SVN
MLT
FRA ITA GBR IRL NLD DNKSWE ESP BEL FIN AUT HUN CZE POL SVK MLT
PRT
slope = 1.03 fit = .75 GRC
EST LVA
SVN
LTU
.5
.05
Japan's 2006 exports (GRC = 1) .1 .5 1 5
BEL
Japan's 2006 imports (GRC = 1) 1 5 10 50 100
10
DEU GBR
CYP
LVA
.05
.1
.5 1 GDP (GRC = 1)
5
10
.05
LTU
.1
.5 1 GDP (GRC = 1)
5
10
Correlation between the Japan-EU trade and the size of partners. The x-axis measure the GDP of each EU members, in relative terms with respect to the Greek one. The y-axis measure the years, size of Japanese in aeach country in concert to establish robustness. In recent estimation has exports become just first step before a (left-hand side) a,dimplications the volume of Japanese imports from each country (rightdeeper analysis of the of the results, notably in terms of welfare. We try to facilitate hand side), again expressed in illustrating relative terms with respect to Greece.cookbook Data are diffusion of best-practice methods by their application in a step-by-step mode for 2006. Source : Head & Mayer (2014). of exposition. G. Corcos & I. Méjean (Ecole polytechnique) International 1.1 Gravity features of trade data Trade: Lecture 6
4 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Trade and distance Figure 2: Trade is inversely proportional to distance (b) France’s imports (2006) French imports Imports/Partner's GDP (%, log scale) .05 .1 .5 1 5 10
slope = -.683 fit = .22
other
500
slope = -.894 fit = .2
EU25 Euro Colony Francophone
.005
EU25 Euro Colony Francophone
.05
Exports/Partner's GDP (%, log scale) 1 .1 .5 5
10
25
France’s exports (2006) French(a)exports
1000
2000 5000 Distance in kms
10000
20000
other
500
1000
2000 5000 Distance in kms
10000
20000
.005
Correlation between the volume of trade and the distance between partners. The x-axis is the distance from France, expressed in kilometers. The x-axis measures theGDP. sizeFor of Japan’s Frenchexports, exports side) andand the trade flow on log the(left-hand GDP elasticity is 1.00 it issize 1.03of forFrench Japan’s imports side),isboth expressed relative terms respect ot the imports. The(right-hand near unit elasticity not unique to the in 2006 data. Over thewith decade 2000–2009, the destination GDP. Data are forintervals 2006. always Sourceincluded : Head1.0. & Import Mayerelasticities (2014). export elasticity country’s averaged 0.98 and its confidence averaged a somewhat higher 1.11 but the confidence intervals included 1.0 in every year except 2000 (when 10 of the EU25 had yet to join). The gravity equation is sometimes disparaged on
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
5 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
The Eaton and Kortum (2002) Model
Eaton & Kortum (2002) : neoclassical trade model in which I I
I
Comparative advantage arises randomly Technological advantage interacts with geography to shape comparative advantage Gravity equation arises structurally
The EK model has been used in quantitative exercises on : I I I I I
gains from trade liberalization (e.g. zero MFN tariffs, NAFTA) the importance/evolution of Ricardian comparative advantage the fall in trade during the 2008-2009 recession the US trade deficit and exchange rate adjustment the volatility of trade and GDP...
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
6 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Outline of this Lecture
Detailed presentation of the Eaton-Kortum model Some extensions Estimation of Eaton-Kortum Numerical applications
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
7 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
The Eaton-Kortum model See analytical details in EatonKortumAnalytics.pdf
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
8 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Main features
Ricardian model (differences in technology) with geography (barriers to trade). The model yields a gravity equation which relates bilateral trade volumes to deviations from purchasing power parity, technology and geographic barriers. The model can be estimated structurally.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
9 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Assumptions I countries (i = 1...I ) Continuum of goods j ∈ [0, 1] CES preferences in each country i : Z
1
Qi (j)
Ui =
σ−1 σ
σ σ−1
dj
0 demand
Goods produced with a bundle of inputs whose price is homogenous within countries ci (first taken as exogenous) Iceberg trade costs dni > 1. Without loss of generality dii = 1. Cross-border arbitrage implies : dni ≤ dnk dki
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
10 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Assumptions (ii) Country i’s efficiency in producing good j : zi (j) ⇒ CIF price of good j produced in country i and sold in country n : pni (j) =
ci z (j) | i{z }
dni |{z}
Trade barrier
Unit cost optimal price
Perfect competition across suppliers ⇒ Price actually paid in country n for good j : pn (j) = min{pni (j); i = 1...I } Note : most results continue to hold with Bertrand competition.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
11 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Assumptions (iii)
Stochastic productivity : zi (j) is the realization of a random variable Zi drawn from a country-specific probability distribution : Fi (z) = Pr [Zi ≤ z] Productivity draws assumed independent across goods and countries Fi assumed to be Fréchet (Type II extreme value) : ∀z > 0, Fi (z) = e −Ti z
Fréchet
−θ
with Ti > 0 and θ > 0 Note : the Fréchet distribution can be shown to be the outcome of a process of innovation and diffusion in which Ti is a national stock of ideas. See Eaton & Kortum (IER, 1999).
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
12 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Interpretation
Fi (z) = e −Ti z
−θ
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Benchmark in red. Doubling of Ti in blue / of θ in green
Ti captures absolute advantage. The higher Ti , the higher that country’s mean z : i is more likely to draw a high z for each product j. θ captures the extent of comparative advantage. The higher θ, the lower Var (z) : trade will be less driven by comparative advantage.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
13 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Price distribution Distribution of prices offered by country i in country n : Gni (p) ≡ Pr [Pni ≤ p] = 1 − e −Ti (ci dni )
−θ p θ
Country n’s actual distribution of prices : Gn (p) ≡ Pr [Pn ≤ p] Y = 1 − [1 − Gni (p)] i
= 1 − e −Φn p where Φn ≡
P
i
θ
Ti (ci dni )−θ
details
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
14 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Price distribution
θ
Gn (p) = 1 − e −Φn p ,
Φn ≡
X
Ti (ci dni )−θ
i
Distribution of prices governed by States of technology around the world {Ti }, Input costs around the world {ci }, Geographic barriers {dni } - If dni = 1, ∀n, i then Φn = Φ, ∀n (Law of One Price) - If dni → ∞, ∀i then Φn = Tn cn−θ (Autarky)
⇒ Φn represents the strength of competition that any firm will encounter in country n
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
15 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Bilateral trade Share of goods that n buys from i = Probability that i provides the lowest price good in country n : Xni πni = Xn = Pr [pni (j) ≤ min{pns (j); s 6= i}] Z ∞Y = [1 − Gns (p)]dGni (p) 0
=
s6=i
Ti (ci dni )−θ Φn
or in log : ln Xni = ln Ti ci−θ + ln Xn Φ−1 −θ ln dni n | {z } | {z } | {z } Gravity Exporter FE
Importer FE
⇒ Gravity-type equation G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
16 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Bilateral trade (ii)
EK structural interpretation of the gravity equation : The trade barriers coefficient relates to heterogeneity in productivity : ⇒ The greater heterogeneity among producers of a commodity (lower θ), the stronger the cost advantage of the lowest cost supplier, the more likely he remains the lowest cost supplier when trade costs increase. ⇒ Trade flows respond to geographic barriers at the extensive margin : As a source becomes more expensive or remote, it exports a narrower range of goods [as in Dornbusch et al. (1977)].
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
17 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Bilateral trade (iii) Country i’s normalized import share in country n : Xni /Xn Φi −θ Sni ≡ d = = Xii /Xi Φn ni
pi dni pn
−θ
Always lower than one due to the triangle inequality (the maximum value for pn is pi dni ) As overall prices in market n fall relative to prices in market i (↑ pi /pn ) or as n becomes more isolated from i (↑ dni ), i’s normalized share in n declines A higher θ means comparative advantage forces weaken relative to trade costs : normalized import shares become more elastic.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
18 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
General equilibrium solution Suppose production is linear in labor (EK has intermediate inputs) : ci = wi ⇒ Price levels as a function of wages : pn = γ
" X
#−1/θ −θ
Ti (dni wi )
i
where γ ≡ Γ
θ+1−σ θ
1/(1−σ)
price index
⇒ Trade shares as a function of wages and prices : Xni = Ti Xn
γdni wi pn
−θ
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
19 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
General equilibrium solution
To close the model, one needs to solve for equilibrium wages across countries This is the trickier part of the exercise ⇒ Numerical solutions Additional simplifying assumptions can help solve the model : Exogenous labor supply, Wages determined in the nonmanufacturing sector, Trade balance.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
20 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Extensions : Multiple sectors
Costinot, Donaldson & Komunjer (2012) reinterpret EK’s model in the context of a multi-country, multi-sector world K industries/goods in each country (k = 1, ..., K ). Within each industry, a continuum of varieties is produced according to the technology described above. With multiple sectors, Ti now has a sector dimension (but, crucially, θ remains common to all countries and industries...) : k z −θ
Fik (z) = e −Ti
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
21 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Extensions : Multiple sectors (ii) The model predicts trade patterns at the industry-country pair level. For any importer j and any pair of exporters i, i 0 6= j, the ranking of relative fundamental productivities determines the ranking of exports : Xji1 XjiK Ti1 TiK ⇒ ≤ ... ≤ ≤ ... ≤ Ti10 Xji10 TiK0 XjiK0 In a 2-country world without heterogeneity, that ranking would imply ziK (j) zi1 (j) < ... < zi10 (j) ziK0 (j) i would specialize in high-k goods and i 0 in low-k goods, just as in Dornbusch et al. (1977). G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
22 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Extensions : Multiple sectors (iii)
In this model with stochastic (Fréchet) productivity differences, we have : zi1 (j) Ti1 TiK ziK (j) ⇒ ≤ ... ≤ ... Ti10 zi10 (j) TiK0 ziK0 (j) where denotes first-order stochastic dominance. Country i 0 is not expected to specialize in high k goods, but to produce and export relatively more of these goods. Unlike Dornbusch et al. (1977) here we can predict trade patterns for more than 2 countries.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
23 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Extensions : Imperfect Competition
Bernard, Eaton, Jensen & Kortum (2003) extend EK to allow for imperfect competition between varieties With imperfect competition, consumer prices are above marginal costs Model predicts a distribution of mark-ups in each market, that is bounded above by the Dixit-Stiglitz constant mark-up Additional predictions on within-country heterogeneity in prices, productivities, etc
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
24 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Bringing the Model to the Data
The model can be estimated to quantitatively assess the role of Ricardian advantages in driving international trade. Structural parameters θ and {Ti } must be estimated. Once those parameters are estimated, it is possible to run various counterfactuals.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
25 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Example : EK Estimation of θ Bilateral trade in manufactures among 19 OECD countries in 1990 (342 bilateral relationships Xni ) Absorption of manufactures as a measure of Xn (STAN, OECD) Proxy for trade barriers : - Distance and other geographic barriers - Retail price differentials measured at the product level (WB) : Interpreted as a sample of pi (j), used to calculate relative prices, which are theoretically bounded above by bilateral trade costs : ln
pi dni max2j {rni (j)} '= = Dni pn mean{rni (j)}
rni (j) = ln pn (j) − ln pi (j) relative price of commodity j ⇒ exp(Dni ) price index in destination n that would prevail if everything was purchased from i, relative to the actual price index in n G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
26 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Estimating θ, EK Method 1 technology, geography, and trade 1755 Normalized import shares and relative prices
Figure 2.—Trade and prices.
Source : Eaton & Kortum, 2002. Unconditional correlation -0.4 we use this value for ! in exploring counterfactuals. This value of ! implies a standard deviation in efficiency (for a given state of technology T ) of 15 percent.
Model :
Xni /Xn Xii /Xi
=
In Section 5 we pursue two alternative strategies for estimating !, but we first complete the−θ full description of the model.
pi dni pn
4 equilibrium input costs
Our exposition so far has highlighted how trade P flows P relate to Xgeography ni /Xn n i ln experiment, and to prices, taking input costs ci as given. In any counterfactual Xii /Xi P P is crucial. however, adjustment of input costs to a new equilibrium [ln d −ln Pi +ln Pn ] n into i laborniand intermediTo close the model we decompose the input bundle ates. We then turn to the determination of prices of intermediates, given wages. Finally we model how wages are determined. Having completed the full model, we illustrate it with two special cases that yield simple closed-form solutions.
Estimated θ (method-of-moments) : θˆ =
⇒ θˆ = 8.28
→ Standard deviation in efficiency at given T = 15%
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6 41 Production
27 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Counterfactuals
Once estimated, the model can be used to run counterfactuals : What are the welfare gains from trade ? (Arkolakis et al, 2012) What is the impact of multilateral/unilateral tariff eliminations ? (Caliendo & Parro, 2015, Alvarez &Lucas 2007) How much does trade spread the benefit of local improvements in technology ? By how much should US real wages fall to restore current account balance ? (Dekle et al. 2007)
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
28 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Application : Evaluating the gains from NAFTA Caliendo & Parro (2015) use a variant of the Eaton-Kortum model to evaluate the trade and welfare impact of NAFTA. NAFTA : A free trade area between the US, Mexico and Canada - Enhance trade within the area / Divert existing trade between the area and the RoW - Increase welfare : Access to cheaper consumption goods plus increased competitiveness through a drop in input prices → Potentially important Ricardian gains since the 3 countries have very different production structures
Main insights : I
Important role of sectoral IO linkages to amplify the trade and welfare effect of the partnership
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
29 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Theoretical framework i. Multiple sectors : Ui
=
K Y
αk Qik i
K X
,
k=1
αik = 1,
Qik =
Z 0
k=1
1
Qik (j)
σ k −1 σk
σk σ k −1
dj
ii. Input-Output linkages : cik
=
γk wi i
K Y k 0 =1
k,k 0
0γ Pik i
,
K X
0
γik,k = 1 − γik
k=1
iii. Non tradable sectors : dnik = +∞ for some k iv. Sector-specific productivity distributions (Fréchet) : k z −θ k
Fik (z) = e −Ti
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
30 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Analytical predictions Equilibrium prices under perfect competition : k " I # h i−θk −1/θ k X c k k pni (j) = k i dnik ⇒ Pnk = Ak Thk chk dnh zi (j) h=1 Expenditure shares : k πni
k k = Pr [pni (j) ≤ mins {pns (j); s 6= i}] k −θ T k ckdk = P i i ni k I k k −θ k h=1 Th ch dnh
k directly (through d k ) and indirectly Changes in tariffs affect πni ni (through the price of inputs encapsulated in cik )
GE solution under the assumption of balanced trade at the world level (but country-specific trade deficits) gives the vector of equilibrium wages w which is specific to a tariff vector G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
31 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Impact of trade liberalization Equilibrium in relative changes implies : ln
w ˆn Pˆn
=
−
K K K J l,k X X X αnk αnk 1 − γnk αnk Y ˆ l ˆ k γn k k ln π ˆ − ln π ˆ − ln P / P nn nn n n θk θk γnk γnk l=1 k=1 k=1 k=1 | {z } | {z } | {z } Final goods
ln π ˆnik
=
−θ
k
h
ln cˆik
+
Intermediate goods
ln dˆnik
−
ln Pˆnk
Sectoral Linkages
i
where xˆ = dx/x, {ˆ cik } and {Pˆnk } are non-linear functions of {w ˆn } k ˆ and {dni } Impact of trade liberalization on real wages can be summarized by the k }) and sectoral price indices impact it has on domestic shares ({πnn k ({Pn })
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
32 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Impact of trade liberalization (ii) Trade liberalization increases real wages by reducing the sectoral k ), i.e. shares of domestic consumption (ln π ˆnn i. Giving consumers access to cheaper imported goods ii. Reducing the cost of same sector imported inputs (Only role of intermediates if γnk 6= 1 and γnk,k = 1 − γnk iii. Reducing the cost of imported inputs for other sectors (when γnk,k 6= 1 − γnk )
Changes in real wages do not directly map into changes in welfare in this model because of trade deficits (Dn ) and tariff revenues (Rn ) : ˆ ˆn Rn Rˆn Dn Dˆn ˆ n = ln In = wn Ln ln w ln W + ln + ln In In In Pˆn Pˆn Pˆn Pˆn
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
33 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Welfare Impact Using the equilibrium conditions of the model : ln
Iˆn Pˆn
=
I X K k X E
hn
h=1 k=1
| +
In
ln cˆnk −
k Mnh ln cˆhk In
{z
Terms of trade
}
I X K k k X dnh Mnh k ˆ nh − ln cˆhk ln M In h=1 k=1 | {z } Volume of trade
Terms of trade effect due to an increase in exporter prices relative to the change in importer prices Volume of trade effect due to the creation of additional trade flows following trade liberalization
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
34 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Empirical strategy Calibration of the observed parameters : I I I
k } calibrated using trade and production data {πni {αik } fitted to data on sectoral absorption 0 {γik } and {γik,k } fitted to IO tables
Estimation of the unobserved parameters {θk } : ln
k dk k Xk Xnik Xim dnik dim mn mn k = −θ ln k Xk k dk dk Xink Xmi d nm in mi nm
k k k = νin + µkn + δik + εkni , νni ln dnik = ln(1 + τnik ) + νni k k )(1 + τ k ) X k X k Xmn (1 + τnik )1 + τim mn k ⇒ ln nik im = −θ ln + εknim k Xk k )(1 + τ k )(1 + τ k ) Xin Xmi (1 + τ nm nm in mi
Use sectoral bilateral trade and tariff data G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
35 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Sectoral trade elasticities Table 1. Dispersion-of-productivity estimates Full sample 99% sample s.e. N θj s.e. N (1.86) 496 9.11 (2.01) 430 (2.76) 296 13.53 (3.67) 178
θj 8.11 15.72 2.55 5.56 10.83 9.07 51.08 4.75 1.66 2.76 7.99 4.30 1.52 12.79 10.60 7.07 8.98 1.01 0.37 5.00
(0.61) (1.14) (2.53) (1.69) (18.05) (1.77) (1.41) (1.44) (2.53) (2.15) (1.81) (2.14) (1.38) (1.72) (1.25) (0.80) (1.08) (0.92)
Test equal parameters Aggregate elasticity j
j
j
j
2.62 8.10 11.50 16.52 64.85 3.13 1.67 2.41 3.28 6.99 1.45 12.95 12.91 3.95 8.71 1.84 0.39 3.98
(0.61) (1.28) (2.87) (2.65) (15.61) (1.78) (2.23) (1.60) (2.51) (2.12) (2.80) (4.53) (1.64) (1.77) (1.56) (0.92) (1.08) (1.08)
429 314 191 352 86 341 272 263 288 314 290 126 269 143 237 126 226 227
F( 17, 7294) = 7.52
4.55 j
495 437 315 507 91 430 376 342 388 404 397 306 343 312 383 237 245 412
(0.35) j
7212
4.49
(0.39)
97.5% sample θj s.e. N 16.88 (2.36) 364 17.39 (4.06) 152 2.46 1.74 11.22 2.57 61.25 2.94 0.60 2.99 -0.05 0.52 -2.82 11.47 3.37 4.82 1.97 -3.06 0.53 3.06
(0.70) (1.73) (3.11) (2.88) (15.90) (2.34) (2.11) (1.88) (2.82) (3.02) (4.33) (5.14) (2.63) (1.83) (1.36) (0.86) (1.15) (0.83)
352 186 148 220 80 220 180 186 235 186 186 62 177 93 94 59 167 135
Prob > F = 0.00 5102
3.29
(0.47)
3482
j
˜ε = εin − εni + εhi − εih + εnh − εhn . Note that all the symmetric and asymmetric components of the Source where : Caliendo & Parro, 2015.j The “99% sample” and “97.5% sample” drop j j j j j iceberg trade costs cancel out. The terms κni /κin , κih /κhi , and κhn /κnh will cancel the symmetric bilateral
j j j j in each j j sector. from the the tradeestimation costs (ν jni , ν jih , and ν jhn ).smallest The terms κcountries ni /κnh , κih /κin , and κhn /κhi cancel the importer fixed effects
(μjn , μji , and μjh ); and the terms κjni /κjhi , κjih /κjnh , and κjhn /κjin cancel the exporter fixed effects (δ ji , δ jh , and j identification restriction is that ˜εj is assumed to be orthogonal to tariffs.40 n ). The only G. Corcos & I. Méjean δ(Ecole polytechnique) International Trade: Lecture 6
Downloaded from http://restud.oxfordjournals.org/ at Yale University on October 2, 201
Sector Agriculture Mining Manufacturing Food Textile Wood Paper Petroleum Chemicals Plastic Minerals Basic metals Metal products Machinery n.e.c. Office Electrical Communication Medical Auto Other Transport Other
36 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Counterfactual analysis
i. Introduce the change in tariffs from 1993 to 2005 between NAFTA members, fix tariffs for the RoW to 1993 levels ii. Introduce the change in tariffs from 1993 to 2005 between NAFTA members as well as observed changes in world tariffs iii. Introduce the change in world tariffs from 1993 to 2005, fixing NAFTA tariffs to 1993 levels. The difference between (ii) and (iii) measures gains from world tariff reductions with and without NAFTA. Note : In principle, trade liberalization affects trade deficits, which are exogenous in the model. This is a limit of the analysis.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
37 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Pre-NAFTA tariffs Fig. A.1. Eective applied tari rates before NAFTA
AppliedtariffratesMexicotoUSA (1993)
20
20
15
15
10
10
%
%
AppliedtariffratesMexicotoCanada(1993)
5
5
0
0
AppliedtariffratesCanadatoUSA (1993) 20
15
15
10
10
%
43 %
AppliedtariffratesCanadatoMexico (1993)
20
5
5
0
0
AppliedtariffratesUSAtoMexico (1993) 20
15
15
10
10
%
%
AppliedtariffratesUSAtoCanada (1993) 20
5
5
0
0
Source:UNCTADͲTRAINS)
Source : Caliendo & Parro, 2015. In 1993, sectoral tariff rates applied by Mexico, Canada and the US to NAFTA members were on average 12.5, 4.2 and 2.7%. By 2005, they dropped to almost zero between NAFTA members but tariffs that Mexico, Canada and the US applied to the RoW were on average 7.1, 2.2 and 1.7%, respectively Downloaded from http://restud.oxfordjournals.org/ at Yale University on October 2, 2015
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
38 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
The role of intermediate goods and sectoral linkages In 1993, the role of intermediate goods is already substantial... I
I
Respectively 68, 61.5 and 64.6% of Mexico’s, Canada’s and the US imports from non-NAFTA countries were intermediate goods Respectively 82.1, 72.3 and 72.8% of Mexico’s, Canada’s and the US imports from NAFTA countries were intermediate goods
... As is the extent of cross-sectoral linkages : I I
I
In the IO Tables, the mean share of own-sector inputs is around 15-20% More than 70% of intermediate consumption is addressed to other sectors Average share of tradables in the production of non-tradables is 23% for the US and 32% for Mexico / Average shares of non-tradables in the production of tradables are 34% for the US and 26% for Mexico
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
39 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
increases by 1.31%. The effects for Canada and the U.S. are smaller. Canada loses 0.06% while the U.S.
gains 0.08%. Still, we find that real wages increase for all NAFTA members and Mexico gains the most, Welfare effect from NAFTA’s Tariff reductions followed by Canada and the U.S. 47
Table 2. Welfare effects from NAFTA’s tariff reductions Welfare Country Total Terms of trade Volume of Trade Real wages Mexico 1.31% -0.41% 1.72% 1.72% Canada -0.06% -0.11% 0.04% 0.32% U.S. 0.08% 0.04% 0.04% 0.11%
Source : Caliendo Parro, Analysis holdsofRoW unchanged Decomposing the welfare effects & into terms2015. of trade and volume trade tariffs underscores the sources of these gains. The third column in Table 2 shows that the major source of gains are increases in volume of trade. The
the most, both in terms of welfare terms of real welfare Mexico gains from gains trade creation for Mexico, Canada and the U.S. are 1.72%,and 0.04%in and 0.04% respectively. We canwages. look deeper and measure the extent to which the welfare effects are a result of trade creation with NAFTAMost members vis-a-vis thesource rest of the is done by applying bilateralofvolume important ofworld. gainsThis is increase in the the volume tradeof trade measures (18) defined before.NAFTA ; (mostly within
trade vis-à-vis the RoW decreases, trade
diversion). Table 3. Bilateral welfare effects from NAFTA’s tariff reductions Terms of trade of Trade US terms-of-trade improved, both vis-à-visVolume NAFTA members and the Country NAFTA Rest of the world NAFTA Rest of the world RoW. Mexico -0.39% -0.02% 1.80% -0.08% -0.09% vary -0.02% 0.08% -0.04% Welfare Canada effects widely across sectors. U.S.
0.03%
0.01%
0.04%
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
0.00%
40 / 52
import tariffs in Petroleum in the year 1993 across NAFTA members was 7%. Finally, NAFTA’sAppendix tariffs The EK model Empirical Applications Conclusion reductions has important effects over the price of intermediate goods traded in some sectors compared to
Introduction
others. This is particularly important for the sectors Electricalreductions Machinery and Autos for reasons we discussed Trade effect from NAFTA’s Tariff in the previous paragraph. The reduction in trade prices in these sectors explains the increase in the volume of trade effect. Table 5. Trade effects from NAFTA’s tariff reductions Mexico Canada U.S. Mexico’s imports 116.60% 118.31% Canada’s imports 58.57% 9.49% U.S.’s imports 109.54% 6.57% -
Source : Caliendo Parro,from 2015. Analysis holds tariffs unchanged Table 5 presents aggregate trade&effects NAFTA. As we can RoW see, NAFTA generated large aggregate trade effects for all members. Mexico’s imports from NAFTA increased by more than 110% and equally so across both partners. For the case of Canada, we find that the percentage increase in imports from Mexico Large aggregate effects for all members is more than five times larger than the percentage increase in imports from the U.S. This results reflect
Canada and the US increased a lot their imports from Mexico : role as a supplier of intermediates to NAFTA
that Mexico’s role as a supplier of intermediate goods to NAFTA members increased as a consequence of NAFTA. In fact, this is even more evident when we look at the case of the U.S. imports. Imports from
Strong impact on the specialization of countries : Mexico becomes more specialized these economies become after the tariff reductions imposed by the agreement. In short, NAFTA strengthened Mexico increase more than 100% while from Canada only 6.57%. These figures reflect how interdependent
the trade dependence that these countries had before the agreement, and as a consequence Canada and the U.S. source more goods from Mexico, while Mexico sources more goods from Canada and the U.S. G. Corcos & I. Méjean (Ecole polytechnique) Trade:Table Lecture 6 NAFTA also had an effect on sectoralInternational specialization. 6 presents
41 / 52 export shares by industry before
the U.S. more diversified. In fact, Mexico’s share of exports from Electrical Machinery increase to 34.07%
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
and the three largest sectors account for 54.95% of total exports after NAFTA. This sectoral concentration is reflected in Mexico’s HHI which increases to 0.138. On the other hand, the HHI indices of Canada and Specialization due to NAFTA the U.S. decrease. 49
Table 6. Export shares by sector before and after NAFTA’s tariff reductions Mexico Canada United States Sector Before After Before After Before After Agriculture 4.72% 3.03% 4.99% 5.04% 6.91% 6.35% Mining 15.53% 7.85% 8.99% 8.96% 1.72% 1.52% Manufacturing Food 2.33% 1.48% 4.82% 4.68% 5.09% 4.73% Textile 4.42% 6.92% 1.05% 1.49% 2.68% 3.49% Wood 0.59% 0.52% 8.12% 8.05% 2.02% 1.98% Paper 0.62% 0.51% 8.34% 8.44% 4.99% 4.89% Petroleum 1.62% 5.28% 0.59% 0.78% 4.30% 5.71% Chemicals 4.40% 2.53% 5.58% 5.40% 10.00% 9.25% Plastic 0.80% 0.48% 2.06% 2.06% 2.28% 2.43% Minerals 1.32% 0.84% 0.81% 0.78% 0.94% 0.92% Basic metals 3.24% 2.00% 10.29% 10.19% 3.05% 3.11% Metal products 1.22% 1.03% 1.47% 1.53% 2.23% 2.59% Machinery n.e.c. 4.30% 2.53% 4.69% 4.49% 10.37% 9.70% Office 3.34% 5.07% 2.44% 2.54% 7.70% 7.29% Electrical 20.79% 34.07% 2.50% 2.35% 6.07% 7.97% Communication 8.57% 7.08% 3.11% 3.02% 7.19% 6.81% Medical 2.48% 3.28% 0.98% 1.03% 5.16% 4.79% Auto 16.43% 13.05% 24.42% 24.07% 8.20% 8.09% Other Transport 0.28% 0.26% 3.21% 3.58% 7.32% 6.65% Other 3.02% 2.20% 1.55% 1.52% 1.77% 1.74% Normalized Herfindahl
0.092
0.138
0.083
0.081
0.042
0.040
rest of the world was hardly by2015. NAFTA’s tariffLecture reductions. Tabletariffs A.3 in Appendix “Additional Source : Caliendo & affected Parro, Analysis holds6 RoW unchanged G. Corcos & The I. Méjean (Ecole polytechnique) International Trade:
42 / 52
Introduction
The EK model
Empirical Applications
Conclusion 54
with no materials used in production (No materials), and with no I-O connections (No I-O).
Appendix
We calibrate
Decomposition trade welfare effects each of these models to of the year 1993 andand compute the welfare and trade responses from NAFTA’s tariff reductions. Table 11 presents the simulated trade and welfare effects implied by the different models. The first column shows the welfare effect from the one sector model. The second column presents the welfare result for the no materials model, and the third column presents the welfare result for the no I-O model.
We find: that for all models the welfare are smaller to theunchanged benchmark model. Source Caliendo & Parro, 2015.effects Analysis holds compared RoW tariffs
Still, in
all cases Mexico gains the most followed by the U.S. then Canada. The results from the one sector model reflect the importance of accounting for sectoral heterogeneity. In fact, recent studies have emphasized
Welfare gains are always reduced in comparison to benchmark
that the sectoral variation in trade elasticities is particularly important for the quantification of the welfare gains.55 The calculations also show that intermediate goods amplify the welfare effects from tariff reductions.
⇒ Trade in intermediates, Sectoral heterogeneity and Sectoral linkages all Mexico’s figure increase from 0.50% to 0.66%, Canada’s deteriorate more from -0.03% to -0.04% and the matter U.S. increase from 0.03% to 0.04% as we move from a model with no materials to a model with materials. We also find that the model with input-output linkages amplifies the effects as well. If we compare the third column on Table 11 to the results from the benchmark model, Table 2, we can clearly see that the welfare effects are substantially larger for the countries that win and lower for the countries that loose. effects(Ecole are also smaller across these models compared the benchmark case. The last four columns of G. Corcos & Trade I. Méjean polytechnique) International Trade: to Lecture 6
43 / 52
Downloaded from http://restud.oxfordjournals.org/ at Yale University o
Country Mexico Canada U.S.
Table 11. Trade and welfare effects from NAFTA across different models Welfare Imports growth from NAFTA members Multi sector Multi sector One sector No materials No I-O One sector No materials No I-O Benchmark 0.41% 0.50% 0.66% 60.99% 88.08% 98.96% 118.28% -0.08% -0.03% -0.04% 5.98% 9.95% 10.14% 11.11% 0.05% 0.03% 0.04% 17.34% 26.91% 30.70% 40.52%
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Concluding remarks
A very elegant way of introducing Ricardo into a multi-country and possibly multi-sector model Analytics strongly rely on some assumptions : Fréchet distribution, Variance of productivities homogenous across industries The model can be parsimoniously calibrated and extended to be brought to the data. Many empirical applications, such as the evaluation of trade agreements. But results can be sensitive to how structural parameters are estimated... [more on this in Lectures 10-11]
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
44 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
References - Alvarez, F. & Lucas, R., 2007. "General equilibrium analysis of the Eaton-Kortum model of international trade," Journal of Monetary Economics, vol. 54(6), pp. 1726-1768 - Arkolakis C., Costinot A., and Rodriguez-Clare A. 2012. “New Trade Models, Same Old Gains ?” American Economic Review, 102(1) : 94-130. - Bernard, A., Eaton J., Jensen, B. & Kortum S., 2003. “Plants and Productivity in International Trade,” American Economic Review 93(4) :1268-1290. - Costinot, Donaldson & Komunjer, 2012, “What Goods Do Countries Trade ? A Quantitative Exploration of Ricardo’s Ideas”, Review of Economic Studies 79(2) :581-608. - Caliendo & Parro, 2015, “Estimates of the Trade and welfare Effects of NAFTA,” Review of Economic Studies 82 (1) : 1-44
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
45 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
References
- Dekle R., J. Eaton & S. Kortum, 2007, "Unbalanced Trade", American Economic Review : Papers and Proceedings, Vol. 97, No. 2, pp. 351-355 - Eaton J. & Kortum S., 1997. “International technology diffusion : Theory and measurement,” International Economic Review 40(3) : 537-570. - Eaton J. & Kortum S., 2002, “Technology, Geography and Trade”, Econometrica 70(5) :1741-1779 - Eaton J. & Kortum S., 2012, “Putting Ricardo to Work”, Journal of Economic Perspectives 26(2) :65-90 - Head, K. & Mayer T., 2014, “Gravity Equations : Workhorse,Toolkit, and Cookbook”, chapter 3 in Gopinath, G, E. Helpman and K. Rogoff (eds), vol. 4 of the Handbook of International Economics, Elsevier : 131-195.
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
46 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Demand functions Consumers solve : max{Q (j)} i
s.t.
R1 0
hR
j∈[0,1]
σ−1 1 σ dj 0 Qi (j)
i
σ σ−1
Pi (j)Qi (j)dj ≤ Ri
Solution of the maximization program is : Pi (j) −σ Ri Qi (j) = Pi Pi with Pi the ideal price index (Ri /Pi = Ui , Z Pi =
1
Pi (j)
1−σ
∀Ri ) : 1 1−σ
dj
0 Back to assumptions G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
47 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Optimal Prices
Firms’ profit : πi (j) =
X n
X ci pni (j)Qni (j) − dni Qni (j) = πni (j) zi (j) n
Under perfect competition : pni (j) =
ci dni zi (j)
and Qin (j) = 0 if pin (j) > pn (j)/Qn (j) otherwise Back to assumptions
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
48 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Price distribution pni (j) = is :
ci zi (j) dni
is the realization of a random variable Pni which cdf
ci dni Gni (p) = Pr [Pni ≤ p] = Pr Zi ≥ p −θ c d ci dni −Ti i pni = 1 − Fi =1−e p pn (j) = min{pni (j); i = 1...I } is the realization of a random variable Pn = min{Pni ; i = 1...I } which cdf is : Gn (p) = Pr [Pn ≤ p] = 1 −
I Y
Pr [Pni > p]
i=1
= 1−
I PI Y θ −θ [1 − Gni (p)] = 1 − e −p i=1 Ti (ci dni ) i=1
Back to the model G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
49 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Price index Cdf / pdf of consumption prices : Fn (p) = 1 − e −Φn p
θ
fn (p) = Φn θp θ−1 e −Φn p
and
θ
Define : y = g (p) = p θ , then Gn (y ) = Fn (g −1 (y ))
−1 ∂g (y ) = Φn e −Φn y gn (y ) = f (g −1 (y )) ∂y
and
Thus the price index : Z Pn
1
=
pn (j)1−σ dj
1 1−σ
0
Z =
1
y
1−σ θ
Φn e
−Φn y
1 1−σ
dy
0
=
Φn−1/θ
1
Z
u
1−σ θ
e
−u
1 1−σ
du
where
u = Φn y
0
=
1 1−σ 1−σ Φn−1/θ Γ −1 θ
Back to the model G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
50 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Fréchet distribution
Generalized extreme value distribution : A family of continuous probability distributions used as an approximation to model the maxima of long (finite) sequences of random variables CDF :
( 1 ) x − µ −ξ F (x; µ, σ, ξ) = exp − 1 + ξ σ
µ a location parameter, σ > 0 the scale parameter, ξ the shape parameter
G. Corcos & I. Méjean (Ecole polytechnique) International Trade: Lecture 6
51 / 52
Introduction
The EK model
Empirical Applications
Conclusion
Appendix
Fréchet distribution In particular : I
I
I
Gumbel or type I extreme value : ξ = 0 x −µ , F (x; µ, σ, 0) = exp −exp − σ
x ∈R
Frechet of type II extreme value : ξ = α−1 > 0 ( 0, n x ≤µ x−µ −α o F (x; µ, σ, ξ) = exp − σ , x >µ Reversed Weibull or type III extreme value : ξ = −α−1 < 0 ( n α o exp − − x−µ , x