.Gravity and Comparative Advantage:
Estimation of Trade Elasticities for theAgricultural Sector
Kari E.R. Heerman, Economic Research Service, USDAIan Sheldon, Ohio State University
2018 IATRC Annual Meeting
Whistler, BC CanadaJuly 25-27, 2018
The analysis and views expressed are the authorsโ and do not represent theviews of the Economic Research Service or USDA.
Heerman and Sheldon July 25-27, 2018
Introduction
Systematic Heterogeneity (SH) Gravity Model
โข Tailored to fundamental features of agriculture & sub-sectors
โ Allows systematic influences on within-sector specialization
Other structural gravity models
โข Intra-sector heterogeneity independently distributed
โ Eaton and Kortum (2002), Chaney (2008) and extensions
โข Multi-sector models address specialization across sectors
โ Independence implies random within-sector specialization
Heerman and Sheldon July 25-27, 2018
Introduction
Does this matter?
โข Allows for more flexible system of bilateral trade elasticities
โ Elasticities drive predicted trade flow responses
โข Standard gravity models impose restrictive elasticities
โ Arkolakis, Costinot and Rodriguez-Clare (2012), Adao,Costinot and Donaldson (2017)
โ โIndependence of Irrelevant Exportersโ (IIE) property
โข Relative demand is unaffected by third-country costs
โข An illustration...
Heerman and Sheldon July 25-27, 2018
Example: U.S. raises tariffs on Costa Rican agriculture
Other 9.1%
Beef 2.8% Fruit, nes
3.2%
Melons 7.5%
Coffee 10.7%
Pineapples 16.8%
Bananas 40.8%
US Ag Imports: Costa Rica
Standard gravity predicts equal increases in trade flows for any two
exporters with the same share of the US ag market
Heerman and Sheldon July 25-27, 2018
Example: U.S. raises tariffs on Costa Rican agriculture
Other 9.1%
Beef 2.8% Fruit, nes
3.2%
Melons 7.5%
Coffee 10.7%
Pineapples 16.8%
Bananas 40.8%
US Ag Imports: Costa Rica
Other 5.0% Coffee 2.6%
Mangoes 2.8%
Fruit, Nes 3.7%
Cocoa 4.7%
Plantains 5%
Bananas 71.4%
Ecuador US Ag Market Share = .001%
Other 4.2%
Eggplants 2.4%
Green Chiles
& Peppers 44.1%
Tomatoes 45.8%
The Netherlands: US Ag Market Share = .001%
Heerman and Sheldon July 25-27, 2018
Roadmap
โข Structural model overview
โข Specification of econometric model
โข Estimation
โข Selected results
โข Conclusion
Heerman and Sheldon July 25-27, 2018
Structural Model Overview
Heerman and Sheldon July 25-27, 2018
About the Model
Environment
โข I countries engaged in bilateral agricultural trade
โ Exporter indexed by i
โ Importer index by n
โข A continuum of products indexed by j
โข Production technology is heterogeneous across products
โ Climate and land characteristics influence which productshave the highest productivity
โข All markets are perfectly competitive
โข Trade occurs as buyers look for the lowest price
Heerman and Sheldon July 25-27, 2018
Model Overview
Production Technology Country i , product j technology
qi (j) = zi (j)ร (Niฮฒi (ai (j)Li )
1โฮฒi )ฮฑi Qi1โฮฑi
โข Input bundle: labor (Ni ), land (Li ), intermediates Qi
โข zi (j) Technological productivity-enhancing Frechet r.v.
Fi (z) = exp{โTizโฮธ}
โ Ti drives average technological productivity in country i
โ ฮธ drives dispersion of technological productivity
โ Independently distributed across products
โข E.g., coffee
โข ai (j) is deterministic variable representing land productivity
Heerman and Sheldon July 25-27, 2018
Model Overview
Production Technology Country i , product j technology
qi (j) = zi (j)ร (Niฮฒi (ai (j)Li )
1โฮฒi )ฮฑi Qi1โฮฑi
โข ai (j) is deterministic variable representing land productivity
โ Value reflects the coincidence of product requirements andcountry ecological characteristics
โข E.g., coffee
โ Country-specific parametric density, independent of zi (j)
Heerman and Sheldon July 25-27, 2018
Trade
Heerman and Sheldon July 25-27, 2018
Model Overview
Comparative Advantage Probability country i has comparativeadvantage in product j in market n
ฯni (j) =Ti (ai (j)ciฯni (j))โฮธ
Nโl=1
Tl(al(j)clฯnl(j))โฮธ
โข Probability country i price offer is lowest in market n
โ ci is the cost of an input bundle
โข ฯni (j) โฅ 1 is exporter i โs cost to export products to market n
โ Deterministic variable with parametric density
โ Independent of zi (j) and ai (j)
Heerman and Sheldon July 25-27, 2018
Model Overview
Market Share Exporter i share in country n agriculture expenditure
ฯni =
โซTi (aiciฯni )
โฮธ
Nโl=1
Tl(alclฯnl)โฮธdFan(a)dFฯ n(ฯ )
โข This is the structural equation from which the SH gravitymodel is derived
โ Fan(a) is the distribution of an = [a1, ..., aI ] across allproducts consumed in market n
โ Fฯ n(ฯ ) is the distribution of ฯ = [ฯn1, ..., ฯnI ] across allproducts consumed in market n
Heerman and Sheldon July 25-27, 2018
Specification
Heerman and Sheldon July 25-27, 2018
Random Coefficients Logit Specification
โข Average productivity and input bundle cost as in EK
lnTi โ ฮธlnci โก Si
โ Country fixed effect
Heerman and Sheldon July 25-27, 2018
Random Coefficients Logit Specification
Land Productivityln(ai (j)) โก Xiฮด(j)
โข Exporter Characteristics
โ Xi =[aLi elvi tropi tempi bori
]โข ali - (log) arable land per capita, World Bankโข elvi share of rural land at high altitude, CIESINโข tropi - share of land in tropical climate zone, GTAP
Heerman and Sheldon July 25-27, 2018
Random Coefficients Logit Specification
Land Productivityln(ai (j)) โก Xiฮด(j)
ฮด(j) = ฮด + (E(j)ฮ)โฒ + (ฮฝE (j)ฮฃE )โฒ
โข Product characteristics
โ โObservableโ production requirements
โข E(j) =[alw(j) elv(j) trop(j) temp(j) bor(j)
]โ Ex., trop(j) - tropical climate intensity of cultivation
โ Trade-weighted averages of country characteristics
โ โUnobservableโ product-specific requirements
โข ฮฝE (j) - vector of normal r.v.โs
Heerman and Sheldon July 25-27, 2018
Random Coefficients Logit Specification
Trade Costsln(ฯni (j)) โก tniฮฒ(j) + exi + ฮพni
ฮฒ(j) = ฮฒ + (ฮฝtn(j)ฮฃt)โฒ
โข Country-pair characteristics
โ tni , exi - border, language, distance, RTA & exporter effects
โข โUnobservableโ product-specific trade costs
โ ฮฝtn(j) - vector of normal r.v.โs
Heerman and Sheldon July 25-27, 2018
Estimation
Heerman and Sheldon July 25-27, 2018
Estimation
Random coefficients logit model
ฯni =1
ns
nsโj=1
exp{Si + Xiฮด(j)โ ฮธ(tniฮฒ(j) + ฮพni )}Iโ
l=1
exp{Sl + Xlฮด(j)โ ฮธ(tnlฮฒ(j) + ฮพnl)}
โข Estimates obtained using simulated method of moments
โ Smooth simulator (Nevo (2000))โ ns draws from each countryโs empirical distribution of
expenditure dFEn(E)dFฮฝn(ฮฝ) More .
โข Dependent variable ฯni calculated from FAO production andtrade data
Heerman and Sheldon July 25-27, 2018
Results
Heerman and Sheldon July 25-27, 2018
Parameter Estimates
Land Productivity Distribution
ln Arable Land per Ag Worker 0.17*** -0.01 -4.51*** 0.42*** 1.81*** 0.33***
High Elevation 1.14*** -0.21 47.96*** 0.44*** 1.31*** -12.32***
Tropical Climate Share 0.7*** -0.16** -3.96*** 0.73*** 6.86*** 0.19
Temp. Climate Share 0.19*** -0.03 1.46*** -0.53*** -2.8*** 0.7***
Boreal Climate Share -0.88*** 0.19** 2.5*** -0.2*** -4.06*** -0.89***
Exporter Characteristics
Mean Effects
Unobserved Reqs
Agro-Ecological Requirements
๐ฟ๐ฟ๐๐ (๐น๐น) (๐บ๐บ๐๐) ๐๐๐๐๐๐(๐๐) ๐๐๐๐๐๐ ๐๐ ๐๐๐๐๐๐(๐๐) ๐๐๐๐๐๐(๐๐)
(๐ฒ๐ฒ)
โข Effect of country characteristics varies significantly with productrequirements โ Reject standard gravity model of agricultural sector
Heerman and Sheldon July 25-27, 2018
Parameter Estimates
Land Productivity Distribution
ln Arable Land per Ag Worker 0.17*** -0.01 -4.51*** 0.42*** 1.81*** 0.33***
High Elevation 1.14*** -0.21 47.96*** 0.44*** 1.31*** -12.32***
Tropical Climate Share 0.7*** -0.16** -3.96*** 0.73*** 6.86*** 0.19
Temp. Climate Share 0.19*** -0.03 1.46*** -0.53*** -2.8*** 0.7***
Boreal Climate Share -0.88*** 0.19** 2.5*** -0.2*** -4.06*** -0.89***
Exporter Characteristics
Mean Effects
Unobserved Reqs
Agro-Ecological Requirements
๐ฟ๐ฟ๐๐ (๐น๐น) (๐บ๐บ๐๐) ๐๐๐๐๐๐(๐๐) ๐๐๐๐๐๐ ๐๐ ๐๐๐๐๐๐(๐๐) ๐๐๐๐๐๐(๐๐)
(๐ฒ๐ฒ)
โข Total effect of high elevation for product j
ฮด(j) = ฮด + (E(j)ฮ)โฒ + (ฮฝE (j)ฮฃE )โฒ
Heerman and Sheldon July 25-27, 2018
Does it matter?
Heerman and Sheldon July 25-27, 2018
Elasticities
SH Model Overcomes Restrictive Elasticities
Source country
Elasticity Mex. Market
Share
Costa Rica 19.41Honduras 18.63Venezuela 18.33Australia 3.35USA 2.22
๐๐๐ ๐ ๐๐๐๐๐๐๐๐๐๐๐๐
๐๐๐๐๐๐๐ ๐ ๐๐๐๐
/๐ ๐ ๐๐๐๐
โข Ex.,1% increase in Mexican trade costs in Canada
Standard Prediction: ElasticityMex .MarketShare = ฮธ
SH Prediction: Disproportionately larger response for closecompetitors
Heerman and Sheldon July 25-27, 2018
Elasticities
Implication: Change in policy can alter relative demand
Source Country
Costa Rica 1.0043Honduras 1.0041Venezuela 1.0041Australia 1.0000USA 0.9997Median 1.0000
๐ ๐ ๐๐๐๐โฒ
๐ ๐ ๐๐๐๐โฒ/๐ ๐ ๐๐๐๐๐ ๐ ๐๐๐๐
โข Ex., Canada raises tariffs on Mexican products
Standard Prediction: Relative demand is constant
SH Prediction: Relative demand for Costa Rican productsincreases, and more than others
Heerman and Sheldon July 25-27, 2018
Conclusion
Heerman and Sheldon July 25-27, 2018
Conclusion
โข Standard gravity models will be misleading if IIE does not hold
โ Systematic forces influence comparative advantage withinagriculture
โข SH gravity generates variation in bilateral elasticities
โ These models and AGE models built on them capture howintra-sector comparative advantage drives the response topolicy change
โข SH gravity permits analysis of policy at the product level
โ Changes in the distribution of trade costs within the sectorcan be analyzed from a single equation
Heerman and Sheldon July 25-27, 2018
Elasticities
Heerman and Sheldon July 25-27, 2018
Trade Elasticities
SH Elasticity Elasticity of market share with respect to competitortrade costs
โฯniโฯnl
ฯnlฯni
=ฮธ
ฯni(cov (ฯni (j), ฯnl(j)) + ฯni ร ฯnl) l 6= i
EK Elasticity Constant elasticity across exporters
โฯniโฯnl
ฯnlฯni
= ฮธ ร ฯnl l 6= i
Heerman and Sheldon July 25-27, 2018
Estimation
Empirical distribution of expenditure: dFEn(E)dFฮฝn(ฮฝ)
โข List of 1000 products purchased in market n
โข Each product is represented in proportion to import share
โ If j=wheat is 50% of country n imports, 500 entries areE (wheat)
โข Each draw from dFEn(E) associated with vector of randomnormal draws
โ โData setโ of ns products for each market: dFEn(E)dFฮฝn(ฮฝ)
Go Back .
Heerman and Sheldon July 25-27, 2018
Parameter Estimates
Variation in effect of high elevation land
0
5
10
15
20
25
-15 -12 -9 -6 -3 0 3 6 9 12 15
Num
ber o
f tra
ded
prod
ucts
, Tho
usan
ds
Product-specific effect
Frequency plot: High elevation effect
Heerman and Sheldon July 25-27, 2018
Parameter Estimates: Trade Costs
Common Border -1.76*** 3.13***
Common Language 1.24*** 0.95***
Common RTA 0.19** -0.11
Distance 1 -5.28*** 2.36***
Distance 2 -7.67*** 2.33***
Distance 3 -7.43*** -0.16
Distance 4 -9.95*** 1.37***
Distance 5 -11.56*** -0.04
Distance 6 -12.94*** 0.64***
Country Pair Characteristics
Mean Effect
Unobserved Heterogeneity
๐๐๐๐๐๐ ฮฒ
๐บ๐บ
๐บ๐บ๐๐
โข Large ฯt implies signifcant unexplained variation