advancements in gravity models of spatial...
TRANSCRIPT
![Page 1: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/1.jpg)
Advancements in Gravity Models of Spatial Economics
Costas Arkolakis1
1Yale University and NBER
Columbia, PER Lecture
January 31, 2017
![Page 2: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/2.jpg)
Frameworks for Spatial Econ (Trade, Geography, Urban)
I Many sciences, as well as disciplines in economics, based on unified setup
I E.g. standard model in physics, growth model in macro etc.
![Page 3: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/3.jpg)
Frameworks for Spatial Econ (Trade, Geography, Urban)
I Many sciences, as well as disciplines in economics, based on unified setup
I E.g. standard model in physics, growth model in macro etc.
![Page 4: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/4.jpg)
Frameworks for Spatial Econ (Trade, Geography, Urban)
I International trade: Heckscher-Ohlin widespread use until mid-90’s
I Geography: Krugman model created an explosion of work in geographyI Urban: Rosen-Roback model main equilibrium framework
I Key shortcoming: intractable with frictions; hard to combine with data
![Page 5: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/5.jpg)
Frameworks for Spatial Econ (Trade, Geography, Urban)
I International trade: Heckscher-Ohlin widespread use until mid-90’s
I Geography: Krugman model created an explosion of work in geographyI Urban: Rosen-Roback model main equilibrium framework
I Key shortcoming: intractable with frictions; hard to combine with data
![Page 6: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/6.jpg)
A Unified Framework for Spatial Economics
I International trade: Heckscher-Ohlin widespread use until mid-90’s
I Geography: Krugman model created an explosion of work in geographyI Urban: Rosen-Roback model main equilibrium framework
I Key shortcoming: intractable with frictions; hard to combine with data
I Recently, trade/geography economists motivated by modeling & databreakthroughs developed an alternative: the so-called ‘gravity framework’
I I argue that this new setup has large advantages for spatial economics
I Offers tractable analytical framework for empirical applications...I ...in addition, can serve as a unified setup for trade, geography, urban
![Page 7: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/7.jpg)
A Unified Framework for Spatial Econ: What do we Need?
I This framework ought to satisfy the following properties
1. Be analytically tractable but rich
I Micro-foundationsI Analytical expressions from consumer/firm choice and nice aggregationI Flexible enough to model complicated spatial linkages
2. Have a clear mapping to the data
I Model variables correspond to national statistics & link to micro dataI Offer an easy setup to estimate key parameters
3. Have desirable theoretical properties
I Positive (existence, uniqueness, ease of computation)I Normative (links welfare and openness, disciplined comparative statics)
![Page 8: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/8.jpg)
A Unified Framework for Spatial Econ: What do we Need?
I This framework ought to satisfy the following properties
1. Be analytically tractable but rich
I Micro-foundationsI Analytical expressions from consumer/firm choice and nice aggregationI Flexible enough to model complicated spatial linkages
2. Have a clear mapping to the data
I Model variables correspond to national statistics & link to micro dataI Offer an easy setup to estimate key parameters
3. Have desirable theoretical properties
I Positive (existence, uniqueness, ease of computation)I Normative (links welfare and openness, disciplined comparative statics)
![Page 9: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/9.jpg)
A Unified Framework for Spatial Econ: What do we Need?
I This framework ought to satisfy the following properties
1. Be analytically tractable but rich
I Micro-foundationsI Analytical expressions from consumer/firm choice and nice aggregationI Flexible enough to model complicated spatial linkages
2. Have a clear mapping to the data
I Model variables correspond to national statistics & link to micro dataI Offer an easy setup to estimate key parameters
3. Have desirable theoretical properties
I Positive (existence, uniqueness, ease of computation)I Normative (links welfare and openness, disciplined comparative statics)
![Page 10: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/10.jpg)
A Unified Framework for Spatial Econ: The Gravity Model
I General equilibrium gravity model easily passes first two tests...
1. Analytically tractable but rich
I Micro-foundations: Anderson, DFS/Eaton Kortum (EK), Melitz/ChaneyI Analytical expressions and nice aggregation: Standard GE frameworkI Flexible enough to model complicated spatial linkages: EK, Allen Arkolakis
2. Have a clear mapping to the data
I Model variables correspond to national statistics & micro data: Dekle EK,EK Kramarz
I Offers easy setup to estimate key parameters: EK
![Page 11: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/11.jpg)
A Unified Framework for Spatial Econ: The Gravity Model
I ...third test is a conundrum.
1. Analytically tractable but rich
I Micro-foundations: Anderson, DFS/Eaton Kortum (EK), Melitz/ChaneyI Analytical expressions and nice aggregation: Standard GE frameworkI Flexible enough to model complicated spatial linkages: EK, Allen Arkolakis
2. Have a clear mapping to the data
I Model variables correspond to national statistics & micro data: Dekle EK,EK Kramarz
I Offers easy setup to estimate key parameters: EK
3. Proving theoretical properties has proven an enormous challenge
I Its empirical success notwithstanding, until recently, little known about itsproperties
![Page 12: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/12.jpg)
A Unified Framework for Spatial Econ: The Gravity Model
I ...third test is a conundrum.
1. Analytically tractable but rich
I Micro-foundations: Anderson, DFS/Eaton Kortum (EK), Melitz/ChaneyI Analytical expressions and nice aggregation: Standard GE frameworkI Flexible enough to model complicated spatial linkages: EK, Allen Arkolakis
2. Have a clear mapping to the data
I Model variables correspond to national statistics & micro data: Dekle EK,EK Kramarz
I Offers easy setup to estimate key parameters: EK
3. Proving theoretical properties has proven an enormous challengeI Its empirical success notwithstanding, until recently, little known about its
properties
![Page 13: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/13.jpg)
What About the Theoretical Properties of the Model?
I The GE gravity model is a formidable system!
I Best case scenario, N locations equations/unknowns+GE interactionsI Black-boxy system with daunting solution. Heterogeneity amplifies
’anything goes’
I Agglomeration externalities in geography (labor mobile) urban (knowledgespillovers) make solution a true nightmare
I Work in past 5 years offers sharp characterization of gravity model:
I Extremely versatile setup. Works well for trade, geography and urban
I A battery of mathematical tools can be used for its characterization, e.g.non-linear equations theory, integral equations, perturbation theory etc.
I Rapidly expanding literature:
I Discussion based on results/model in Allen Arkolakis (AA) ’14, AATakahashi ’14 (AAT), AA and Li ’14 (AAL14), AA and Li ’15 (AAL15)and some earlier results by Arkolakis, Costinot Rodriguez-Clare (ACR) ’12
![Page 14: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/14.jpg)
What About the Theoretical Properties of the Model?
I The GE gravity model is a formidable system!
I Best case scenario, N locations equations/unknowns+GE interactions
I Black-boxy system with daunting solution. Heterogeneity amplifies’anything goes’
I Agglomeration externalities in geography (labor mobile) urban (knowledgespillovers) make solution a true nightmare
I Work in past 5 years offers sharp characterization of gravity model:
I Extremely versatile setup. Works well for trade, geography and urban
I A battery of mathematical tools can be used for its characterization, e.g.non-linear equations theory, integral equations, perturbation theory etc.
I Rapidly expanding literature:
I Discussion based on results/model in Allen Arkolakis (AA) ’14, AATakahashi ’14 (AAT), AA and Li ’14 (AAL14), AA and Li ’15 (AAL15)and some earlier results by Arkolakis, Costinot Rodriguez-Clare (ACR) ’12
![Page 15: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/15.jpg)
What About the Theoretical Properties of the Model?
I The GE gravity model is a formidable system!
I Best case scenario, N locations equations/unknowns+GE interactionsI Black-boxy system with daunting solution. Heterogeneity amplifies
’anything goes’
I Agglomeration externalities in geography (labor mobile) urban (knowledgespillovers) make solution a true nightmare
I Work in past 5 years offers sharp characterization of gravity model:
I Extremely versatile setup. Works well for trade, geography and urban
I A battery of mathematical tools can be used for its characterization, e.g.non-linear equations theory, integral equations, perturbation theory etc.
I Rapidly expanding literature:
I Discussion based on results/model in Allen Arkolakis (AA) ’14, AATakahashi ’14 (AAT), AA and Li ’14 (AAL14), AA and Li ’15 (AAL15)and some earlier results by Arkolakis, Costinot Rodriguez-Clare (ACR) ’12
![Page 16: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/16.jpg)
What About the Theoretical Properties of the Model?
I The GE gravity model is a formidable system!
I Best case scenario, N locations equations/unknowns+GE interactionsI Black-boxy system with daunting solution. Heterogeneity amplifies
’anything goes’
I Agglomeration externalities in geography (labor mobile) urban (knowledgespillovers) make solution a true nightmare
I Work in past 5 years offers sharp characterization of gravity model:
I Extremely versatile setup. Works well for trade, geography and urban
I A battery of mathematical tools can be used for its characterization, e.g.non-linear equations theory, integral equations, perturbation theory etc.
I Rapidly expanding literature:
I Discussion based on results/model in Allen Arkolakis (AA) ’14, AATakahashi ’14 (AAT), AA and Li ’14 (AAL14), AA and Li ’15 (AAL15)and some earlier results by Arkolakis, Costinot Rodriguez-Clare (ACR) ’12
![Page 17: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/17.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 18: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/18.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 19: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/19.jpg)
Gravity: The Idea
I The idea: We can write bilateral flows (trade, commuting etc) as
Xij = Kijγiδj
I Kij can describe bilateral frictions etc.I Motivated by empirical success of reduced form gravity (Tinbergen ’62):
ln Xij = a ln τij + b ln Ei + c ln Ej
where τij is bilateral trade cost, Ej is country spending
![Page 20: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/20.jpg)
Example: Trade Gravity
I The idea: We can write bilateral flows (trade, commuting etc) as
Xij = Kijγiδj
I Anderson ’79: each location differentiated variety, CES w/ elasticity σI Bilateral trade given by
Xij =
(wi
Aiτij
)1−σ∑
k
(wk
Akτkj
)1−σ︸ ︷︷ ︸share of market i
Ej
I Ai is productivity, τij is iceberg cost, wi , is wage rateI We call ε ≡ 1− σ the ‘trade elasticity’
![Page 21: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/21.jpg)
Example: Trade Gravity and the Data
I The idea: We can write bilateral flows (trade, commuting etc) as
Xij = Kijγiδj
I Anderson ’79: each location differentiated variety, CES w/ elasticity σI Bilateral trade given by
Xij = (τij)1−σ︸ ︷︷ ︸×τεij
(wi
Ai
)1−σ
︸ ︷︷ ︸γi
× 1∑k
(wk
Akτkj
)1−σ Ej
︸ ︷︷ ︸δj
I You can formulate fixed effects specification (EK/Redding Venables)
Xij = γiδjτεij =⇒ ln Xij = ε ln τij + ln γi + ln δj
I Thousands of applications.I Empirical counterpart for aggregate variables (GDP, trade, prices, tariffs)I Micro-data can be used without affecting macro structure
![Page 22: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/22.jpg)
Example: Trade Gravity and the Data
I The idea: We can write bilateral flows (trade, commuting etc) as
Xij = Kijγiδj
I Anderson ’79: each location differentiated variety, CES w/ elasticity σI Bilateral trade given by
Xij = (τij)1−σ︸ ︷︷ ︸×τεij
(wi
Ai
)1−σ
︸ ︷︷ ︸γi
× 1∑k
(wk
Akτkj
)1−σ Ej
︸ ︷︷ ︸δj
I You can formulate fixed effects specification (EK/Redding Venables)
Xij = γiδjτεij =⇒ ln Xij = ε ln τij + ln γi + ln δj
I Thousands of applications.I Empirical counterpart for aggregate variables (GDP, trade, prices, tariffs)I Micro-data can be used without affecting macro structure
![Page 23: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/23.jpg)
Example: Trade Gravity and the Data
I The idea: We can write bilateral flows (trade, commuting etc) as
Xij = Kijγiδj
I Anderson ’79: each location differentiated variety, CES w/ elasticity σI Bilateral trade given by
Xij = (τij)1−σ︸ ︷︷ ︸×τεij
(wi
Ai
)1−σ
︸ ︷︷ ︸γi
× 1∑k
(wk
Akτkj
)1−σ Ej
︸ ︷︷ ︸δj
I You can formulate fixed effects specification (EK/Redding Venables)
Xij = γiδjτεij =⇒ ln Xij = ε ln τij + ln γi + ln δj
I Thousands of applications.I Empirical counterpart for aggregate variables (GDP, trade, prices, tariffs)I Micro-data can be used without affecting macro structure
![Page 24: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/24.jpg)
Trade Gravity: Other Micro-foundations
I Many microfoundations for the gravity trade model
I Perfect competition: Armington/Anderson ’79. EK w/ Frechet for DFSI Monopolistic competition homog. firms: Bergstrand ’85, Arkolakis Klenow
Demidova Rodriguez-Clare ’08 in GE (Krugman ’80)I Monopolistic competition heterogenous. firms: Chaney ’08 (Melitz ’03)I Bertrand competition: Bernard, Eaton, Jensen and Kortum ’03
I This model has been recently adapted to
I economic geography (AA, Redding ’16, Ramondo et al, Caliendo et al)I urban (Ahlfeldt, Redding, Sturm, Wolf ’15, AAL15, Monte et al ’15)
I Gravity flows in commuting.
![Page 25: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/25.jpg)
Trade Gravity: Other Micro-foundations
I Many microfoundations for the gravity trade model
I Perfect competition: Armington/Anderson ’79. EK w/ Frechet for DFSI Monopolistic competition homog. firms: Bergstrand ’85, Arkolakis Klenow
Demidova Rodriguez-Clare ’08 in GE (Krugman ’80)I Monopolistic competition heterogenous. firms: Chaney ’08 (Melitz ’03)I Bertrand competition: Bernard, Eaton, Jensen and Kortum ’03
I This model has been recently adapted to
I economic geography (AA, Redding ’16, Ramondo et al, Caliendo et al)I urban (Ahlfeldt, Redding, Sturm, Wolf ’15, AAL15, Monte et al ’15)
I Gravity flows in commuting.
![Page 26: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/26.jpg)
Urban Gravity: Analytics and Micro-foundations
I Workers choose residential (i) and working (j) location after observing
Frechet preference shock as in Ahlfeldt et al: F (v) = e−v−θ
I In AAL15 we consider this idea in a GE urban framework
I Obtain bilateral commuting flows in a ’gravity’ form
Lij =
(uiwj
Piµij
)θ∑i ′j′
(ui ′wj′
Pi ′µi ′j′
)θ︸ ︷︷ ︸
W
L = W−θ (µij)−θ(
ui
Pi
)θwθj L
where µij commuting utility cost,W is an endogenous object (expectedwelfare)
![Page 27: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/27.jpg)
Urban Gravity: Analytics and Micro-foundations
I Workers choose residential (i) and working (j) location after observing
Frechet preference shock as in Ahlfeldt et al: F (v) = e−v−θ
I In AAL15 we consider this idea in a GE urban framework
I Obtain bilateral commuting flows in a ’gravity’ form
Lij =
(uiwj
Piµij
)θ∑i ′j′
(ui ′wj′
Pi ′µi ′j′
)θ︸ ︷︷ ︸
W
L = W−θ (µij)−θ(
ui
Pi
)θwθj L
where µij commuting utility cost,W is an endogenous object (expectedwelfare)
![Page 28: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/28.jpg)
Next Step: General Equilibrium
I We discussed how to create spatial linkages across locations (e.g. cities,countries) using gravity equations for Xij , Lij
I Trade or commuting are just two examples
I Next step: use spatial linkages+equilibrium equations to close model
I Goal: formulate a computable system of equations/unknowns.I Assume labor is only factor (can be relaxed). Yi = wiLi
![Page 29: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/29.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 30: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/30.jpg)
Equilibrium Equations: Trade Module
I A GE model with trade needs to satisfy two equilibrium conditions
I “Goods market clearing”:
Yi =∑j∈S
Xi j ∀i ∈ S
I “Budget Balance”:
Ei =∑j∈S
Xj i ∀i ∈ S
I Note: it may be that Ei 6= Yi
![Page 31: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/31.jpg)
Equilibrium Equations: Urban Module
I GE model with urban flows satisfies commuting equilibrium equations
I Total output in i is equal to total earnings:
Yi =∑j∈S
wiLj i
I Total spending in i is equal to what earned everywhere:
Ei =∑j∈S
wjLi j
I We developed essential components for trade, geography, urban models
I Gravity (trade and commuting)I Equilibrium modules (trade and urban)I Now we define and analyze the GE of these models
![Page 32: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/32.jpg)
Equilibrium Equations: Urban Module
I GE model with urban flows satisfies commuting equilibrium equations
I Total output in i is equal to total earnings:
Yi =∑j∈S
wiLj i
I Total spending in i is equal to what earned everywhere:
Ei =∑j∈S
wjLi j
I We developed essential components for trade, geography, urban models
I Gravity (trade and commuting)I Equilibrium modules (trade and urban)
I Now we define and analyze the GE of these models
![Page 33: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/33.jpg)
Equilibrium Equations: Urban Module
I GE model with urban flows satisfies commuting equilibrium equations
I Total output in i is equal to total earnings:
Yi =∑j∈S
wiLj i
I Total spending in i is equal to what earned everywhere:
Ei =∑j∈S
wjLi j
I We developed essential components for trade, geography, urban models
I Gravity (trade and commuting)I Equilibrium modules (trade and urban)I Now we define and analyze the GE of these models
![Page 34: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/34.jpg)
Closing the Trade Model
I In trade models (with no deficit) we have Ei = Yi
I Equilibrium is trade gravity+trade module.
I Solve wi ,Pi using
wσi =
∑j∈S
(τij)1−σ L−1i Aσ−1i uj
σ−1LjwjPσ−1j
P1−σi =
∑j∈S
(τji )1−σ Aσ−1j (wj)
1−σ
I We intentionally avoided substituting the price index.
I Crucial to write it this way, as it is much easier to characterize
![Page 35: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/35.jpg)
Closing the Trade Model
I In trade models (with no deficit) we have Ei = Yi
I Equilibrium is trade gravity+trade module.
I Solve wi ,Pi using
wσi =
∑j∈S
(τij)1−σ L−1i Aσ−1i uj
σ−1LjwjPσ−1j
P1−σi =
∑j∈S
(τji )1−σ Aσ−1j (wj)
1−σ
I We intentionally avoided substituting the price index.
I Crucial to write it this way, as it is much easier to characterize
![Page 36: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/36.jpg)
Geography Model: Local Spillovers and Isomorphisms
I In economic geography, Li is endogenous
I We model in addition, as in AA, local spillovers:
I productivity Ai = AiLαi , amenity ui = uiL
βi
I Different α, β isomorphic to different economic geography models
I E.g. Monopolistic competition with free entry: α = 1σ−1 .
I As in Krugman ’91
I Cobb-Douglas preferences over non-tradable sector: β = − 1−γγ .
I As in Helpman ’98, Redding Sturm ’08
I Also utility shocks (Redding ’16) land in production (Donaldson Hornbeck)
![Page 37: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/37.jpg)
Geography Model: Local Spillovers and Isomorphisms
I In economic geography, Li is endogenous
I We model in addition, as in AA, local spillovers:
I productivity Ai = AiLαi , amenity ui = uiL
βi
I Different α, β isomorphic to different economic geography models
I E.g. Monopolistic competition with free entry: α = 1σ−1 .
I As in Krugman ’91
I Cobb-Douglas preferences over non-tradable sector: β = − 1−γγ .
I As in Helpman ’98, Redding Sturm ’08
I Also utility shocks (Redding ’16) land in production (Donaldson Hornbeck)
![Page 38: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/38.jpg)
Geography Model: Local Spillovers and Isomorphisms
I In economic geography, Li is endogenous
I We model in addition, as in AA, local spillovers:
I productivity Ai = AiLαi , amenity ui = uiL
βi
I Different α, β isomorphic to different economic geography models
I E.g. Monopolistic competition with free entry: α = 1σ−1 .
I As in Krugman ’91
I Cobb-Douglas preferences over non-tradable sector: β = − 1−γγ .
I As in Helpman ’98, Redding Sturm ’08
I Also utility shocks (Redding ’16) land in production (Donaldson Hornbeck)
![Page 39: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/39.jpg)
Geography Model: Equilibrium Equations
I Equilibrium is trade gravity+trade module+
I Utility equalization Wi ≡ wi
Piui = W
I Aggregate labor clears∑
i Li = L
I Solve wi , Li ,W using
W σ−1Li1−α(σ−1)wσ
i =N∑j=1
τ 1−σij Ai
σ−1ujσ−1Lj
1+β(σ−1)wσj
W σ−1wi1−σLi
β(1−σ) =N∑j=1
τ 1−σji Aj
σ−1uiσ−1wj
1−σLα(σ−1)j
and of course∑
i Li = L.
I Existence and uniqueness in AA and AAT: notice same mathematicalstructure as in the trade model.
I Except now welfare is the eigenvalue of the system
![Page 40: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/40.jpg)
Summary of GE Gravity Trade & Geography Models
I GE gravity trade (Anderson ’79: solve for wi ,Pi )
wσi =
N∑j=1
τ1−σij Aσ−1i LjPσ−1j wj
P1−σi =
N∑j=1
τ1−σji Aσ−1j w1−σj
I GE geography (AA: welfare equalizes, solve for W ,wi , Li )
W σ−1Li1−α(σ−1)wσ
i =N∑j=1
τ1−σij Aiσ−1uj
σ−1Lj1+β(σ−1)wσ
j
W σ−1wi1−σLi
β(1−σ) =N∑j=1
τ1−σji Ajσ−1ui
σ−1wj1−σL
α(σ−1)j
and total population constraint∑
j Lj = L
![Page 41: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/41.jpg)
Comparison: Kernel
I GE gravity trade (Anderson ’79: solve for wi ,Pi )
wσi =
N∑j=1
τ1−σij Aσ−1i LjPσ−1j wj
P1−σi =
N∑j=1
τ1−σji Aσ−1j w1−σj
I GE geography (AA: welfare equalizes, solve for W ,wi , Li )
W σ−1Li1−α(σ−1)wσ
i =N∑j=1
τ1−σij Aiσ−1uj
σ−1Lj1+β(σ−1)wσ
j
W σ−1wi1−σLi
β(1−σ) =N∑j=1
τ1−σji Ajσ−1ui
σ−1wj1−σL
α(σ−1)j
and total population constraint∑
j Lj = L
![Page 42: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/42.jpg)
Comparison: Global Parameters
I GE gravity trade (Anderson ’79: solve for wi ,Pi )
wσi =
N∑j=1
τ1−σij Aσ−1i LjPσ−1j wj
P1−σi =
N∑j=1
τ1−σji Aσ−1j w1−σj
I GE geography (AA: welfare equalizes, solve for W ,wi , Li )
W σ−1Li1−α(σ−1)wσ
i =N∑j=1
τ1−σij Aiσ−1uj
σ−1Lj1+β(σ−1)wσ
j
W σ−1wi1−σLi
β(1−σ) =N∑j=1
τ1−σji Ajσ−1ui
σ−1wj1−σL
α(σ−1)j
and total population constraint∑
j Lj = L
![Page 43: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/43.jpg)
Comparison: Eigenvalues
I GE gravity trade (Anderson ’79: solve for wi ,Pi )
1 wσi =
N∑j=1
τ1−σij Aσ−1i LjPσ−1j wj
1 P1−σi =
N∑j=1
τ1−σji Aσ−1j w1−σj
I GE geography (AA: welfare equalizes, solve for W ,wi , Li )
W σ−1Li1−α(σ−1)wσ
i =N∑j=1
τ1−σij Aiσ−1uj
σ−1Lj1+β(σ−1)wσ
j
W σ−1wi1−σLi
β(1−σ) =N∑j=1
τ1−σji Ajσ−1ui
σ−1wj1−σL
α(σ−1)j
and total population constraint∑
j Lj = L
![Page 44: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/44.jpg)
Urban Model: Spatial Spillovers
I We now turn to consider the urban model (trade+commuting)
I Agglomerations are important for cities’ economic activity (Fujita Thisse’02, Glaeser Gottlieb ’09, Moretti ’11, Davis Dingel ’12)
I A most crucial: spatial knowledge spillover (Fujita-Ogawa ’82)I Turns out: easy to extend this framework to introduce this spatial spilloverI Assume that productivity in a location depend on the number of spatial
interactions with other nearby workers, Lj
Ai =∑j
Kij (Lj)η
I Kij represents spatial knowledge links. η is the degree of spillover
I Microfoundations for this functional form presented in AAL15
![Page 45: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/45.jpg)
Urban Model: Spatial Spillovers
I We now turn to consider the urban model (trade+commuting)
I Agglomerations are important for cities’ economic activity (Fujita Thisse’02, Glaeser Gottlieb ’09, Moretti ’11, Davis Dingel ’12)
I A most crucial: spatial knowledge spillover (Fujita-Ogawa ’82)I Turns out: easy to extend this framework to introduce this spatial spilloverI Assume that productivity in a location depend on the number of spatial
interactions with other nearby workers, Lj
Ai =∑j
Kij (Lj)η
I Kij represents spatial knowledge links. η is the degree of spillover
I Microfoundations for this functional form presented in AAL15
![Page 46: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/46.jpg)
Closing the Urban Model (AAL15)
I Equilibrium isI gravity for trade+trade module+I gravity for commuting+urban module+I spatial spillovers
I Solve for Ei ,Yi , Li ,wi ,Ai ,W , in the following 5 sets of equations
trade module: Yi =∑j∈S
Xij , Ei =∑j∈S
Xji
urban module: Ei =∑j
wjLij , Yi =∑j
wiLji
spatial spillover: Ai =∑j
Kij (Lj)η
This general structure incorporates all the previous models as subcasesI Trade module: Armington, AA, AATI Urban module: Ahfeldt, Redding, Sturm, Wolf (two factors)I Trade+urban+spatial spillovers module: AAL15 (Skip illustration)
![Page 47: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/47.jpg)
Closing the Urban Model (AAL15)
I Equilibrium isI gravity for trade+trade module+I gravity for commuting+urban module+I spatial spillovers
I Solve for Ei ,Yi , Li ,wi ,Ai ,W , in the following 5 sets of equations
trade module: Yi =∑j∈S
Xij , Ei =∑j∈S
Xji
urban module: Ei =∑j
wjLij , Yi =∑j
wiLji
spatial spillover: Ai =∑j
Kij (Lj)η
This general structure incorporates all the previous models as subcasesI Trade module: Armington, AA, AATI Urban module: Ahfeldt, Redding, Sturm, Wolf (two factors)I Trade+urban+spatial spillovers module: AAL15 (Skip illustration)
![Page 48: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/48.jpg)
Visualization of the Spatial Links
![Page 49: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/49.jpg)
Visualization of Additional Countries
![Page 50: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/50.jpg)
Visualization of Additional Spatial Links
![Page 51: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/51.jpg)
Visualization of Additional Spatial Links
![Page 52: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/52.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 53: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/53.jpg)
Analytical Solution of a Geography Model
I I now provide example of how distance costs/border change geographyI Consider trade on the line S = [−π, π],I Global parameters: α = β = 0, σ > 0I Kernel: u (i) = A (i) = 1, τ (i , j) = eτ |i−j| for all i , j ∈ S .
I Equilibrium written as an integral equation or a differential equation
I Same differential equation in space as the pendulum in timeI Like a pendulum, strength of agglomeration force proportional to distance
from center and symmetric.
I In this special case, there exists a closed form solution (!):
L (i) = c1 cos (ki)2σ−1σ−1
I c1, k depend on eigenvalue. Agglomeration force increases with τ .
![Page 54: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/54.jpg)
Analytical Solution of a Geography Model
I I now provide example of how distance costs/border change geographyI Consider trade on the line S = [−π, π],I Global parameters: α = β = 0, σ > 0I Kernel: u (i) = A (i) = 1, τ (i , j) = eτ |i−j| for all i , j ∈ S .
I Equilibrium written as an integral equation or a differential equation
I Same differential equation in space as the pendulum in timeI Like a pendulum, strength of agglomeration force proportional to distance
from center and symmetric.
I In this special case, there exists a closed form solution (!):
L (i) = c1 cos (ki)2σ−1σ−1
I c1, k depend on eigenvalue. Agglomeration force increases with τ .
![Page 55: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/55.jpg)
Increasing trade costs τ
![Page 56: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/56.jpg)
Increasing trade costs τ
![Page 57: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/57.jpg)
Increasing trade costs τ
![Page 58: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/58.jpg)
Increasing trade costs τ
![Page 59: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/59.jpg)
Increasing trade costs τ
![Page 60: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/60.jpg)
Increasing trade costs τ
![Page 61: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/61.jpg)
Building a Border
I Now add a border in the middle (on top of trade cost)
I The solution becomes
L (i) = (c1 cos (ki) + c2 sin (ki))2σ−1σ−1
I Same differential equation in space as the spring in timeI Like a spring, strength of agglomeration force proportional to distance but
border introduces assymetry.
![Page 62: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/62.jpg)
Building a border
![Page 63: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/63.jpg)
Building a border
![Page 64: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/64.jpg)
Building a border
![Page 65: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/65.jpg)
Building a border
![Page 66: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/66.jpg)
Urban Model: Analytical Solution in the Circle
I We consider the urban model on a circle [−π, π]
I No trade costs, but any commuting costs with distance.
I Spatial knowledge spillovers Kij = cos2(
i−j2
), i , j ∈ [−π, π]
I Solution L (i) = (C1 + C2 sin (i + C3))(σ−1)
I C1 > C2 ≥ 0 are determined by normalization C3 can be arbitrary.
I Larger η more agglomeration
I η (σ − 1) ≤ 1 unique equilibrium, |η (σ − 1)| > 1 multiple (any point canbe center)
![Page 67: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/67.jpg)
Analytical Solution
![Page 68: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/68.jpg)
Analytical Solution
![Page 69: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/69.jpg)
Analytical Solution
![Page 70: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/70.jpg)
Analytical Solution
![Page 71: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/71.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 72: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/72.jpg)
Positive Properties
I 1 or 2-equation versions of GE gravity spatial model very tractableI Based on tools by Karlin Nirenberg ’67, see AAT
I ’universal gravity’: map trade-geography gravity models to ’aggregatefactor mobility’ economy
I Qi : factor available in i , pi its factor price.I Aggregate factor demand
piτijQij =(piτij)
−φ∑k∈S (pkτkj)−φ
pjQj ; Pi ≡
(∑k∈S
(pkτkj)−φ
)− 1φ
I Aggregate factor supply
Qi =Ci (pi/Pi )
ψ∑Nk=1 Ck (pk/Pk)ψ
Q
I Aggregate factor market clearing Qi =∑N
k=1 τijQij
I Key parameters: aggregate demand, supply elasticities, φ, ψ
![Page 73: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/73.jpg)
Positive Properties
I 1 or 2-equation versions of GE gravity spatial model very tractableI Based on tools by Karlin Nirenberg ’67, see AAT
I ’universal gravity’: map trade-geography gravity models to ’aggregatefactor mobility’ economy
I Qi : factor available in i , pi its factor price.I Aggregate factor demand
piτijQij =(piτij)
−φ∑k∈S (pkτkj)−φ
pjQj ; Pi ≡
(∑k∈S
(pkτkj)−φ
)− 1φ
I Aggregate factor supply
Qi =Ci (pi/Pi )
ψ∑Nk=1 Ck (pk/Pk)ψ
Q
I Aggregate factor market clearing Qi =∑N
k=1 τijQij
I Key parameters: aggregate demand, supply elasticities, φ, ψ
![Page 74: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/74.jpg)
Positive Properties
I 1 or 2-equation versions of GE gravity spatial model very tractableI Based on tools by Karlin Nirenberg ’67, see AAT
I ’universal gravity’: map trade-geography gravity models to ’aggregatefactor mobility’ economy
I Qi : factor available in i , pi its factor price.I Aggregate factor demand
piτijQij =(piτij)
−φ∑k∈S (pkτkj)−φ
pjQj ; Pi ≡
(∑k∈S
(pkτkj)−φ
)− 1φ
I Aggregate factor supply
Qi =Ci (pi/Pi )
ψ∑Nk=1 Ck (pk/Pk)ψ
Q
I Aggregate factor market clearing Qi =∑N
k=1 τijQij
I Key parameters: aggregate demand, supply elasticities, φ, ψ
![Page 75: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/75.jpg)
Aggregate Factor Mobility: Trade Model
Aggregate Factor Demand, elasticity φ=σ-1
Aggregate Factor Supply, elasticity ψ=0
Ln Qi
Ln Pi
![Page 76: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/76.jpg)
Aggregate Factor Mobility: Trade Model w/ Intermediates
Aggregate Factor Demand, elasticity φ=σ-1
Aggregate Factor Supply, elasticity ψ=(1-ζ)/ζ
Ln Qi
Ln Pi
![Page 77: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/77.jpg)
Aggregate Factor Mobility: Geography Economy
Aggregate Factor Demand, elasticity φ=σ-1
Aggregate Factor Supply
No extern, α+β=0-> ψ=-1/(α+β)=+∞
Ln Qi
Ln Pi
![Page 78: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/78.jpg)
Aggregate Factor Mobility: Geography w/ spillovers
Aggregate Factor Demand, elasticity φ=σ-1
α+β<0-> ψ=-1/(α+β)>0
Ln Qi
Ln Pi
α+β>0-> ψ=-1/(α+β)<0
![Page 79: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/79.jpg)
Positive Properties
I Straightforward and intuitive resultsI If σ 6= −ψ then equilibrium exists (non existence requires equal slopes)I If σ − 1 > −ψ uniqueness is guaranteed (demand steeper than supply)
I To analyze general urban model, bigger systems need new tools
I So far analysis in urban focused on constructing possible equilibria (FujitaOgawa ’82, Lucas Rossi-Hansberg ’03, Rossi-Hansberg ’05) typicallyassuming symmetry
I How much can we hope for with asymmetry and more equations?
I Turns out, a lot: Existence, Uniqueness, Computation of the Equilibrium!
![Page 80: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/80.jpg)
Positive Properties
I Straightforward and intuitive resultsI If σ 6= −ψ then equilibrium exists (non existence requires equal slopes)I If σ − 1 > −ψ uniqueness is guaranteed (demand steeper than supply)
I To analyze general urban model, bigger systems need new tools
I So far analysis in urban focused on constructing possible equilibria (FujitaOgawa ’82, Lucas Rossi-Hansberg ’03, Rossi-Hansberg ’05) typicallyassuming symmetry
I How much can we hope for with asymmetry and more equations?
I Turns out, a lot: Existence, Uniqueness, Computation of the Equilibrium!
![Page 81: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/81.jpg)
Positive Properties
I Straightforward and intuitive resultsI If σ 6= −ψ then equilibrium exists (non existence requires equal slopes)I If σ − 1 > −ψ uniqueness is guaranteed (demand steeper than supply)
I To analyze general urban model, bigger systems need new tools
I So far analysis in urban focused on constructing possible equilibria (FujitaOgawa ’82, Lucas Rossi-Hansberg ’03, Rossi-Hansberg ’05) typicallyassuming symmetry
I How much can we hope for with asymmetry and more equations?
I Turns out, a lot: Existence, Uniqueness, Computation of the Equilibrium!
![Page 82: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/82.jpg)
A Generalized Gravity ‘Model’
I Suppose equilibrium of a model reduces to a system of eqns where wedenote locations (or sectors/location-sectors) with i , j ∈ {1, ...,N}, eqnswith k, type of variable with h; k, h ∈ {1, ...,H}
λkH∏
h=1
(xhi
)γkh =N∑j=1
K kij
[H∏
h=1
(xhj
)βkh] (1)
I Equilibrium variables xhi : # to be solved H " N (wage, price, labor etc)
I Eigenvalue λk : Its role across models varies (typically welfare)I Kernel K k
ij ≥ 0: spatial links (trade/commuting costs, productivity decayetc)
I Global parameters γkh, βkh ≥ 0 :(EoS, Frechet elast., spillovers etc)
I Γ = {γkh}, B = {βkh} are the corresponding matrices
![Page 83: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/83.jpg)
Theorem: Allen Arkolakis Li ’14
TheoremConsider the system of equations (1).If Γ is invertible then:
(i) If K kij > 0, then there exists a strictly positive solution,
{xhi , λ
k}
Define A ≡ BΓ−1, with element Aij & Ap ≡ {|Aij |}(ii) If K k
ij ≥ 0 and the maximum of the eigenvalues of Ap, ρ (Ap) ≤ 1, thenthere exists at most one strictly positive solution (up-to-scale)
(iii) If K kij > 0 and ρ (Ap) < 1 the unique (up-to-scale) solution can be
computed by a simple iterative procedure
![Page 84: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/84.jpg)
Theorem: Allen Arkolakis Li ’14
TheoremConsider the system of equations (1).If Γ is invertible then:
(i) If K kij > 0, then there exists a strictly positive solution,
{xhi , λ
k}
Define A ≡ BΓ−1, with element Aij & Ap ≡ {|Aij |}(ii) If K k
ij ≥ 0 and the maximum of the eigenvalues of Ap, ρ (Ap) ≤ 1, thenthere exists at most one strictly positive solution (up-to-scale)
(iii) If K kij > 0 and ρ (Ap) < 1 the unique (up-to-scale) solution can be
computed by a simple iterative procedure
![Page 85: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/85.jpg)
Application on Geography and Urban Model
I Note: Convenient conditions on global parameter vector not on KernelI Can handle large dimensionality (many locations etc) like a charm
I The theorem is extremely powerful for economic geography model
I In AA you can prove that equilibrium always exists; is unique if α + β ≤ 0
I Comes also quite handy for urban model with spatial spillovers
I You can prove that existence is always guaranteed (not trivial)I With no trade costs, uniqueness also holds as long if η small
![Page 86: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/86.jpg)
Application on Geography and Urban Model
I Note: Convenient conditions on global parameter vector not on KernelI Can handle large dimensionality (many locations etc) like a charm
I The theorem is extremely powerful for economic geography model
I In AA you can prove that equilibrium always exists; is unique if α + β ≤ 0
I Comes also quite handy for urban model with spatial spillovers
I You can prove that existence is always guaranteed (not trivial)I With no trade costs, uniqueness also holds as long if η small
![Page 87: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/87.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity Equations, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 88: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/88.jpg)
Welfare and Gains from Openness
I What about welfare?I We may distinguish the ex-post and ex-ante evaluation of a policy change
I Ex-post: For trade and geography model ACR formula applies(Arkolakis, Costinot, Rodriguez-Clare ’12)
I Ex-post welfare d ln Wj = − d lnλjj
ε (note: welfare equalizes in econgeography)
I Formula remarkably robust to changes in preferences, intermediateinputs/sectors, market structure (Costinot Rodriguez-Clare, ACDR,Midrigan Xu)
![Page 89: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/89.jpg)
Analytical Characterization of Counterfactuals
I Ex-ante: Can we characterize comparative statics/counterfactuals?I GE Elasticities for welfare
εWij =d ln W
d ln τij
or other variables (e.g. wage)
εwl
ij =d ln wl
d ln τij
I General equilibrium theory instills pessimism.
I In addition, characterization requires harnessing network effects of tradeI These effects a core mechanism of gravity models but poorly understood
![Page 90: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/90.jpg)
Analytical Characterization of Counterfactuals
I Ex-ante: Can we characterize comparative statics/counterfactuals?I GE Elasticities for welfare
εWij =d ln W
d ln τij
or other variables (e.g. wage)
εwl
ij =d ln wl
d ln τij
I General equilibrium theory instills pessimism.
I In addition, characterization requires harnessing network effects of tradeI These effects a core mechanism of gravity models but poorly understood
![Page 91: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/91.jpg)
Total Welfare Counterfactuals
I In trade models the derivative is simple (see Atkeson Burstein, Lai et al)
I d lnWd ln τij
=Xij
YW (W here is expenditure weighted welfare, Y W : world GDP)
I Much harder characterization in geography models because of eigenvalueI Need to consider a perturbation.
I With no externalities, α + β = 0, we can write system as
λx = Kx,
λy′ = y′K,
x ≡ [Liwσi ] and y ≡
[w1−σi
], N × 1 vectors, λ ≡W σ−1
I A standard result in perturbation theory is ∂λ∂Kij
=yixjyT x
I Using that you can prove the formula is the same in economic geographyI Note: Things get remarkably more tricky when α + β 6= 0.
![Page 92: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/92.jpg)
Total Welfare Counterfactuals
I In trade models the derivative is simple (see Atkeson Burstein, Lai et al)
I d lnWd ln τij
=Xij
YW (W here is expenditure weighted welfare, Y W : world GDP)
I Much harder characterization in geography models because of eigenvalueI Need to consider a perturbation.
I With no externalities, α + β = 0, we can write system as
λx = Kx,
λy′ = y′K,
x ≡ [Liwσi ] and y ≡
[w1−σi
], N × 1 vectors, λ ≡W σ−1
I A standard result in perturbation theory is ∂λ∂Kij
=yixjyT x
I Using that you can prove the formula is the same in economic geographyI Note: Things get remarkably more tricky when α + β 6= 0.
![Page 93: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/93.jpg)
Sufficient Statistics for Ex-ante Counterfactuals
I 3 types of results can be proven for comparative statics in gravity models
1. Changes in all variables x = x ′/x pinned down by data+globalparameters (e.g. trade, commuting etc)
I Hat algebra: Dekle EK, ACR, Caliendo Parro, Costinot Rodriguez-Clare
2. Total elasticities εWij , εwlij pinned down by data+global parameters
I AAT, AAL. Valid locally with multiple equilibria
3. Total elasticities εWij , εwlij can be represented as geometric series
I In AAT we solve term as geometric sum. E.g. for Armington
εwl
ij =µij
0.5σ
(1 +
(σ − 1
σ
)µli +
(σ − 1
σ
)2
µ2li + ...
),
where {µnil} ≡ X n is the i , lth element of the export share matrix raised to
the power n
![Page 94: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/94.jpg)
Sufficient Statistics for Ex-ante Counterfactuals
I 3 types of results can be proven for comparative statics in gravity models
1. Changes in all variables x = x ′/x pinned down by data+globalparameters (e.g. trade, commuting etc)
I Hat algebra: Dekle EK, ACR, Caliendo Parro, Costinot Rodriguez-Clare
2. Total elasticities εWij , εwlij pinned down by data+global parameters
I AAT, AAL. Valid locally with multiple equilibria
3. Total elasticities εWij , εwlij can be represented as geometric series
I In AAT we solve term as geometric sum. E.g. for Armington
εwl
ij =µij
0.5σ
(1 +
(σ − 1
σ
)µli +
(σ − 1
σ
)2
µ2li + ...
),
where {µnil} ≡ X n is the i , lth element of the export share matrix raised to
the power n
![Page 95: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/95.jpg)
Sufficient Statistics for Ex-ante Counterfactuals
I 3 types of results can be proven for comparative statics in gravity models
1. Changes in all variables x = x ′/x pinned down by data+globalparameters (e.g. trade, commuting etc)
I Hat algebra: Dekle EK, ACR, Caliendo Parro, Costinot Rodriguez-Clare
2. Total elasticities εWij , εwlij pinned down by data+global parameters
I AAT, AAL. Valid locally with multiple equilibria
3. Total elasticities εWij , εwlij can be represented as geometric series
I In AAT we solve term as geometric sum. E.g. for Armington
εwl
ij =µij
0.5σ
(1 +
(σ − 1
σ
)µli +
(σ − 1
σ
)2
µ2li + ...
),
where {µnil} ≡ X n is the i , lth element of the export share matrix raised to
the power n
![Page 96: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/96.jpg)
Sufficient Statistics for Ex-ante Counterfactuals
I 3 types of results can be proven for comparative statics in gravity models
1. Changes in all variables x = x ′/x pinned down by data+globalparameters (e.g. trade, commuting etc)
I Hat algebra: Dekle EK, ACR, Caliendo Parro, Costinot Rodriguez-Clare
2. Total elasticities εWij , εwlij pinned down by data+global parameters
I AAT, AAL. Valid locally with multiple equilibria
3. Total elasticities εWij , εwlij can be represented as geometric series
I In AAT we solve term as geometric sum. E.g. for Armington
εwl
ij =µij
0.5σ
(1 +
(σ − 1
σ
)µli +
(σ − 1
σ
)2
µ2li + ...
),
where {µnil} ≡ X n is the i , lth element of the export share matrix raised to
the power n
![Page 97: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/97.jpg)
Why is this Characterization Useful
I Extremely useful in policyI Consider a weighted graph with infrastructure matrix T = {tij} denoting
the cost of two connected points.I Bilateral trade costs τij depends on tkl on the realized path
I e.g. τij = ti1 × t1k × ...× tlj
I Example: Infrastructure Investment. Want to measure
d ln W
d ln tij=
N∑k=1
N∑l=1
d ln W
d ln τkl× d ln τkl
d ln tij
I We have an analytical characterization of all d lnWd ln τkl
!
I What about d ln τkld ln tij
? Two ways to tackle this.
1. Black box (but cool): Djikstra (Donaldson), Fast Marching Method (AA)2. Analytical characterization (but super cool): New AA
![Page 98: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/98.jpg)
Why is this Characterization Useful
I Extremely useful in policyI Consider a weighted graph with infrastructure matrix T = {tij} denoting
the cost of two connected points.I Bilateral trade costs τij depends on tkl on the realized path
I e.g. τij = ti1 × t1k × ...× tlj
I Example: Infrastructure Investment. Want to measure
d ln W
d ln tij=
N∑k=1
N∑l=1
d ln W
d ln τkl× d ln τkl
d ln tij
I We have an analytical characterization of all d lnWd ln τkl
!
I What about d ln τkld ln tij
? Two ways to tackle this.
1. Black box (but cool): Djikstra (Donaldson), Fast Marching Method (AA)2. Analytical characterization (but super cool): New AA
![Page 99: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/99.jpg)
Why is this Characterization Useful
I Extremely useful in policyI Consider a weighted graph with infrastructure matrix T = {tij} denoting
the cost of two connected points.I Bilateral trade costs τij depends on tkl on the realized path
I e.g. τij = ti1 × t1k × ...× tlj
I Example: Infrastructure Investment. Want to measure
d ln W
d ln tij=
N∑k=1
N∑l=1
d ln W
d ln τkl× d ln τkl
d ln tij
I We have an analytical characterization of all d lnWd ln τkl
!
I What about d ln τkld ln tij
? Two ways to tackle this.
1. Black box (but cool): Djikstra (Donaldson), Fast Marching Method (AA)
2. Analytical characterization (but super cool): New AA
![Page 100: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/100.jpg)
Why is this Characterization Useful
I Extremely useful in policyI Consider a weighted graph with infrastructure matrix T = {tij} denoting
the cost of two connected points.I Bilateral trade costs τij depends on tkl on the realized path
I e.g. τij = ti1 × t1k × ...× tlj
I Example: Infrastructure Investment. Want to measure
d ln W
d ln tij=
N∑k=1
N∑l=1
d ln W
d ln τkl× d ln τkl
d ln tij
I We have an analytical characterization of all d lnWd ln τkl
!
I What about d ln τkld ln tij
? Two ways to tackle this.
1. Black box (but cool): Djikstra (Donaldson), Fast Marching Method (AA)2. Analytical characterization (but super cool): New AA
![Page 101: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/101.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity, Modules, and Models
I Analytical Solution of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 102: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/102.jpg)
1. The Fast Marching Method for Spatial Economics
![Page 103: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/103.jpg)
1. The Fast Marching Method for Spatial Economics
![Page 104: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/104.jpg)
1. The Fast Marching Method for Spatial Economics
![Page 105: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/105.jpg)
1. The Fast Marching Method with an Example
![Page 106: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/106.jpg)
1. The Fast Marching Method with an Example
![Page 107: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/107.jpg)
1. Trade costs with FMM: transportation networks
![Page 108: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/108.jpg)
2. Characterization of Transportation Costs (new AA)
I Assume iceberg cost of directly connected locations tijI For path p = {i = p0, p1, ..., pK = j} realized cost is τij (p) =
∏Kk=1 tpk−1,pk
I Assume each trader incurs path-specific shock, εij (p, ν) ∼Frechet(θ)
I Expected cost: τij = minp∈PK ,K≥0
τij (p) εij (p, ν) ∝∑∞
K=0 AKij , Aij = t−θij
I If spectral radius of A <1, can directly link trade costs to observables:
∞∑K=0
AK = (I− A)−1 ≡ B
I Using this and previous results we can link effect of policy to observables
d ln W
d ln tij=
N∑k=1
N∑l=1
(Xkl
Y W
)(tijτji
c
)−θX
θ1−σ
ji
(XkiXjl
Xkl
) θσ−1
︸ ︷︷ ︸d ln τkl/d ln tij
![Page 109: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/109.jpg)
2. Characterization of Transportation Costs (new AA)
I Assume iceberg cost of directly connected locations tijI For path p = {i = p0, p1, ..., pK = j} realized cost is τij (p) =
∏Kk=1 tpk−1,pk
I Assume each trader incurs path-specific shock, εij (p, ν) ∼Frechet(θ)
I Expected cost: τij = minp∈PK ,K≥0
τij (p) εij (p, ν) ∝∑∞
K=0 AKij , Aij = t−θij
I If spectral radius of A <1, can directly link trade costs to observables:
∞∑K=0
AK = (I− A)−1 ≡ B
I Using this and previous results we can link effect of policy to observables
d ln W
d ln tij=
N∑k=1
N∑l=1
(Xkl
Y W
)(tijτji
c
)−θX
θ1−σ
ji
(XkiXjl
Xkl
) θσ−1
︸ ︷︷ ︸d ln τkl/d ln tij
![Page 110: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/110.jpg)
2. Characterization of Transportation Costs (new AA)
I Assume iceberg cost of directly connected locations tijI For path p = {i = p0, p1, ..., pK = j} realized cost is τij (p) =
∏Kk=1 tpk−1,pk
I Assume each trader incurs path-specific shock, εij (p, ν) ∼Frechet(θ)
I Expected cost: τij = minp∈PK ,K≥0
τij (p) εij (p, ν) ∝∑∞
K=0 AKij , Aij = t−θij
I If spectral radius of A <1, can directly link trade costs to observables:
∞∑K=0
AK = (I− A)−1 ≡ B
I Using this and previous results we can link effect of policy to observables
d ln W
d ln tij=
N∑k=1
N∑l=1
(Xkl
Y W
)(tijτji
c
)−θX
θ1−σ
ji
(XkiXjl
Xkl
) θσ−1
︸ ︷︷ ︸d ln τkl/d ln tij
![Page 111: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/111.jpg)
Roadmap
I Analytical Gravity and Mapping to the Data
I Gravity, Modules, and Models
I Analytical Characterization of Equilibrium
I Positive Properties and Computation of the Equilibrium
I Welfare and Counterfactuals
I Applications
![Page 112: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/112.jpg)
Estimating trade costs with FMM: mode-specific trade
0.2
.4.6
.81
Mod
e-sp
ecifi
c sh
are
ofbi
late
ral t
rade
val
ue
0 .2 .4 .6 .8Normalized straight-line distance
Road Air
Rail Water
Observed trade shares
11.
21.
41.
61.
8E
stim
ated
iceb
erg
trad
e co
st
0 .2 .4 .6 .8Normalized straight-line distance
Road Air
Rail Water
Estimated trade costs
![Page 113: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/113.jpg)
Removing the IHS: Estimated increase in P
![Page 114: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/114.jpg)
Removing the IHS: Cost-benefit analysis
Estimated annual cost of the IHS: ≈ $100 billion
Annualized cost of construction: ≈ $30 billion ($560 billion @5%/year)(CBO, 1982)
Maintenance: ≈ $70 billion (FHA, 2008)
Estimated annual gain of the IHS: ≈ $150− 200 billion
Welfare gain of IHS: 1.1− 1.4%.Given homothetic preferences and holding prices fixed, can multiply welfaregain by U.S. GDP.
Suggests gains from IHS substantially greater than costs.
![Page 115: Advancements in Gravity Models of Spatial Economicska265/teaching/Lectures/Advancements_in_Gravity... · Advancements in Gravity Models of Spatial Economics Costas Arkolakis1 1Yale](https://reader030.vdocuments.site/reader030/viewer/2022040418/5d66528a88c9937a098bd364/html5/thumbnails/115.jpg)
Conclusion
I We developed an analytical GE framework with tight connection to data
I We showed that we can go a long way characterizing this setupI Robust and appealing framework: works equally well for trade, geography,
urbanI Can be extended to keep main properties and model additional phenomena
I What are the future applications?
I Trade and Inequality (Burstein at al ’15, Lee ’15, Gale et al ’15)I Non-CES (Parenti et al ’15, Okubo Picard Thisse ’10, Arkolakis et al)I Spatial sorting (Costinot Vogel ’10, Davis Dingel ’15, Ziv ’15, Gaubert)I Dynamics (EKNR, Takahashi ’16, Caliendo et al ’15, AA ’16)I Infrastructure investment (Fajgelbaum Schaal ’16, new AA)