International Trade, Technology, and the Skill Premium
Ariel Burstein
Jonathan Vogel
UCLA and NBER
Columbia and NBER
January 4, 2013
(UCLA and NBER
Columbia and January
NBER) 4, 2013
1 / 43
Introduction
What are consequences of + in trade costs for skill premium?
Two mechanisms linking ∆trade to ∆s/w
I
Heckscher-Ohlin (H-O)
I
Skill-biased technology
Embed into otherwise standard quantitative trade model
Discipline key parameters: …rm-level skill intensity & other facts
No analytic gravity — alternative approach to match bilateral trade
Counterfactuals: + trade costs, China growth, skill-biased tech ∆
Revisit previous approaches: understimate role of trade on s/w
Model
Technologies
Consumption in merchandise and services
Merchandise and services each aggregator over sectors j
Each sector aggregate of a continuum of varieties (ω, j )
Within each variety, 2 potential producers x country, Bertrand
Iceberg transport cost τ in
1 of shipping from i to n
Firms
Production function
Country n …rm in (ω, j ) with productivity z produces
1
y = An (j ) αjρ z 2φ h
ρ 1
ρ
+ (1
1
α j ) ρ z 2 (1
φ)
l
ρ 1
ρ
I
αj determines relative importance of skilled labor in sector j
I
An (j ) Hicks-neutral sectoral TFP
F
A n (j ) = T n
T n (j )
I
φ determines skill bias of technology
I
z =u
θ,
where u
exp (1)
ρ
ρ 1
Firms
Skill bias of technology
h
=
l
2 (2φ
ϕ
1) ( ρ
wi
si
ρ
αj
1
αj
zϕ
1) skill-bias of technology
I
if ϕ = 0 we say technology is Hicks neutral
I
if ϕ > 0 we say technology is skill biased
Two ways reallocation a¤ects demand for skill
1
2
Across …rms between sectors
Across …rms within sectors
Firms
Skill bias of technology
h
=
l
2 (2φ
ϕ
1) ( ρ
wi
si
ρ
αj
1
αj
zϕ
1) skill-bias of technology
I
if ϕ = 0 we say technology is Hicks neutral
I
if ϕ > 0 we say technology is skill biased
Two ways reallocation a¤ects demand for skill
1
2
Across …rms between sectors
Across …rms within sectors
In an extension, we allow H-O to be active within sectors
General equilibrium
Goods-market clearing
yi (ω, j ) =
∑n τin qn (ω, j ) Iin (ω, j )
Factor-market clearing with inelastic supplies Hi and Li
Z 1
Z 1
Li =
∑j
0
li (ω, j ) d ω and Hi =
∑j
0
hi (ω, j ) d ω
Trade imbalances (where NXi are net exports in i )
Pi Qi = (si Hi + wi Li + Πi ) 1
I
NXi
Outputi
We treat NXi /Outputi as a parameter
Also consider no labor mobility between merchandise & service sectors
Parameterization
Connecting model and data
64 countries + rest of the world ROW (aggregate of 89 countries)
I
64 countries account for approx 93% of world GDP
Data averaged over 2005-2007 (if possible)
Skilled worker: completed tertiary degree (i.e. in US, college degree)
98 merchandise sectors = goods producing industries
155 services industries include construction, exclude government
Parameterization basics
Parameters assigned directly from data
I
Hn / (Hn + Ln ) = % with tertiary degree from Barro Lee
I
αj = % w/ tertiary degree in US, American Community Survey
σ = η median 5-digit SITC, Broda Weinstein
Choose Tn , τ in to match relative country size and bilateral trade
ρ, θ, ϕ, tn to target speci…c moments
Target moment 1
Aggregate elasticity of substitution btw HUS and LUS in US, b
ρ = 1.6
I
Katz and Murphy 92 estimate elasticity = 1.4
I
Acemoglu and Autor 10 estimate elasticity 2 [1.6, 1.8]
In baseline parameterization, we " HUS by 10% and calculate
b
ρ = ∆ log
HUS
LUS
If ϕ = 0 and only one sector ) b
ρ=ρ
log
With ϕ > 0 and many sectors ) ρ = 1.4
wUS
sUS
Target moment 2
Elasticity of trade with respect to variable trade cost, bε = 5
I
Eaton and Kortum 2002 preferred estimate 8.28
I
Donaldson 2010 preferred estimate 4
I
I
Simonovska and Waugh 2011 estimate [2.47, 5.51]
Eaton, Kortum, and Kramarz 2011 preferred estimate 5
I
Costinot, Donaldson, and Komunjer 2012 preferred estimate 6.53
Run a gravity equation on data generated by our model
log (Expin ) = Importern FE + Exporteri FE
If ϕ = 0 ) θ = 1/bε
With ϕ > 0 ) θ = 0.25
bε ln τ in
Target moment 3
log
hi
hi + li
= β0 + β1 log salesi + IndustryFEi + εi
In Mexico, β1 = 0.136; unreported result from Verhoogen (2008)
I
1998 Encuesta Industrial Anual (EIA) w/ large manufacturing plants
In the model: ϕ = 0 ) β1 = 0
I
β1 is increasing in ϕ
ϕ
Elasticity
0
0
0.08
0.05
0.24
0.085
0.4 (φ = 0.75)
0.139
0.64
0.213
0.72
0.23
Note: If ϕ = 0 and αs vary within sector, then elasticity in skill-scarce
countries is negative
Target moment 4
Between sector trade patterns
For each n = 1, ..., 64, regress
Net exportsn (j )
HUS (j )
= β0i + βn
+ εn (j )
Exportsn (j ) + Importsn (j )
HUS (j ) + LUS (j )
Target moment 4
Between sector trade patterns
For each n = 1, ..., 64, regress
Net exportsn (j )
HUS (j )
= β0i + βn
+ εn (j )
Exportsn (j ) + Importsn (j )
HUS (j ) + LUS (j )
Comparative advantage determined by Hn /Ln and relative tn ’s in
Tn (j ) = 1 + (αj α¯ ) tn
Alternative 1: Choose tn to match βn
∑i wiout βi
Alternative 2: Choose tn = 0 (Morrow 2010)
Target moment 4
Between sector trade patterns in the data and matched in the model
ISR
IRL
1
CHE
GBR
DOM
CYP
CRI
JPN
MEX
NOR
βn - world average β
0.5
FRA
JAM
0
HRV
SVN
HND
PRT
IDN
THA
ITA
CZE
DEU
FIN
GRC
LTU
UKR
CAN
AUS
ESP
SRB
GTM
PRY
ECU
LVA
VNM
EST
NZL
CHL
POL
BLR
CHN
-1.5
RUS
PHL
ISL
SLV
LKAAUT
ROM
ROW
BGR
SVK
-0.5
-1
HUN
DNK
KAZ
MYS
USA
NLD
KOR SWE
ARG
URY
IND
BRA
-2
PER
COL
TUR
0.05
0.1
0.15
0.2
0.25
fraction of college workers in country n
0.3
Target moment 4
Between sector trade pattern if we do and if we do not target moment 4
2
1.5
β
1
0
-0.5
-1
n
β - world average
0.5
-1.5
-2
-2.5
data & model calibrated t
model t=0
-3
0.05
0.1
0.15
0.2
0.25
fraction of college workers in country n
0.3
Solution Algorithm
Solution Algorithm
Overview of three loops
Outer loop: iterate over ϕ, θ, ρ
Middle loop: iterate over τ in , Tn , tn
I
I
I
I
Match Exportsin / (Outi + Outn ) , Outn /World Out, and target
moment 4
Update τ in using excess bilateral exports data - model
Update Tn using excess outputn data - model
Update tn using excess βn data - model
Inner loop: iterate over wn , sn , π n
I
Extends Alvarez and Lucas
F
I
I
no analytic gravity, 2 factors, Πn 6= 0, & trade imbalances
no proof of uniqueness
numerical demonstration of existence
Moments targeted and not targeted
Trade ‡ows and output: Data versus model
To H/L and trade ‡ows
Log bilateral exports
Log output
-2
25
R2 = 0.998
model
model
-3
20
-4
-5
R2 = 1
-6
15
-7
-8
10
10
15
20
25
-8
-6
data
-4
-2
data
Imports / output merchandise
Exports / output merchandise
0.9
0.8
0.8
model
model
0.7
0.6
R2 = 1
0.4
0.6
0.5
0.4
R2 = 1
0.3
0.2
0.2
0.2
0.4
0.6
data
0.8
0.2
0.4
0.6
data
0.8
Gravity
Plot log [Xin Xni / (Xii Xnn )] and log (τ in τ ni )
With ϕ = 0, constant elasticity
Skill-biased technology ,
-5
-5
))
0
nn
φ > 0.5
ii
-1 0
slope = -4.9 , trade elast = -4.9
-1 0
slope = -5.1 , trade elast = -5.1
ni
ni
-1 5
in
in
log ((X * X ) / (X * X
ii
log ((X * X ) / (X * X
nn
))
Hicks neutral technology , φ = 0.5
0
-2 0
-2 5
0
0 .5
1
1 .5
2
2 .5
3
log (τ * τ )
in
ni
3 .5
4
4 .5
5
5 .5
-1 5
-2 0
-2 5
0
0 .5
1
1 .5
2
2 .5
3
log ( τ * τ )
in
ni
3 .5
4
4 .5
5
5 .5
Trade costs
We project τ in onto standard “gravity” variables
I
distance, distance squared, common language, common border,
exporter and importer FEs
F
only using those τ in s not set to +∞
) R 2 = 0.74 with expected signs and statistical signi…cance
Trade costs
We project τ in onto standard “gravity” variables
I
distance, distance squared, common language, common border,
exporter and importer FEs
F
only using those τ in s not set to +∞
) R 2 = 0.74 with expected signs and statistical signi…cance
Do poor countries face higher export and/or import costs
conditioning on other observables?
I
Regressing importer FEs on importer GDP per capita ) negative
coe¢ cient highly signi…cant
I
Regressing exporter FEs on exporter GDP per capita ) negative
coe¢ cient signi…cant at 10% level
Similar results if we directly include exporter & importer GDP per
capita in gravity regression
Other moments not targeted: Mexico
Exporter skill-intensity premium, controlling for industry
ln
I
I
hi
hi + l i
= β0 + β1 Exporteri + IndustryFEi + εi
in model β1 = 0.25 in merchandise
in data β1 = 0.21, 1998 EIA unreported from Verhoogen (2008)
Other moments not targeted: Brazil
Elasticity of skill intensity to …rm i size controlling for industry
log
I
I
hi
hi + li
= β0 + β1 log salesi + IndustryFEi + εi
in model β1 = 0.24 in merchandise
in data β1 = 0.36, 1995 Pesquisa Industrial Anual (PIA) sample (large
manuf …rms) unreported from Menezes-Filho et. al. (2008)
Elasticity of skill intensity to domestic sales controlling for industry
log
hi
hi + l i
= β0 + β1 log (domestic sales)i + IndustryFEi + εi
I
in model β1 = 0.34 in merchandise
I
in data β1 = 0.34, 1995 PIA sample unreported from Menezes-Filho et.
al. (2008)
Other moments not targeted: US
% of exporters = 0.51 too high, as in BEJK
I
need …xed cost
However
I
I
share of aggregate revenues by exporters
F
in model = 65% in merchandise
F
in data = 60%, 1992 Census of Manuf, BEJK
VA per worker exporter premium in US
ln (VA per workeri ) = β0 + β1 Exporteri + IndustryFEi + εi
F
in model β1 = 0.135 in merchandise
F
in data β1 = 0.11, 2002 Census of Manuf, Bernard et. al. (2007)
Other moments not targeted: US
Exporter skill-intensity premium, controlling for industry
ln
I
I
I
hi
hi + l i
= β0 + β1 Exporteri + IndustryFEi + εi
in model β1 = 0.14 in merchandise
in data β1 = 0.11, 2002 Census of Manuf, Bernard et. al. (2007)
Imperfect comparison: Bernard et. al. (2007) use non-production
worker share
Other moments not targeted: US
Regress
I
I
I
Exp US (j )+Imp US (j )
Absorption US (j )
on j skill intensity in US merchandise js
in data, coe¢ cient on skill intensity = 0.70
F
signi…cant at 1% level
F
use BEA’s detailed IO tables for 2002 Benchmark
in model, coe¢ cient on skill intensity = 0.88
re-parameterize model imposing φ = 1/2, coe¢ cient =
0.06
Other moments not targeted: US
Regress
I
I
I
Exp US (j )+Imp US (j )
Absorption US (j )
on j skill intensity in US merchandise js
in data, coe¢ cient on skill intensity = 0.70
F
signi…cant at 1% level
F
use BEA’s detailed IO tables for 2002 Benchmark
in model, coe¢ cient on skill intensity = 0.88
re-parameterize model imposing φ = 1/2, coe¢ cient =
0.06
Intuition: interaction between the two mechanisms
φ > 1/2 ) unit costs more sensitive to z in high αj sectors
d d log [unit cost (ω, j )]
> 0 , φ > 1/2
d αj
d log z
) more dispersed distribution of unit costs in high αj sectors
I
even though same distribution of productivities across sectors
) more trade in high αj sectors
Counterfactuals
Counterfactuals
Range of counterfactuals:
I
autarky
I
10% reduction in trade costs
I
Growth in China
I
F
Both with factor mobility and limited factor mobility, labor …xed in
merchandise and services at baseline levels
F
In 10% and China experiments, keep (Net Exports)i /Outputi …xed
Skill-biased technical change
Revisit previous approaches using data generated by model and show
why they would predict small e¤ects of trade
Real wage changes from autarky to baseline
Large di¤erences in real wage changes across factors
0.35
MYS
Skilled
Unskilled
0.3
Log change in real wage
0.25
mean skilled = 0.12
max skilled = 0.34
min skilled = 0.029
CRI
mean unskilled = 0.038
0.2 max unskilled = 0.16
min unskilled = 0.0081
0.15
0.1
0.05
HND
BGR
EST
HUN
LVA
THA
SVN
LTU NLD
SVK
PHL
JAM
PRY CHE IRLCZE
BLR
KAZ
AUT
ECU
LKA
SLV
ISL
ROM UKR
ISR
GTMHRV
SRB
FIN
SWE
DNK
DOM
NOR
MEX
ROW
IDN CHL
POL
PRT KOR
URY
DEU
NLD EST
BGR
CAN
GRC ARG
LVA
NZL
FRA
GBR
LTU
PER
HUN
ESP
RUS
SVN
THA
CZE
TUR
ITA CHN
BLR
PRY AUT
SVK
COL
AUS
SRB JAM ROW
HND
IND
HRV
JPN
CHE
CRI
POLROM
SLVCYP
DEU
USA
UKR
DNK
PRTGTM
CAN
ISL
PHL
MEX
LKA
SWE
ECU
GRC
TUR
BRA
FIN
IRL
KAZ
ESP
FRA
KOR
CHL
URY
ITA
GBR
DOM
NZL
ISR
ARG
CHN
IND
IDN
AUS
NOR
PER
USA COL
RUS
BRA JPN
0
0.05
VNM
MYS
CYP
0.1
0.15
0.2
0.25
0.3
Trade share, 2005-2007
VNM
0.35
0.4
From autarky to baseline
Change in skill premium vs 2005-07 trade share, correlation = 0.70
0.2
CRI
MYS
0.18
VNM
0.16
HND
∆ log s/w
0.14
BGR
PHL
IRL
SVK
SVN
LVA
LTU
0.12
KAZ
CHE
CYP JAM
PRY
ISR
ECU
LKA
BLR CZE
ISL
UKR
SLVNOR
FIN
DOM
SWE
GTM
IDN
AUT
ROM
CHL
DNK
HRV
MEX
KOR
URY
SRB
0.1
0.08
0.06
0.04
0.02
PRT
ARG
RUS
PER
NZL
GRC
GBR FRA
ESPCHN
COL
AUS ITA
JPN
IND
TUR
USA
BRA
0.05
0.1
POL
EST
THA
NLD
mean = 0.08
max = 0.2
min = 0.021
ROW
CAN
0.15
HUN
DEU
0.2
0.25
0.3
Trade share, 2005-2007
0.35
0.4
From autarky to baseline
Change in skill premium vs 2005-07 country size, correlation = -0.62
0.2
CRI
MYS
0.18
0.16
V NM
HND
∆ log s/w
0.14
HUN
P HL
0.12
S VN
LV A
LTU
0.1
JA M
IRL
THA
S VK
K AZ
CYP
P RY
CHE
IS R
IS L
S LV
0.08
URY
0.06
LK A E CU
B LR
DOM
GTM
NLD
CZE
UKR
NOR
S WE
AIDN
UT
DNK
FIN
ROM
CHL
HRV
MEX K OR
S RB
P ER NZL
0.04
P RT
A RG
GRC
COL
0.02
-9
mean = 0.08
max = 0.2
min = 0.021
B GR
E ST
P OL
ROW
DEU
GBR
FRA
CAN
E
SP
CHN
A US
ITA
JP N
IND
TUR
RUS
USA
B RA
-8
-7
-6
-5
-4
-3
Log relative output, 2005-2007
-2
-1
From autarky to baseline: strength of H-O
Correlation change skill premium & H/L = -0.16
0.2
CRI
MY S
0.18
0.16
mean = 0.08
max = 0.2
min = 0.021
V NM
HND
0.14
B GR
∆ log s/w
HUN
0.12
0.1
S V NLV A
0.04
CY P
IS R
LKEACU
B LR
IDN GTM
A UT
CHL
DNK
ME X
ROM
URY
P RT
A RG
IND
P OL
ROW
0.02
0.05
ITA
NLD
UK R
S WE
K OR
S RB
FRA
CHN
IS L
NOR
FIN
S LV
DOM
HRV
0.06
LTU
K A Z CHE
JA M
P RY
CZE
0.08
P HL
E SIRL
T
THA
SVK
COL
GB R DE U
PER
ESP
RUS
NZL
GRC
CA N
A US
JP N
TUR
B RA
US A
0.1
0.15
0.2
Fraction college, 2005
0.25
0.3
From autarky to baseline: strength of H-O
No skill bias, low prctivity dispersion, tn=0: correl change skill premium & H/L = 0.75
UKR
0.04
PHL
EST
CYP
LTU
0.02
ESP
KOR
NLD
ISL
LVA
SRB
GBR
FIN
JAMLKA HUN
DEU
SVN
AUT
HRV
FRA
CHE
DNK
POL
ROM SLV
KAZ
PRYPRT TUR
COL
URYROW
BRAITA
CRINOR
IND
CHL
0
∆ log s/w
GRC
ISR
USA
CAN
BGR
-0.02
DOM
ARG
GTM
-0.04
CZE
SVK
SWE
NZL
RUS
JPN
AUS
PER
IRL
MEX
BLR
ECU
mean = -0.0082
max = 0.044
min = -0.088
THA
CHN
-0.06
IDN
HND
-0.08
VNM
MYS
0.05
0.1
0.15
0.2
Fraction college, 2005
0.25
0.3
10% fall in trade costs from baseline parameterization
Skill premium with full and limited mobility
Mobility
Limited Mobility
JAM
0.1
CYP
∆ log s/w
0.08
0.06
mean mobility = 0.016
max mobility = 0.039
min mobility = -0.0046
ISL
PRY
HND
SLV
LVA
URY
BGR HRV
LKA
LTU
SRBDOM
EST
CRI
0.04
JAM
HND
GTM
SVN
ECU
BLR
0.02
0
GRC
PRT
NZL
KAZ
PHL
UKR
VNM
FINDNK
SWE
PER HUN
SVK
THA
CHL CZE
NOR
POL
AUT
CHE
COL
ARG
IDN
VNM
LKA
CYP
SLV
BGR
DOM
GTM
LTU
ISL CRI SRB
HRV
BLR
SVN
URY
ECU
LVA
EST
PRY
mean limited = 0.042
max limited = 0.1
min limited = -0.0048
ISR
ROM
GBR
AUS
MEX
KOR
ESP
CAN
TUR
RUS
PHL
ROM
THA
KAZ UKR
IND
IRL
ISR
HUN
SVK
MEX KOR
FIN
PRT
CZE
PER NZL
IDNSWE
NOR
CHL
BRA
POL
DNK
COL
GRC
CHE
ARG
AUT
RUS
TUR AUSINDCAN
ESP
IRL
NLD
BRA
NLD
FRA
JPN
ROW
DEU
USA
GBR ROW
FRA
JPN
DEU CHN
ITA
CHN
USA
ITA
MYS
-8
-7
-6
-5
-4
-3
Log Output n / Output World, 2005-07
-2
Three-fold increase in China’s TFP
Skill premium change in China’s trading partners, with full and limited mobility
Increase in TFP China
P HL
CRI
0.03
mean mobility = 0.0044
max mobility = 0.016
min mobility = 0
K OR
KAZ
P RY
∆ log s/w
0.025
0.02
THA
mean limited = 0.013
max limited = 0.034
min limited = 0
FIN
0.015
0.01
0.005
0
V NM
A US
0.02
MY S
CHL
JAMP E R
NZL
RUS IS R
CY P
US A
IS L
CA
DENUA RG
IRL IDN
S W EA UT B GR
HUN
UK RHRVDNK
GB R
NOR SFRA
V KCZE
CHE
DOM
COL S RBURY
GTM
IND
P RT
B RAP OL
SEVSNP
ITA
SROM
LV
PER
FIN
B LR
HND
GRC RUS IDN
TUR
A RG
JAM
ES
RCU
NZL
CA
N
V K HUN
LTU UK
LKA
DERU
IRL IS
WE
IS
LCZE
BNOR
LRSIND
USEAS T
B GR
A UT
DOM
DNK
COL
CHE
HRV
GTM
B
RA
S
VPN
SOL
RB
ROM
URY
FRA
CY
P R
GB
HND
ITAE S
P RT
EP
CU
TUR
LTU
EST
S
LV GRC
LVA
LVA LKA
0.01
JPN
CRI
KAZ
NLD
JPN
ROW V NM
THA
P HL
K OR
MY S
P RY
CHL
A US
ROW
NLD
0.03 0.04 0.05 0.06 0.07 0.08
Trade with China / Merchandise Output
0.09
Increase in China’s skill abundance to equal US
Skill premium change in China’s trading partners, with full mobility
Increase in H/L China
0
-0.005
HND
GRC
ISL
CYP
SRB
LTU
A ESP
TUR
PRT LV
PER
BLR
SLV
USA
NZL
POL
AUTDNK
SVN
HRV
UKRFRA
CAN
GBR
ITA
BGR
EST
BRA
LKECU
A
ROM
COL
SVKRUSFIN
SWE
IND URY
HUN
ISR JA M
CZE
NOR
DEU
GTM CHE
DOM
IRL
ARG
AUS
NLD
JP N
CHL
KAZ
ROW
PRY
CRI
-0.01
IDN
KOR
-0.015
THA
-0.02
VNM
PHL
mean mobility = -0.0051
max mobility = 0.00058
min mobility = -0.028
-0.025
MYS
0.02
0.04
0.06
0.08
Trade with China / Merchandise Output
0.1
Skill-biased technical change in all countries
s/w rises by 25% in median country
h
=
l
wi
si
ρ
Ah α j
zϕ
Al 1 α j
Hicks-neutral technology, ϕ = 0
I
Trade share for median country rises by 0.1%
Skill-biased technology, ϕ > 0
I
Trade share for median country rises by 4.5%
Skill-biased technical change induces aggregate outcomes that look
like reductions in international trade costs
Intuition: with φ > 0, elasticity of unit costs with respect to
productivity " if Ah /Al "
I
same intuition for why more trade in high αj sectors
Other approaches
Other approaches
Factor content of trade (FCT)
Between-sector price changes
Between-sector factor reallocation
Standard measure of factor content of trade
0.2
0.18
0.16
∆ log s/w
0.14
0.12
0.1
0.08
0.06
0.04
0.02
-0.06
-0.04 -0.02
0
0.02 0.04 0.06
Change in standard measure of FCT
0.08
0.1
Correct measure of factor content of trade
0.2
0.18
0.16
∆ log s/w
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0.05
0.1
0.15
Change in correct measure of FCT
0.2
Changes in domestic prices by sector
∆ log sectoral price index / ∆ log s/w
2.5
φ>0.5, bertrand
φ=0.5, bertrand
2
1.5
1
0.5
0
-0.5
-1
-1.5
0.1
0.2
0.3
0.4
H/(H+L) by sector
0.5
0.6
Between sector factor reallocation
Model’s implication for Chile: from autarky to baseline (s/w rises 7.5%)
1.6
Employment reallocation
1.4
1.2
1
Net change
Intrasector all firms
Intrasector continuing firms
0.8
0.6
0.4
0.2
0
-0.2
0.05
0.1
0.15
0.2
0.25
0.3
H/(H+L) by sector
0.35
0.4
0.45
Conclusion
Embed into otherwise standard quantitative trade model 2 central
mechanisms in theoretical and empirical trade literature through
which trade shapes skill premium
Much of gains from trade accrue to skilled labor bc skill premium in
most countries in response to changes in trade costs
Use computational approach to accurately match bilateral exports,
does not require analytic gravity at any level of aggregation
Conclusion
Embed into otherwise standard quantitative trade model 2 central
mechanisms in theoretical and empirical trade literature through
which trade shapes skill premium
Much of gains from trade accrue to skilled labor bc skill premium in
most countries in response to changes in trade costs
Use computational approach to accurately match bilateral exports,
does not require analytic gravity at any level of aggregation
Multinational production is another major form of globalization
I
MP may strengthen H-O mechanism, high productivity …rms can
produce in countries with comparative advantage in their sector
I
MP may strengthen SBT mechanism, promotes international di¤usion
of best technologies
Sensitivity
Perfect competition
Same fρ,ϕ,θ g, redo middle and inner loops
Move countries to autarky, full factor mobility, change in skill premium (%)
mean
max
min
Baseline
+8.00
+19.65
+2.12
Perfect competition
+7.89
+19.82
+1.88
Alternative trade cost parameterization
Same fρ,ϕ,θ g, redo middle and inner loops
Move countries to autarky, full factor mobility, change in skill premium (%)
Baseline
mean
max
min
+8.00
+19.65
+2.12
symm trade
costs in ROW
+8.00
+19.63
+2.12
symm trade
costs in US
+8.00
+19.63
+2.12
symm trade
costs in all n
+8.08
+19.47
+2.12
Sectoral comparative advantage
Same fρ,ϕ,θ g, redo middle and inner loops
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
mean
max
min
Baseline
+8.00
+19.65
+2.12
Setting ti = 0
+9.27
+23.23
+0.81
Measure of skill endowment
Same fρ,ϕ,θ g, redo middle and inner loops
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
mean
max
min
Baseline
+8.00
+19.65
+2.12
Hi
Li
Hi
Li
avg yrs of educ.
+7.90
+19.40
+2.01
avg yrs of educ.
and setting ti = 0
+9.80
+22.63
+1.84
Skill bias of technology
Same fρ,θ g, redo middle and inner loops
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
mean
max
min
Baseline
ϕ = 0.4
+8.00
+19.65
+2.12
ϕ=0
ϕ = 0.08
ϕ = 0.24
ϕ = 0.64
ϕ = 0.72
0.2
+2.67
2.56
+1.14
+4.05
1.01
+4.28
+11.41
+0.6
+13.83
+33.19
+3.04
+15.64
+37.99
+3.28
Heterogeneity of productivity within sectors
Same fρ,ϕg, redo middle and inner loops
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
mean
max
min
Baseline
θ = 0.25
+8.00
+19.65
+2.12
θ = 0.125
θ = 0.17
θ = 0.3
+3.60
+10.34
0
+5.15
+13.56
+0.93
+9.74
+23.20
+2.45
Heterogeneity of alpha within sectors
Aggregation bias in skill intensities: Feenstra 2010
αj (ω ) = min fα¯ j exp (ε) , 1g
ln N (0, σα )
ε
Stronger H-O mechanism (now also operates within sector)
If impose ϕ = 0, exporters exhibit low h/l in high s/w countries
I
Negative elasticity of …rm’s skill intensity to …rm’s sales
Heterogeneity of alpha within sectors
αj (ω ) = min fmax f0, α¯ j exp (ε)g , 1g , ε
N (0, σα )
Redo outer, middle and inner loops
I
Require lower ρ (more within reallocation)
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
St. dev log h/l: (median
sector within) / btw
mean
max
min
Baseline
σα = 0
0.21
σα = 0.05
0.66
σα = 0.1
2
σα = 0.2
4.2
+8.00
+19.65
+2.12
+8.32
+20.26
+2.09
+9.64
+24.07
+1.73
+10.84
+28.62
1.67
Elasticity of substitution across goods
Lower σ # btw sector reallocation induced by SBT e¤ect
Redo outer, middle and inner loops, keeping η = 2.7.
From 2006 parameterization, move countries to autarky, full factor
mobility, change in skill premium (%)
Baseline
σ = η = 2.7
mean
max
min
+8.00
+19.65
+2.12
σ = 2.2
η = 2.7
BW 3 digits
+6.94
+17.9
+1.53
σ=1
η = 2.7
+3.96
+12.10
0.3
σ=1
η = 2.7
base ρ (b
ρ = 1.38)
+6.24
+18.21
+0.3
Skill premium decomposition
De…ne:
I
I
I
I
Lk ,i = employment of factor k in country i
Lk ,in (j ) = employment of k in country i sector j used in goods bound
for country n
wk ,i avg wage paidhto factor k in country i
i
Λ (j ) w k ,ii (j )
FCTi (k ) = ∑j ∑n Lk ,in (j ) Lk ,ii (j ) Λni (j ) w
k ,i
ii
F
F
I
wk ,ii (j ) = wage paid to factor k employed in sector j used to supply
domestic mkt
Λni (j ) share of i ’s expenditure in sector j from country n
Φi (k ) = ∑j [wk ,ii (j ) Lk ,ii (j )] /Λii (j )
Accounting identity Lk ,i = ∑j ∑n Lk ,in (j ) implies
wk ,i Lk ,i = wk ,i FCTi (k ) + Φi (k )
Skill premium decomposition
Can express Φi (k ) and FCTi (k ) as
Φi (k ) =
wk ,i FCTi (k ) =
I
I
I
∑j λii (j ) αk ,ii (j ) Ei (j )
∑j ,n
αk ,in (j ) λin (j ) Λin (j ) En (j )
αk ,ii (j ) λii (j ) Λni (j ) Ei (j )
αk ,in (j ) = share of factor payments paid to k, in j prodn bound for n
λin (j ) = share of i sales in country n in sector j paid to all factors
En (j ) = n’s expenditure in j
If αk ,in (j ) and λin (j ) …xed across destinations
) FCTi (k ) = ∑j Lk ,i (j ) ω i (j )
I
I
ω i (j ) = (Net Expi (j )) (Revi (j ))
) Component 1 easily measured using sector-level data
If λii (j ) and αk ,ii (j ) …xed and Ei (j ) /Ei (j 0 ) …xed )
back to FCT
) Component 2 constant across equilibria
Do H/L’s play large role in shaping bilateral exports?
Set Hn /Ln = Hworld /Lworld for all n, keep Hn + Ln = 1
Other parameters (incl. calibrated tn , Tn , τ in ) unchanged
Log bilateral exports
Log output
-2
model common H/L
model common H/L
0
-5
R 2 = 0.998
-10
-15
-3
-4
-5
R2 = 1
-6
-7
-8
-15
-10
-5
0
-8
model baseline
0.8
0.8
0.7
0.6
R 2 = 0.997
0.4
-4
-2
Exports / output merchandise
0.9
0.3
0.2
model common H/L
model common H/L
Imports / output merchandise
0.9
0.5
-6
model baseline
0.7
0.6
0.5
R 2 = 0.995
0.4
0.3
0.2
0.2
0.4
0.6
model baseline
0.8
0.2
0.4
0.6
model baseline
0.8
Do H/L’s play large role in shaping bilateral exports?
Set Hn /Ln = Hworld /Lworld for all n, keep Hn + Ln = 1
Other parameters (incl. tn = 0, Tn , τ in ) unchanged
Log bilateral exports
Log output
-2
model common H/L
model common H/L
0
-5
R 2 = 0.985
-10
-15
-3
-4
-5
R 2 = 0.998
-6
-7
-8
-15
-10
-5
0
-8
model baseline
-6
-4
-2
model baseline
Imports / output merchandise
Exports / output merchandise
0.8
0.6
R 2 = 0.967
0.4
0.2
model common H/L
model common H/L
0.9
0.8
0.7
0.6
0.5
0.4
R 2 = 0.96
0.3
0.2
0.2
0.4
0.6
model baseline
0.8
0.2
0.4
0.6
model baseline
0.8
Do H/L’s play large role in shaping bilateral exports?
Set Hn /Ln = Hworld /Lworld for all n, keep Hn + Ln = 1
Other parameters (incl. tn = 0, θ = 0.1, ϕ = 0, Tn , τ in ) unchanged
back to trade ‡ows
Log bilateral exports
Log output
-2
model common H/L
model common H/L
0
-5
R 2 = 0 .99 8
-10
-15
-3
-4
-5
R2 = 1
-6
-7
-8
-15
-10
-5
0
-8
model baseline
-6
-4
-2
model baseline
Imports / output merchandise
Exports / output merchandise
0 .6
R 2 = 0 .99 7
0 .4
0 .2
model common H/L
model common H/L
0 .9
0 .8
0 .8
0 .7
0 .6
0 .5
0 .4
R 2 = 0 .99 6
0 .3
0 .2
0 .2
0 .4
0 .6
model baseline
0 .8
0 .2
0 .4
0 .6
model baseline
0 .8
From autarky to baseline: strength of H-O
Skill bias, high prctivity dispersion, tn=0: correl change skill premium & H/L = 0.04
CRI
0.2
mean = 0.093
max = 0.23
min = 0.0081
B GR
EST
P HL
HND
SV
LV
AN
LTU
∆ log s/w
0.15
URY
KUK
APZRY
R
CY P
M
LKISAR EJA
CU
CHL
IS L
0.1
PER
S LV FIN
SWE
THA
SVK
BIRL
LR
CHE
CZE
MY S
V NM
HUN
NLD
A UT
S
RB
DNK
NOR
ROM
K OR
CA N
HRV
P OL
GTM
A RG
DOM
RUS
GRC
P RT
COL
A US
ME X
ESP
DE U ROW
IDN
TUR
JP N GB R FRA
ITA
US AIND
B RA
NZL
0.05
CHN
0
0.05
0.1
0.15
0.2
0.25
Trade share
0.3
0.35
0.4
From autarky to baseline: strength of H-O
Skill bias, low prctivity dispersion, tn=0: correl change skill premium & H/L = 0.60
0.14
P HL
EST
CRI
B GR
0.12
UK R
LTU
0.1
IS R
CY P
0.08
LV A
S V N HUN
∆ log s/w
IR L
KAZ
FIN
CHE
URY
0.04
S RB
SVK
THA
S LV
CZE
P OL
HRV ROM
MY S
COL
P RT
A RG TUR
HND
ITA
B RA
V NM
DOM
GTM
IN D
PER
K OR
CHL
E CU
B LR A UT
DNK
P RY
0.02
0
NLD
IS L
LK A
JA M
0.06
NOR
SWE
NZL
A US
ESP
JP N
US A
DE U
GB R
FRA
ROW
ME X
mean = 0.041
max = 0.14
min = -0.042
-0.02
-0.04
CA N
GRC
RUS
ID N
CHN
0.05
0.1
0.15
0.2
Fraction college, 2005
0.25
0.3
Target moment 4: Alternative parameterizations
Skill bias
2
0
-1
-2
n
β - world average β
1
-3
data and model with calib t
model θ=0.2 , t=0
model θ=0.125 , t=0
-4
-5
0.05
0.1
0.15
0.2
0.25
fraction of college workers in country n
0.3
10% fall in trade costs from baseline parameterization
Real wages: large di¤erence between skilled & unskilled workers
Skilled
Unskilled
0.07
VNM
MY S
HND
0.06
mean skilled = 0.034
max skilled = 0.066
min skilled = 0.011
MY S
BGR
Log change in real wage
THA
0.05
JA M
PRY
CRI
ES T
SV N
LTU BLR
LV A
0.02
SV K
CY P
SLV
IS L
URY
ES
T
HND
PRY
CRI
LV A
KA Z
VNM
GTM
SRBECU HRV
DOM
BGR
SV N
BLR
LTU
0.01
URY SRB
LKGTM
A
CY
P
IS
LSLV
HRV
DOM
UKROM
R
CHE
THA AUT
FIN
IRL IDNS W E
PRT
POL
DNK
NOR
CZE
SV
PEKR
PHL ARG
NZL
IRL AUT
CHE
IS R
CHL
HUN
KA Z
ECU
JA M
mean unskilled = 0.018
max unskilled = 0.063
min uns killed = 0.0061
CZE
LK A
0.04
0.03
HUN
PHL
NLD
NLD
ME X
PE R
NZL
COL
GRC
CA N
TUR
RUS
ES P
AUS CA
N
ME X
IND
KOR
TUR
RUS
ES P
BRA
AUSIND
BRA
UK R COL
CHL GRC
SWE
FINDNK POL
ROM
IDN
ARG NOR
IS R PRT
ROW
KOR
DE U
ROW
FRA
R U
ITA GBDE
CHN
JP N
CHN
FRA
ITA
GB R
US A
JP N
US A
0
-8
-7
-6
-5
-4
-3
Log Output n / World output, 2005-2007
-2
Costs and prices
Let cink (ω, j ) denote τ in
the unit cost of production of the k’th
most productive (ω, j ) …rm in country i
i 1
ϕ
ϕ
τ in h
1 ρ
1 ρ 1 ρ
cink (ω, j ) =
α j z 2 + ρ 1 si
+ (1 αj ) z ρ 1 2 wi
Ai ( j )
where z is the productivity of this …rm
Denote 1st - and 2nd -lowest costs of supplying (ω, j ) to n by
C1n (ω, j ) = min fcin1 (ω, j )g
i
C2n (ω, j ) = min ci
where i satis…es C1n (ω, j ) = ci
Price of (ω, j ) in country n is
n1
n2 , min fcin1
i 6 =i
(ω, j )
pn (ω, j ) = min C2n (ω, j ) ,
back to Trade
(ω, j )g
η
η
1
C1n (ω, j )
The strength of the mechanisms
What determines strength of H-O mechanism?
If ϕ = 0, then only H-O mechanism is active
Assume marginal cost pricing; i = 1, 2; j = x, y ; & σ = ρ = 1
I
Let i = 1 have comparative advantage in skill-intensive sector x
The strength of the mechanisms
What determines strength of H-O mechanism?
If ϕ = 0, then only H-O mechanism is active
Assume marginal cost pricing; i = 1, 2; j = x, y ; & σ = ρ = 1
I
Let i = 1 have comparative advantage in skill-intensive sector x
Proposition: Rise (fall) in s1 /w1 (s2 /w2 ) caused by moving from
autarky to …xed trade share decreasing in θ & increasing in
A1 (x ) A2 (y ) / [A1 (y ) A2 (x )]
Intuition 1: Higher θ ) …rm productivities more dispersed
) in relative …rm costs, z more important vs. Ai (j ) and wages
) comparative advantage mitigated
) less btw sector reallocation ) smaller wage changes
The strength of the mechanisms
What determines strength of H-O mechanism?
If ϕ = 0, then only H-O mechanism is active
Assume marginal cost pricing; i = 1, 2; j = x, y ; & σ = ρ = 1
I
Let i = 1 have comparative advantage in skill-intensive sector x
Proposition: Rise (fall) in s1 /w1 (s2 /w2 ) caused by moving from
autarky to …xed trade share decreasing in θ & increasing in
A1 (x ) A2 (y ) / [A1 (y ) A2 (x )]
Intuition 1: Higher θ ) …rm productivities more dispersed
) in relative …rm costs, z more important vs. Ai (j ) and wages
) comparative advantage mitigated
) less btw sector reallocation ) smaller wage changes
Intuition 2: Higher A1 (x ) A2 (y ) / [A1 (y ) A2 (x )] strengthens 1’s
comparative advantage in x
) more btw sector reallocation ) bigger wage changes
The strength of the mechanisms
Skill-biased technology and trade
If ϕ > 0 then skill-biased technology and trade interact
h
=
l
wi
si
ρ
αj
1
αj
zϕ
What shapes the strength of this mechanism?
I
I
h (z 0 )
l (z 0 )
.
h (z )
l (z )
is increasing in ϕ for all z 0 > z
avg di¤erence btw expanding z 0 & contracting z increasing in θ
Shown quantitatively: strength of mechanism " in θ and ϕ
Back to h/l
Skill Intensities
Five most and least skill-intensive merchandise sectors
Most skill intensive
Pharma. & medicine manuf.
Aerospace product and parts manuf.
Computer and peripheral equip. manuf.
Commun., audio, & video equip. manuf.
Forestry except logging
Intensity
.611
.561
.553
.465
.455
Least skill intensive
Logging
Animal slaughtering, processing
Fiber, yarn, and thread mills
Carpets and rug mills
Turned product, screw, nut, bolt manuf.
Intensity
.040
.073
.075
.085
.086
Back to Parameterization Basics
Inner loop: factor prices and pro…t shares
Inner loop kI : given ϕ, θ, ρ, τ, Tn , tn
Initial guesses fwn , sn , π n g from inner loop (kI
1)
Solve for
I
Pn Qn = wn Ldn + sn Hnd (1 + π n ) 1
I
pn (ω, j ) , Iin (ω, j ) , Pn (j ) , Pn ( price equations
I
I
nxnd
Qn , qn (ω, j ) ( price and demand equations
yn (ω, j ) , ln (ω, j ) , hn (ω, j ) ( production fcn, h/l, qn (ω, j ),
Iin (ω, j )
Back to inner loop