Transcript Structural Transformations

Structural Transformation
Ricardo Hausmann
Kennedy School of Government and
Center for International Development
Harvard University
Development seems to be more than
producing more of the same


Increasing diversity
Changing what you produce




Self-discovery externalities
Coordination failures
Progress when progress is easy: quality
improvements
Growth collapses
Development entails diversification, not specialization
Source: Imbs and Wacziarg (2003)
Rich countries produce rich-country goods…
Background: Hausmann, Hwang & Rodrik

Measuring the revealed ‘sophistication’ of
exports

How sophisticated is a particular product?

Using this measure, how sophisticated is a
country’s export basket?
You become what you export: initial level of
sophistication and subsequent growth
Residuals
Linear prediction
IRL
e( growthgdp | X,lexpy1992 ) + b*lexpy1992
.429625
CHN
KOR
SGP
TTO
CYP
CHL
PER BLZ
LKA
BGD
IDN
JAM
BOL LCA
COL
ECU
OMN
TUR
AUS
HUN
GRCHRV
MYS
PRT
THA
IND
ROM
BRA
FIN
CANUSA
SWE
NZL
DNK
ESP
NLD
DEU
ISL
CHE
MEX
DZA
SAU
PRY
KEN
MDG
HTI
.31443
8.10487
9.83871
lexpy1992
What you produce is determined by a lot
more than “fundamentals” (I)
Residuals
Linear prediction
Residuals
CHN
.813688
.657409
IND
NPL
IDN
PAK
BGD
TGO
SEN
CMR
GTM
DZA
CAF
NAM
TUR
BRA
RWA
SDN
PNG
SGP EGY
ROM
AUT
FRA
ESP
CRI
ITA
LKA
CIV
SLVPRT
JOR
IRN
B
OL
COL
SYR
KEN
HND
VEN
ECU
MDG
MUS
OMN
GUY
PRY
NLDBEL
DNK
CAN
AUS
ARG
CHLFJI
NOR
PER
PHL
NGA
NZL
USA
HKG
URY CYP
GRC BRB
MAR
NIC
UGA TZA
MWI
GAB
MEX
ISL LUX
CHE
RUS IRL
POL
KOR
FIN
DEU
JPN
ISR
GBR SWE
ZAF
THA
HUN
MYS
CZE
SVK
ISLLUX
CHE
POL
IRL
BGR
KOR
EST
NAM
EGY
FIN
SVN
TUR
DEU
ROM
SEN
BRA
SGP
ISR
CIV
HRV
SWE AUT
LVA
JPN
CRI
GBR
GEOLKA
KGZ
ESPFRA
CMR
SLV
LTU WSM
DNK
ITA
BEL
NLD
CAN
PRT
NZL
KEN
JOR
IRN
SYRBOL
USA
COL KAZ
MKD
HKG
MAR URY
CYP
GTM
DZA NIC UGA
GRC
BRB
TZA
MWI
MDG
ECU
VEN
AUS
LBN
HND
ARG MNG
ALB FJI
CHL
GAB
MUS
GUY
OMN
PAN
NOR
PER
RWA
PNG
PRY
TTO
SDN
HUN
PHL
PAN
TTO
IDN
RUS
BLR
PAK
BGD
AZE
TGO
e( lexpy2003 | X,rule ) + b*rule
e( lexpy2003 | X,loghl ) + b*loghl
THA
MYS
IND
ARM
BIH
ZAF
NGA
Linear prediction
CHN
MDA
MEX
BLZ
NER
ETH
-.508204
NER
-.875729
.07236
1.21472
ln human capital
BHR
-1.20609
1.90945
rule of law
Partial associations between EXPY and human capital (left panel) and institutional
quality (right panel)
Problems with structural
transformation

Information Externalities:


Self-discovery spillovers
Coordination Externalities

Public inputs and training externalities
Coordination externalities and
the evolution of comparative
advantage
Hausmann and Klinger (2007)



Every product requires a number of factors of
production that are relatively specific
E.g. producing asparagus requires a certain type
of soil, mechanized farming equipment,
agribusinesses firms that know the market,
but also such “public goods” such as port
infrastructure, road system, cold-storage facilities,
phytosanitary regulations, market access
agreements, etc.
Implication


The distance from the products in which a
country has accumulated its specific human
capital to alternative products may affect the
speed of its structural transformation
But what do we mean by “distance” and how
would we measure it empirically?
Monkeys & the Product Space

Our metaphor:



Products are like trees
Firms are like monkeys
Growth can happen by:


Having more monkeys in the same trees: more of the same
Improved quality in the same trees: move up the tree



Hwang 2006 finds rapid and unconditional convergence within
trees
Or structural transformation: jumping to more valuable
trees
HHR (2006) show that this last step drives growth in
a significant fashion
Empirical implementation


Monkeys tend to jump short distances
Control for any time-varying national
characteristic


Human capital, rule of law, financial conditions
Control for any time-varying product
characteristic

Price, PRODY
Implementing the model

The ‘proximity’ (φ ) of two products captures how
easily the capabilities to produce one can be used to
produce the other: measure of the cost of jumping.

φAB = min {P(RCA A | RCA B),P(RCA B | RCA A)}

Proximity of Cotton Undergarments to:




Synthetic undergarments: 0.78
Overcoats: 0.51
Centrifuges 0.02
Proximity of CPUs to:



Digital central storage units: 0.56
Epoxide resins: 0.50
Unmilled rye: 0
Visual Representation of the
Product Space
New Work

“The Product Space and its Consequences for
Economic Growth” with Hidalgo, Klinger & LazloBarabasi

How can we map this product space visually?

Could the topography of the export space help
explain bimodal income distribution and the lack
of convergence?
Step 1: Maximum Spanning Tree
Step 2: Overlay Strong Links
0.4 >
0.4 – 0.55
0.55 – 0.65
0.65 <
Step 3: Add Products
Nodes sized according to World Exports, darker links are stronger (red is strongest)
Nodes sized according to World Exports, darker links are stronger (red is strongest)
Step 3: Add Products
Regions Produce in Different Areas of the
Space
Malaysia: 1975-2000
Malaysia 1975
Malaysia 1980
Malaysia 1985
Malaysia 1990
Malaysia 1995
Malaysia 2000
Monkeys jump to nearby trees
1
2
3
4
DEU
ITA
ESP
AUTUSA
POL CZE
GBR
NLD
SVK
DNK
CHN
TUR
HUN PRT SWE
JPN
IND
GRC
CAN
BRA
THA HRV
IDN UKR ROM
KOR
ZAF ARG
FIN
HKG
NZL
MEX
URY
ISR AUS
LTU
EGY
COL
PER
RUS
MDA
LVA
SGPIRL
MAR
CHL
LBN
MYS
PAK
NOR
LKA
ALB
PHL
ZWE
DOM
KGZGEO
PRY
GTM
ECU
KAZ
BLR
BOL
KEN
ARM
MOZ
MDG
VEN
TZA
HND
AZE
BGD
NPL
SLV
GHA
JAM
NIC
SYR
CIV
TJK
SEN
HTI
SLE
MLI
BFA
ZMB
TKM IRN
BEN
ETH
MWI
UGA SDN PNG
SAU
TGO
GIN
NGA
DZA
CMR
NER RWA
CAF
BDI
0
lnavgpaths
Average Paths vs. GDP per capita (logs), 2000
6
7
8
9
lngdppcppp
10
11
Measuring density around a tree

We use these pairwise distances to measure
how close a country’s entire export basket is
to an unoccupied tree: Density
0
0
.5
.5
.6
.4
.5
.3
For all the surrounding trees you occupy, add their “proximity” (conditional
probability) to the new tree, divided by the total number of ‘roads leading to Rome
’
This is a measure of the ‘density’ around a particular good
5
0
Density
10
Density for jumps (green) versus nonjumps (brown)
0
.2
.4
density1b
Density
.6
Density
.8
Does the product space matter?

More formally, we estimate:
xi , c ,t 1    xi , c ,t  densityi , c ,t  X  
where X is a vector of country+year and
product+year dummies, controlling for all
time-varying country and product-level
characteristics.

Standard errors clustered at the country level,
density normalized into units of standard
deviation
x
density
(1)
(2)
x
0.657
(66.27)**
0.062
(7.03)**
x
0.655
(67.44)**
0.056
(6.36)**
0.004
(7.46)**
0.008
(6.19)**
389092
0.56
RCA_lall
RCA_leamer
Observations
R-squared



398362
0.56
1 standard deviation increase in density associated
with 6.2 percentage point increase in the probability
of having RCA in that good in the next period
The unconditional probability is 1.27%: almost 5-fold
increase
This effect dominates the influence of having RCA in
the Leamer or Lall category
The model at the country
level
How green is your valley?
Proposition

It is easier for a country to move to a higher
EXPY if the unoccupied trees are near and
fruity

We need an equivalent measure of “density”
at the country level

We call it “open forest”
Open_forest


Open forest measures the value of the option to
move to a higher EXPY
It calculates the value of the unoccupied trees, by
weighing their proximity and their PRODY
2,000 x 0
1,000 x .6
1,600 x .3
Take the scaled distance from the tree you occupy to trees you don’t
Multiplied by the ‘fruitiness’ of the potential tree
And add that together for the whole export basket
15
open_forest vs. GDP p.c. (logs), 2000
12
13
14
POL CZE ESPITA
AUT
HUN
DEU
TUR
NLD
SVK GRC SWE
GBR
DNK
USA
UKR
CHN ROM
IND
PRT
HRV
BRA
JPN
FIN
THA
KOR
CAN
IDN
ISR HKG
RUS ZAF ARG
MEX
LTU
NZL AUS
URY
IRLNOR
MDA
EGY
COL MYS
SGP
PER
MAR
JOR
LBN PANLVACHL
PAK
LKA
PHLBLR
ALB
GEO ZWE
KGZ
KAZ
PRY DOM
ECU
GTM
BOL
CRI
ARM
VEN
MOZ
KEN
AZE
BGD
SYR
MDG
NPL
TZA
GHAHND
SLV
CIV
SLE
MNG
JAM
TJK
IRN
MLI
NIC
HTI
BFA SEN
ZMB
TKM
ETH BEN
SDN
SAU
MWI
UGA
PNG
TGO
NGA
OMN
GIN
DZA
NER
CMR
RWA
CAF
11
BDI
6
7
8
9
lngdppcppp
10
11
Open Forest & EXPY Growth
Table 5: Open_Forest and EXPY Growth, 1985-2000
lnEXPYc,t
lnGDPpcc,t
lnopen_forestc,t
(1)
FE
EXPY
growth
-0.185
(9.36)**
0.025
(1.48)
0.027
(3.67)**
(2)
RE
EXPY
growth
-0.059
(5.69)**
0.010
(2.75)**
0.016
(4.14)**
(3)
FE
EXPY
growth
-0.229
(10.86)**
0.009
(0.53)
(4)
RE
EXPY
growth
-0.068
(6.35)**
0.012
(3.22)**
lnopen_forest_sizec,t
0.006
0.010
(0.79)
(2.38)*
lnopen_forest_valuec,t
0.329
0.145
(5.95)** (3.51)**
Constant
1.085
0.242
-1.111
-0.865
(5.81)** (4.99)**
(2.53)*
(2.43)*
Observations
1434
1434
1434
1434
Number of countryid
106
106
106
106
R-squared
0.06
0.09
Growth rate is between t and t+1 (annual observations)
Absolute value of t statistics in parentheses
* significant at 5%; ** significant at 1%
1-standard deviation in open forest is associated with higher EXPY growth of
1.6 percentage points per year.
Quality improvements and
convergence
What happens when countries can
upgrade within the same products?
Based on Hwang (2007)
There is no unconditional convergence of
GDP per capita
But there is unconditional convergence
given the within-product quality distance
to the frontier (Hwang 2006)
The evolution of within-product quality
(Hwang 2006)





Quality in any particular product converges to the
frontier at a rate of 5-6% per year
This happens unconditionally
Countries that are further away from the quality
frontier grow faster
When a country develops a new product, it tends to
enter at a lower quality
Therefore, the development of new products creates
more room for within-product quality upgrading, and
subsequently faster growth
Africa and LAC have the lowest gaps in
the products they are in
290%
270%
250%
230%
210%
190%
EAP
ECA
MNA
SAS
LAC
AFR
CZE
MYS
POL
MEX
CHN
CHL
ROM
ARG
HUN
BRA
URY
150%
TUR
170%
Recent work by Kugler, Stein
and Wagner
Does quality matter for jumping to new
trees?
RIP
Not really
a good project !
But height will help you !
Safe landing !
Growth collapses
Based on Hausmann, Rodriguez and
Wagner (2006)
Question: How many industrial
countries had their highest GDP
per capita before 2000
None
10
5
3
1
0
Frequency
15
20
20
2000
2001
2002
MAXPCTIME
2003
2004
Out of 112 developing countries
with data since 1980, how many
had their maximum GDP per
capita before 2000?
50
67 (58 percent) had their peak before 2000
30
20
17
16
10
10
6
4
2
4
4
2
2
0
Frequency
40
49
1960
1970
1980
MAXPCTIME
1990
2000
How deep have recessions
been?
Developing countries: peak to trough fall
in GDP per capita in long recessions
25
26
15
15
14
10
11
9
8
7
6
5
6
3
2
1
2
2
1
2
1
0
Frequency
20
52 countries in excess of 20 percent
21 countries in excess of 40 percent
0
.2
.4
.6
LGAPPCGDP
.8
1
Implication



Many countries have seen negative per
capita growth for a very long time
This has happened in spite of improvements
in schooling attainment, life expectancy and
global technological possibilities
In fact, most developing countries have seen
declines in GDP per capita lasting more than
10 years
Question #1: Why do countries fall into
crises?


Probit analysis: We study the determinants of
the probability of countries falling into crises.
Usual suspects:





Wars
Natural disasters
Export collapses
Sudden Stops
Unusual suspect:

Open Forests
Most growth collapses coincide with
export collapses
1990
1970
1980
BDI
NCL
KEN
BHS
ROM
JOR
CMR
DZA
COGCOM
VUT
GMB
RWA
SLE
PRYPER
ANT
ZAF
GTM TGO
AGO
HTI
NAM
HND
IRQ
CIV SLV
AFG
SUR
VEN NIC
BOL
CAF SAU
NGA
IRN
GAB
KIR
ZAR
ZWE
ARE
LBR
JAM
GHAMDG
NER ZMB
1960
Date of growth collapse
2000
USA
ATG
CYP
CRI
LKA
MEX
MRT
PHL
VCT
PANFJI
WSM
BLZ
AUT
AUS
BGR
BGD
BFA
BEN
BEL
BRA
BWA
BTN
CAN
CHN
CHL
DEU
DNK
FIN
ESP
EGY
ECU
F
GBR
GRC
HKG
GUY
HUN
IRL
IND
ISL
ITA
JPN
KOR
LBY
MAR
LSO
MOZ
MUS
MYS
NOR
NZL
PAK
SDN
SGP
SWZ
SWE
TCD
TUR
TUN
TTO
THA
RA
BHR
MLI
DOM
NPLMMRPRT
CHE
NLD
OMN
BRBDMAMLT
PYF
GRD
ISR
SYC
LCA MWI
ARG
KNA
URY
SYR
BMU
COL IDN
GNB
SLB
PNG
KWT
SEN
1960
1970
1980
LOCMAXTIMEX
MAXPCTIME
1990
2000
LOCMAXTIMEX
Date of export collapse
1
Collapses in exports were typically larger
than those in output
LBR
.8
IRQ
KWT
ZAR
.6
ARE
NIC
NER
GAB
CIVIRN
MDG
HTI
.2
.4
CMR
ISR
DOM
.2
DMA
MLT
NPL
.4
PRY
HND
KEN
NCL
LCA
GRDMWI
SYR
BRB
KNA
OMN
.6
GAPXPC
GAPPCGDP
SLE
SAUZMB
RWA
TGO
SURVEN GHA
COG JAM
CAF
SLBNGAGNB
BDI
SLV
PER
ROM
BOL
ZWE
VUT
SEN
DZA NAM
PNG
GTM
ZAF
COM
URY
IDN
0
Fall of output
KIR
.8
GMB
BHS
ANT
AFG
1
GAPXPC
Fall of exports
of real GDP5.6per capita 5.7
Log5.5
1965
1972 1967 1974
1968
1976
1970
1973
1966 1969
1975
1971
1964
1960
1977 1961
1978
1963
1962
1981
1980 1979
1982
1986
1993
2004
1992
1983
1984
1985 1988
1987
1989
1990
1991
2003
2002
2001
19971994
1996
2000
1999
1998
1995
5.4
LYPCLCUK
The growth collapse in Zambia
4
5
Log of
6
realLXPCKUS
exports
per capita
7
3.5
Growth collapse in Bolivia
1978 1977
1976
1979
1980
1981
1974
2004
1998 2003
2002
2000
1999 2001
1997
1982
3.4
1996
1995
1983
1994
1993
1984
1991
1992
1985 1990
3.35
LYPCLCUK
3.45
1975
1989
1988
1987
4.5
1986
5
5.5
LXPCKUS
6
Baseline results: Random Effects Probit
Table 4: Random Effects Probit Regressions, All Countries
Dependent Variable: Probability of Falling into a Crisis
(1)
Log GDP per Working Age Person
-0.017
(1.78)*
(2)
(3)
(4)
(5)
-0.007
(0.66)
-0.266
(4.03)***
-0.001
(0.06)
-0.430
(4.85)***
0.732
(3.66)***
-0.037
(0.3)
0.167
(2.19)**
-0.004
(0.17)
-0.410
(2.94)***
0.467
(1.81)*
1062
83
6.1%
7.9%
-1.173
(11.46)***
-1.301
(10.82)***
-1.366
(9.37)***
0.000
(0.01)
-0.422
(3)***
0.415
(1.56)
0.063
(0.37)
0.240
(2.36)**
1.017
(3.31)***
0.312
(2.45)**
-0.144
(1.69)*
-0.002
(0.18)
0.370
(0.29)
3344
187
0.0%
2.3%
2785
169
0.0%
3.8%
1872
145
1.6%
5.5%
1054
83
5.1%
7.4%
Log Change in Real Merchandise Exports
War
Natural Disaster
Sudden Stop
Log of Inflation
Change in Polity Indicator
Open Forest
Democracy
Constant
Observations
Countries
Percent crises predicted
Pseudo-R^2
0.229
(2.27)**
1.020
(3.35)***
0.362
(2.92)***
-0.158
(1.89)*
1.068
(0.9)
Question 2: What determines how long a
country stays in a crisis?

None of the usual suspects





Wars
Natural disasters
Export collapses
Sudden Stops
…but the impact of open forest is very robust
Duration analysis

Two types of specifications.

Parametric with frailty (Weibull + others).
hi (t | X )  h0t vi exp( X )


Cox with corrected variance (models for multiple spells).
hi (t | X )  h(t ) exp( X )
Parametric may be more adequate to precisely
estimate the hazard function.
Table 12: Duration Regressions, Weibull Specification with Frailty
Dependent Variable: Years in crisis.
(1)
(2)
(3)
Representation, Hazard rate with Region and Decade dummies (not shown)
Log GDP per Working Age Person
0.024
0.030
0.056
(1.2)
(1.4)
(1.21)
Openforest
0.533
(3.61)***
Democracy (Polity Index)
Sudden Stop
(4)
(5)
(6)
(7)
0.057
(1.15)
0.558
(3.13)***
0.031
(1.48)
-0.092
(0.45)
0.046
(0.75)
0.712
(3.36)***
0.028
(1.07)
-0.223
(1)
0.287
(0.83)
-0.584
(1.11)
0.055
(0.14)
-0.262
(0.4)
-0.186
(0.65)
0.049
(1.07)
0.438
(2.49)**
0.055
(1.1)
0.494
(2.36)**
0.032
(1.35)
-1.648
(0.64)
-1.345
(0.38)
0.136
(0.66)
0.104
(0.38)
-0.003
(0.06)
-0.003
(0.12)
-8.138
(2.82)***
191
Log Change in Real Merchandise Exports
War
Natural Disaster
Log of Inflation
Change in Polity Indicator
Change in Exports*Open Forest
Polity*Change in Merchandise Exports
Polity*Sudden Stops
Constant
N
-1.679
(11.74)***
535
-0.796
(2.64)***
535
-8.117
(3.89)***
233
-8.982
(3.63)***
191
-10.640
(3.3)***
175
-6.859
(2.81)***
230
Conclusion: a common cause of
protracted growth collapses


Adverse shock to the earning capacity of
exports
…in a country with low open forest
(connectedness)