Economic Consequences of Unrecognized States

Evidence from Nagorno-Karabakh in Azerbaijan

Masayuki Kudamatsu

15 June 2018

Unrecognized states in former Soviet Union (1/5)

When Soviet Union broke up into 15 republics in 1991...

4 wars of secession broke out in 3 newly independent republics

Unrecognized states in former Soviet Union (2/5)

Image source: http://www.economist.com/node/3110979

Secessionists won the war, declaring independence

Unrecognized states in former Soviet Union (3/5)

What are the "frozen conflicts"? (4/5)

But...

not a single country around the world

recognized them as sovereign states

Image source: http://www.economist.com/node/3110979

They remain unrecognized until today

(Russia recognized Abkhazia & South Ossetia in 2008)

Unrecognized states in former Soviet Union (5/5)

What is the impact

of being unrecognized as a state

on economic activities?

Research question

Lack of data

Empirical challenge

DHS Survey areas

for Moldova in 2005

Solution

Use of satellite images

Estimation strategy

Spatial Regression Discontinuity

Why should we care?

Reason #1 for why we should care

Unrecognized states are not uncommon

Not recognized by any

   Transnistria (1991-present)

   Nagorno-Karabakh (1994-present)

   Somaliland (1991-present)

Recognized by a few

   Taiwan (1971-present)

   Northern Cyprus (1974-present)

   South Ossetia (1992-present)

   Abkhazia (1993-present)

Recognized by many but not all

   Kosovo (1999-present)

Since 2014, two more unrecognized states in Ukraine

Interest appears to be slowly emerging even in Japan

What aspect of states

promote development?

Recent literature's focus:

State Capacity

Besley and Persson (2009, 2011)

Acemoglu, Garcia-Jimeno, and Robinson (2015) 

etc.

This study:

International recognition

Reason #2 for why we should care

Unrecognized states = A result of civil wars

But so far ignored

in the (emerging) academic literature

on the impact of civil wars

It's an ignorable part

of the impact of civil wars

Reason #3 for why we should care

We have two contributions to this literature...

Contribution #1: Deal with sample selection bias

Literature finds that output loss is temporary

Davis and Weinstein (2002) on bombs in Hiroshima & Nagasaki

Miguel and Roland (2011) on Vietnam wars

Cerra and Saxena (2008) on cross-country panel regressions

These findings may be due to sample selection bias (Blattman and Miguel 2010)

Unrecognized states

is an example of out-of-the-sample cases

Unrecognized states

is a political "institution" relevant for post-conflict recovery

Contribution #2:

Role of Political Institutions

in post-conflict recovery

cf. Casey et al. (2012) on post-conflict Sierra Leone

Road Map

Potential Consequences of non-recognition

 

Background on Nagorno-Karabakh

 

Data

 

Empirical Strategy

 

Validity Checks

 

Main Results

Potential consequences

of non-recognition

Cost of Non-recognition

Foreign direct investment 

& International NGOs' support

Military expenditure 

Kolsto (2006)

65 soldiers per 1,000 inhabitants

in Nagorno-Karabakh (International Crisis Group 2005)

e.g.

For Kosovo to be recognized as a state,

the UN introduced the "Standards before status" policy.

2003

Incentive for other unrecognized states

to build a functional state

Benefit of Non-recognition

e.g.

In 2016, Freedom House ranks

Nagorno-Karabakh

Azerbaijan

as Not Free

as Partly Free

Background on Nagorno-Karabakh

Nagorno-Karabakh:

Armenian enclave within Azerbaijan

Azerbaijanis

Armenians

Source:

Atlas Narodov Mira (1964)

Cultural differences

Armenians

Christian (Armenian Orthodox)

Armenian language

Azerbaijanis

Muslim (85% Shia; 15% Sunni)

A Turkish language

1923: Nagorno-Karabakh Autonomous Oblast

Created by Soviet Union

with the border drawn from scratch

Source: Wikimedia Commons

1923: Nagorno-Karabakh Autonomous Oblast

Made part of Azerbaijan, not of Armenia, because (de Waal 2013: 144-145)

Source: Wikimedia Commons

1. Moscow wanted to appease Azeris, to control their oil fields

2. Economic consideration overrode ethnic integration

Azerbaijan's Soviet-era policies

on Nagorno-Karabakh

Encouraging the migration of Azerbaijanis

1926 1979
Armenians 117,000 123,000
Azerbaijanis 13,000 37,000

Population by ethnicity in Nagorno-Karabakh (de Waal 2013: 153-154)

Neglecting the Armenian culture

No Armenian-language television

Armenian history not taught in Armenian-language schools

Medieval Armenian churches closed and crumbling

1991: Collapse of Soviet Union

Azerbaijan declared independence from Soviet Union

Sept 2

Aug 30

Karabakh declared its secession from Azerbaijan

Dec 10

Karabakh voted yes at a referendum on independence

1992-1994: War of Secession

1992-1994: War of Secession

1992-1994: War of Secession

1992-1994: War of Secession

1992-1994: War of Secession

This map is

not completely

accurate, though

Boundary of Nagorno-Karabakh

= 1994 Ceasefire Line

Source: Wikimedia Commons

(different from the Soviet-era autonomous oblast boundary)

What determines the 1994 ceasefire line?

"Defensible border" for Karabakh against Azerbaijani forces

The 1994 ceasefire line is "remarkably similar" to

what the Karabakh leader (Artur Mkrtchian) sketched out

as "defensible frontiers" in the winter of 1991

(de Waal 2013: 240)

To minimise the length of the front line

Ever since 1994, the ceasefire line is heavily fortified

"the most militarized zone in Europe" (Thomas de Waal)

Consequences of Secession War

Death toll: around 35,000

480,000 Azerbaijanis displaced

from the occupied part of Azerbaijan

300,000 Armenians displaced

from the rest of Azerbaijan

Ethnically homogenized

on both sides of the border

Political institutions in Nagorno-Karabakh

Presidential and parliamentary elections: regularly held since 1994

Generally deemed free and fair

But the ruling party coalition always won

Media is largely controlled by the government

Freedom House ranks Nagorno-Karabakh as "Partly Free" for 2016

Singapore (127)  Malaysia (134)  Turkey (142)  Armenia (146)

Myanmar (152)  Thailand (157)  Iraq (163)  Egypt (165)

Afghanistan (171)  Azerbaijan (188)

151st out of 211 "countries"

Threats to identification

What can be different across the border? (1/3)

Initial level of state capacity

Nagorno Karabakh: zero(?)

Azerbaijan: legacy of Union Republic

New oil fields off the coast of Baku

Extraction began in 1996

2006-2008: oil export boom in Azerbaijan

But does it reach the border zone?

(can be checked by DHS surveys in 2006)

Human capital of Armenians vs Azerbaijanis

Literacy rate is 99.9%

in Soviet Union in 1989 (Mironov 1991)

cf.

Damages during the 1992-94 civil war

Need to focus on the segments of the ceasefire line

drawn in the middle of battle fields

What can be different across the border? (2/3)

Market access

Azerbaijan side has Baku

Possibility of being invaded

Karabakh yes

but Azerbaijan no?

What can be different across the border? (3/3)

Karabakh side has Armenia proper

Data

Map available

on the website of Armenia's

Ministry of Foreign Affairs

 

Digitized by myself with ArcMap

Data source #1

The 1994 Ceasefire Line

Difference from Natural Earth data

Old

(Natural Earth)

New

Data source #2

Nighttime light

VIIRS (2015-2016)

DMSP-OLS (1992-2013) used by previous studies in economics

 

Top-coding

 

Bottom-

censoring

 

Blurring

 

Known issues on DMSP-OLS (not for VIIRS)

30 arc-second (around 1km)

15 arc-second (around 0.5km)

Watt per steradian per square cm (W/cm2*sr)

Digital number 0 to 63

Not

comparable

across years

Data source | VIIRS Day/Night Band Nighttime Lights 2015 Annual Composite

2015

1992

Data source | DMSP-OLS Nighttime Lights Time Series

Data source | DMSP-OLS Nighttime Lights Time Series

1993

Data source | DMSP-OLS Nighttime Lights Time Series

1994

Data source | DMSP-OLS Nighttime Lights Time Series

1995

Data source | DMSP-OLS Nighttime Lights Time Series

1996

Data source | DMSP-OLS Nighttime Lights Time Series

1997

Data source | DMSP-OLS Nighttime Lights Time Series

1998

Data source | DMSP-OLS Nighttime Lights Time Series

1999

Data source | DMSP-OLS Nighttime Lights Time Series

2000

Data source | DMSP-OLS Nighttime Lights Time Series

2001

Data source | DMSP-OLS Nighttime Lights Time Series

2002

Data source | DMSP-OLS Nighttime Lights Time Series

2003

Data source | DMSP-OLS Nighttime Lights Time Series

2004

Data source | DMSP-OLS Nighttime Lights Time Series

2005

Data source | DMSP-OLS Nighttime Lights Time Series

2006

Data source | DMSP-OLS Nighttime Lights Time Series

2007

Data source | DMSP-OLS Nighttime Lights Time Series

2008

Data source | DMSP-OLS Nighttime Lights Time Series

2009

Data source | DMSP-OLS Nighttime Lights Time Series

2010

Data source | DMSP-OLS Nighttime Lights Time Series

2011

Data source | DMSP-OLS Nighttime Lights Time Series

2012

Data source | DMSP-OLS Nighttime Lights Time Series

2013

Data sources #3:

Covariates (measured at 30 arc-second cell level)

Elevation

Ruggedness

Temperature

Precipitation

Predicted vegetation (biome)

(Temperate broadleaf / Desert)

Distance to coast

Distance to lake (i.e. Caspian Sea)

Distance to port

Distance to river

Henderson et al. (2018) show

these variables predict the spatial distribution of nighttime light 

SRTM30

 

WorldClim 2

 

Olson et al.

(2001)

GSSHG

GLWD

World Port Index

Natural Earth

Summary statistics

% of lit cells in 2015

Nagorno-
Karabakh     
 		 Azerbaijan
proper
All cells 3.9 21.3
Observations 70,941 420,708
Cells
within 10km
of border		 2.7 24.4 
Observations 13,110   14,432 

Azerbaijan

Nagorno-Karabakh

Azerbaijan

Nagorno-Karabakh

Dashed lines: Cells within 10km of border

Empirical Strategy

Estimation Equation

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Indicator of being lit in cell \(i\)

Indicator of being an unrecognized state

Estimation Equation

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Distance to the border

(Negative for \(T_i = 0\))

Estimation Equation

T_i = 1
Ti=1T_i = 1
T_i = 0
Ti=0T_i = 0
0
00

Nagorno Karabakh

Azerbaijan

d_i
did_i
y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Standard practice in the literature

(Lee and Lemieux 2010)

Estimation Equation

Allow distance-outcome relation

to differ across the border

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Border segment fixed effect

To make sure comparing k with j, not with i

Estimation Equation

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Border segment fixed effect

Estimation Equation

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Denote cell \(i\)'s nearest border segment by \(s\)

Estimation Equation

This equation is estimated for each year of 1992-2013 and 2015

Identification in RDD does not require panel estimation

Interested in the dynamics of the impact

y_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}
yis=βTi+γdi+δTidi+ηs+εisy_{is} = \beta T_i + \gamma d_i + \delta T_i d_i + \eta_s + \varepsilon_{is}

Estimation procedure

Start with all locations within Azerbaijan and Nagorno-Karabakh

Find the optimal bandwidth to restrict the sample

Observations far away from the border

are irrelevant to estimate the outcome at the border

For how to choose the optimal bandwidth

we follow Calonico et al. (2014) [                    in Stata]

rdrobust

For inference

we also follow Calonico et al. (2014)

Calonico et al. (2014)

When the treatment effect is estimated by local linear regressions,

its confidence interval is approximately correct when 

nh^5 \to 0
nh5→0nh^5 \to 0

h: Bandwidth

n: # of observations

To minimize the mean squared error of the estimator

the bandwidth takes the form of

h^* = cn^{-1/5}
h∗=cn−1/5h^* = cn^{-1/5}

c: constant

Thus, the C.I. with h* is not correct because

n(h^*)^5 \nrightarrow 0
n(h∗)5↛0n(h^*)^5 \nrightarrow 0

1

2

3

Calonico et al. (2014) (cont.)

Estimate the bias in the RD estimator

Subtracting the estimated bias from the RD estimator

Correct the variance of the RD estimator

by taking into account the finite sample variability

in the estimated bias

Validity checks

Data sources #3:

Covariates (measured at 30 arc-second cell level)

Elevation

Ruggedness

Temperature

Precipitation

Predicted vegetation (biome)

(Temperate broadleaf / Desert)

Distance to coast

Distance to lake (i.e. Caspian Sea)

Distance to port

Distance to river

Henderson et al. (2018) show

these variables predict the spatial distribution of nighttime light 

SRTM30

 

WorldClim 2

 

Olson et al.

(2001)

GSSHG

GLWD

World Port Index

Natural Earth

Terrain

Elevation     
 		 Ruggedness
In Karabakh 36.02 18.26** 
(28.27) (6.92) 
Optimal bandwidth 7804.8 11751.8 
Observations 5444 8033

Climate

Temperature       Precipitation
In Karabakh -0.16 0.24
(0.15) (0.27)
Optimal bandwidth 7182.5 6671.5
Observations 5020 4664

Predicted vegetation (Biome)

Broadleaf
      		 Desert
In Karabakh -0.01 0.01
(0.01) (0.01)
Optimal bandwidth 8571.0 8549.9
Observations 5944 5925

Trade cost

Distance to
 		 Coast Lake Port River
In Karabakh -139.25 -57.12 -180.60 106.30
(523.00) (553.75) (521.33) (523.19)
Optimal bandwidth 11684.8 7811.4 12500.8 9785.9
Observations 7991 5449 8519 6733

Ruggedness at cell \(ij\)

\equiv \sqrt{\sum_{i=x-1}^{x+1} \sum_{j=y-1}^{y+1} (e_{ij}-e_{xy})^2}
≡∑i=x−1x+1∑j=y−1y+1(eij−exy)2\equiv \sqrt{\sum_{i=x-1}^{x+1} \sum_{j=y-1}^{y+1} (e_{ij}-e_{xy})^2}

\(e_{xy}\): elevation at cell \((x, y)\)

e_{ij}
eije_{ij}
e_{i-1,j}
ei−1,je_{i-1,j}
e_{i+1,j}
ei+1,je_{i+1,j}
e_{i+1,j-1}
ei+1,j−1e_{i+1,j-1}
e_{i,j-1}
ei,j−1e_{i,j-1}
e_{i-1,j-1}
ei−1,j−1e_{i-1,j-1}
e_{i,j+1}
ei,j+1e_{i,j+1}
e_{i-1,j+1}
ei−1,j+1e_{i-1,j+1}
e_{i+1,j+1}
ei+1,j+1e_{i+1,j+1}

Data source | SRTM30

Ruggedness

Image template

Source: https://nelson.wisc.edu/sage/data-and-models/atlas/maps.php?datasetid=19&includerelatedlinks=1&dataset=19

Findings

for 2015

Data source | VIIRS Day/Night Band Nighttime Lights 2015 Annual Composite

2015

Image template

Source: https://nelson.wisc.edu/sage/data-and-models/atlas/maps.php?datasetid=19&includerelatedlinks=1&dataset=19

Image template

Source: https://nelson.wisc.edu/sage/data-and-models/atlas/maps.php?datasetid=19&includerelatedlinks=1&dataset=19

Being lit in 2015

Local polynomial Linear   
 		 Quadratic
In Karabakh 0.01 -0.00
(0.01) (0.01) 
Optimal bandwidth 2814.2 6941.8 
Observations 8071 19403

Data source | VIIRS Day/Night Band Nighttime Lights 2015 Annual Composite

2015 (north-east)

Findings

for 1992-2013

Bandwidth choice for 1992-2013

Optimal bandwidth for each year

Median optimal bandwidth for all years

Fixed bandwidths for all years

1

2

3

From 5106.43 to 14435.77 meters

7450.34 meters

2500, 5000, 10000 meters

Data source | DMSP-OLS Nighttime Lights Time Series

2012

Data source | DMSP-OLS Nighttime Lights Time Series

2013

Bandwidth choice for 1992-2013

Optimal bandwidth for each year

Median optimal bandwidth for all years

Fixed bandwidths for all years

1

2

3

From 5106.43 to 14435.77 meters

7450.34 meters

2500, 5000, 10000 meters

Data source | DMSP-OLS Nighttime Lights Time Series

1995

What goes wrong?

Issue #1

Measurement error of the border

Solution to Issue #1

Google Map

Issue #2

RD design

for discrete dependent variable

Issue #3

External validity

The area around the border: different from the rest of the country

e.g. Cab drivers dim their lights at night, to avoid Armenian sniper fire from a few kilometers away

Gunfire and occasional shell explosions are routine for frontline residents, with up to 200 people killed each year

In frontline areas the Azeri government retains ownership of all land and artesian wells for irrigation

farmers cannot take loans from banks because they have no land to put down as collateral.

Source: EurasiaNet (2007)

Mail me your comments to m.kudamatsu@gmail.com

Made with Slides.com