Group Project

7.pdf

Home >Business & Finance homework help >Economics homework help >Group Project

CONTENTS PAGE

I. Introduction .......................................................................................................................... 4

II. Export Sophistication and Growth: Measurement and Stylized Facts ................................. 5

III. Sharp vs. Blunt Instrument ................................................................................................. 8

IV. Empirical Results ................................................................................................................ 9

V. Robustness ......................................................................................................................... 12

VI. Conclusion ....................................................................................................................... .19

TABLES

Table 1. Growth and Export Sophistication: OLS and Fixed Effects (5-Year Panel) ............ 10

Table 2. Growth and Export Sophistication: IV Estimation (5-Year Panel) .......................... 11

Table 3. Other Measures of Export Sophistication: IV Estimation (5-Year Panel).................13

Table 4. The Spillover Effect: IV Estimation (5-Year Panel) ................................................ 14

Table 5. Robustness: IV Estimation (5-Year Panel) ............................................................... 15

Table 6. GMM-System Estimation (5-Year Panel) ................................................................ 16

Table 7. Dissecting GMM (5-Year Panel) .............................................................................. 17

Table 8. IV Estimation (10-Year Panel) ................................................................................. 18

Table 9. IV Estimation, Instrument Set: Median (5-Year Panel) ........................................... 18

Table 10. IV Estimation, Instrument Set: Weighted Mean (5-Year Panel) ............................ 19

FIGURES

Figure 1. Log Initial Export Sophistication vs. 5-Year Ahead Log Real GDP per Capita Growth

Conditional on Initial Real GDP per Capita ............................................................................. 6

Figure 2. Log Real Manufacturing Exports per Capita vs. 5-Year Ahead Log Real GDP per

Capita Growth Conditional on Initial Real GDP per Capita..................................................... 6

References ............................................................................................................................... 21

Appendix ................................................................................................................................. 24

©International Monetary Fund. Not for Redistribution

Underline

I. INTRODUCTION

Determining the causes of economic growth is the grail sought by a large empirical growth

literature. To understand the causes of economic growth, what is sought is not an exhaustive list

of growth drivers which is probably unfathomable but rather a brief list of key determinants

of growth. However, uncovering key factors of growth is hard in practice. Data have

measurement errors; there are too many growth drivers to consider compared to data available;

and causation is hard to distinguish from correlation. In this paper, we attempt to tackle one of

the key problems in the empirical growth literature, namely, the causation vs. correlation or

endogeneity problem, in determining the key causes of growth in a cross-country setting.

The standard technique used in the literature to tackle the endogeneity problem is to use an

instrumental variable (IV) estimation. This technique relies on finding an instrumental variable

outside the model that is both relatively strongly correlated with the endogenous variable of

interest ( strength) and at the same time uncorrelated with the residual or innovation

term of the growth regression ( validity). Many growth studies seem to suffer from a

violation of one or both of these two conditions. Bazzi and Clemens (2013) showed that some

prominent growth studies focus on different determinants of growth while using the same

instrumental variables. As a result, they collectively problem.

In this case, at least one of the instrumental variable estimations must be invalid and possibly all

could be invalid.

Over the last two to three decades, the number of determinants of growth explored in the

literature grew much faster than the stock of instrumental variables available, which makes

tackling the blunt instrument issue crucial. For example, population size was used as an

instrument in a myriad of growth regressions to instrument for different endogenous variables

such as trade (Spolaore and Wacziarg 2005), international aid (Rajan and Subramanian 2008) or

export sophistication (Hausmann, Hwang, and Rodrik 2007) without necessarily controlling for

other studies explanatory variables.2 Moreover, even when the instrument does not suffer from

the problem, it could be weak, producing estimates that could misinform the reader

about the true effects on growth.3 It is not an exaggeration to say that thanks to Bazzi and

Clemens (2013) we know that we may not know much about key growth determinants from

instrumental variable growth regressions. We suggest a way to address their criticism.

In this paper we revisit the study of the main determinants of growth while avoiding the blunt

and weak instrument problems. Our instrumentation technique consists in using, as an instrument

for each endogenous variable, the average of the same variable in the neighboring marine and

land countries. The instruments we propose have the advantage of being variable-specific and

time-varying and the method produces strong instruments. The relatively

strong correlations of growth determinants between a country and its neighbors suggest that

geographic proximity can lead to imitation in trade openness, quality of institutions, education,

2 Bazzi and Clemens (2013) also argue that even when multiple instruments are used for the same endogenous

variable, in many cases population contains all the relevant information and other instruments are weak. 3 Kraay (2015), following the approach suggested in Bazzi and Clemens (2013), finds a similar weak instrument

problem in several studies of growth and inequality. A recent study by Berg et. al. (forthcoming) on growth and

inequality uses a variety of robustness checks to address this problem.

©International Monetary Fund. Not for Redistribution

financial development, and other policies. Moreover, we show evidence that the spillover effect

from neighbors or time-invariant country features do not affect our main conclusions.

We find that export sophistication is the only robustly significant determinant of growth among

the standard determinants such as human capital, trade, financial development, and institutions.

Moreover, in the presence of export sophistication, other standard growth determinants mostly

become statistically insignificant. One potential implication of our result is that improvements in

human capital, trade, financial development or institutions would raise economic growth to the

extent that they contribute to increasing export sophistication. We also show evidence that export

orientation of domestic production, as opposed to domestic production per se or specialization in

manufacturing, is critical.

In their seminal paper, Hausmann, Hwang, and Rodrik (2007) proposed a measure of export

sophistication and argued that it was a key causal factor of growth. The measure is based on the

weighted average of the degree of sophistication of the goods exported, which is measured by the

average GDP per capita of all the countries exporting such a good. Replicating Hausmann,

Hwang, and Rodrik (2007), Bazzi and Clemens (2013), show that in addition to the blunt

instrument problem of using the population variable as an instrument for export sophistication,

the problem of weak instruments could not be readily dismissed in the estimation. In this paper,

we first recalculate the export sophistication variable, extending it to 2014. Then we not only

resurrect the result of Hausmann, Hwang, and Rodrik (2007), but also show that export

sophistication is the only robust variable when the standard factors of growth are included in the

regression and the averages of variables in the neighboring countries are used as instruments.

Moreover, we propose other proxies for export sophistication such as real manufacturing exports

per capita or the share of manufacturing exports in total exports of goods and find broadly

similar results.

II. EXPORT SOPHISTICATION AND GROWTH: MEASUREMENT AND STYLIZED FACTS

As the experience of many oil-exporting countries shows, in the absence of improvement in

export sophistication, economic growth is fleeting (Cherif and Hasanov 2016). Although many

oil exporters have grown for periods of time on the back of large oil income flows, sustained

growth has not materialized as productivity growth has been stagnant or even negative. The

authors argue that the main source of productivity gains stems from the production of

sophisticated tradable goods, which in turn could be proxied by the degree of sophistication of

exports. This type of production and exports have been lacking in many oil exporters.

Export sophistication has a strong positive association with the 5-year ahead real GDP per capita

growth controlling for the level of initial GDP per capita (Figure 1).4 The level of sophistication

of each good is measured as the weighted average of real GDP per capita of all countries that

export that good a proxy for the level of sophistication. If a good is typically exported by rich

countries (poor countries), it will have a high (low) sophistication level. Export sophistication

4 The plot represents the residuals of the pooled OLS regression of growth on the initial logarithm of real GDP per

capita vs. the residuals of the pooled regression of export sophistication on the initial logarithm of real GDP per

capita. The slope of the fitted line should be equal to the coefficient of export sophistication in the pooled growth

regression controlling for initial income. The plot excludes a few outliers with 5-year growth rates over 20 percent.

©International Monetary Fund. Not for Redistribution

(EXPY in Hausmann, Hwang, and Rodrik 2007) is defined as the export-share weighted average

of sophistication levels of the count

Figure 1. Log Initial Export Sophistication vs. 5-Year Ahead Log Real GDP per Capita Growth

Conditional on Initial Real GDP per Capita

We also use alternative proxies of export sophistication such as the share of manufacturing in

goods exports and real manufacturing exports per capita. Both measures have high coefficients

of correlation with EXPY, about 60 and 75 percent, respectively. These measures have also a

strong positive correlation with the 5-year ahead real GDP per capita growth controlling for the

level of initial real GDP per capita (Figure 2).

Figure 2. Log Real Manufacturing Exports per Capita vs. 5-Year Ahead Log Real GDP per

Capita Growth Conditional on Initial Real GDP per Capita

AGO

AGO AGO

AGOALB ALB

ALB ALB ALB

ALB ALBALB

ALB

ARE

AREARGARG ARGARG

ARG ARG

ARG

ARGARG

ARG

ARM

ARMARM ATGATG

ATG ATG

ATG ATGATG

ATG ATG

AUS AUS

AUSAUSAUSAUS AUS

AUSAUS AUSAUS

AUT AUT

AUTAUT

AUT AUT

AUTAUTAUT

AZE

BDI

BDI BDI

BDI

BDI BDI

BDI

BDI BEL BEL

BELBEL

BEL

BEL BEL BELBELBEL

BENBEN BEN

BEN BEN

BEN

BEN BEN

BEN

BFA

BFA BFA BFA

BFA BFA

BFA

BFA BFA BFA

BFA BGD

BGD BGD BGD

BGD BGD BGDBGD

BGR

BGR BGR

BGR

BGR BHR

BHR

BHR BHR BHR

BHR

BHS

BHS BHS

BHS

BHS BHS

BHS

BIH BIH

BIH

BLR

BLR BLR

BLRBLZBLZ

BLZ

BLZBLZBLZ

BLZ BLZ

BOL

BOL BOLBOL

BOL BOL

BRA

BRA BRA

BRA

BRA BRB

BRB

BRBBRB BRB

BRN

BRN BRN

BRN BRNBRN

BTN

BTN BTN

BTN

BTNBTNBTN

BTN

BWABWA BWA

CAF

CAF CAFCAFCAF

CAFCAF

CAF

CAN CANCANCAN

CANCAN CAN

CAN

CAN CAN

CAN CHECHE

CHE

CHE CHE CHE

CHE

CHE CHECHECHECHL

CHL

CHL CHL CHL

CHL

CHN

CHN CHN

CHN

CHNCHN

CHN

CIV CIV

CIV

CIV CIV

CIV

CIV CMR

CMR

CMRCMR CMR

CMR

CMR CMR

CMR

COD

COD COD

COD

COD COD

COD

COG

COG COG

COG

COG COG

COG

COG COGCOG

COL

COL COLCOL

COL

COL COL

COL

COL COL COL

COM

COM COM

COM COM COM

COM COMCOM

CPV CPV

CPV

CPV CPV

CPV

CPV CPV

CPV

CRI CRI CRI CRI

CRI

CRI CRICRI CRI

CRICRI CYP

CYP

CYP CYP

CYPCYPCYP

CYP

CZE

CZE CZE

DEU DEU

DEUDEU DEU

DEUDEU

DJI

DJI DJI

DJI

DJI DJI

DMA

DMA DMA

DMA

DNKDNK

DNK DNKDNK

DNK DNK DNK

DNK DNKDNK

DOM

DOM DOM

DOM

DZA

DZA DZA

DZA

DZADZA

DZA

DZADZAECUECU

ECU

ECUECU ECU

ECU

ECU ECU

ECU

EGY

EGY EGY

EGYEGY EGY

EGY

ESP ESP

ESP

ESPESP

ESP

ESP ESP

EST EST

EST

ETH ETH

ETH

ETH ETH

FIN FIN

FIN FINFIN

FIN

FIN FINFJI

FJI FJI

FJI

FJIFJI FJI

FJI

FRAFRA

FRA

FRA FRA

FRAFRA

GAB

GAB GAB

GAB

GAB GAB GAB

GABGBR GBRGBRGBRGBR

GBR

GBR GBRGBR

GBR

GEO

GEOGEO

GHA

GHA GHA

GHA

GHA GHA GHA

GHA GHA

GHA

GIN GIN

GIN

GIN GIN

GMB

GMB GMB

GMB

GMB GMB

GMB

GNB

GNB GNB

GNQGNQ

GNQ

GNQ GNQ

GNQ

GRC

GRC GRCGRC

GRC GRC

GRC

GRD GRD

GRD

GTM GTM GTMGTM

GTM

GTM GTMGTM

HKG

HKG HKG

HKG

HKG HKG

HKG HND

HND HND

HND

HND HND

HNDHND

HND HNDHND

HRVHRV

HRV HRV

HTI

HTI HTI

HTI

HTI HTI

HTI

HUN

HUN HUN

HUN

HUN HUN

HUN

IDN

IDN IDN

IDN

IDN IDN

IDN

IDN IDNIDN

IND IND

INDIND

IND IND

IND

IRL IRL

IRL

IRLIRL

IRL

IRN

IRN IRN

IRN

IRQ

IRQ IRQ

ISL

ISLISL

ISL ISL

ISL

ISL ISL

ISL

ISL ISR

ISR ISR

ISR ISRISR

ISR

ISR ISR ITA

ITA

ITA ITA ITA

ITAITA

JAM JAM

JAM

JAM JAM

JOR

JOR JOR

JOR

JORJOR

JOR

JPN

JPNJPNJPN JPN

JPNJPNJPNJPNJPN

KAZ

KAZ KAZ

KEN

KEN KEN

KEN

KEN KEN KEN

KENKEN KGZ

KGZKGZKGZ

KHM

KHMKHMKNA

KNA

KNAKNA KNA

KNA

KOR

KORKOR KOR

KOR

KOR KOR

KWT

LAO

LAO LAO

LAO

LAOLAO

LBN LBN

LBN

LBNLBN

LBN

LBR

LBR LBR

LBR

LCA

LCA LCA

LCA

LCA LCA

LCALCA LCA

LKA

LKA LKA

LKA

LKALKA LKA

LKA LKA

LSO

LSO LSO

LTU

LUX

LUXLUX

LVA

LVA MAR

MAR MARMAR

MAR MAR

MAR

MARMAR MAR

MDA

MDA MDA

MDG

MDG MDG

MDG

MDG MDG

MDG MDG MDV

MDV MDV

MDV

MDV MDVMEX

MEX MEX MEX

MEXMEX MEX

MEX

MEXMEX MEX

MKD MKD

MKD

MLI

MLI MLI MLI

MLI

MLI MLI

MLI

MLT

MLT MLTMLT

MLTMLT MLTMMR

MMR

MMR MMR

MMR

MNE MNG MNG

MNG

MNGMNG

MNG

MOZ MOZ

MOZ

MOZ MOZMOZ

MRT

MRT MRT

MRT MRTMRT

MRT MRT

MRT

MUS

MUS MUS

MUS

MUS MUS

MUS

MUS MUS

MWI

MWIMWI

MWI MWIMWI

MWI

MYS

MYS MYS

MYS

MYSMYSMYS MYS

NAMNAMNAMNER

NER

NER NER

NER NERNER

NER

NGANGA

NGA

NGA NGA

NGA

NIC

NIC NIC

NIC

NIC NLDNLD

NLDNLD NLD

NLD NLD

NLD

NLDNLD NLD

NOR NOR

NORNOR NOR

NOR

NORNOR NOR

NOR NORNPLNPL

NPLNPL

NPLNPLNPLNPL NPL NPLNPL

NZL

NZLNZL NZL

NZL

OMN OMN

OMN

OMN OMN

OMN

PAKPAK

PAK

PAKPAKPAK PAK

PAK

PAKPAK

PAN

PAN PANPAN

PAN PAN

PERPER PER

PER PER

PER

PHLPHL PHLPHL

PHL

PHL PHL

PHL PHL PHL

POL

POLPOL

POL

PRT

PRT PRT

PRT

PRTPRT PRT

PRYPRY

PRY

PRY PRY

PRY

PRYPRY PRY

QAT

QAT QAT QAT

ROMROM

ROM

ROM ROM

ROM

ROM ROM

RUS

RUS RUS

RWA

RWA RWA

RWA

RWARWA RWA

SAU SAU

SDN

SDN SDN

SDNSDN

SDN SDN

SDN

SDNSENSEN SEN SEN

SENSEN SEN

SENSEN SENSEN

SGP

SGP SGP

SGP

SGP SGP

SGP

SLE

SLE SLESLE SLE

SLE

SLV

SLVSLV SLV

SLV SRB

SRB STP

STP

STP STP

STP

SUR

SURSUR

SUR

SUR SUR

SUR SVK

SVKSVK

SVK

SVN SVN

SWE SWE

SWESWE

SWE

SWE SWE SWESWE

SWZ SWZSWZ

SYC

SYRSYR

SYR

SYR SYR

SYR

TCD

TCD TCD

TCD

TCDTCD

TCD TCD

TCD

TGO

TGO TGO

TGO

TGO TGO

TGO

THA

THA THA

THA

THA THA THA

TJK

TJK TJK

TKM TKM

TTO

TTO TTO

TTO

TTOTTO

TTO

TUN

TUN TUN

TUN

TUN TUN

TUN

TUN TUNTUN

TUN

TUR TUR

TUR

TUR TUR

TUR

TWN TWN

TWN

TWNTWN TWN

TZA TZA TZA TZA

TZA TZA

TZA

TZA TZA

TZAUGA

UGA

UGA UGA

UGA

UKR

UKR UKR

URY

URYURY

URY

URYURY

URY URY

URY

URYUSA

USA USA USAUSAUSA

USA USA

USA

USA USA

UZB

UZB UZB

VCT

VCT VCTVCTVCT

VCT

VCT VCT

VEN

VEN VEN VEN

VEN

VEN VEN

VEN

VEN VEN

VEN VNM

VNM VNM

VNM

VNM VNM

VNM

YEM

YEM YEM

YEM

ZAF

ZAF ZAFZAF

ZAF ZAF ZAF

ZAF ZAFZAF

ZAF ZMB

ZMB

ZMB ZMB

ZMB

ZMB ZWE

ZWE

ZWE ZWE

ZWEZWE

ZWE

-. 2

-. 1

0 .1

.2 5

-Y e

a r

re a

l G

D P

p e

r c a

p it a

g ro

w th

| i n

it ia

l in

c o

m e

-2 -1 0 1 2 Log initial export sophistication | initial income

AGO AGO

AGO

ALBALB

ALB

ARE

ARGARG ARG ARG

ARG ARG

ARG

ARGARG

ARG

ARM

ARMARM ATG

ATG

ATG ATG

AUS

AUSAUSAUSAUS AUS

AUSAUS AUSAUS

AUT AUTAUT

AUT AUT

AUTAUTAUT

AZE

BDI

BDI BDI

BDI

BDI BEL BEL BELBEL

BEL

BEL BEL

BELBELBEL

BENBEN BEN

BEN BEN BENBEN

BEN

BFA

BFA BFA

BFA

BFABFA BFA

BFA

BGD BGDBGD

BGD BGD

BGDBGD

BGR

BGR BHR

BHR

BHR BHR

BHR

BHS

BHS BHS

BHS

BIH

BLR BLR

BLR BLZ

BLZ

BLZBLZBLZ

BLZBLZ

BOL

BOL BOLBOL

BOL BOL

BRA

BRA BRA BRA BRA BRA

BRA

BRB

BRBBRBBRB BRN

BTN

BWABWA BWA

CAF

CAF CAF CAF

CAF

CAN CANCANCAN

CANCAN CAN

CAN

CAN CAN

CAN CHECHE

CHE

CHE CHE

CHE

CHE CHECHECHECHL

CHL

CHL CHL

CHN

CHNCHN

CHN

CIV CIV

CIV

CIVCIV CIV

CIV

CIV CMR

CMR

CMRCMRCMR

CMR

CMR CMR CMR

CMR

COG

COGCOG

COG

COG COL

COLCOL COL

COL

COL COL

COL

COL COL

COL COM

COM COM COM

CPV CPV

CPV

CRICRI CRI

CRI

CRI CRICRI CRI

CRICRI

CYP

CYP CYP

CYPCYPCYP

CYP

CZE

CZE CZE

DEU DEU

DEUDEU DEU

DEUDEU

DJI

DMA DMA

DMA DMA DMA

DMA

DNKDNK

DNK DNKDNK

DNK DNK DNK

DNK DNKDNK

DOM

DOM DOM

DOMDOM

DOM DZA

DZA

DZADZA

DZA

DZADZAECUECU

ECU

ECUECU ECU

ECU

ECU ECU

ECU EGY

EGY

EGY EGY

EGY EGY EGY

EGY

ESP ESP

ESP

ESP ESP

ESP

ESP ESP

EST EST

EST

ETH

ETH ETH

FIN FIN

FINFIN FIN

FIN

FIN FIN

FJI

FJI FJI

FJI

FRAFRA

FRA

FRA FRA FRAFRA

GAB

GAB GAB

GBR GBRGBRGBRGBR

GBR

GBR GBRGBR

GBR

GEO

GEOGEO

GHA

GHA GHA

GHA

GHA GHA

GHA

GINGIN

GIN GMB

GMB

GMB GMB

GMB

GNB

GNQ

GRC

GRC GRCGRC

GRC GRC

GRC

GRD

GTM GTMGTM

GTM

GTM GTMGTM

HKG

HKG HKG

HKG

HKG HKG

HKG

HND HND

HND

HND HND

HNDHND

HND HNDHND

HRVHRV

HRV HRV

HTI

HUN

HUN HUN

HUN

HUN HUN

HUN

IDN

IDN IDN

IDN

IDN IDN

IDN

IDN IDNIDN

IND IND

INDIND

IND IND

IND

IND IRL

IRL IRL

IRL

IRLIRL

IRL

IRN

IRNIRN

IRN

IRQ

ISL

ISLISL

ISL ISL

ISL

ISL ISL

ISL

ISL ISR

ISR ISR

ISR ISRISR

ISR

ISRISR ITA

ITA

ITA ITA

ITA ITAITA

JAM

JAM JAM

JOR JOR

JOR

JORJOR

JOR

JPN

JPNJPNJPN JPN

JPNJPNJPNJPNJPN

KAZ

KAZ KAZ

KEN

KEN KEN

KEN KENKEN

KGZ KGZKGZ KGZ

KHM

KHMKHM

KNA KNA

KNA

KOR

KOR KOR KOR

KOR

KOR KOR KOR KOR

KWT

LAO

LBN

LBR LBR

LBR

LCA LCA

LCA

LCA LCA

LCALCA LKA

LKA

LKA LKA

LKA

LKA LKA

LSO LSO

LTU

LUX

LUXLUX

LVA

LVA MAR

MAR MAR MAR

MAR MAR

MAR

MARMAR MAR

MDA

MDA MDA

MDG

MDG MDG

MDG

MDG MDG

MDGMDG

MDV

MDV MEX

MEXMEX MEX

MEX MEX MEX

MEX

MEXMEX MEX

MKD MKD

MKD

MLI

MLI MLI MLI

MLI MLI

MLI

MLT

MLT MLT

MLTMLT MLTMMR

MMR

MMR MNGMNG

MOZ

MOZ MOZMOZ

MRT

MRT MRT

MUS

MUS MUS MUS

MUS

MUS MUS

MWI

MWIMWI

MWI MWIMWI

MWI

MYS

MYS MYS

MYS

MYSMYSMYS MYS

NAMNAMNAMNER

NER

NER NER NER

NER

NGANGA

NGA

NIC

NIC NIC

NIC

NIC NLDNLD

NLDNLD NLD

NLD NLD

NLD

NLDNLD NLD

NOR NOR

NORNOR NOR

NOR

NORNOR NOR

NOR NOR

NPL

NPL NPLNPLNPLNPL NPL

NZL

NZLNZL NZL

NZL

OMN

OMN OMN

OMN

PAKPAK

PAK

PAKPAKPAK PAK PAK

PAK

PAKPAK

PAN

PAN PANPAN

PAN PAN

PERPER PER

PER PER

PER

PHLPHL PHL PHL

PHL

PHL PHL

PHL

POL

PRT

PRT PRT

PRT

PRTPRT PRT

PRYPRY

PRY

PRY PRY

PRY

PRYPRYPRY

QAT

QAT QAT

QAT

RUS

RUS RUS

RWA RWA

SAU SAU

SDN

SDN SDN

SDN

SDN SDN

SDN

SDN SENSEN SEN

SEN SEN

SEN

SEN SENSEN

SGP

SGP SGP

SGP

SGP SGP

SGP

SLE

SLV

SLVSLV SLV SLVSTP

STP STP

SUR

SUR SUR

SUR SVK

SVKSVK

SVK

SVN SVN

SWE SWE

SWESWE

SWE

SWE SWE SWESWE

SWZ SWZ

SYC

SYCSYR

SYR

SYR SYR

SYR

TCD

TCD TCD

TCD

TGO

TGO TGO

TGO

TGO TGO

TGO

THA

THA THA

THA

THA THA THA

TJK

TKM

TTO

TTO TTO

TTO

TTOTTO

TTO

TUN

TUNTUN

TUN

TUN TUN

TUN

TUN TUNTUN

TUN

TUR TUR

TUR

TUR TUR

TUR TUR TUR

TUR TUR

TZA

TZA TZA

TZAUGA UGA

UGA

UKR

UKR UKR

URY

URYURY

URY URY

URY

URYUSA

USA USA

USAUSAUSA USA USA

USA

USA USA

VCT VCT

VCT

VCTVCT

VEN

VENVEN VEN

VEN

VEN VEN

VEN

VEN VEN

VEN

VNM VNM

VNM YEM

YEM YEM

YEM

ZAF

ZAF ZAFZAF

ZAF

ZMB

ZWE

ZWEZWE

ZWE

-. 2

-. 1

0 .1

.2 5

-Y e

a r

re a

l G

D P

p e

r c a

p it a

g ro

w th

| i n

it ia

l in

c o

m e

-6 -4 -2 0 2 4 Log initial real manufacturing exports per capita | initial income

©International Monetary Fund. Not for Redistribution

In a macro setting, finding valid and strong instruments is not a straightforward task and the

GMM estimation, using own lagged variables, should bypass this problem (and in

theory could help avoid the blunt instrument issue). Unfortunately, many of these instruments

turn out to be weak, especially when using a smaller number of instruments. Bazzi and Clemens

(2013) show that several of the seminal papers they examine do not survive the opening of the

, use their term. In particular, the authors analyze the excludability of the

country size (population and area) instrument and examine the instrument validity (e.g. using the

test of Hahn, Ham, and Moon 2011), perform various tests of underidentification of the

instrument set and tests of weak instruments (e.g. using Kleibergen-Paap and Cragg-Donald test

statistics), and use estimation methods robust to weak instruments (e.g. a testing procedure using

Kleibergen 2002). Interestingly, the only determinant of growth, which seemed to broadly

survive their comprehensive analysis, is the export sophistication variable of Hausmann, Hwang,

and Rodrik (2007). However, even in this case, the regressions do not pass all the tests. Weak

instrument tests fail in the case of smaller or collapsed number of instruments, and the validity

of the population variable is questioned. However, the weak-instrument robust confidence set

estimation indicates that the export sophistication variable has a positive and statistically

significant effect.

In the following section, we study the standard determinants of growth put forth by the literature

in addition to export sophistication human capital, quality of institutions, trade openness,

financial development, foreign direct investment (FDI), saving rates, investment, government

size, and the Gini coefficient. We use real GDP per capita for 1960-2014 from the Penn World

Tables 9.0 (Feenstra, Inklaar, and Timmer 2015). Years of schooling come from Barro and Lee

(2013).

Export sophistication data (EXPY) of Hausmann, Hwang, and Rodrik (2007) are computed using

the World Trade Flows data (Feenstra and Romalis 2014) for 1962-2014. We also compute a

structural EXPY measure S-EXPY which discounts the high share of commodity exports of

high-income commodity exporters correcting the artificially high EXPY of commodity exporters

(see Appendix for details).

trade (exports plus imports, percent of GDP) as a measure of trade openness, domestic credit to

the private sector (percent of GDP) as a measure of financial development, manufacturing

exports (in constant USD and percent of GDP), manufacturing production (real value added in

local currency and percent of GDP), FDI (percent of GDP), government consumption (percent of

GDP) as a measure of government size, and gross fixed capital formation (percent of GDP). The

Economic Outlook database, and the Gini

coefficient is from SWIID v4. The law and order indicator, measuring the strength and

impartiality of the legal system and the assessment of popular observance of the law, is used as a

proxy of the quality of institutions and

database (the data start in 1984). We also use a corruption indicator from the same source.

©International Monetary Fund. Not for Redistribution

III. SHARP VS. BLUNT INSTRUMENT

In this section, we describe our instrumental variable methodology. Finding valid and strong

instruments in the cross-country setting is challenging. As argued by Bazzi and Clemens (2013),

many papers use the same instruments such as population and area for different variables. In

addition, these instruments suffer from validity and possibly weak instrument problems. To

illustrate the blunt instrument problem, suppose that growth could be (potentially) explained by

two factors and such that:

Let us assume that two studies use growth regressions (A) and (B) which have the following

forms:

Let us suppose that one study uses (A) and instrumental variable , and a second study uses (B)

while relying on the same instrumental variable . Assuming that is a valid instrument in both

(A) and (B) and that and/or are significant determinants of growth, is problematic. Indeed,

(A) could be re-written as , while (B) could be re-written as

. If is correlated with , then the latter is also correlated with the error term of

(A), and the same logic applies to (B). In other words, at least one of the studies relies on an

invalid instrumental variable (and it could be the case for both).

To remedy the blunt instrument problem, we propose the sharp instrument solution. Our

method instruments for variables of a country with the average values of these variables in its

neighboring countries. The advantage of this IV method is that it generates variable-specific

instruments and can be applied to a wide range of explanatory variables thus bypassing the

problem of blunt instruments described above. We also test for the strength of our instruments

(correlation with the variables for which they are instruments).

We argue that using the average of a variable in neighboring countries as an instrument is likely

to satisfy the exclusion restriction from the growth equation (validity of instruments) while at the

same time, it should be variable. The exclusion

restriction requires that the innovation or error term in the growth regression be uncorrelated

with the instruments for explanatory variables the average values of those variables in

neighboring countries. If the validity assumptio

In contrast, some

the growth regressions. Chua (1993) and Ades and Chua (1993) show that various practices and

traits that are unfavorable to growth could spill over from neighboring countries and add simple

regression. Easterly and Levine (1998) control

for the neighbor th regression and instrument it with the

Growth and its determinants in neighboring countries could be related in several ways. We

©International Monetary Fund. Not for Redistribution

propose different methods and controls to verify that our instruments are not invalid due to some

unaccounted correlation with the residuals of the growth regression. Governments, firms and

households in neighboring countries could imitate each other because of regional competition,

common languages, or cultural proximity.5 In particular, Bahar, Hausmann, and Hidalgo (2014)

show that a country is more likely to export a product if its neighbor is exporting it. This type of

effect would explain the strength of our instrument without invalidating it.

A country could be affected by spillovers from its neighbors mostly, but not exclusively, through

trade and finance. Being close to a country that is growing fast could encourage FDI and

technological transfers as was the case in East Tigers. 6 We offer several types of

robustness checks to verify that our instrument remains valid (see the next section). First, we

control for the average real GDP per capita or real GDP in neighboring countries as a proxy for

the spillover effect. Second, we modify the weighting of the instrument to mitigate a potential

violation of the exclusion restriction of instruments based on simple averages. We use the

median of variables of neighboring countries and the weighted average of variables of neighbors,

in which weights are inversely proportional to real GDP. The median neighbor is less likely to be

the main trading partner of a country, while the weights based on the inverse of real GDP

mitigate the impact of large neighbors on the construction of instruments. This weighting scheme

is inversely related to the size, a key predictor of trade links in the gravity model, assigning a

smaller weight to bigger neighbors.

Neighboring countries in general share and climate,

which are likely to affect growth. This could invalidate our instruments if not accounted for. We

use latitude, ethnic fractionalization, and a dummy for Sub-Saharan Africa to control for some of

these features. We also run a fixed effect IV regression. If we properly control for spillovers,

common traits

iction.

IV. EMPIRICAL RESULTS

Running ordinary least squares (OLS) and fixed effects (FE) regressions (Table1), we find that

many standard growth determinants are correlated with the growth rate. Regressing growth on

initial log real GDP per capita and export sophistication and controlling for each standard

determinant of growth separately (columns 1-5) yields mostly highly statistically significant

coefficients (law and order is, however, not statistically significant) with the expected signs

except for private sector credit. The coefficient on private sector credit is negative but this could

be due to potential nonlinearities in the finance-growth nexus found in the literature (e.g. Arcand,

Berkes, and Panizza 2015 and Demetriadis and Rousseau 2016). Increasing private credit could

5 Riva

in a sporting contest. 6 Typically, emerging and low-income countries have strong trade and financial links with advanced countries or

large emerging markets, which are remote. Meanwhile, a developing economy is usually surrounded by other

developing economies with little trade and financial links. In the absence of such links, it is plausible that there is no

spillover effect from neighboring countries.

be correlated with higher vulnerabilities, financial instability and lower growth (e.g. Popov 2014

and Levine, Lin, and Xie 2016).

The coefficient estimates for export sophistication we find when including more controls, are

consistent with the relationship shown in Figure 1. A 10 percent increase in export sophistication

is associated with about 0.2-0.3 percent increase in the annual growth rate. The regression with

all variables (columns 6-7) also yields statistically significant coefficients with the expected

signs for all the variables. The same regressions with fixed effects result in a similar statistically

significant estimate on export sophistication as in OLS regressions (columns 8-11). However,

adding law and order reduces the sample size substantially and makes the export sophistication

parameter statistically insignificant (columns 12-14). These regressions, however, show us

correlations between growth and standard growth determinants and to infer causality, we turn to

instrumental variable estimations.

Table 1. Growth and Export Sophistication: OLS and Fixed Effects (5-Year Panel)

Using an instrumental variable estimation, based on the average of variables of neighboring

countries as an instrument for each explanatory variable, we find that export sophistication is the

key determinant of growth (Table 2). A 10 percent increase in export sophistication, measured by

EXPY, increases the average annual growth rate in the next 5 years by about 0.6-0.7 percent.

This result, namely the statistical significance and the magnitude of the coefficient on export

sophistication, is robust across most estimations and is about as robust as the initial real GDP per

capita.7 The coefficient obtained is two to three times larger than the coefficient in OLS or FE

regressions suggesting a large downward bias. In column 1, real GDP per capita is assumed to be

exogenous, while it is not assumed to be exogenous in column 2. In each case, our instruments

are

value of the instrument for export sophistication for Mexico is the average of export

7 All estimations include a constant and time dummies (not shown). In this specification, fixed effects are subsumed

variables. It is a less

-LEV

Dependent variable: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

5-year ave. annual growth rate OLS OLS OLS OLS OLS OLS OLS FE FE FE FE FE FE FE

Log export sophistication 0.027*** 0.024*** 0.027*** 0.029*** 0.029*** 0.024*** 0.022*** 0.020*** 0.026*** 0.024*** 0.012 0.008

(0.004) (0.004) (0.004) (0.004) (0.006) (0.005) (0.006) (0.005) (0.006) (0.006) (0.010) (0.010)

Log real GDP per capita -0.010*** -0.015*** -0.012*** -0.010*** -0.012*** -0.010*** -0.015*** -0.044*** -0.050*** -0.052*** -0.049*** -0.097*** -0.072*** -0.070***

(0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) (0.006) (0.008) (0.006) (0.005) (0.023) (0.012) (0.012)

Years of schooling 0.003*** 0.004*** 0.004*** -0.003 -0.003 -0.002

(0.001) (0.001) (0.001) (0.002) (0.002) (0.002)

Trade (% of GDP) 0.017*** 0.005** 0.004** 0.037** 0.010 0.014**

(0.005) (0.002) (0.002) (0.015) (0.008) (0.007)

Credit to private sector (% of GDP) -0.007** -0.010*** -0.013*** -0.010* -0.028*** -0.028***

(0.003) (0.003) (0.003) (0.005) (0.005) (0.005)

Law and order 0.001 0.003*** 0.002** -0.000 0.004** 0.004**

(0.001) (0.001) (0.001) (0.002) (0.002) (0.002)

Observations 1,592 1,226 1,376 1,333 748 609 601 1,592 1,226 1,376 1,333 748 609 601

Adjusted R-squared 0.082 0.136 0.118 0.092 0.088 0.119 0.159 0.156 0.216 0.207 0.211 0.284 0.301 0.300

\# of countries 171 137 168 167 134 117 117

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

i.e. Belize, Cuba, Guatemala, Honduras, and the U.S. In

both specifications, tests for weak instruments suggest that the instruments are strong. The

conditional likelihood ratio (CLR) confidence set of Moreira (2003), which is robust to the

weak-instrumentation of one endogenous variable (column 1), indicates that the coefficient on

export sophistication is in the range of 0.06 to 0.09. We find that the coefficient estimates stay

within this range as we add more variables to the regression (columns 3-5) and increase to about

0.1-0.15 in other specifications (columns 6-9).

Table 2. Growth and Export Sophistication: IV Estimation (5-Year Panel)

Controlling for other determinants of growth (one at a time) such as human capital, law and

order, trade, and financial development do not much affect the coefficient or significance of

export sophistication or initial income (columns 3-6). All variables are considered endogenous

and are instrumented using the average values of those variables in neighboring countries. Years

of schooling and trade are not statistically significant (columns 3-4). The effects of credit and

law and order are negative (columns 4-6), albeit at lower significance levels than those for export

sophistication or initial income. The negative IV coefficient on credit is also obtained in OLS

and FE regressions. The negative coefficient on law and order is more surprising, especially

since the coefficient in OLS and FE regressions, is positive. One potential explanation is that it

could also exhibit some nonlinearities similar to the private credit variable. In addition, this

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent var.: 5-year ave. annual growth rate IV IV IV IV IV IV IV IV IV-FE

Log export sophistication 0.072*** 0.063*** 0.065*** 0.074*** 0.077*** 0.105*** 0.148*** 0.154**

(0.009) (0.009) (0.013) (0.013) (0.012) (0.028) (0.055) (0.070)

Log real GDP per capita -0.024*** -0.020*** -0.024*** -0.024*** -0.023*** -0.021*** -0.007* -0.026*** -0.116***

(0.003) (0.003) (0.004) (0.005) (0.003) (0.006) (0.004) (0.008) (0.043)

Years of schooling 0.001 0.004** -0.003 0.057

(0.001) (0.002) (0.004) (0.046)

Trade (% of GDP) -0.017 -0.006 -0.047

(0.013) (0.011) (0.032)

Credit to private sector (% of GDP) -0.023* -0.001 -0.003

(0.013) (0.013) (0.023)

Law and order -0.016** 0.001 -0.013*

(0.006) (0.003) (0.007)

Observations 1,590 1,590 1,216 1,369 1,319 748 606 598 983

\# of endogenous variables 1 2 3 3 3 3 5 6 3

\# of instruments 13 13 13 14 14 9 11 12 14

\# of excluded instruments 1 2 3 3 3 3 5 6 6

Cragg-Donald F stat 347.0 136.2 40.7 12.9 37.9 9.6 4.4 0.8 1.4

Kleibergen-Paap F stat 267.5 104.8 38.4 10.8 24.8 7.7 2.5 1.0 1.3

Kleibergen-Paap LM test p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.11

H_0: t-test size>10% (p-value) | KP 0.00 0.00 0.00 0.17 0.00 0.47 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.01 0.79 1.00 0.99

H_0: t-test size>10% (p-value) | CD 0.00 0.00 0.00 0.07 0.00 0.26 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.00 0.33 1.00 0.99

H_0: t-test rel-bias>10% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.04 0.95 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.00 0.43 0.94 0.88

H_0: t-test rel-bias>10% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.01 0.68 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.00 0.08 0.97 0.85

Hansen J-test p-value 0.17

Lower CLR bound 0.06

Upper CLR bound 0.09

H0: Beta_EXPY=0 | CLR p-value 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

regression exhibits a potential weak instrument problem. Kleibergen-Paap (KP) Wald statistic is

much smaller, and the null hypothesis that the actual size of the t-test of the coefficients equal

zero at the 5 percent level is greater than 10 percent cannot be rejected (but is rejected if the size

is greater than 25 percent). The tests for weak instruments in other estimations suggest that the

instruments are not weak.

In the IV regression with all the determinants of growth while excluding export sophistication

(column 7), only years of schooling and initial income remain significant although the coefficient

on initial income drops sharply. With export sophistication (column 8), years of schooling

becomes insignificant; law and order has a negative coefficient at 10 percent significance level;

and export sophistication and initial income are strongly significant with a larger coefficient of

0.15 on export sophistication. In these two regressions, we no longer obtain favorable statistics

for strong instruments.8 It is likely that including several endogenous explanatory variables in the

growth regression results in the weak instrument problem as endogenous variables and

instruments are probably all correlated with each other resulting in weak identification.

Nonetheless, the coefficient on export sophistication is statistically significant at 1 percent level

in this specification as well. Since the effect of years of schooling in the regression without

export sophistication is statistically significant and has a meaningful sign, we specify our

baseline regression with initial income, years of schooling and export sophistication. Lastly, we

add fixed effects to this specification and confirm our previous finding that export sophistication

remains statistically significant and robust with a positive and relatively large effect on growth.

However, the fixed effects IV regression, in which the equation is differenced and the dependent

variable is the change in the growth rate, has the weak instrument problem as well. This suggests

that it is harder to predict endogenous variables that are growth rates rather than levels using

as instruments.

V. ROBUSTNESS

We experiment with alternative proxies of export sophistication manufacturing exports as a

regression, we find that both the share of manufacturing exports and real manufacturing exports

per capita have all significant and positive coefficients (Table 3, columns 2 and 4). Including

both EXPY and another measure of manufacturing exports results in quasi-multicollinearity and

insignificant coefficients (columns 3 and 5). The weak instrument tests show that the regressions

with both measures are plagued with the weak instrument problem.

Adding a control for manufacturing production, we find that export sophistication seems to be

more important than manufacturing production in affecting growth. Manufacturing value added

as a share of GDP is statistically significant in the regression with EXPY (columns 6-7) but real

manufacturing value added per capita (in logs) is not statistically significant (columns 8-9).

However, with other proxies for export sophistication, manufacturing value added as a share of

GDP is no longer statistically significant while the coefficient of log real manufacturing value

added per capita is negative, which seems to pick up the effect of the initial income variable

8 Since the reported test statistics are based on 2-3 endogenous variables from Stock and Yogo (2005), and we have

a total of 6 endogenous variables, the thresholds used are relatively conservative.

(columns 10-11). Export sophistication proxies have positive and statistically significant

estimates in all regressions. This suggests that export orientation is important in the growth

process and that producing manufacturing, and potentially sophisticated, goods without

exporting them may not be sufficient to increase long-run growth.

Table 3. Other Measures of Export Sophistication: IV Estimation (5-Year Panel)

We control for the average logarithm of real GDP (or real GDP per capita) in the neighboring

countries to capture directly spillover effects. Doing so should also mitigate a potential violation

e average GDP or GDP per capita in

initial income.9

statistically significant at the 10 percent level (Table 4, column 1). Excluding schooling, it

becomes statistically insignificant (column 2). The coefficient on EXPY is statistically

significant and similar to other estimates. In the baseline regression (Table 4, column 2), the

coefficient of the spillover effect as measure by average real GDP of neighbors is positive and

strongly significant while EXPY is no longer significant (column 3). However, when we exclude

schooling, which is not significant in most of our regressions (see Tables 2-10), including

column 3 regression when EXPY is not included, we obtain a positive and statistically

significant coefficient on EXPY (column 4).

9 We exclude the duplicate countries and the country in question for which the instrument is calculated when

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Dependent var.: 5-year ave. annual growth rate

Log export sophistication 0.065*** 0.053 0.209 0.063*** 0.083***

(0.013) (0.051) (0.830) (0.021) (0.020)

Log real GDP per capita -0.024*** -0.012*** -0.026* -0.027*** -0.025*** -0.003 -0.023*** -0.002 -0.029** -0.013*** -0.005

(0.004) (0.003) (0.014) (0.005) (0.009) (0.004) (0.008) (0.009) (0.013) (0.003) (0.009)

Years of schooling 0.001 0.004*** 0.003** 0.001 0.008 -0.003 -0.003 0.003 0.002 0.003* 0.004*

(0.001) (0.001) (0.001) (0.002) (0.028) (0.003) (0.003) (0.002) (0.003) (0.002) (0.003)

Manufactures exports (% of merchandise exports) 0.000*** -0.000 0.000***

(0.000) (0.000) (0.000)

Log real manufacturing exports per capita 0.012*** -0.034 0.012***

(0.003) (0.180) (0.004)

Manufacturing value added (% of GDP) 0.004*** 0.003** 0.001

(0.002) (0.001) (0.001)

Log real manufacturing value added per capita -0.004 -0.003 -0.022**

(0.008) (0.009) (0.011)

Observations 1,216 947 947 946 946 828 799 799 770 671 651

\# of endogenous variables 3 3 4 3 4 3 4 3 4 4 4

\# of instruments 13 13 14 13 14 13 14 13 14 14 14

\# of excluded instruments 3 3 4 3 4 3 4 3 4 4 4

Cragg-Donald F stat 40.7 109.7 2.6 13.0 0.0 6.0 5.1 13.1 6.6 5.9 5.5

Kleibergen-Paap F stat 38.4 100.0 1.7 11.6 0.0 4.9 4.0 15.1 5.7 4.0 5.9

Kleibergen-Paap LM test p-value 0.000 0.000 0.013 0.000 0.800 0.000 0.000 0.000 0.000 0.000 0.000

H_0: t-test size>10% (p-value) | KP 0.000 0.000 1.000 0.117 1.000 0.822 0.996 0.021 0.968 0.996 0.961

H_0: t-test size>25% (p-value) | KP 0.000 0.000 0.822 0.000 1.000 0.077 0.286 0.000 0.080 0.289 0.067

H_0: t-test size>10% (p-value) | CD 0.000 0.000 1.000 0.062 1.000 0.689 0.982 0.060 0.931 0.962 0.974

H_0: t-test size>25% (p-value) | CD 0.000 0.000 0.607 0.000 1.000 0.031 0.125 0.000 0.038 0.069 0.093

H_0: t-test rel-bias>10% (p-value) | KP 0.000 0.000 0.960 0.002 1.000 0.247 0.608 0.000 0.288 0.611 0.256

H_0: t-test rel-bias>30% (p-value) | KP 0.000 0.000 0.634 0.000 1.000 0.054 0.122 0.000 0.022 0.123 0.017

H_0: t-test rel-bias>10% (p-value) | CD 0.000 0.000 0.865 0.001 1.000 0.132 0.381 0.001 0.176 0.262 0.317

H_0: t-test rel-bias>30% (p-value) | CD 0.000 0.000 0.374 0.000 1.000 0.021 0.040 0.000 0.008 0.018 0.027

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

The coefficient for the spillover effect as measured by real GDP per capita of neighbors is

statistically insignificant using median or weighted averages of neighbors for the instruments

(columns 5-6 and 9-10). However, the estimates when using real GDP of neighbors as the

spillover effect are similar irrespective of the weighting schemes for the instruments used

(columns 7-8 and 11-12). The tests show that instruments are not weak when using real GDP as a

measure of the spillover effect. Overall, we find that the spillover effect, even if present, does not

invalidate our initial finding that export sophistication is a key growth determinant.10

Table 4. The Spillover Effect: IV Estimation (5-Year Panel)

Further, we explore other potential explanatory variables in the baseline regression and examine

the robustness of our results (Table 5). We add such variables as investment to GDP ratio, the

national saving rate, FDI, government consumption, the Gini coefficient, and corruption

(columns 3-9). The coefficient on export sophistication varies in the range of 0.05-0.08 and is

statistically significant in line with the previous results. However, some of these regressions

suffer from the weak instrument problem. We also include a measure of export sophistication

Ding and Hadzi-Vaskov (2017) that results in a robust and positive estimate

which is even larger than in previous regressions (column 2).11 Another measure of EXPY we

use structural EXPY, or S-EXPY which corrects for the share of commodities in exports also

produces a strong and positive coefficient although it is two to three times smaller in magnitude

(column 1).

10 The results with other proxies for export sophistication are broadly the same. We also control for the growth rates

of neighbors in the previous 5-year period, but obtain statistically insignificant results and weak instruments.

Regression results are not included in Table 4 and are available upon request. 11 The same study also computes a standardized EXPY but it is highly correlated with the original EXPY and

produces similar results (with a coefficient closer to our estimates).

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Dependent var.: 5-year ave. annual growth rate

Log export sophistication 0.084*** 0.065*** 0.002 0.043** 0.109*** 0.066** 0.010 0.057** 0.045*** 0.066*** -0.001 0.033*

(0.019) (0.015) (0.016) (0.020) (0.032) (0.027) (0.016) (0.024) (0.012) (0.015) (0.016) (0.019)

Log real GDP per capita -0.057*** -0.005 -0.016*** -0.018*** -0.080* 0.007 -0.018*** -0.023*** -0.023 -0.029* -0.019*** -0.017***

(0.018) (0.029) (0.004) (0.004) (0.041) (0.047) (0.004) (0.006) (0.015) (0.015) (0.004) (0.004)

Years of schooling 0.001 0.002* 0.001 0.001 0.003** 0.005***

(0.001) (0.001) (0.002) (0.001) (0.001) (0.001)

Log real GDP per capita of neighbors 0.025* -0.015 0.038 -0.027 0.001 0.005

(0.014) (0.025) (0.030) (0.036) (0.010) (0.009)

Log real GDP of neighbors 0.012*** 0.005 0.013*** 0.005 0.011*** 0.007**

(0.002) (0.003) (0.002) (0.003) (0.002) (0.003)

Observations 1,216 1,489 1,216 1,489 1,216 1,489 1,216 1,489 1,216 1,489 1,216 1,489

\# of endogenous variables 4 3 4 3 4 3 4 3 4 3 4 3

\# of instruments 14 13 14 13 14 13 14 13 14 13 14 13

\# of excluded instruments 4 3 4 3 4 3 4 3 4 3 4 3

Cragg-Donald F stat 4.1 3.2 18.7 17.8 1.0 1.1 18.8 15.6 2.8 7.8 20.3 22.6

Kleibergen-Paap F stat 3.8 2.4 16.7 15.3 0.9 0.8 17.1 12.9 2.6 7.0 17.2 18.9

Kleibergen-Paap LM test p-value 0.00 0.01 0.00 0.00 0.06 0.11 0.00 0.00 0.00 0.00 0.00 0.00

H_0: t-test size>10% (p-value) | KP 1.00 0.98 0.05 0.02 1.00 1.00 0.04 0.06 1.00 0.56 0.04 0.00

H_0: t-test size>25% (p-value) | KP 0.32 0.42 0.00 0.00 0.96 0.85 0.00 0.00 0.59 0.01 0.00 0.00

H_0: t-test size>10% (p-value) | CD 1.00 0.96 0.02 0.00 1.00 1.00 0.02 0.02 1.00 0.46 0.01 0.00

H_0: t-test size>25% (p-value) | CD 0.27 0.27 0.00 0.00 0.94 0.78 0.00 0.00 0.55 0.01 0.00 0.00

H_0: t-test rel-bias>10% (p-value) | KP 0.64 0.70 0.00 0.00 0.99 0.96 0.00 0.00 0.85 0.07 0.00 0.00

H_0: t-test rel-bias>30% (p-value) | KP 0.14 0.36 0.00 0.00 0.88 0.81 0.00 0.00 0.36 0.01 0.00 0.00

H_0: t-test rel-bias>10% (p-value) | CD 0.59 0.54 0.00 0.00 0.99 0.93 0.00 0.00 0.83 0.04 0.00 0.00

H_0: t-test rel-bias>30% (p-value) | CD 0.11 0.21 0.00 0.00 0.84 0.73 0.00 0.00 0.32 0.00 0.00 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Instruments: Average Instruments: Median Instruments: Weighted Average

Table 5. Robustness: IV Estimation (5-Year Panel)

We also use the GMM methodology using our sharp instrument as an additional instrument in

the regression. By doing so, we are revisiting the main specification of Hausmann, Hwang, and

Rodrik (2007). We present our results in Table 6 (constant and time dummies are not shown).

The standard GMM instruments are two lags of the explanatory variables (levels in the

difference equation, DIF) and one lagged difference of the variables in the level equation (LEV).

We add the neighbor of the same explanatory variables assumed exogenous (that is,

IV instruments) in the GMM setting (implying the difference of the variables in the difference

equation and the levels of variables in the level equation). We also experiment with using sharp

instruments as GMM instruments and excluding standard GMM instruments from the first stage.

Export sophistication has a highly significant and positive coefficient in all specifications.

Moreover, the magnitude of the coefficient remains relatively stable around 0.05-0.1 and is

similar to that found in IV regressions.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent var.: 5-year ave. annual growth rate

Log export sophistication 0.061** 0.067*** 0.078*** 0.053*** 0.098*** 0.079*** 0.055***

(0.024) (0.018) (0.020) (0.012) (0.022) (0.020) (0.017)

Log real GDP per capita -0.010*** -0.024*** -0.026*** -0.027*** -0.027*** -0.021*** -0.042*** -0.020*** -0.029***

(0.002) (0.006) (0.005) (0.004) (0.006) (0.005) (0.008) (0.005) (0.008)

Years of schooling 0.000 0.008*** 0.002 0.001 0.002 0.002 0.004** 0.001 0.001

(0.001) (0.002) (0.002) (0.002) (0.002) (0.001) (0.001) (0.002) (0.002)

Log structural export sophistication, S-EXPY 0.022***

(0.004)

Log export sophistication (Ding and Hadzi-Vaskov) 0.443***

(0.147)

Gross fixed capital formation (% of GDP) 0.000

(0.001)

National saving rate (% of GDP) 0.000

(0.000)

Foreign direct investment (% of GDP) -0.487*

(0.291)

Government consumption (% of GDP) -0.001

(0.001)

Gini coefficient (net) 0.001**

(0.000)

Corruption -0.010***

(0.003)

Ethnic fractionalization 0.000*

(0.000)

Latitude 0.000

(0.000)

Sub-Saharan Africa dummy -0.036***

(0.007)

Observations 1,216 1,200 1,004 1,047 936 1,052 716 675 980

\# of endogenous variables 3 3 4 4 4 4 4 4 5

\# of instruments 13 13 14 14 13 14 14 10 16

\# of excluded instruments 3 3 4 4 4 4 4 4 5

Cragg-Donald F stat 88.0 23.2 3.2 11.0 0.2 22.8 11.3 16.5 15.2

Kleibergen-Paap F stat 88.4 17.2 2.6 5.1 0.8 19.4 6.9 16.5 13.5

Kleibergen-Paap LM test p-value 0.00 0.00 0.00 0.00 0.06 0.00 0.00 0.00 0.00

H_0: t-test size>10% (p-value) | KP 0.00 0.01 1.00 0.98 1.00 0.01 0.92 0.06 0.51

H_0: t-test size>25% (p-value) | KP 0.00 0.00 0.61 0.13 0.97 0.00 0.03 0.00 0.00

H_0: t-test size>10% (p-value) | CD 0.00 0.00 1.00 0.49 1.00 0.00 0.45 0.06 0.31

H_0: t-test size>25% (p-value) | CD 0.00 0.00 0.45 0.00 1.00 0.00 0.00 0.00 0.00

H_0: t-test rel-bias>10% (p-value) | KP 0.00 0.00 0.87 0.39 1.00 0.00 0.15 0.00 0.00

H_0: t-test rel-bias>30% (p-value) | KP 0.00 0.00 0.38 0.04 0.91 0.00 0.01 0.00 0.00

H_0: t-test rel-bias>10% (p-value) | CD 0.00 0.00 0.76 0.01 1.00 0.00 0.01 0.00 0.00

H_0: t-test rel-bias>30% (p-value) | CD 0.00 0.00 0.24 0.00 0.99 0.00 0.00 0.00 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 6. GMM-System Estimation (5-Year Panel)

Comparing the GMM estimation of the difference equation and the level equation, we find

favorable test statistics using the level equation specification. The GMM-LEV estimator that

uses smaller or collapsed number of instruments (Table 7, column 7) than the usual GMM

estimator (Table 7, column 6) does not suffer from the weak instrument problem and produces a

positive and statistically significant coefficient on EXPY, similar to the IV estimates. Using own

explanatory variables and sharp instruments as GMM instruments results in the weak instrument

problem (columns 8-9). However, it seems strong identification comes from using sharp

instruments as IV instruments (column 10). The J-test (column 7) suggests that overidentifying

restrictions are not valid, but using Hahn, Ham, and Moon (2011) test for instrument validity, we

find that the instruments based on the averages are valid (p-value of about 0.7). In the

difference equation specification (columns 1-5), the tests for weak instruments suggest that the

instruments are weak. The coefficient on the initial log real GDP per capita is statistically

significant and has the usual negative sign. However, the coefficient on schooling becomes

negative and is statistically significant in a few estimations (columns 1-3). The identification

seems to come from sharp instruments used as IV instruments (column 5) in which the

coefficient on EXPY is statistically significant at 10 percent. However, the weak instrument

problem is present in this specification as well. These results indicate that it is the cross-country

variation stemming from the level equation estimation rather than the time series variation in the

difference equation estimation that produces parameter identification and favorable test

statistics.12

12 If we assume fixed effects are correlat

not produce robust estimates of the parameters and the differenced variables result in the weak instrument problem.

Table 2, column 9, and Table 4, column 12, show estimations considering fixed effects. The coefficient on export

sophistication is similar to that in other estimations.

(1) (2) (3) (4) (5)

Dependent var.: 5-year ave. annual growth rate

Own variable 2-

4 period lags,

GMM instr

Neighbor

averages, IV

instr

Neighbor

averages and 1-

period lags, IV

instr

Neighbor

averages and 1-

2 period lags,

GMM instr

Neighbor

averages and 2-

4 period lags,

GMM instr

Log export sophistication 0.052*** 0.059*** 0.084*** 0.111*** 0.098**

(0.018) (0.022) (0.026) (0.033) (0.042)

Log real GDP per capita -0.027** -0.022*** -0.029*** -0.030*** -0.017**

(0.011) (0.006) (0.007) (0.011) (0.008)

Years of schooling 0.008*** 0.001 0.000 -0.003 -0.005

(0.003) (0.002) (0.002) (0.004) (0.005)

\# of observations 1226 1216 1119 1226 1226

\# of countries 137 136 136 137 137

\# of instruments 22 13 15 19 22

\# of overidentifying restrictions 9 0 1 6 9

Hansen J-test p-value 0.146 . 0.005 0.025 0.012

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 7. Dissecting GMM (5-Year Panel)

averages are used, and

the 10-year panel regressions, we find that export sophistication is statistically significant in most

estimations (Table 8). The coefficients are similar in magnitude and the tests for weak

instruments suggest that in several estimations, the instruments are as strong as in the 5-year

panel case.

erform the same estimations IV using 5-

year panels with the instrument set that uses the median of and the weighted mean (with

weights equal to the . The results are broadly

unchanged compared to the instrument set based on simple averages (Tables 9-10). This shows

that our results, tests and coefficients, are mostly robust to changing the aggregation method for

the instrument calculation. Using the weights of the inverse of real GDP for our instruments

interest. Even though the weak instrument tests are not favorable in a few estimations, the overall

conclusion remains the same (Table 10). Export sophistication is still a key determinant in

growth regressions.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

DIF DIF-Collapse DIF-Collapse DIF-Collapse DIF-Collapse LEV LEV-Collapse LEV-Collapse LEV-Collapse LEV-Collapse

Dependent var.: 5-year ave. annual growth rate

Own vars,

GMM instr

+ Neighbor

averages,

IV instr

Own vars,

GMM instr +

Neighbor

averages, IV

instr

Own vars,

GMM instr

Neighbor

averages 1-2

period lags,

GMM instr

Neighbor

averages 1-2

period lags, IV

instr

Own vars,

GMM instr

+ Neighbor

averages,

IV instr

Own vars,

GMM instr +

Neighbor

averages, IV

instr

Own vars,

GMM instr

Neighbor

averages,

GMM instr

Neighbor

averages, IV

instr

Log export sophistication 0.036*** 0.038*** 0.068 0.228*** 0.065***

(0.011) (0.011) (0.068) (0.077) (0.013)

Log real GDP per capita -0.018*** -0.018*** -0.126 -0.052*** -0.024***

(0.004) (0.004) (0.121) (0.018) (0.004)

Years of schooling 0.003*** 0.003*** 0.044 -0.018 0.001

(0.001) (0.001) (0.039) (0.014) (0.001)

Differenced log export sophistication 0.021 0.057* 0.039 -0.067 0.173*

(0.019) (0.030) (0.057) (0.103) (0.093)

Differenced log real GDP per capita -0.142*** -0.110*** -0.193*** -0.060* -0.126**

(0.025) (0.033) (0.048) (0.034) (0.052)

Differenced years of schooling -0.037*** -0.056** -0.111*** -0.055 0.060

(0.013) (0.029) (0.034) (0.043) (0.060)

Observations 1,080 1,080 1,089 944 944 1,216 1,216 1,226 1,080 1,216

\# of endogenous variables 3 3 3 3 3 3 3 3 3 3

\# of instruments 75 21 18 14 14 37 16 13 12 13

\# of excluded instruments 66 12 9 6 6 27 6 3 3 3

Cragg-Donald F stat 1.9 1.8 1.5 1.2 0.9 9.7 39.0 0.5 2.3 40.7

Kleibergen-Paap F stat 2.5 1.7 1.6 1.5 0.9 10.4 36.8 0.3 2.5 38.4

Kleibergen-Paap LM test p-value 0.01 0.03 0.06 0.09 0.23 0.00 0.00 0.32 0.01 0.00

C-stat (p-value) 0.00 0.03 . . . 0.00 0.00 . . .

Hansen J-test p-value 0.00 0.03 0.47 0.31 0.52 0.00 0.00 . . .

H_0: t-test size>10% (p-value) | KP 1.00 1.00 1.00 1.00 1.00 1.00 0.00 1.00 0.98 0.00

H_0: t-test size>25% (p-value) | KP 1.00 1.00 1.00 0.99 1.00 1.00 0.00 0.97 0.40 0.00

H_0: t-test size>10% (p-value) | CD 1.00 1.00 1.00 1.00 1.00 1.00 0.00 1.00 0.99 0.00

H_0: t-test size>25% (p-value) | CD 1.00 1.00 1.00 1.00 1.00 1.00 0.00 0.94 0.46 0.00

H_0: t-test rel-bias>10% (p-value) | KP 1.00 1.00 1.00 1.00 1.00 0.10 0.00 0.99 0.68 0.00

H_0: t-test rel-bias>30% (p-value) | KP 0.97 0.93 0.89 0.82 0.96 0.00 0.00 0.95 0.33 0.00

H_0: t-test rel-bias>10% (p-value) | CD 1.00 1.00 1.00 1.00 1.00 0.25 0.00 0.99 0.73 0.00

H_0: t-test rel-bias>30% (p-value) | CD 1.00 0.91 0.93 0.91 0.95 0.00 0.00 0.92 0.39 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 8. IV Estimation (10-Year Panel)

Table 9. IV Estimation, Instrument Set: Median (5-Year Panel)

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent var.: 10-year ave. annual growth rate IV IV IV IV IV IV IV IV IV-FE

Log export sophistication 0.085*** 0.074*** 0.061*** 0.091*** 0.081*** 0.142** 0.225 0.020

(0.013) (0.013) (0.016) (0.024) (0.020) (0.062) (0.180) (0.052)

Log real GDP per capita -0.030*** -0.026*** -0.027*** -0.030*** -0.024*** -0.032*** -0.007 -0.035 -0.062**

(0.005) (0.005) (0.005) (0.009) (0.005) (0.012) (0.005) (0.022) (0.026)

Years of schooling 0.002 0.003 -0.009 -0.000

(0.001) (0.002) (0.014) (0.034)

Trade (% of GDP) -0.042** -0.015 -0.096

(0.021) (0.019) (0.112)

Credit to private sector (% of GDP) -0.033 0.017 0.014

(0.025) (0.018) (0.044)

Law and order -0.021 -0.002 -0.026

(0.013) (0.004) (0.021)

Observations 724 724 596 627 618 365 306 301 363

\# of endogenous variables 1 2 3 3 3 3 5 6 3

\# of instruments 7 7 8 8 8 6 8 9 9

\# of excluded instruments 1 2 3 3 3 3 5 6 6

Cragg-Donald F stat 147.6 53.3 20.8 6.3 19.3 2.5 1.5 0.2 0.4

Kleibergen-Paap F stat 103.7 38.9 17.9 4.8 11.3 2.2 0.9 0.2 0.4

Kleibergen-Paap LM test p-value 0.00 0.00 0.00 0.00 0.00 0.01 0.03 0.28 0.66

H_0: t-test size>10% (p-value) | KP 0.00 0.00 0.00 0.83 0.13 0.99 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | KP 0.00 0.00 0.00 0.08 0.00 0.48 0.99 1.00 1.00

H_0: t-test size>10% (p-value) | CD 0.00 0.00 0.00 0.66 0.00 0.98 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | CD 0.00 0.00 0.00 0.03 0.00 0.40 0.95 1.00 1.00

H_0: t-test rel-bias>10% (p-value) | KP 0.00 0.00 0.00 0.26 0.00 0.75 1.00 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | KP 0.00 0.00 0.00 0.06 0.00 0.41 0.93 1.00 1.00

H_0: t-test rel-bias>10% (p-value) | CD 0.00 0.00 0.00 0.11 0.00 0.68 0.99 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | CD 0.00 0.00 0.00 0.02 0.00 0.33 0.76 1.00 1.00

Hansen J-test p-value . . . . . . . . 0.11

Lower CLR bound 0.06

Upper CLR bound 0.12

H0: Beta_EXPY=0 | CLR p-value 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent var.: 5-year ave. annual growth rate IV IV IV IV IV IV IV IV IV-FE

Log export sophistication 0.076*** 0.078*** 0.076*** 0.082*** 0.082*** 0.122*** 0.116*** 0.108

(0.010) (0.011) (0.013) (0.014) (0.013) (0.032) (0.030) (0.067)

Log real GDP per capita -0.025*** -0.026*** -0.028*** -0.029*** -0.025*** -0.025*** -0.005 -0.025*** -0.109**

(0.004) (0.004) (0.004) (0.005) (0.004) (0.007) (0.005) (0.006) (0.043)

Years of schooling 0.001 0.003 -0.001 -0.007

(0.001) (0.002) (0.003) (0.037)

Trade (% of GDP) -0.003 -0.016 -0.022

(0.011) (0.012) (0.016)

Credit to private sector (% of GDP) -0.021* 0.004 -0.006

(0.013) (0.014) (0.016)

Law and order -0.018** 0.001 -0.009*

(0.007) (0.003) (0.005)

Observations 1,590 1,590 1,216 1,369 1,319 748 606 598 983

\# of endogenous variables 1 2 3 3 3 3 5 6 3

\# of instruments 13 13 13 14 14 9 11 12 14

\# of excluded instruments 1 2 3 3 3 3 5 6 6

Cragg-Donald F stat 313.1 100.7 43.3 17.7 40.1 8.3 4.4 2.3 1.4

Kleibergen-Paap F stat 232.7 71.0 39.3 14.5 23.9 6.9 2.4 2.3 1.3

Kleibergen-Paap LM test p-value 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.09

H_0: t-test size>10% (p-value) | KP 0.00 0.00 0.00 0.03 0.00 0.57 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.01 0.82 0.92 0.99

H_0: t-test size>10% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.39 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.00 0.34 0.93 0.99

H_0: t-test rel-bias>10% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.08 0.96 0.99 1.00

H_0: t-test rel-bias>30% (p-value) | KP 0.00 0.00 0.00 0.00 0.00 0.01 0.48 0.53 0.87

H_0: t-test rel-bias>10% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.03 0.69 0.99 1.00

H_0: t-test rel-bias>30% (p-value) | CD 0.00 0.00 0.00 0.00 0.00 0.00 0.08 0.55 0.85

Hansen J-test p-value . . . . . . . . 0.10

Lower CLR bound 0.06

Upper CLR bound 0.09

H0: Beta_EXPY=0 | CLR p-value 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 10. IV Estimation, Instrument Set: Weighted Mean (5-Year Panel)

VI. CONCLUSION

This paper explores the determinants of growth based on an instrumental variable technique that

factor of growth is instrumented by its average in the neighboring countries. We show that export

sophistication whether proxied by EXPY, the share of manufacturing exports

exports, or real manufacturing exports per capita stands out as an important and robust

determinant of growth. This is further confirmed by verifying the strength of the

Bazzi and Clemens (2013).

Although standard growth determinants are not robust in the regressions, they may be important

to the extent they help improve export sophistication.

The technique we propose could be applied to other empirical studies suffering from the blunt

instrument problem. It offers a variable-specific, dynamic and plausibly valid instrument for as

many variables as needed. The striking result in our study is that overall, the instruments passed

the instrument strength tests. Correlations among neighborin

mimetic forces could be at play, where economic agents learn from across the borders in formal

and informal ways. It suggests that competition with immediate neighbors is a potent factor in

the diffusion of technologies and policies. Perhaps a , far away from advanced

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dependent var.: 5-year ave. annual growth rate IV IV IV IV IV IV IV IV IV-FE

Log export sophistication 0.062*** 0.055*** 0.031** 0.060*** 0.072*** 0.077** 0.269 0.157**

(0.010) (0.011) (0.013) (0.020) (0.016) (0.036) (0.444) (0.072)

Log real GDP per capita -0.021*** -0.018*** -0.015*** -0.015** -0.020*** -0.013** 0.004 -0.009 -0.135**

(0.004) (0.004) (0.004) (0.007) (0.005) (0.006) (0.011) (0.026) (0.059)

Years of schooling 0.003** 0.000 -0.024 0.082

(0.001) (0.004) (0.052) (0.085)

Trade (% of GDP) -0.059 -0.047 -0.215

(0.046) (0.050) (0.443)

Credit to private sector (% of GDP) -0.027** -0.008 0.111

(0.014) (0.025) (0.246)

Law and order -0.013* 0.004 -0.033

(0.008) (0.006) (0.061)

Observations 1,590 1,590 1,216 1,369 1,319 748 606 598 983

\# of endogenous variables 1 2 3 3 3 3 5 6 3

\# of instruments 13 13 13 14 14 9 11 12 14

\# of excluded instruments 1 2 3 3 3 3 5 6 6

Cragg-Donald F stat 260.4 95.7 38.0 1.6 29.6 4.4 0.6 0.0 1.0

Kleibergen-Paap F stat 192.8 72.8 33.9 1.5 20.5 3.3 0.4 0.0 1.1

Kleibergen-Paap LM test p-value 0.00 0.00 0.00 0.03 0.00 0.00 0.18 0.60 0.13

H_0: t-test size>10% (p-value) | KP 0.00 0.00 0.00 1.00 0.00 0.95 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | KP 0.00 0.00 0.00 0.67 0.00 0.25 1.00 1.00 1.00

H_0: t-test size>10% (p-value) | CD 0.00 0.00 0.00 1.00 0.00 0.87 1.00 1.00 1.00

H_0: t-test size>25% (p-value) | CD 0.00 0.00 0.00 0.62 0.00 0.12 1.00 1.00 1.00

H_0: t-test rel-bias>10% (p-value) | KP 0.00 0.00 0.00 0.88 0.00 0.52 1.00 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | KP 0.00 0.00 0.00 0.61 0.00 0.20 0.99 1.00 0.92

H_0: t-test rel-bias>10% (p-value) | CD 0.00 0.00 0.00 0.85 0.00 0.32 1.00 1.00 1.00

H_0: t-test rel-bias>30% (p-value) | CD 0.00 0.00 0.00 0.56 0.00 0.08 0.98 1.00 0.94

Hansen J-test p-value . . . . . . . . 0.02

Lower CLR bound 0.04

Upper CLR bound 0.08

H0: Beta_EXPY=0 | CLR p-value 0.00

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1