IV Estimation of a Panel Threshold Model of Tourism

Specialization and Economic Development*

Chia-Lin Chang
Department of Applied Economics
National Chung Hsing University
Taichung, Taiwan

Thanchanok Khamkaew
Faculty of Economics
Chiang Mai University
Thailand
and
Faculty of Economics
Maejo University
Thailand

Michael McAleer
Econometric Institute
Erasmus School of Economics
Erasmus University Rotterdam
and
Tinbergen Institute
The Netherlands

Revised: April 2010

____________________

* The authors wish to thank Francesso Pigliaru and Shu Lin for helpful comments and suggestions, Ravee
Phoewhawm for technical support, and Bruce Hansen for providing the Matlab codes to implement the threshold
analysis. For financial support, the first author is grateful to the National Science Council, Taiwan, the second
author would like to thank the Office of Higher Education Commission, Ministry of Education, Thailand, for a
CHE-PhD 2550 scholarship, and the third author wishes to acknowledge the Australian Research Council, National
Science Council, Taiwan, and the Center for International Research on the Japanese Economy (CIRJE), Faculty of
Economics, University of Tokyo. An earlier version of the paper was presented at the second conference of the
International Association for Tourism Economics (IATE 2009).

Electronic
Electroniccopy
copyavailable
availableat:
at:[Link]
[Link]
 Abstract

The significant impact of international tourism in stimulating economic growth is especially

important from a policy perspective. For this reason, the relationship between international
tourism and economic growth would seem to be an interesting and topical empirical issue. The
purpose of this paper is to investigate whether tourism specialization is important for economic
development in 159 countries over the period 1989-2008. The results from panel threshold
regressions show a positive relationship between economic growth and tourism. Instrumental
variable estimation of a threshold regression is used to quantify the contributions of tourism
specialization to economic growth, while correcting for endogeneity between the regressors and
error term. The significant impact of tourism specialization on economic growth in most
regressions is robust to different specifications of tourism specialization, as well as to differences
in real GDP measurement. However, the coefficients of the tourism specialization variables in
the two regimes are significantly different, with a higher impact of tourism on economic growth
found in the low regime. These findings do not change with changes in the threshold variables.
The empirical results suggest that tourism growth does not always lead to substantial economic
growth.

Keywords: International tourism, economic development, tourism specialization, threshold

regression, instrumental variables, panel data, cross-sectional data.

JEL Classifications: C33, L83, O10, O40, O57.

Electronic
Electroniccopy
copyavailable
availableat:
at:[Link]
[Link]
1. Introduction

A compelling reason to analyse tourism is its purported positive effect on economic

development in destinations throughout the world. On a global scale, tourism has become one of
the major international trade categories that generate foreign exchange earnings, which leads to a
positive contribution to the national balance of payments and in the travel account. Tourism is
also an effective source of income and employment. Tourism is a major source of income and
employment for many local economies.

The contribution of tourism to world GDP is estimated to be approximately 5%.

Tourism’s contribution to employment tends to be slightly higher, and has been estimated in the
order of 6-7% of the overall number of jobs (direct and indirect) worldwide. For advanced and
diversified economies, the contribution of tourism to GDP ranges from approximately 2% for
countries where tourism is a comparatively small sector, to over 10% for countries where
tourism is an important pillar of the economy. For small islands and developing countries, or
specific regional and local destinations where tourism is a key economic sector, the importance
of tourism tends to be even higher (UNWTO (2009)).

According to the World Travel and Tourism Council (WTTC), in many developing
regions the travel and tourism sectors have contributed a relatively larger total share to GDP and
employment than the world average. The travel and tourism economy GDP, the share to total
GDP, the travel and tourism economy employment for all regions in 2009, as well as future
tourism in real growth that has been forecast by the WTTC for the next ten years, are presented
in Table 1.

[Insert Table 1 here]

The success of economic development attributed to the tourism sector depends on

different aspects. More precisely, the extent of a country’s specialization in tourism may have a
different effect on economic growth. In this respect, we intend to examine empirically whether

Electronic copy available at: [Link]

tourism’s contribution to economic growth can be characterized by three different
macroeconomic threshold variables.

The relationship between tourism and development, and implications for an

understanding of the potential contribution to the development of destination areas, are
conceptualized in the model of Sharpley and Telfer (2004). The model demonstrates not only the
interdependence between tourism and the broad socio-culture, but also the political and
economic context within which it operates. The relationship between the potential
developmental role of tourism and the consequences of development are recognized as a
dynamic tourism-development system in which a multi-directional relationship exists. Therefore,
an essential issue is the potential endogeneity associated with the purported contribution of
tourism to development. In this scenario, it is important to clarify the relationship between
tourism specialization, economic development, and the correction for statistical bias that arises
from the endogeneity problem in economic growth models. Therefore, the instrumental variable
estimation method is used to accommodate this potentially serious problem.
The main contributions of this paper are as followed. First, no previous studies have
rigorously evaluated whether the relationship between economic growth and tourism is different
in each sample grouped on the basis of three macroeconomic variables, namely the degree of
trade openness, investment share to GDP, and government consumption as a percentage of GDP.
Second, we examine the nonlinear relationship between economic growth and tourism
specialization through two powerful methods, namely the panel threshold model of Hansen
(2000) and instrumental variable (IV) estimation of a threshold model of Caner and Hansen
(2004). These two models are used to deal with the potential endogeneity of the level of tourism
specialization in empirical growth regressions.

The remainder of the paper is organized as follows: Section 2 presents a literature

review, Section 3 describes the growth model, Section 4 describes the data, Section 5 presents
the empirical specification and methodology, Section 6 reports the empirical results from the
panel threshold and IV threshold models, Section 7 gives some concluding remarks.

2. Literature Review

Electronic copy available at: [Link]

 In the economic growth literature, tourism’s contribution to economic development has
been well documented, and is important from a policy perspective. There are two main steams of
thought stemming from the Export-Led Growth (ELG) hypothesis. The strong association
between tourism and economic growth is often attributed to two main economic channels.
Nowak et al. (2007) explained the so called “two-gap” hypothesis, whereby tourism export
promotion permits accumulation of foreign exchange that can be used to import essential inputs
and capital goods not produced domestically. This can, in turn, be used to expand the host
nation’s production possibilities, which is generally known as Tourism Capital Imports to
Growth (TKIG). The importance of the two-link chain between tourism and growth through
imports of capital goods has not been well explored in previous empirical papers, with an
exception of a case in Spain.

Second, the influence of tourism activities can generate additional demand of goods and
services, incomes and new employment opportunities. The direct effect of increasing
international tourism promotes economic growth as a non-traditional export, which is known as
the Tourism-Led-Growth (TLG) hypothesis. Balaguer and Cantavella-Jordá (2002) were the first
authors to mention this concept. International tourism can be considered as either a non-
traditional export which implies a source of receipts, or as a potential strategic factor to
development and economic growth. The empirical literature on a reciprocal causal relationship
between tourism and economic development may be considered in several classifications,
depending on the techniques applied. Most historical studies have been based on various
econometric techniques, such as causality testing, application of the cointegration and error
correction models, and relying mainly on regional analysis. Various results might be obtained
according to the method used, period analysed, and the variables selected.

Empirical work which demonstrates that tourism is considered as a main factor in

economic growth are Balaguer and Cantavella-Jordá (2002, 2004) for Spain, Dritsakis (2004) for
Greece, Durbarry (2004) for Mauritius Gunduz and Hatemi (2005) for Turkey, Oh (2005) for
Korea, Kim et al. (2006) for Taiwan, Louca (2006) for Cyprus, Brida et al. (2008) for Mexico,
Noriko and Motosugu (2007) for the Amami Islands in Japan, Gani (1998) for some South
Pacific islands, Cortés-Jiménez (2009) for Spanish and Italian regions, and Cortes-Jimenez and
Pulina (2010) for Spain and Italy. Among these empirical studies, it is worth mentioning that
Durbarry (2004) is clearly innovative when considering tourism as one type of export. This

Electronic copy available at: [Link]

paper, which was inspired by the Export-Led Growth (ELG) hypothesis, attempted to verify both
the ELG and TLG hypotheses for Mauritius. The relationship between disaggregated exports,
including international tourism, and economic growth is investigated through a production
function, where economic growth is explained by physical and human capital, and is compatible
with the new growth theory.

Several recent studies have delved deeper into cross-sectional analysis. Eugenio-Martı´n
et al. (2004) investigated the impact of the tourism industry on economic growth and
development in seventeen Latin American countries within the framework of the conventional
neoclassical growth model, from 1995 to 2004. The empirical results show that revenues from
the tourism industry made a positive contribution to the current level of GDP and economic
growth of LACs. Sequeria and Campos (2007) used tourism receipts as a percentage of exports
and as a percentage of GDP as proxy variables for tourism. A sample of 509 observations from
1980 to 1999 was divided into several smaller subsets of data. Their results from pooled OLS,
random effects and fixed effects models showed that growth in tourism was associated with
economic growth only in African countries. A negative relationship was found between tourism
and economic growth in Latin American countries, and in the countries with specialization in
tourism. However, they did not find any evidence of a significant relationship between tourism
and economic growth in the remainder of the groups.

Lee and Chang (2008) applied the heterogeneous panel cointegration technique to
investigate the long-run comovements and causal relationships between tourism development
and economic growth for OECD and non-OECD countries for the 1990-2002 period. A
cointegrated relationship between GDP and tourism development was substantiated.
Furthermore, the panel causality test provided an unidirectional causality relationship from
tourism development to economic growth in OECD countries, and bidirectional relationships in
non-OECD countries.

Regarding previous research on the importance of tourism as a significant growth-

enhancing factor, there have been few studies which highlighted a possibility that the difference
in the comparative advantage in a less productive sector, such as tourism, might lead the country
to grow at a different rate. Lanza and Pigliaru (2000) used an analytical framework based on
Lucas’s two-sector endogenous growth model, in which the growth-effect of different

Electronic copy available at: [Link]

specialization can be easily compared. Based on their work, the model pointed to an important
reason as to why tourism specialization is not harmful for growth. They noticed that countries
with relatively high tourism specialization are likely to grow fast, and are generally small.
Moreover, their analysis suggested that what matters for explaining specialization in tourism is a
country’s relative endowment of the natural resources, rather than its absolute size. Therefore,
countries with relative abundance of a natural resource will be more specialized in tourism, and
are likely to grow faster.

Brau, Lanza and Pigliaru (2007) investigated the relative economic performance of
countries that have specialized in tourism, from 1980 to 2003. Tourism specialization and small
countries are defined simply as the ratio of international tourism receipts to GDP and to
countries with an average population of less than one million, during 1980-2003. They found
that tourism could be a growth-enhancing factor for small countries, which are likely to grow
faster only when they are highly specialized in tourism. Although the paper considered the
heterogeneity among countries in terms of the degree of tourism specialization and country size,
the threshold variables were not based on any selection criteria. It would be preferable to use
selection criteria to separate the whole sample into different subsets in which tourism may
significantly affect economic growth.

Algieri (2006) analyzed the linkages between economic growth and tourism-based
economies. The results showed that tourism can be a significant engine of economic growth
when the elasticity of substitution between manufacturing goods and tourism services is less
than 1. There are two stylized facts: (1) countries specialized in tourism register good economic
performances; (2) these same countries have small dimensions, as defined by international trade
theory. Po and Huang (2008) use cross-section data (1995-2005 yearly averages) for 88
countries to investigate the nonlinear relationship between tourism development and economic
growth when the degree of tourism specialization (defined as receipts from international tourism
as a percentage of GDP) is used as the threshold variable. The results of the nonlinear threshold
model indicated that the data for 88 countries should be divided into three regimes to analyze the
tourism-growth nexus. The results of the threshold regression showed that a significantly
positive relationship between tourism and economic growth is found only in the low and high
regimes. However, the potential endogeneity is not taken into account in their economic growth
regression.

Electronic copy available at: [Link]

 Arezki et al. (2009) quantified the relationship between tourism specialization and
growth while correcting for endogeneity by using the instrumental variables technique (IV) for a
cross-section of up to 127 countries, over the period 1980 to 2002. The instrument for tourism is
the number of UNESCO sites per 100,000 inhabitants in 2002. They showed that the gains from
tourism specialization can be significant, and that the result holds against a large array of
robustness checks. Adamou and Clerides (2009) investigated the relationship between tourism
and specialization, and economic growth. It was found that tourism specialization is associated
with higher rates of economic growth at relatively low levels of specialization. The independent
tourism’s contribution will become minimal at high levels of specialization and tourism can even
become a hindrance to further growth. Finally, Figini and Vici (2010) provided an empirical
assessment of the relationship between tourism specialization and economic growth. They found
that tourism-based countries did not grow at a higher rate than non-tourism based countries,
except for the 1980-1990 periods.

Thus, the influence of tourism specialization on economic growth has received much
attention in recent studies. To the best of our knowledge, there has not been any analysis that
identifies the existence of threshold effects of tourism specialization on economic growth, with
correction for potential endogeneity. Unlike previous studies, this paper uses endogenous
threshold regression analysis rather than arbitrarily assuming a cut-off point. Furthermore,
special attention is given to identify the relationship between tourism specialization with
different possible threshold variables which are commonly used in the macroeconomics
literature.

3. The Growth Model

This paper assesses the determinants of growth, where the focus is on the role of tourism
specialization based upon the Cobb-Douglas production function within the neoclassical
framework. The augmented version of the Solow-Swan neoclassical growth model, developed
by Mankiw, Romer and Weil (1992), hereafter MRW, is of interest. Adopting the MRW
neoclassical approach has one advantage in which a simple theoretical framework for empirical
growth regression is explicitly derived. Hence, following the MRW framework is a useful
foundation for empirical work on economic growth.

Electronic copy available at: [Link]

 Although the Solow model, in which the rates of saving and population growth are taken
as exogenous, accurately predicts the direction of the effects of saving and population growth,
the magnitude of such effects is too large. MRW extended the Solow model by considering a
broader measure of the capital stock that includes both human and physical capital, in which
both are augmented by investment of a fraction of GDP, while maintaining the assumptions of
exogenous technological progress and diminishing returns to all capital. The exclusion of human
capital from the Solow model can potentially explain why the estimated influences of saving and
population growth appear too large. MRW gave two reasons regarding this point. They found
that accumulation of human capital is, in fact, correlated with saving and population growth.
Including human capital in an aggregate production function as a separate factor of production
lowers the estimated effects of saving and population growth roughly to the value predicted by
the augmented Solow model. This slows the rate of convergence to the steady state, thereby
allowing the transitional dynamics to be more important in explaining differences in growth.
However, the MRW model still suggests that when economies have reached their steady states,
they will experience the same growth rates in output per worker; which is equal to the common
exogenously determined rate of technological progress.
Including human capital can potentially alter not only the theoretical modelling, but also
the empirical analysis of economic growth. At the theoretical level, properly accounting for
human capital may change the nature of the growth process. At the empirical level, the existence
of human capital can alter the analysis of cross-country differences. Thus, the empirical results
are likely to be biased from the omitted variables problem.

MRW start from a Cobb-Douglas production function with constant returns to scale:

(1)

where Y is output, K is physical capital, H is human capital, L is labour supply, and A is the
level of technology. MRW assume that investment rates in physical and human capital are
constant at and , respectively, and that both types of capital depreciate at a common rate .
Technology grows at the same exogenous rate, g, across countries, while the labour force grows
at differing rates n. The initial level of efficiency, A(0), is assumed to vary randomly across

Electronic copy available at: [Link]

countries, and this can be used to justify the error term. In addition, is assumed to
represent decreasing returns to all capital.
The dynamic equations for k and h are given by

(2)

(3)

where , , and are the levels of output per effective unit of labour, the

stock of physical capital per effective unit of labour, and the stock of human capital per effective
unit of labour, respectively. Equations (2) and (3) imply that k and h converge to their steady
state values, and , defined by

(4)

Substituting (4) into the production function and taking logarithms gives the following
expression for steady state income per capita:

(5)

This equation shows how income per capita depends on population growth and the
accumulation of physical and human capital. In the empirical growth literature, the physical
capital saving rate was approximated by the investment share in GDP, while human capital is
essentially a linear function of the rate of secondary school enrolments. Nonetheless, there is an
alternative way to express the role of human capital in determining income in this model.
Combining (5) with the equation for the steady-state level of human capital given in (4), yields

10

Electronic copy available at: [Link]

an equation for income as a function of the rate of investment in physical capital, the rate of
population growth, and the level of human capital:

(6)

Equations (5) and (6) are virtually identical, except that the level of human capital is a
component of the error term in (5). As the saving and population growth rates influence ,
human capital should be positively correlated with the saving rate, and negatively correlated
with population growth. The model with human capital provides two possible ways to estimate
the steady-state of income per capita. One can choose either (5) or (6), depending on whether the
available data on human capital correspond more closely to the rate of accumulation or to
the level of human capital (h).

After developing and testing the augmented Solow model, MRW examined the dynamics
of the economy when it is not in a steady state. Let be the steady state level of income per
effective worker given by equation (5), and let be the actual value at time t. Approximating
around the steady state, the speed of convergence is given by

where
(7)

The model suggests a logarithmic regression to examine the rate of convergence.

Equation (7) implies that
(8)
where is income per effective worker at some initial point of time and .

Subtracting from both sides so as to obtain a partial adjustment process,

11

Electronic copy available at: [Link]

 (9)

Equation (11) can be rearranged as follows:

(10)

Let , and substitute intio equation (5):

(11)

It is obvious that, in the augmented Solow model or MRW model, the growth of income
is a function of the determinants of the ultimate steady state and the initial level of income. The
negative coefficient of the initial income implies the convergence process. In contrast to
endogenous growth models, the MRW model predicts that countries with similar technologies
and rate of accumulation and population growth should converge in income per capita. Yet this
convergence occurs more slowly than the Solow model suggests.

We can express equation (11) in the form of a panel specification, as can be

treated as a time-invariant individual country-effect term and gt as the time specific effect. Islam
(1995) noted that equation (11) was based on an approximation around the steady state, and was
supposed to capture the dynamics toward the steady state. If the character of approaching the
steady state of the convergence process remains unchanged over the period as a whole, then
considering that process in consecutive shorter time interval should also reflect the same
dynamic process.

As noted in Temple (1999), in the absence of a suitable proxy for technical efficiency, A,
the only way to obtain consistent estimates of a conditional convergence regression is to use
panel data methods, as it fundamentally allows one to control for the effects of omitted variables
that persist over time. By moving to a panel data framework, at least unobserved heterogeneity
in the initial level of efficiency can be controlled. Moreover, several lags of the regressors can be

12

Electronic copy available at: [Link]

used as instruments, where required, which can alleviate measurement error and endogeneity
biases. The panel specification of growth model is generally expressed as follows:

(12)

where is the average growth rate of income per effective worker over a shorter time interval,
which is normally a 5- or 10-year average, is an initial level of income per effective

worker (5-year average of income per effective worker, from the previous period), is a
vector of control variables, is a country specific effect, is time specific effect, is
transitory error term that varies across countries and time (a serially uncorrelated measurement
error), sub-index i denotes different countries, and the sub-index t refers to different time
periods.

4. Data

The countries in the sample were selected based on data availability. Tourism data cause
the main constraint in our analysis. Subject to such criteria, 159 countries are used in the sample,
as given in Table 2. Annual data from 1989 to 2008 for 159 countries and 20 annual
observations were organized in a five-year averaged panel data format in order to smooth out
business cycle fluctuations and the effects of particular events. The empirical literature on
economic growth usually emphasizes the reduction in measurement errors, as well as avoiding
problems associated with missing observations in a specific year for a country in the sample. We
have four periods, namely 1989-1993, 1994-1998, 1999-2003, and 2004-2008, in which the
procedure of directly averaging the values of the variables has been taken. In addition to a broad
panel of 159 countries, we also have a pure cross-section averaged over the same period in order
to identify the threshold effects in the tourism and growth relationship through a cross-sectional
instrument variable (IV) threshold approach.

[Insert Table 2 here]

13

Electronic copy available at: [Link]

 Economic growth is specified using the growth rates of three different GDP
measurements, namely real GDP chain per worker (rgdpwok), real GDP chain per capita
(rgdpch), and real GDP per capita (Constant Prices: Chain series), and real GDP Laspeyres per
capita (rgdpl) or real GDP per capita (Constant Prices: Laspeyres), derived from the growth rates
of c, g and i. These variables are obtained from the Penn World Tables version 6.3, which is
available online at the Center for International Comparisons of Production, Income and Prices,
University of Pennsylvania (see Heston et al. (2009)). Initial income is defined as the 5-year
average of real GDP per capita in the previous period in the case of panel threshold analysis, and
as the real GDP per capita in the initial year (1989) in the case of cross-sectional instrumental
variable threshold analysis. This variable is used to capture the convergence process in the
economic growth model.

The physical investment variable comes from the investment share of real GDP per
capita (ki); population (POP), and openness in current prices (OPENK), which is total trade (the
value of exports plus imports) as a percentage of GDP, and is used as a proxy for the trade
openness variable. These are also obtained from the Penn World Tables version 6.3.

Public expenditure in education is used as a proxy for human capital, government

consumption as a percentage of GDP, surface area (sq. km), and three tourism specialization
variables, tourist arrivals, and tourism receipts as a share of exports of goods and services,
tourism receipts as a share of exports of GDP, as an indication of the degree of tourism
specialization, are obtained from the World Bank’s World Development Indicators (WDI)
database.

For the institutional variables, we obtained the “Worldwide Governance Indicators

(WGI) project” for 1996-2008 from the World Bank. It consists of six different indicators of
institutional quality referring to six dimensions of governance, namely voice and accountability,
political stability and absence of violence, government effectiveness, regulatory quality, rule of
law, and control of corruption. These indicators are available biannually since 1996, and
annually since 2002. In our analysis, the first available data (that is, 1996) are used for the values
in the initial 5-year averaged period (1989-1993).

The descriptions for all six institutional variables are as follows (Kaufmann, et al. (2009);

14

Electronic copy available at: [Link]

 (1) Voice and accountability: captures perceptions of the extent to which a country’s citizens
are able to participate in selecting their government, as well as freedom of expression,
freedom of association, and a free media.

(2) Political stability and absence of violence: captures perceptions of the likelihood that the
government will be destabilized or overthrown by unconstitutional or violent means,
including politically-motivated violence and terrorism.

(3) Government effectiveness: captures perceptions of the quality of public services, the
quality of the civil service and the degree of interdependence from political pressures, the
quality of policy formation and implementation, and the credibility of the government’s
commitment to such policies.

(4) Regulatory quality: captures perceptions of the ability of the government to formulate
and implement sound policies and regulations that permit and promote private sector
development.

(5) Rule of law: captures perceptions of the extent to which agents have confidence in and
abide by the rules of society and, in particular, the quality of contract enforcement,
property rights, the police and the courts, as well as the likelihood of crime and violence.

(6) Control of corruption: captures perceptions of the extent to which public power is
exercised for private gain, including both petty and grand forms of corruption, as well as
the impact on the state by the elite and private interests.

The UNESCO World Heritage List (WHL) per country is obtained from a website of
UNESCO ([Link] The World Heritage List includes 890 properties
forming part of the cultural and natural heritage, which the World Heritage Committee considers
as having outstanding universal value. This includes 689 cultural, 176 natural and 25 mixed
properties in 148 States Parties. As of April 2009, 186 States Parties had ratified the World
Heritage Convention. The details of the variables and data sources are provided in Table 3.

[Insert Table 3 here]

15

Electronic copy available at: [Link]

5. Empirical Specification

5.1 Panel Threshold Model

The main purpose of this paper is to use a threshold variable to investigate whether the
relationship between tourism specialization and economic growth is different in each sample
grouped on the basis of certain thresholds. This is to determine if the existence of threshold
effects between two variables is different from the traditional approach, in which the threshold
level is determined exogenously. If the threshold level is chosen arbitrarily, or is not determined
within an empirical model, it is not possible to derive confidence intervals for the chosen
threshold. The robustness of the results from the conventional approach is likely to be sensitive
to the level of the threshold. The econometric estimator generated on the basis of exogenous
sample splitting may also pose serious inferential problems (for further details, see Hansen
(1999, 2000)).

The critical advantages of the endogenous threshold regression technique over the
traditional approach are as follows: (1) it does not require any specified functional form of non-
linearity, and the number and location of thresholds are endogenously determined by the data;
and (2) asymptotic theory applies, which can be used to construct appropriate confidence
intervals. A bootstrap method to assess the statistical significance of the threshold effect is also
available in order to test the null hypothesis of a linear formulation against a threshold
alternative.

For the reasons given above, we used the panel threshold regression method developed
by Hansen (1999) to search for multiple regimes, and to test the threshold effect in the tourism
and economic growth relationship within a 5-year panel data set. The possibility of endogenous
sample separation, rather than imposing a priori an arbitrary classification scheme and the
estimation of a threshold level, are allowed in the model. If a relationship exists between these
two variables, the threshold model can identify the threshold level and permit testing of such a
relationship over different regimes categorized by the threshold variable.

Although the Hansen (2000) approach is commonly used in cross-sectional analysis, it

can also be extended to a fixed effect panel, provided that no endogenous problem exists.

16

Electronic copy available at: [Link]

Specifically, the method requires that all explanatory variables are exogenous. In some
circumstances, especially in empirical growth models, the key variables for economic growth are
likely to be endogenous. In an economic model, a variable is endogenous when there is a
correlation between the variable and the error term. Endogeneity can arise as a result of
measurement error, autoregression with auto correlated errors, simultaneity, omitted variables,
and sample selection errors. The problem of endogeneity occurs when one or more regressors
are correlated with the error term in a regression model, which implies that the regression
coefficient in an OLS regression is biased. Thus, the Hansen (2000) approach will no longer be
applicable. In order to overcome the endogeneity problem, instrumental variable estimation of
the cross-sectional threshold model introduced by Caner and Hansen (2004) is also used.

Hansen (1999) developed econometric techniques appropriate for threshold regression

with a panel data. Allowing for fixed individual effects, the panel threshold model divides the
observations into two or more regimes, depending on whether each observation is above or
below a threshold level. The observed data are from a balanced panel
( . The subscript i indexes the individual and t indexes time. The

dependent variable, , is scalar, the threshold variable is scalar, and the regressor is a k
vector. The structural equation of interest is

(13)

where I() is an indicator function.

The observations are divided into two regimes, depending on whether the threshold
variable, , is smaller or larger than the threshold, . The regimes are distinguished by
different regression slopes, and . For the identification of and , it is necessary that
the elements of are not time-invariant. The threshold variable, , is not time invariant.
is the fixed individual effect, and the error is assumed to be independently and identically
distributed (iid), with mean zero and finite variance .

The threshold value ( ) is estimated using the least squares method developed by
Hansen (2000). A bootstrap procedure is used to obtain approximate critical values of the test

17

Electronic copy available at: [Link]

statistics which allows one to perform the hypothesis test for the threshold effect. If the bootstrap
estimate of the asymptotic p-value is smaller than the desire critical value, then the null
hypothesis of no threshold effect is rejected. After a threshold value is found, the confidence
intervals for the threshold value and slope coefficients are then estimated. A similar procedure
can also be conducted to deal with the case of multiple thresholds. The possibility of existence of
more than one threshold represents another advantage of this method over the traditional
approach.

Our focus is to assess the role of tourism specialization on economic growth. The
economic growth regression based on the neoclassical growth model described in the previous
session is augmented with the tourism variables in order to investigate empirically the
relationship between tourism specialization and economic growth varies across subsamples
grouped on the basis of various threshold variables. The empirical specification of the economic
growth regression, with tourism specialization within the panel threshold model framework, is
represented as follows:

(14)

where

) is the indicator function;

is the growth rate of real GDP chain per worker (rgdpwok). We also use different
definitions for income, namely real GDP chain per capita (rgdpch) and real GDP Laspeyres per
capita (rgdpl) in order to check whether the result is robust to the different specifications of the
real GDP growth rate;

is the tourism specialization variable that is widely used as a proxy for the
influence of international tourism in most empirical tourism studies. There are several
alternatives to measure the volume of tourism discussed by Gunduz and Hatemi (2005). One is
tourism receipts, which is the volume of earnings generated by foreign visitors, a second is the
number of nights spent by visitors from abroad, and a third is the number of tourist arrivals.
Depending on the availability of data for most countries in our sample, the second cannot be

18

Electronic copy available at: [Link]

considered. As a result, three measures of tourism are used to check whether the impact on
economic growth is sensitive to different specifications of tourism measurement.

The selected tourism variables are as follows (Sequeira and Campos, 2007):

(1) tourist arrivals as population proportion (TA);

(2) tourism receipts as a share of exports of goods and services (TRE);
(3) tourism receipts as a share of real GDP (TRG).

is the threshold variable used to examine whether tourism plays a different role in the
growth process due to the differing regimes endogenously categorized by three criteria, namely
degree of trade openness ( ), investment share to GDP ( ), and the government
consumption expenditure as a percent of GDP ( ). These threshold variables are highly
related to international tourism policies. Specifically, the degree of trade openness could be used
to capture the relevance of a country to international trade. Clearly, international tourism and
international trade are two major sources of foreign currency for small, as well as larger
economies. We use trade openness as the criteria to verify whether the impact of tourism
specialization on economic growth differs across regimes. The investment share to GDP is also
used as a threshold variable as investment is an important factor to support tourism expansion.
The extent of government consumption involvement in the economy represents government-
induced distortions. In our analysis, we consider whether the impact of tourism specialization at
different levels of government-induced distortions different across countries.

represents the vector of other explanatory variables and control variables which are:

is the 5-year average of real GDP chain per worker for panel threshold analysis
(and real GDP chain per capita and real GDP Laspeyres per capita, depending on which
specification is used as the dependent variable) from the previous period, which is used to
capture the convergence process. It is also defined as the real GDP chain per worker (or real
GDP chain per capita and real GDP Laspeyres per capita) in the initial year (1989) for
instrumental variable threshold analysis (a negative sign is expected);

19

Electronic copy available at: [Link]

 is the investment share of real GDP per capita, which is used as a proxy for physical
capital investment (a positive sign is expected);

is the stock of human capital (currently, a common proxy is the average years of
schooling in the population, but there might be a problem with this proxy due to excluding the
quality of education: omitting the quality may decrease human capital accumulation, and bias the
results, so we use an alternative proxy for human capital, which is public spending on education
as a percentage of GDP, and can be used to capture the quality of education as well as human
capital investment);

is the population growth rate (a negative sign is expected);

is trade openness in constant prices, which is used to measure the impact of

openness of the economy in its growth performance, and is consistent with the current emphasis
on the export-led growth hypothesis (a positive sign is expected);

is the ratio of government consumption to GDP, which measures the extent of

government involvement in the economy, and can also capture the effects of distortions induced
by government);

The six institutional variables used in the model are as follows:

is an indicator of voice and accountability;

is an indicator of political stability and absence of violence;

is an indicator of government effectiveness;

is an indicator of regulatory quality;

is an indicator of the rule of law;

is an indicator of the control of corruption.

20

Electronic copy available at: [Link]

 The inclusion of institutional variables in empirical growth studies has recently been
taken into consideration because the quality of institutions is regarded as a pre-condition to
exploit natural and/or historical endowments which tourism development relies on (Brau et al.,
2009); moreover, the inclusion of such an important explanatory variable identifies a further
possible channel whereby tourism could affect economic growth through institutions (a positive
impact is expected);

is the individual (country) effect, is a time effect, and is independently and

identically distributed across countries and years.

5.2 Instrumental Variables (IV) Threshold Model

Next, we briefly introduce the Instrumental Variable (IV) threshold model developed by
Caner and Hansen (2004). This approach is carried out with the pure cross-sectional data
averaged over 1989-2008, such that there is one observation per country.

The observed sample is , where is real valued, is a m-vector, and is a

k-vector, with . The threshold variable, , is an element or a function of the
vector , and must have a continuous distribution. The data are either a random sample or a
weakly dependent time series, so that unit roots and stochastic trends are excluded.
The structural equation of interest is

which may also be written in the form

(15)

21

Electronic copy available at: [Link]

 The threshold parameter is , where is a strict subset of the support of . This
parameter is assumed to be unknown and is to be estimated.

The reduced form is a model of the conditional expectation of , given :

where , and the parameter is unknown.

The reduced form threshold parameter, , may equal the threshold, , in the structural equation,
but this is not necessary, and this restriction will not be used in estimation. Caner and Hansen
(2004) estimate the parameter sequentially. First, they estimate the reduced form parameter by
OLS. Second, they estimate the threshold, , using predicted values of the endogenous variable,
. Third, the slope parameters, and , are estimated by 2SLS or GMM on the split samples
implied by the estimate of .

It is widely perceived that the effect of tourism specialization on economic growth gives rise to
the possibility of both endogeneity and thereby a reverse relationship. Unobservable variables
such as managerial skills that are crucial inputs in tourism activities, could directly explain both
high economic growth and a high level of tourism. Moreover, security and health issues, such as
political stability, criminality and malaria, are detrimental to both tourism and growth (Arezki et
al., 2009). We then apply instrumental variable estimation of a threshold model proposed by
Caner and Hansen (2004) to avoid the endogeneity problem and to investigate the threshold
effect of tourism specialization on economic growth. The IV threshold regression takes the form:

(16)

(17)

where ) is the indicator function, is the vector of keys variables which are , ,
, , , , , , , , , and is the threshold variable, which
is also contained in , namely degree of trade openness ( ), investment share to GDP

22

Electronic copy available at: [Link]

( ) and the level of government consumption , is the number of the UNESCO
World Heritage List per surface area, which is an instrumental variable , is the threshold
value, and , and , are two sets of slope parameters corresponding to the low and high
regimes, respectively. Equation (17) is estimated using OLS by substituting the fitted values of
the endogenous variable, , into (16). Then the threshold parameter, , is estimated using
OLS. Finally, the slope coefficients are estimated using GMM on the split samples.

6. Empirical Results

The main objective is to investigate the threshold effect of tourism specialization on

economic growth by applying endogenous threshold regression techniques rather than arbitrarily
assuming cut-off points through a theoretical specification within the panel and cross-sectional
growth regression frameworks. In both frameworks, we select three key variables as threshold
variables for tourism specialization and growth relationship. Specifically, the selected threshold
variables are the degree of trade openness, investment share to GDP, and the government
consumption expenditure as a percentage of GDP.

The robustness of the tourism specialization and growth relationships is checked by using
different definitions of tourism specialization and the growth rate of real GDP per capita. Three
tourism specialization definitions are used to quantify the impact of international tourism
specialization on economic growth, namely tourist arrivals as a proportion of the population
(TA), tourism receipts as a share of exports of goods and services (TRE), and tourism receipts as
a share of real GDP (TRG). We also use various measurements of real GDP per capita, namely
growth rate of real GDP chain per capita (rgdpch), growth rate of real GDP chain per worker
(rgdpwok), and growth rate of real GDP (Laspeyres) per capita (rgdpl), which are obtained from
the Penn World Table 6.3 (PWT).

6.1 Results from panel threshold regression

The descriptive statistics for the variables used in the 5-year panel threshold model are
reported in Table 4. We first conduct the panel threshold analysis, in which the slope estimates
of the tourism specialization variables switch between regimes over different thresholds. The

23

Electronic copy available at: [Link]

other variables are omitted as their coefficients do not change significantly from the linear
specification model. Any results discussed in this section but not presented are available from
the authors upon request.

[Insert Table 4 here]

Before estimating the threshold regression model, we test for the existence of a threshold
effect between economic growth and tourism. This paper uses the bootstrap method to
approximate the F statistic, and then calculates the bootstrap p-value. The results are estimated
over three economic growth specifications, with three different tourism specialization measures
over three possible thresholds. The test statistic for a single threshold is significant for all
models, while the test statistics for double and triple thresholds are insignificant. Thus, we may
conclude that there is strong evidence that there is a single threshold in the relationship between
economic growth and tourism within the 5-year panel data context. Given a single threshold
effect between economic growth and tourism, the whole sample is split into two regimes, where
three variables, namely degree of trade openness, investment share to GDP and government
consumption as a percentage of GDP, are used as the threshold variables. When a threshold is
found, a simple regression can be used to yield consistent estimates. The threshold regression
estimates for the economic growth-tourism model, using the Hansen (2000) method, are reported
in Tables 6, 7 and 8.

These three tables report the results from the threshold regression, where trade openness,
investment share to GDP and government consumption are used as the threshold variables. The
first conclusion to be drawn from Tables 6,7 and 8 is that the initial value of real GDP per capita
is negative and statistically significant for all growth regressions for each possible threshold. The
coefficients are not especially different across these three models. Moreover, the coefficients of
initial income in models 1, 2 and 3 do not differ across three possible thresholds. The negative
and significant effect of initial income is evidence of conditional convergence in the growth
process, which predicts higher growth in response to a lower starting real GDP per capita, and
the important influence on the growth rate (see Barro and Sala-i-Martin (2003)).

Second, the estimated coefficient of trade openness is found to be positive and significant
for all regressions across the three threshold variables, which supports the positive influence of

24

Electronic copy available at: [Link]

trade on economic growth. Third, the investment share to GDP, which is a proxy for physical
investment, has a positive effect on economic growth. Fourth, the population growth rate and
government consumption are found to have negative effects on economic growth, but are not
statistically significant in all regressions. Fifth, the coefficients of public investment in education
are statistically significant and positive for most regressions over the three threshold variables.
This finding is consistent with the literature on human capital and growth, which suggests that
the accumulation of human capital can promote economic growth. Finally, the role of
institutions, by introducing various measures of institutional quality, is found to be positive and
significant for some regressions across the three thresholds. This confirms that a strong social
infrastructure is influential in explaining the economic development process.

[Insert Tables 6, 7 and 8 here]

Focusing on the coefficients of tourism specialization, namely TA, TRE and TRG, the
results for the three economic growth models indicate that there is a positive (and typically
significant) relationship between tourism specialization and economic growth across three
different economic growth specifications. The coefficients associated with tourism specialization
are robust to different tourism specialization definitions in each growth model, and this is
consistent with the different threshold variables. Tourism specialization alone plays an
ambiguous role in contributing to economic growth for two regimes. The higher impact of
tourism on economic growth is sometimes found in regimes 1 and 2.

6.2 Results from IV threshold regression

In order to examine the contribution of tourism specialization to economic growth with

different thresholds and regimes, the potential endogeneity of the level of tourism specialization
in the growth regression needs to be taken into account. Ignoring this issue can lead to biased
estimates of the coefficient associated with tourism in the growth regression, in which several
explanatory variables are likely to be endogenous. Therefore, we apply instrumental variable
estimation of an endogenous threshold model, as recently developed by Caner and Hansen
(2004), to the pure cross-sectional data averaged over 1989-2008. The possible threshold effect
of tourism specialization on economic growth is estimated, while the endogeneity problem is
mitigated. The estimator for the threshold value involves two stage least squares (2SLS), and the

25

Electronic copy available at: [Link]

estimates of the slope parameters are obtained by using generalized method of moment (GMM).
Following Arezki et al. (2009), the number of UNESCO sites for each country’s surface area is
used as the instrumental variable. [In their study, the instrument for tourism is the number of
UNESCO sites per 100,000 inhabitants in the year 2002, kilometers of coastal area, and related
interactions as additional instruments. They further test the robustness of the results by using
different versions of the UNESCO World Heritage List, and the number of sites per surface area
is also included in their analysis.]

The descriptive statistics for the variables used in the cross-sectional IV threshold model are
reported in Table 5. Tables 9, 10 and 11 report the results from the IV threshold model. Three
different growth specifications, with three alternative measures of degree of tourism
specialization, as well as the set of control variables in the economic growth literature, are
investigated in the threshold effect of tourism specialization on economic growth. The two
regimes are based on different threshold variables, namely the degree of trade openness,
investment share to GDP, and government consumption as a percentage of GDP. In contrast to
the panel threshold analysis in the previous session, the slope coefficients of the tourism
variables, as well as other control variables, switch between regimes. We consider whether the
coefficients of these key variables change between regimes after taking account of endogeneity
in the cross-sectional regression.

[Insert Table 5 here]

Tables 9, 10 and 11 show the results from three different definitions of the economic
growth regressions, namely growth rate of real GDP chain per capita (rgdpch), growth rate of
real GDP chain per worker (rgdpwok), and growth rate of real GDP (Laspeyres) per capita
(rgdpl). The whole sample is grouped by the degree of trade openness, the investment share to
GDP, and the ratio of government consumption to GDP. In each table, regressions (1a)-(1c) are
growth regressions of rgdpch augmented with three tourism variables, namely tourist arrivals as
a proportion of population (TA), tourism receipts as a share of exports of goods and services
(TRE), and tourism receipts as a share of real GDP (TRG), respectively. Regressions (2a)-(2c)

26

Electronic copy available at: [Link]

and (3a)-(3c) are organized in the same manner for the rgdpwok and rgdpl growth regressions,
respectively.
In Table 9, the threshold values for trade openness are as follows: 91.872 for the rgdpch
per capita growth regression (model 1), where 97 countries have a smaller value and 62
countries have a larger value; 105.486 for the rgdpwok per capita growth regression (model 2),
where 115 countries have a smaller value and 44 countries have a larger value; and 74.056 for
the rgdpl per capita growth regression (model 3), where 74 countries have a smaller value and
85 countries have a larger value.

The relationship between tourism specialization and economic growth is found

empirically. The coefficients associated with the tourism development variables range from
0.0145 to 0.029 in the lower trade openness regime, from 0.0051 to 0.00948 in the higher trade
openness regime, and are significant across different growth rate specifications. These results
suggest that tourism development has a positive growth-boosting effect on the open economy,
though this contribution may not be sustained as the economy reaches very high trade openness.
According to Brau et al.(2009), who define a group of states with a degree of tourism
specialization greater than 8%, on average, over the period 1980-2004 as tourism countries, the
results here suggest that 33 countries can be characterized as “tourism countries”. Most of these
tourism specialized countries have a degree of trade openness higher than the estimated
threshold value for trade openness, particularly the small tourism specialized countries. About
41.07% (or 34.92%) of countries with trade openness greater than 105.49% (or 91.87%) are
tourism countries. In other words, several countries with a relatively high degree of tourism
specialization (tourism country) generally involve a higher degree of trade openness, yet they
have not been able to achieve the desired consequences of this particular characteristic of
economic growth.

The results obtained by Adamou and Clerides (2009) are supportive in this respect. They
find that specialization in tourism adds to a country’s rate of economic growth, but it does so at a
diminishing rate. This means that, at high levels of specialization the independent contribution of
tourism to economic growth becomes minimal, and tourism can even become a hindrance to
further growth. This interesting finding can be explained by the fact that the tourism destinations
which have already achieved higher tourism specialization may import capital goods in order to

27

Electronic copy available at: [Link]

support tourism expansion which, in turn, leads to a higher degree of trade openness.
Furthermore, a sub-optimal use of natural resources of a country with relative endowment of
natural resources might induce the country’s loss of comparative advantage in tourism with a
lower contribution of tourism, and possibly also cause unsustainable economic growth in the
long run.

The negative sign associated with initial income (the natural logarithm of real GDP per
capita in 1989) supports the convergence hypothesis, some of which are significant. Regarding
the influence of initial income on the growth rate, two estimation methods yield substantially
different results. Such differences arise because initial income is measured differently based on
alternative estimation methods. The initial income in a 5-year panel (a fixed effect panel), for
instance, is defined as the 5-year average of income from the previous period. However, the
initial income commonly used to check for convergence in the growth process in a pure cross-
sectional analysis is income in the initial year. The difference in the coefficients of initial income
in both methods emerges from differences in specification.

Trade openness provides evidence of the positive impact on economic growth. Note that
the slightly greater magnitude is found in the higher-trade opening regime, which implies that
the more open countries exert a powerful impact on economic prosperity. Investment share to
GDP is found to be positive across all three models, but only a few are found to be statistically
significant. The regressions also provide evidence of the negative impact of the population
growth rate, the negative impact of government consumption, and the positive impact of six
measures of institutional quality on economic growth. The coefficients of public investment in
education for economic growth are found to be significantly positive for most regressions. This
confirms that human capital plays a crucial role for economic growth, and that the inclusion of
public expenditure in education in the economic growth regression is an accurate measure of
human capital. The finding that human capital accumulation promotes economic growth is
supported by several studies (see, for example, Barro (1991) and Barro and Lee (2001)).

Differences in the coefficients of the key variables between regimes are of particular
interest. It is observed that the coefficients of all variables in the low regime are similar in

28

Electronic copy available at: [Link]

magnitude to those in the high regime for each corresponding economic growth specification.
This empirical finding does not change as the threshold variable under consideration changes.

[Insert Table 9 here]

In Table 10, investment share to GDP is used as a threshold variable. The threshold
values for the three growth specifications are similar. The threshold value for the rgdpch per
capita growth regression (model 1) is 17.526, where 62 countries have a smaller value and 97
countries have a larger value; 13.1726 for the rgdpwok per capita growth regression (model 2),
where 39 countries have a smaller value and 120 countries have a larger value; and 13.0743 for
the rgdpl per capita growth regression (model 3), where 38 countries have a smaller value and
121 countries have a larger value. The estimates in each model are in line with the economic
growth literature. Initial GDP has the expected negative coefficient, and the magnitude is similar
to those obtained from Table 9. With respect to the sign of the other coefficients, trade openness,
investment share to GDP, and institutional variables have a positive impact on economic growth,
while population growth and government consumption have a negative impact. As in Table 9,
public investment in education typically has a positive impact on economic growth. It is
observed that the coefficients of all variables in the low regime are similar in magnitude to those
in the high regime for each corresponding economic growth specification.

The impact of tourism and economic growth seems consistent with the results in Table 9.
The three tourism variables yield similar impacts on economic growth in each model. This
implies that the impact of tourism specialization on economic growth is robust to the various
specifications of tourism specialization. Although the significantly positive impact on economic
growth is found, such impacts in different regimes are not the same. Tourism specialization has a
slight effect on economic growth in the high-investment share countries, while the lower-
investment share countries have a higher impact. The coefficients associated with the three
tourism variables range from 0.0129 to 0.025 for the low-investment share regime, and from
0.00402 to 0.0062 for the high-investment share regime. Examining the list of countries with the
investment share to GDP is greater than the estimated threshold value, it is found that 23.71%

29

Electronic copy available at: [Link]

(or 21.66%) of countries with investment share to GDP greater than 17.5268% (or 13.1726%),
for example, are identified as “tourism countries”.

[Insert Table 10 here]

The results from three different growth specifications with government consumption
expenditure as a percent of GDP as a threshold variable, are reported in Table 11. The crucial
role of tourism expansion has been quantified through three different growth regressions. The
empirical evidence from most regressions (a)-(c) in each economic growth specification strongly
confirms the significantly positive impact of tourism specialization and economic growth. Only
a few regressions are insignificant. The estimates of all three tourism effects range from 0.0175
to 0.0198 for the lower-government spending regime, and from 0.0044 to 0.00593 for the higher-
government spending regime. All the tourism variables used to measure the reliance of a country
on tourism yield similar findings for each empirical growth model.

Overall, the sign of the coefficients of the common regressors for economic growth are
consistent with those reported in the previous tables. Moreover, similar magnitudes of the
coefficients of all the variables across the two regimes in each corresponding economic growth
specification are observed. In addition, it is found that government consumption has a largely
negative impact in the high-government spending regime, while the low-government spending
regime experiences lower negative impact on economic growth. This finding is of interest in the
government spending and economic growth relationship. Economic theory does not automat-
ically generate strong conclusions about the impact of government outlays on economic
performance. Indeed, there are circumstances in which lower levels of government spending
might enhance economic growth and other circumstances in which higher levels of government
spending would be desirable.

The “Rahn Curve” measures the relationship between different levels of government
spending and economic performance. The growth-maximizing point on the Rahn Curve is the
subject of considerable research. Experts generally conclude that this point is somewhere
between 15%-20% of GDP, although it is possible that these estimates are too high since

30

Electronic copy available at: [Link]

statistical studies are constrained by a lack of data for countries with limited governments
(Larson, 2007). The threshold estimates for government spending in our case are 21.7132 for the
rgdpch per capita growth regression (model 1), 17.6995 for the rgdpwok per capita growth
regression (model 2), and 15.2363 for the rgdpl per capita growth regression (model 3).
Therefore, countries in the high government-spending regime can be considered as countries
where higher government spending leads to a lower growth performance.

[Insert Table 11 here]

7. Concluding Remarks

Tourism specialization has significant potential beneficial economic impacts on the overall
economy of tourism destinations. This paper has not investigated the direction of the relationship
between economic growth and tourism, but whether tourism specialization has the same impact
on economic growth in countries that differ in their degree of trade openness, investment share
to GDP, and government consumption as a percentage of GDP. In order to examine the
contribution of tourism specialization to economic growth, the analysis is undertaken with
different threshold variables and regimes through the panel threshold regression model of
Hansen (2000) and IV threshold model of Caner and Hansen (2004). A 5-year averaged panel
data set and a pure cross-sectional data set of 159 countries over the period 1989-2008 were
used.

The results obtained from the panel threshold model of Hansen (2000) showed that economic
growth is boosted by means of trade openness, investment share, public investment in education,
and institutional variables, while population growth and government consumption have negative
effects. Initial income, trade openness, and public investment in education are significant in most
regressions, and this remains unchanged as the threshold variable changes. However, the degree
of influence of tourism specialization on economic growth in different regimes does not hold for
several regressions or for different threshold variables. As a result, there is no consensus
regarding whether tourism specialization has the same impact on economic growth for different
values of the threshold variables.

31

Electronic copy available at: [Link]

 We applied instrumental variable estimation of a threshold regression approach,
developed by Caner and Hansen (2004), to quantify the contributions of tourism specialization
on economic growth, while correcting for endogeneity. The number of UNESCO World
Heritage List per surface area is used as the instrumental variable. The results of the instrumental
variable threshold estimation reveal that the estimates in each model are similar to those found in
the economic growth literature. Initial GDP has the expected negative effect, implying the
existence of conditional convergence in the economic growth process. Trade openness,
investment share to GDP, and institutional variables have a positive impact on economic growth,
while population growth and government consumption have a negative impact, and are
insignificant in most regressions. Public investment in education typically has a positive impact
on economic growth. It is observed that the coefficients of all variables in the low regime are
similar in magnitude to those in the high regime for each corresponding economic growth
specification. These empirical findings do not change as the threshold variable under
consideration changes.

Focusing on the coefficients of tourism specialization, namely TA, TRE and TRG, the
results for the three economic growth models indicate that there is a significant and positive
relationship between tourism specialization and three economic growth specifications. The
robustness of such a relationship is illustrated by the qualitatively unchanged direction of the
coefficients associated with the tourism specialization variables. The significant impact of
tourism specialization on economic growth in most regressions is robust to the different
specifications of tourism specialization, as well as to the different real GDP measures. However,
the coefficients of these tourism specialization variables in the two regimes are significantly
different, with the higher impact of tourism on economic growth found in the lower regime.
These findings do not change as the threshold variables under consideration change.

The greater reliance on tourism through three tourism specialization definitions increases
the economic growth rate, but relatively less than that of the countries in the lower-trade
openness regime or lower-investment regime. Countries with a higher degree of trade openness
and investment are tourism countries. By listing countries with trade openness and investment
share to GDP greater than the threshold values, about 41.07% with trade openness greater than
105.486%, and 23.71% with investment share to GDP greater than 17.5268%, are identified as

32

Electronic copy available at: [Link]

“tourism countries”. Moreover, as the threshold variable is changed to government consumption
expenditure as a percentage of GDP, countries in the high government-spending regime can be
considered as countries where the higher government spending leads to a lower growth
performance.

Countries with a very high degree of trade openness and investment share to GDP are
likely to experience lower benefits from tourism development on economic growth. This could
be explained by the fact that the development of the tourism sector in these countries possibly
relies on investment in fixed capital formation in order to provide the necessary supply of
tourism. Furthermore, there is supporting evidence to suggest that many destinations,
particularly emerging tourism countries, have attempted to overcome the lack of financial
resources to speed up the process of tourism-specific infrastructure development. With limited
opportunities for local public sector funding, these countries have been offered funding by
international development organizations, or international companies, to make them more
attractive as tourism destinations. Although foreign capital investment can generate extra income
and growth from international tourist earnings for the host country, it can also generate greater
leakages than domestic capital investment from local private and government sources. In
addition to the leakages being remitted to the source of international funds, more imported goods
may be used to support the tourism industry. As a result, these factors could cause the
contribution of tourism to GDP to be lower than expected.

On the other hand, countries with relatively low trade openness, investment share to
GDP, and government consumption share to GDP, are possibly developed or developing, and
their economies may not be so heavily dependent on the tourism sector. Conversely, they might
be able to develop other non-tourism sectors that could make a greater contribution to overall
economic growth. However, we have seen that tourism development in some countries,
especially in regime 1, may make substantial contributions to economic growth.

In summary, tourism growth does not always lead to substantial impacts on economic
growth. If the economy is too heavily dependent on the tourism sector, tourism development
may not lead to impressive economic growth since the overall contribution of tourism to the
economy could be reduced by many factors. It is important to consider the overall balance

33

Electronic copy available at: [Link]

between international tourism receipts and expenditures, the degree of development of domestic
industries, and their ability to meet tourism requirements from domestic production. Should
these issues be constantly ignored, then such a country would likely experience lower benefits
than might be expected, regardless of whether they are considered to be a country with a high
degree of tourism specialization.

34

Electronic copy available at: [Link]

References

Adamou, A. and S. Clerides (2009), Prospects and limits of tourism-led growth: The
international evidence, Rimini Centre for Economic Analysis (RCEA) Working Paper
WP 41-09.
Algieri, B. (2006), International tourism specialization of small countries, International Journal
of Tourism Research, 8(1), 1-12.
Arezki, R., R. Cherif and J. Piotrowski (2009), Tourism specialization and economic
development: Evidence from the UNESCO World Heritage List, IMF working paper,
WP/09/176.
Balaguer, J. and M. Cantavella-Jordá (2002), Tourism as a long-run economic growth factor:
The Spanish case, Applied Economics, 34, 877-884
Balaguer, J. and M. Cantavella-Jordá (2004), Export composition and Spanish economic growth:
Evidence from the 20th Century, Journal of Policy Planning, 26, 165-179.
Barro, R.J. (1991), Economic growth in a cross-section of countries, Quarterly Journal of
Economics, 45, 407-443.
Barro, R.J. and Lee, J. (2001), International data on educational attainment: updates and
implications, Oxford Economic Papers, 53, 541-563.
Barro, R.J. and X. Sala-i-Martin (2003), Economic Growth, 2nd edition, MIT Press.
Brau, R., A. Lanza and F. Pigliaru (2007), How fast are small tourism countries growing? The
international 1980-2003 evidence, Milan, Italy, Fondazione Eni Enrico Mattei Nota di
Lavoro, No.1.
Brau, R., A.D. Liberto, and F. Pigliaru. (2009), Endowments vs institutions: the recent episode
of tourism development, Presented at the second conference of the International
Association for Tourism Economics (IATE), Chiang Mai, Thailand.
Brida, J.G., E.J. Sanchez Carrera and W.A. Risso (2008), Tourism’s impact on long run Mexican
economic growth, Economics Bulletin, 3(21), 1-8.
Caner, M. and B.E. Hansen (2004), Instrumental variable estimation of a threshold model,
Econometric Theory, 20, 813-843.
Cortés-Jiménez, I. (2009), Which type of tourism matters to the regional economic growth? The
cases of Spain and Italy, International Journal of Tourism Research, 10(2), 127-139.

35

Electronic copy available at: [Link]

Cortes-Jimenez, I. and M. Pulina (2010), Inbound tourism and long-run economic
growth, Current Issues in Tourism, 13(1), 61 – 74.
Dritsakis, N. (2004), Tourism as a long-run economic growth factor: An empirical investigation
for Greece using causality analysis, Tourism Economics, 10, 305-316.
Durbarry, R. (2004), Tourism and economic growth: The case of Mauritius, Tourism Economics,
10, 389-401.
Eugenio-Martı´n, J.L, N.M. Morales and R. Scarpa (2004), Tourism and economic growth in
Latin American countries: A panel data approach, Tourism and Sustainable Economic
Development - Macro and Micro Economic Issues, International Conference, Sardinia,
Italy, Fondazione Eni Enrico Mattei Nota di Lavoro, No. 26.
Figini, P. and L. Vici (2010), Tourism and growth in a cross-section of countries, to appear in
Tourism Economics, available at Rimini Centre for Economic Analysis, WP01-09.
Gani, A. (1998), Macroeconomic determinants of growth in the South Pacific island economies,
Applied Economics Letters, 5, 747-749.
Governance Matters (2009), Worldwide Governance Indicators, World Bank Development
Economics Research Group, World Bank institute, Available at
[Link]
Gunduz, L. and A. Hatemi (2005), Is the tourism-led growth hypothesis valid for Turkey?,
Applied Economics, 12, 499-504.
Hansen, B.E. (1999), Threshold effects in non-dynamic panels: Estimation, testing and inference,
Journal of Econometrics, 93, 345-368.
Hansen, B.E. (2000), Sample splitting and threshold estimation, Econometrica, 68, 575-603.
Heston, A., R. Summers and B. Aten. (2009), Penn World Table Version 6.3, Center for
International Comparisons of Production, Income and Prices, University of
Pennsylvania.
Islam, N. (1995), Growth empirics: a panel data approach, Quarterly Journal of Economics,
110(4), 1127-1170.
Kaufmann, D., Kraay, A. and Mastruzzi M. (2009), Governance matters VIII: Aggregate and
individual governance indicators 1996-2008, World Bank Policy Research Working
Paper4978.

36

Electronic copy available at: [Link]

Kim, H.J., Chen, M.H., and Jang, S.C. (2006), Tourism expansion and economic development:
The case of Taiwan, Tourism Management, 27, 925-933.
Lanza, A. and F. Pigliaru (2000), Why are tourism countries small and fast-growing?, in A.
Fossati and G. Panella (eds), Tourism and Sustainable Economic Development, Kluwer
Academic, pp. 57-69.
Larson, S.R. (2007), The economic case for limited government, Prosperitas: A Policy Analysis,
Center for Freedom and Prosperity Foundation, April, 7(3).
Lee, C.C. and Chang, C.P. (2008), Tourism development and economic growth: A closer look at
panels, Tourism Management, 29, 180-192.
Louca, C. (2006), Income and expenditure in the tourism industry: Time series evidence from
Cyprus, Tourism Economics, 12, 603-617.
Mankiw, N.G., Romer, D., and Weil, D.N. (1992), A contribution to the empirics of economic
growth, Quarterly Journal of Economics, 107(2), 407-437.
Noriko, I. and Mototsugu, F. (2007), Impacts of tourism and fiscal expenditure to remote islands:
The case of the Amami islands in Japan, Applied Economics Letters, 14, 661-666.
Nowak, J.J., Sahli, M., and Cortés-Jiménez, I. (2007), Tourism, capital good imports and
economic growth: Theory and evidence for Spain, Tourism Economics, 13, 515-536.
Oh, C.O. (2005), The contribution of tourism development to economic growth in the Korean
economy, Tourism Management, 26, 39-44.
Po, W.C. and Huang, B.N. (2008), Tourism development and economic growth - A nonlinear
approach, Physica A: Statistical Mechanics and its Applications, 387(22), 5535-5542.
Sequeira, T.N. and C. Campos (2007), International tourism and economic growth: A panel data
approach, Advances in Modern Tourism Research: Economic Perspectives, Physica-
Verlag.
Sharpley, R. and Telfer, D.J. (2004), Tourism and development: Concepts and issues, Great
Britain, Cromwell Press Ltd.
Temple, J. (1999), The new growth evidence, Journal of Economic Literature, 37(1), 112-156.
UNESCO (2008), World Heritage List, UNESCO World Heritage Center. Available at
[Link]
UNWTO World Tourism Barometer (2009), International Tourism Challenged by Deteriorating
World Economy, 7(1), January 2009, Available at [Link].

37

Electronic copy available at: [Link]

World Bank (2008), World Development Indicators, Database.
World Travel and Tourism Council (2009), Methodology document for WTTC 2009 Travel and
Tourism Economic Research, available at
[Link]
World Tourism Organization (2008), Tourism and the world economy, in the Facts & Figures
section, available at [Link]

38

Electronic copy available at: [Link]

Table 1: Contribution of Tourism to the Overall Economy GDP and Employment in 2009,
and Projection of Travel & Tourism Economy Real Growth, by Global Regions

2009 Travel Travel &

2009 Travel 2009 Travel
2009 Visitor &Tourism Tourism
&Tourism &Tourism
Regions Exports Economy Economy Real
Economy GDP Economy GDP
(US$ Mn) Employment Growth
(US$ Mn) % share
(Thous of jobs) (2010-2019)
Caribbean 39,410.668 30.312 24,154.262 2,042.512 3.568

Central and Eastern Europe 142,439.966 9.580 36,940.472 6,797.150 5.741

European Union 1,667,656.460 10.716 423,685.250 23,003.960 3.808

Latin America 176,954.984 8.729 30,223.315 12,421.720 4.031

Middle East 158,112.740 11.457 50,738.918 5,130.767 4.564

North Africa 62,893.900 12.164 25,622.089 5,440.087 5.417

North America 1,601,235.000 10.492 188,517.700 21,130.230 4.031

Northeast Asia 1,053,780.332 18.333 114,400.124 70,512.123 5.488

Oceania 115,902.843 18.558 38,403.241 1,701.315 4.394

Other Western Europe 150,082.280 10.207 42,694.005 2,277.688 2.642

South Asia 84,223.460 14.846 14,904.677 37,174.593 4.970

South-East Asia 155,158.492 10.478 65,765.366 23,231.522 4.415

Sub-Saharan Africa 65,866.259 9.047 23,392.256 8,948.552 4.718

World 5,473,717.384 1,079,441.62 219,812.220

Source: World Travel and Tourism Council (2009).

39

Electronic copy available at: [Link]

 Table 2: Countries in the Sample

Countries
Albania Guinea Paraguay
Angola Guinea-Bissau Peru
Antigua and Barbuda Guyana Philippines
Argentina Haiti Poland
Armania Honduras Portugal
Australia Hong Kong Romania
Austria Hungary Russia
Azerbaijan Iceland Rwanda
Bahamas India Samoa
Bahrain Indonesia Sao Tome and Principe
Bangladesh Iran Saudi Arabia
Barbados Ireland Senegal
Belarus Israel Seychelles
Belgium Italy Sierra Leone
Belize Jamaica Singapore
Benin Japan Slovak Rep.
Bolivia Jordan Slovenia
Bosnia and Herzegovina Kazakstan Solomon Islands
Botswana Kenya South Africa
Brazil Korea [Link] Spain
Brunei Darussalam Kuwait Sri Lanka
Bulgaria Kyrgyzstan [Link]
Burkina Faso Laos PDR. [Link]&Grenadines
Burundi Latvia Sudan
Cambodia Lebanon Suriname
Cameroon Lesotho Swaziland
Canada Libya Sweden
Cape Verde Lithunia Switzerland
Chile Luxembourg Syrian Arab Rep.
China Macao Tanzania
Colombia Macedonia, FYR Thailand
Congo Rep. Madagascar Togo
Costa Rica Malawi Tonga
Croatia Malaysia Trinidad&Tobago
Cyprus Maldives Tunisia
Czech Rep. Mali Turkey
Denmark Malta Uganda
Dominica Mauritania U.K.
Dominican Rep. Mauritius Ukraine
Ecuador Mexico United Arab Emirates
Egypt Moldova United States
Elsalvador Mongolia Uruguay
Eritrea Morocco Vanuatu
Estonia Mozambique Venezuela
Ethiopia Namibia Vietnam
Fiji Nepal Yemen [Link]
Finland Netherlands Zambia
France New Zealand
Gabon Nicaragua
Gambia Niger
Georgia Nigeria
Germany Norway
Ghana Oman
Greece Pakistan
Grenada Panama
Guatemala Papua New Guinea

40

Electronic copy available at: [Link]

 Table 3: Data Description and Sources

Variables Definition Description Data source

Dependent variables: Growth variables

growth rate of real GDP chain log difference of real GDP chain per worker at time t PWT 6.3
per worker and t-1

growth rate of real GDP chain log difference of real GDP per capita (Constant Prices: PWT 6.3
per capita Chain series) at time t and t-1

growth rate of real GDP log difference of real GDP per capita (Constant Prices: PWT 6.3
Laspeyres per capita Laspeyres), derived from growth rates of c, g, i, at time t
and t-1
Explanatory
Variables:
Tourism variables
TA tourist arrivals tourist arrivals as a percentage of total population WDI 2008
TRE tourism receipts tourism receipts as a percentage of exports of goods and WDI 2008
services
TRG tourism receipts tourism receipts as a percentage of GDP WDI 2008

Control variables
, , initial income 5-year average of real GDP per capita in the previous PWT 6.3
period (panel threshold analysis) or real GDP per capita
1989 (IV threshold)
physical capital investment investment share of real GDP per capita WDI 2008
public spending on education public spending on education as a percentage of GDP WDI 2008
growth rate of population log difference of population at time t and t-1 PWT 6.3
trade openness openness in current prices (% in current prices); total PWT 6.3
trade (export plus imports) as a percentage of GDP
government consumption government consumption as a percentage of GDP WDI 2008
voice and accountability* see details on page 10-11 World Bank 2008
political stability and absence see details on page 10-11 World Bank 2008
of violence*
government effectiveness* see details on page 10-11 World Bank 2008
regulatory quality* see details on page 10-11 World Bank 2008
rule of law* see details on page 10-11 World Bank 2008
control of corruption* see details on page 10-11 World Bank 2008

Threshold 1. trade openness openness in constant price PWT 6.3

variables: 2. physical capital investment investment share of real GDP per capita WDI 2008
3. government consumption government consumption as a percentage of GDP WDI 2008

Instrument UNESCO number of UNESCO World Heritage List (WHL) per UNESCO 2009
variable: surface area**
UNESCO
Note: *The six governance indicators are measured in units ranging from about -2.5 to 2.5, with higher values
corresponding to better governance outcomes. **surface area ([Link]) is obtained from WDI 2008 database.

41

Electronic copy available at: [Link]

 Table 4: Summary Statistics: 5-year Panel Dataset
Full Sample Summary Statistics
Variables
Mean [Link]. Minimum Maximum Observations*
0.0249 0.4165 -3.3670 8.8020 N=636, n=159,
(rgdpwok) T=4
0.0288 0.1347 -0.221 2.2170 N=636, n=159,
(rgdpch) T=4
0.0304 0.3570 -1.9410 7.9450 N=636, n=159,
(rgdpl) T=4
54.4223 13.3426 0.0390 2082.955 N=636, n=159,
(TA) T=4
15.2337 16.3920 0.0530 76.7100 N=636, n=159,
(TRE) T=4
3.1792 5.5017 0.003 46.534 N=636, n=159,
(TRG) T=4
9.5248 1.0725 6.8550 11.987 N=636, n=159,
T=4
8.6443 1.1264 5.8840 11.0610 N=636, n=159,
T=4
8.6418 1.1274 5.8840 11.0610 N=636, n=159,
T=4
21.3671 11.4698 -2.3420 84.2340 N=636, n=159,
T=4
4.4079 1.8508 0.8310 13.574 N=636, n=159,
T=4
0.0193 0.0251 -0.369 0.2210 N=636, n=159,
T=4
86.5657 50.4278 14.3770 443.1870 N=636, n=159,
T=4
16.4026 6.4296 3.8450 54.9830 N=636, n=159,
T=4
0.0506 0.9129 -2.0380 1.6330 N=636, n=159,
T=4
0.0218 0.8894 -2.5000 1.6300 N=636, n=159,
T=4
0.0913 0.9561 -1.763 2.3360 N=636, n=159,
T=4
0.1193 0.8663 -2.1500 2.4130 N=636, n=159,
T=4
0.0450 0.9416 -1.8500 2.0420 N=636, n=159,
T=4
0.0678 0.9739 -1.7568 2.4649 N=636, n=159,
T=4
UNESCO 0.000124 0.00082 0 0.0093 N=636, n=159,
T=4

42

Electronic copy available at: [Link]

 Table 5: Summary Statistics: Cross-sectional Dataset

Full Sample Summary Statistics

Variables
Mean [Link]. Minimum Maximum Observations*
0.0249 0.2329 -1.6725 2.2594 159
(rgdpwok)
0.0289 0.0704 -0.0609 2.5904 159
(rgdpch)
0.0303 0.1838 -0.4989 2.0532 159
(rgdpl)
54.4223 131.4667 0.0559 1376.0350 159
(TA)
15.2337 16.0551 0.4479 72.8091 159
(TRE)
3.1792 5.4034 0.0136 35.0176 159
(TRG)
9.5248 1.0653 7.1821 11.7081 159

8.6443 1.1184 6.4326 10.8721 159

8.6442 1.1191 6.4368 10.8739 159

21.367 10.5891 4.3893 69.6619 159

4.4079 1.66431 0.83944 11.2392 159

0.0193 0.01565 -0.0192 0.0637 159

86.5657 47.8855 20.9003 359.7687 159

16.4026 5.9844 4.8312 39.9588 159

0.05059 0.9011 -1.7828 1.5972 159

0.02184 0.8597 -2.2944 1.4487 159

0.09132 0.9406 -1.3772 2.3677 159

0.1193 0.8290 -1.7719 1.8854 159

0.0450 0.9254 -1.5362 1.9756 159

0.0679 0.95621 -1.3186 2.3498 159

UNESCO 0.000124 0.00082 0 0.00938 159

43

Electronic copy available at: [Link]

 Table 6: Results from Panel Threshold Regression
Threshold Variable: Trade Openness

Model Model 1: growth rate of real GDP chain per capita Model 2: growth rate of real GDP chain per worker Model 3: growth rate of real GDP (Laspeyres) per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2
0.0010** 0.0021 0.0082** 0.0081 0.0221** 0.0021
TA
(0.036) (0.644) (0.022) (0.641) (0.039) (0.803)
0.0018 0.0010 0.0084 0.0118** 0.0068* 0.0029
TRE
(0.381) (0.5985) (0.664) (0.049) (0.086) (0.501)

0.0013** 0.0028 0.0090 0.0098 0.0144 0.0026

TRG
(0.013) (0.608) (0.455) (0.234) (0.281) (0.851)
-0.1052*** -0.1058*** -0.1084*** -0.4091** -0.4075*** -0.4208** -0.3657*** -0.3566*** -0.3641***
(0.0060) (0.0017) (0.0045) (0.0116) (0.0087) (0.0113) (0.0003) (0.0004) (0.0003)
0.0008** 0.0008*** 0.0009** 0.0023** 0.0037*** 0.0025** 0.0032*** 0.0024*** 0.0025***
(0.0353) (0.0005) (0.0147) (0.0391) (0.0029) (0.0247) (0.0006) (0.0099) (0.0072)
0.0011 0.0014 0.0012 0.0048 ** 0.0056** 0.0053** 0.0061* 0.0057* 0.0056*
(0.3617) (0.2897) (0.2976) (0.0237) (0.0209) (0.0176) (0.0515) (0.0700) (0.0715)
-0.2836 -0.2885 -0.2818 -0.6536 -0.5792 -0.6459 -0.6499 -0.4804 -0.4945
(0.2939) (0.1221) (0.2977) (0.3917) (0.4264) (0.3956) (0.3640) (0.5007) (0.4889)
0.0324*** 0.0319** 0.0337*** 0.0945* 0.0828* 0.0949* 0.0857*** 0.0845*** 0.0871***
(0.0000) (0.0298) (0.0000) (0.0522) (0.0877) (0.0566) (0.0000) (0.0000) (0.0000)
-0.0002 -0.0001 -0.0008 0.0099 0.0093 0.0099 0.0076 0.0079 0.0077
(0.9466) (0.9705) (0.7418) (0.1117) (0.1315) (0.1133) (0.2149) (0.1938) (0.2134)
0.0610* 0.0635 0.0605* 0.0190 0.0373 0.0210 0.0103 0.0137 0.0182
(0.0878) (0.1148) (0.0913) (0.7606) (0.5663) (0.7385) (0.9134) (0.8850) (0.8478)
0.0075 0.0085 0.0148 0.1482 0.1420 0.1391 0.0454 0.0549 0.0485
(0.7534) (0.5792) (0.5353) (0.1143) (0.1403) (0.1344) (0.4734) (0.3857) (0.4442)
0.0075 0.0081 0.0053 0.1043 0.0727 0.1158 0.1074 0.0892 0.0820
(0.8233) (0.7563) (0.8742) (0.1336) (0.2247) (0.1164) (0.2296) (0.3145) (0.3580)
0.0411* 0.0452** 0.0424* 0.1076 0.0892 0.1024 0.1420 ** 0.1414** 0.1395**
(0.0722) (0.0374) (0.0642) (0.2781) (0.3849) (0.3059) (0.0192) (0.0195) (0.0214)
0.0013 0.0057 0.0084 0.0096 0.0237 0.0151 0.0466 0.0391 0.0531
(0.9669) (0.8060) (0.7952) (0.8417) (0.6751) (0.7427) (0.5844) (0.6469) (0.5357)
0.0391 0.0411* 0.0376 0.0148 0.0136 0.0225 0.0280 0.0225
0.0145 (0.6720)
(0.2111) (0.0858) (0.2304) (0.6958) (0.7105) (0.7851) (0.7350) (0.7857)
Threshold 114.6010 114.6010 114.6010 116.7580 116.7580 116.7580 103.6650 103.6650 103.6650

Note: p-values are reported in parentheses. *, ** and *** indicate statistical significance at the 1%, 5% and 10% levels, respectively.

44

Electronic copy available at: [Link]

 Table 7: Results from Panel Threshold Regression
Threshold Variable: Investment Share to GDP

Model Model 1: growth rate of real GDP chain per capita Model 2: growth rate of real GDP chain per worker Model 3: growth rate of real GDP (Laspeyres) per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2
0.0094** 0.0001* 0.0026** 0.0040 0.0026 0.0001
TA
(0.017) (0.0974) (0.0297) (0.9791) (0.229) (0.7083)
0.0086 0.0040 0.0017 0.0045 0.0019 0.0008**
TRE
(0.638) (0.2023) (0.1126) (0.5098) (0.7064) (0.0180)

0.0078** 0.0159 0.0022* 0.0042 0.0028 0.0009

TRG
(0.0344) (0.244) (0.0880) (0.6499) (0.5386) (0.951)
-0.1044*** -0.1107*** -0.0968*** -0.4245*** -0.4063*** -0.4190*** -0.3638*** -0.3586*** -0.3515***
(0.0064) (0.0018) (0.0012) (0.0005) (0.0080) (0.0006) (0.0004) (0.0004) (0.0005)
0.0010*** 0.0010*** 0.0011*** 0.0030*** 0.0029** 0.0030** 0.0029*** 0.0028*** 0.0029***
(0.0029) (0.0022) (0.0023) (0.0031) (0.0128) (0.0034) 90.0014) (0.0017) (0.0013)
0.0020* 0.0000 0.0006 0.0056 0.0049** 0.0072* 0.0054* 0.0041 0.0066**
(0.0960) (0.9893) (0.5295) (0.1129) (0.0175) (0.0512) (0.0865) (0.1918) (0.0445)
-0.2278 -0.2091 -0.2569 -0.6359 -0.5654 -0.6235 -0.5423 -0.4353 -0.5319
(0.3996) (0.1557) (0.1161) (0.4307) (0.4122) (0.4397) (0.4488) (0.5403) (0.4577)
0.0291*** 0.0307** 0.0302** 0.0895*** 0.0873* 0.0906*** 0.0832*** 0.0812*** 0.0842***
(0.0000) (0.0262) (0.0273) (0.0000) (0.0534) (0.00000 (0.0000) (0.0000) (0.0000)
-0.0002 0.0005 -0.0007 0.0097 0.0102 0.0098 0.0077 0.0088 0.0072
(0.9175) (0.7790) (0.7057) (0.1599) (0.1082) (0.1602) (0.2083) (0.1508) (0.2449)
0.0689* 0.0573 0.0588 0.0202 0.0076 0.0368 0.0111 0.0156 0.0027
(0.0539) (0.1280) (0.1174) (0.8505) (0.9060) (0.7294) (0.9077) (0.8685) (0.9770)
0.0083 0.0070 0.0115 0.1360* 0.1784 0.1406** 0.0409 0.0510 0.0456
(0.7281) (0.6418) (0.4914) (0.0560) (0.1300) (0.0482) (0.5197) (0.4183) (0.4732)
0.0051 0.0175 0.0031 0.0909 0.1352* 0.0905 0.0902 0.0602 0.0981
(0.8800) (0.5404) (0.9016) (0.3658) (0.0984) (0.3686) (0.3118) (0.4999) (0.2725)
0.0393* 0.0408** 0.0430** 0.1176* 0.1062 0.0973 0.1519** 0.1237** 0.1323**
(0.0861) (0.0332) (0.0341) (0.0961) (0.3090) (0.1540) (0.0157) (0.0403) (0.0291)
0.0056 0.0000 0.0052 0.0180 0.0043 0.0146 0.0409 0.0164 0.0452
(0.8614) (0.9999) (0.8174) (0.8511) (0.9244) (0.8792) (0.6320) (0.8474) (0.5986)
0.0493 0.0379* 0.0376* 0.0058 0.0195 0.0091 0.0133 0.0120 0.0215
(0.1157) (0.0908) (0.0858) (0.9500) (0.6091) (0.9228) (0.8725) (0.8842) (0.7962)
Threshold 24.103 24.1030 24.103 11.849 11.849 11.849 18.849 18.849 18.849

Note: p-values are reported in parentheses. *, ** and *** indicate statistical significance at the 1%, 5% and 10% levels, respectively.

45

Electronic copy available at: [Link]

 Table 8: Results from Panel Threshold Regression
Threshold Variable: Government Consumption as a Percentage of GDP

Model Model 1: growth rate of real GDP chain per capita Model 2: growth rate of real GDP chain per worker Model 3: growth rate of real GDP (Laspeyres) per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2 RG1 RG2
0.0001* 0.0026 0.0055** 0.0011 0.0001* 0.0087
TA
(0.077) (0.4026) (0.3266) (0.1898) (0.0790) (0.375)
0.0016** 0.0028 0.0068* 0.0003* 0.0015* 0.0095**
TRE
(0.0428) (0.2171) (0.0720) (0.0942) (0.0729) (0.0235)

0.0073** 0.00158 0.0067 0.0139 0.0063 0.0176

TRG
(0.0252) (0.175) (0.6503) (0.424) (0.6205) (0.2475)
-0.0987*** -0.0972** -0.1054 *** -0.4000*** -0.4221*** -0.4250*** -0.3493*** -0.3377*** -0.3380***
(0.0006) (0.0112) (0.0018) (0.0010) (0.0005) (0.0005) (0.0006) (0.0008) (0.0009)
0.0011*** 0.0010** 0.0011*** 0.0032*** 0.0027*** 0.0029*** 0.0028*** 0.0027*** 0.0029***
(0.0030) (0.0042) (0.0025) (0.0018) (0.0067) (0.0048) (0.0020) (0.0028) (0.0018)
0.0012 0.0015 0.0016 0.0057 0.0057 0.0058 0.0055* 0.0058* 0.0055*
(0.3004) (0.2180) (0.2551) (0.1062) (0.1010) (0.1017) (0.0801) (0.0631) (0.0779)
-0.2895 -0.2712 -0.2912 -0.6512 -0.7341 -0.6586 -0.4947 -0.5405 -0.4717
(0.1213) (0.3185) (0.1106) (0.4230) (0.3593) (0.4140) (0.49020 (0.4475) (0.5108)
0.0304** 0.0290*** 0.0317** 0.0909*** 0.0983*** 0.0928*** 0.0843*** 0.0859*** 0.0867***
(0.0280) (0.0000) (0.0251) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000)
-0.0004 -0.0004 -0.0004 0.0106 0.0041 0.0084 0.0087 0.0096 0.0082
(0.8306) (0.8837) (0.8263) (0.1367) (0.5667) (0.2344) (0.1623) (0.1167) (0.1879)
0.0588 0.0645* 0.0617 0.0157 0.0117 0.0277 0.0044 0.0055 0.0067
(0.1119) (0.0724) (0.1195) (0.8834) (0.9120) (0.7950) (0.9627) (0.9537) (0.9437)
0.0121 0.0074 0.0196 0.1355* 0.1308* 0.1361* 0.0470 0.0552 0.0477
(0.4606 (0.7575) (0.3053) (0.0567) (0.0640) (0.0555) (0.4598) (0.3816) (0.4525)
0.0106 0.0046 0.0082 0.0886 0.0671 0.0900 0.0864 0.0819 0.0858
(0.7003) (0.8921) (0.7555) (0.3778) (0.5013) (0.3705) (0.3325) (0.3553) (0.3375)
0.0434** 0.0420* 0.0461** 0.1006 0.0982 0.0977 0.1348** 0.1336** 0.1326**
(0.0407) (0.0677) (0.0344) (0.1413) (0.1470) (0.1521) (0.0263) (0.0266) (0.0286)
0.0028 0.0033 0.0043 0.0303 0.0075 0.0046 0.0339 0.0268 0.0271
(0.9004) (0.9201) (0.8519) (0.7534) (0.9372) (0.9624) (0.6929) (0.7525) (0.7523)
0.0367* 0.0389 0.0325 0.0024 0.0255 0.0139 0.0184 0.0056 0.0213
(0.0966) (0.2156) (0.1227) (0.9794) (0.7839) (0.8820) (0.8244) (0.9458) (0.7976)
Threshold 24.642 24.642 24.642 16.769 16.769 16.769 18.2770 18.2770 18.2770
Note: p-values are reported in parentheses. *, ** and *** indicate statistical significance at the 1%, 5% and 10% levels, respectively

46

Electronic copy available at: [Link]

 Table 9: Results from Cross-section Instrumental Variable (IV) Threshold Model
Threshold Variable: Trade Openness

Model Model 1: growth rate of rgdpch per capita Model 2: growth rate of rgdpwok per capita Model 3: growth rate of rgdpl per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
L H L H L H L H L H L H L H L H L H
0.016 0.0063 0.0187 0.00948 0.0156 0.00544
TA
(0.008) (0.013) (0.0096) (0.0064) (0.0072) (0.0037)
0.0208 0.0051 0.0220 0.00887 0.0185 0.0058
TRE
(0.0088) (0.0015) (0.0114) (0.0063) (0.0079) (0.0029)
0.0215 0.0053 0.0145 0.0082 0.0255 0.0067
TRG
(0.0148) (0.0026) (0.0081) (0.0042) (0.0067) (0.0038)
-0.0065 -0.0218 -0.0128 0.0274 -0.0132 -0.0235 -0.0158 -0.0261 -0.0131 -0.0097 -0.0147 -0.0177 -0.0114 -0.0051 -0.0112 -0.0059 0.0170 -0.0065
(0.0023) (0.0024) (0.0039) (0.0029) (0.0018) (0.0059) (0.00157) (0.0155) (0.0168) (0.0663) (0.0188) (0.0185) (0.0275) (0.0056) (0.0165) (0.0051) (0.0185) (0.0023)
0.0034 0.0045 0.0039 0.0041 0.0043 0.0049 0.00347 0.00396 0.0034 0.0042 0.00351 0.0036 0.0033 0.0041 0.0035 0.0047 0.0033 0.0041
(0.0013) (0.0024) (0.0117) (0.0023) (0.0029) (0.0027) (0.0018) (0.00158) (0.0181) (0.0025) (0.0018) (0.0016) (0.0026) (0.0017) (0.0026) (0.0001) (0.0025) (0.002)
0.0032 0.0026 0.0090 0.0054 0.0018 0.0029 0.0073 0.0048 0.0049 0.0061 0.0057 0.0041 0.0071 0.0027 0.0053 0.00069 0.0062 0.0026
(0.0029) (0.0049) (0.0158) (0.0778) (0.0034) (0.0008) (0.0049) (0.0074) (0.0062) (0.0699) (0.0034) (0.0075) (0.0074) (0.0176) (0.0044) (0.0018) (0.0057) (0.0007)
-0.6304 -0.1992 -0.3814 -0.3602 -0.6198 -0.3331 -0.5783 -0.2978 -0.5486 -0.2508 -0.5788 -0.2433 -0.3185 -0.197 -0.3670 -0.1856 -0.3678 -0.247
(0.2881) (0.379) (0.441) (0.4592) (0.2339) (0.472) (0.3722) (0.525) (0.4074) (0.546) (0.4017) (0.543) (0.638) (0.746) (0.691) (0.889) (0.644) (0.557)
0.0048 0.0113 0.0227 0.0184 0.0230 0.0093 0.0443 0.0453 0.0415 0.0471 0.0461 0.0402 0.0124 0.0174 0.0111 0.0086 0.0102 0.0182
(0.00219) (0.0066) (0.0027) (0.0075) (0.0188) (0.00532) (0.021) (0.0164) (0.0251) (0.0202) (0.0253) (0.0162) (0.0051) (0.0057) (0.0135) (0.0071) (0.0125) (0.0061)
-0.006 -0.0491 -0.0104 -0.0486 0.0068 -0.0464 -0.0108 -0.0248 -0.0064 -0.0206 -0.0068 -0.0302 0.0155 -0.0435 -0.0079 -0.0480 -0.0086 -0.0408
(0.008) (0.0272) (0.014) (0.0249) (0.016) (0.225) (0.0348) (0.0067) (0.031) (0.0170) (0.028) (0.0083) (0.0102) (0.0548) (0.047) (0.0529) (0.0054) (0.0517)
0.0241 0.0148 0.0108 0.0204 0.0108 0.0121 0.0194 0.0474 0.01594 0.0439 0.0488 0.0358 0.0992 0.0118 0.0972 0.1568 0.0817 0.0155
(0.01334) (0.0120) (0.0106) (0.0172) (0.0081) (0.0120) (0.0335) (0.0410) (0.0428) (0.8816) (0.0389) (0.0588) (0.1225) (0.0119) (0.0923) (0.0177) (0.0891) (0.0126)
0.0125 0.0317 0.0228 0.05418 0.0100 0.1631 0.2056 0.0713 0.1559 0.4722 0.1584 0.1032 0.1542 0.0317 0.1229 0.0112 0.1233 0.0484
(0.0057) (0.0276) (0.0086) (0.0651) (0.0052) (0.0694) (0.0289) (0.0788) (0.0261) (0.5017) (0.0284) (0.1412) (0.0695) (0.0190) (0.0586) (0.0346) (0.0587) (0.0184)
0.0538 0.0168 0.10583 0.0093 0.0419 0.28201 0.1781 0.2320 0.2311 0.4534 0.2059 0.1708 0.1105 0.1501 0.1310 0.1808 0.1101 0.1991
(0.0178) (0.0709) (0.0352) (0.1102) (0.1104) (0.08749) (0.0544) (0.1647) (0.0689) (1.0873) (0.0604) (0.2569) (0.1021) (0.0758) (0.1110) (0.0610) (0.1035) (0.0841)
0.0289 0.0411 0.1128 0.0063 0.1914 0.0304 0.0561 0.0475 0.1829 0.1753 0.1406 0.1994 0.0896 0.1737 0.1731 0.0514 0.1458 0.1802
(0.0151) (0.0369) (0.0165) (0.0335) (0.0337) (0.0346) (0.0668) (0.0703) (0.0387) (0.0680) (0.0420) (0.1146) (0.1021) (0.0624) (0.0655) (0.0610) (0.0562) (0.0591)
0.0619 0.1086 0.0743 0.273 0.0268 0.0152 0.0046 0.1777 0.0106 0.2294 0.2367 0.2234 0.2255 0.0541 0.3747 0.2913 0.1674 0.0718
(0.0457) (0.0401) (0.0490) (0.0591) (0.0080) (0.0496) (0.0411) (0.1289) (0.0382) (0.7099) (0.0374) (0.1897) (0.1033) (0.284) (0.2097) (0.337) (0.1205) (0.329)
0.0179 0.0127 0.0139 0.0052 0.0049 0.0659 0.0733 0.0478 0.0666 0.0097 0.0719 0.0962 0.1503 0.0518 0.0945 0.0425 0.0992 0.0529
(0.0116) (0.0569) (0.0202) (0.0643) (0.0084) (0.0564) (0.0558) (0.2085) (0.0469) (0.3446) (0.0524) (0.2293) (0.1425) (0.0442) (0.0956) (0.0651) (0.1049) (0.0413)
[Link] obs. 97 62 97 62 97 62 115 44 115 45 115 45 74 85 74 85 74 85
Threshold 91.8722 91.8722 91.8722 105.4862 105.4862 105.4862 74.0565 74.0565 74.0565

Note: Robust standard errors in parentheses. Boldface values indicate statistical significance at the 1% level.

47

Electronic copy available at: [Link]

 Table 10: Results from Cross-section Instrumental Variable (IV) Threshold Model
Threshold Variable: Investment Share to GDP
Model Model 1: growth rate of rgdpch per capita Model 2: growth rate of rgdpwok per capita Model 3: growth rate of rgdpl per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
L H L H L H L H L H L H L H L H L H
0.0181 0.0057 0.025 0.0045 0.0175 0.00402
TA
(0.005) (0.0025) (0.0068) (0.002) (0.0077) (0.0022)
0.0176 0.0062 0.0146 0.0046 0.0238 0.0043
TRE
(0.002) (0.0035) (0.0078) (0.00218) (0.0119) (0.0018)
0.0252 0.0048 0.0179 0.0042 0.0129 0.0047
TRG
(0.011) (0.0143) (0.0089) (0.024) (0.0065) (0.0463)
-0.0122 -0.02032 -0.01176 -0.0102 -0.0190 -0.0156 -0.0287 -0.0073 -0.0144 -0.0175 -0.0262 -0.0117 -0.0294 -0.02135 -0.0798 -0.01992 -0.0196 -0.0062
(0.0027) (0.0056) (0.0023) (0.0156) (0.0022) (0.0041) (0.0034) (0.0065) (0.0014) (0.0022) (0.0064) (0.0079) (0.0097) (0.0038) (0.0616) (0.0063) (0.0059) (0.0039)
0.0015 0.00057 0.0002 0.0001 0.0046 0.00056 0.0093 0.0004 0.00034 0.0031 0.0099 0.0034 0.0028 0.0004 0.0126 0.0028 0.0097 0.0004
(0.0003) (0.00018) (0.0001) (0.0004) (0.0012) (0.0002) (0.0141) (0.0003) (0.0001) (0.0001) (0.0016) (0.0002) (0.0004) (0.0002) (0.0055) (0.0008) (0.0084) (0.0002)
0.0017 0.0074 0.0015 0.0069 0.0019 0.0079 0.0015 0.0085 0.0021 0.0129 0.0040 0.0094 0.0021 0.0153 0.0018 0.0198 0.0019 0.0169
(0.0016) (0.0209) (0.0017) (0.0104) (0.0006) (0.0227) (0.1716) (0.0065) (0.0024) (0.0490) (0.0720) (0.0049) (0.0349) (0.0206) (0.0029) (0.0633) (0.0026) (0.0618)
-0.2222 -0.1759 -0.1451 -0.1830 -0.1264 -0.1024 -0.4176 -0.1755 -0.2926 -0.3264 -0.4624 -0.3172 -0.1453 -0.1813 -0.1995 -0.1258 -0.1418 -0.1702
(0.0346) (0.859) (0.219) (0.456) (0.218) (0.273) (0.154) (0.2869) (0.315) (0.2076) (0.1998) (0.2115) (0.342) (0.376) (0.971) (0.588) (0.520) (0.379)
0.0255 0.0131 0.0059 0.0088 0.0212 0.0096 0.0103 0.0299 0.0138 0.0214 0.0204 0.0078 0.0144 0.0229 0.02621 0.01607 0.0274 0.0065
(0.0050) (0.0024) (0.0023) (0.0306) (0.0025) (0.0031) (0.0051) (0.0096) (0.0099) (0.0061) (0.0079) (0.0038) (0.0044) (0.0042) (0.00752 (0.0060) (0.0124) (0.0033)
-0.0037 -0.0165 -0.0016 - 0.0168 0.0106 -0.0011 -0.0105 -0.0049 -0.0012 -0.0128 0.0556 -0.0046 -0.0170 -0.0087 -0.0342 -0.0008 -0.0552 -0.0010
(0.0011) (0.016) (0.014) (0.038) (0.0044) (0.016) (0.0043) (0.0037) (0.0015) (0.1095) (0.0646) (0.0034) (0.0086) (0.013) (0.0525) (0.0019) (0.0616) (0.0013)
0.0439 0.0109 0.0155 0.0369 0.0658 0.0378 0.0892 0.0324 0.0268 0.07404 0.0238 0.0130 0.04572 0.0155 0.4468 0.0160 0.2346 0.0132
(0.0164) (0.0114) (0.0046) (0.0327) (0.0046) (0.0128) (0.0127) (0.0174) (0.0205) (0.0639) (0.3214) (0.0279) (0.314) (0.0082) (0.4597) (0.0211) (0.3347) (0.0093)
0.0103 0.0806 0.00767 0.0727 0.1412 0.0057 0.0141 0.0402 0.0176 0.0574 0.0434 0.0382 0.024446 0.0224 0.0408 0.0128 0.0408 0.0232
(0.0069) (0.111) (0.0044) (0.3631) (0.0046) (0.0128) (0.0465) (0.0301) (0.0222) (0.0272) (0.0091) (0.0281) (0.83127) (0.0137) (0.0098) (0.0567) (0.0095) (0.0137)
0.1225 0.2652 0.05697 0.1489 0.1346 0.1071 0.1429 0.1697 0.0704 0.10385 0.3517 0.1872 0.1329 0.0409 0.0464 0.0431 0.04115 0.0501
(0.0236) (0.441) (0.0121) (0.7659) (0.0121) (0.0571) (1.7200) (0.1146) (0.0641) (0.2799) (0.4806) (0.1246) (0.0654) (0.0544) (0.0644) (0.1751) (0.0579) (0.0602)
0.0740 0.1418 0.0249 0.0835 0.1288 0.0982 0.07029 0.0592 0.0444 0.0817 0.0318 0.0567 0.1190 0.0074 0.0456 0.0121 0.04132 0.00784
(0.0242) (0.0230) (0.0078) (0.2528) (0.0075) (0.0287) (0.0932) (0.0479) (0.0425) (0.0492) (0.4155) (0.0438) (1.7531) (0.0273) (0.3677) (0.0332) (0.3678) (0.0271)

0.0555 0.0889 0.0490 0.0242 0.0172 0.0171 0.1587 0.0422 0.0178 0.0433 0.4746 0.0302 0.0394 0.0428 0.0220 0.0406 0.0320 0.0487
(0.0167) (0.0364) (0.0102) (0.1120) (0.0099) (0.0425) (0.0902) (0.0545) (0.0324) (0.0109) (0.5593) (0.0510) (0.0198) (0.0504) (0.3683) (0.0803) (0.5581) (0.0546)

0.0202 0.0296 0.0478 0.0042 0.0105 0.02003 0.0118 0.01603 0.0174 0.01644 0.0339 0.0158 0.0077 0.0093 0.0177 0.0109 0.0095 0.0102
(0.0144) (0.0267) (0.0122) (0.2194) (0.0117) (0.0276) (0.0462) (0.0109) (0.0591) (0.5101) (0.0537) (0.0105) (0.4342) (0.0237) (0.6628) (0.0603) (0.5388) (0.0251)
[Link] obs. 62 97 62 97 62 97 39 120 39 120 39 120 38 121 38 121 38 121
Threshold 17.5268 17.5268 17.5268 13.1726 13.1726 13.1726 13.0743 13.0743 13.0743

Note: Robust standard errors in parentheses. Boldface values indicate statistical significance at the 1% level.

48

Electronic copy available at: [Link]

 Table 11: Results from Cross-section Instrumental Variable (IV) Threshold Model
Threshold Variable: Government Consumption as a Percentage of GDP

Model Model 1: growth rate of rgdpch per capita Model 2: growth rate of rgdpwok per capita Model 3: growth rate of rgdpl per capita
1a 1b 1c 2a 2b 2c 3a 3b 3c
variable
L H L H L H L H L H L H L H L H L H
0.0179 0.0047 0.0186 0.00485 0.01979 0.00545
TA
(0.007) (0.004) (0.0098) (0.0024) (0.0106) (0.0019)
0.0185 0.0045 0.01757 0.0046 0.0188 0.00574
TRE
(0.009) (0.0021) (0.0076) (0.002) (0.008) (0.0029)
0.0195 0.0044 0.0198 0.0048 0.0191 0.00593
TRG
(0.0012) (0.042) (0.0045) (0.021) (0.008) (0.018)
-0.0054 -0.0196 0.0042 -0.0219 -0.0047 0.0195 -0.0186 -0.0043 -0.0396 -0.0061 -0.0428 -0.0027 -0.0318 -0.0104 -0.2942 -0.0043 -0.0302 -0.0101
(0.0025) (0.0116) (0.0110) (0.0117) (0.0019) (0.0160) (0.3883) (0.0041) (0.0223) (0.0338) (0.0233) (0.0035) (0.0367) (0.0063) (0.7662) (0.0051) (0.0356) (0.0065)
0.0012 0.0014 0.00064 0.00124 0.00014 0.00142 0.0035 0.0019 0.0025 0.0013 0.00295 0.0014 0.00092 0.00064 0.0025 0.0006 0.00106 0.00058
(0.0001) (0.00073) (0.0005) (0.0005) (0.00016) (0.00038) (0.0214) (0.0003) (0.0003) (0.0105) (0.0005) (0.0108) (0.0005) (0.0051) (0.0045) (0.0001) (0.0008) (0.0002)
0.0059 0.0044 0.0013 0.0021 0.0036 0.0028 0.0128 0.0015 0.0143 0.0077 0.0141 0.00295 0.0277 0.00104 0.0115 0.0009 0.0056 0.00088
(0.0004) (0.0017) (0.0155) (0.0014) (0.0004) (0.0016) (0.0177) (0.0010) (0.0791) (0.0525) (0.0075) (0.0017) (0.0053) (0.0006) (0.0616) (0.0014) (0.0031) (0.0007)
-0.1932 -0.0654 -0.158 -0.0660 -0.0527 -0.0721 -0.0685 -0.0475 -0.0343 -0.1111 -0.1258 -0.1199 -0.0509 -0.0863 -0.0701 -0.0741 -0.0740 -0.0877
(0.523) (0.0397) (0.404) (0.0499) (0.269) (0.0367) (0.0486) (0.187) (0.175) (0.0831) (0.0891) (0.0722) (0.215) (0.0426) (0.0598) (0.0673) (0.0675) (0.342)
0.04703 0.0384 0.0109 0.0202 0.0633 0.0299 0.0725 0.0395 0.0780 0.0133 0.0781 0.0121 0.0247 0.0062 0.0161 0.0052 0.0139 0.0059
(0.024) (0.0095) (0.0053) (0.0079) (0.017) (0.0072) (0.0392) (0.0033) (0.0142) (0.1291) (0.0125) (0.0037) (0.0130) (0.0051) (0.0061) (0.0025) (0.0221) (0.0018)
-0.00131 -0.00453 -0.0021 -0.00496 -0.0012 -0.00403 -0.0028 -0.0194 -0.0053 -0.0082 -0.0046 -0.0061 -0.0142 -0.0047 -0.0278 -0.0041 -0.0070 -0.0047
(0.0075) (0.0026) (0.0085) (0.00491) (0.0009) (0.00512) (0.0026) (0.0091) (0.0614) (0.0083) (0.013) (0.0032) (0.0122) (0.012) (0.1222) (0.019) (0.0091) (0.021)
0.0115 0.0164 0.0285 0.05265 0.0113 0.01833 0.0256 0.0102 0.0798 0.0188 0.0086 0.0063 0.0171 0.0202 0.0182 0.0149 0.0084 0.0236
(0.0061) (0.0319) (0.1658) (0.02661) (0.0065) (0.0278) (0.0177) (0.0106) (0.0670) (0.0785) (0.0696) (0.0159) (0.0389) (0.0155) (0.7681) (0.0164) (0.0561) (0.0165)
0.013 0.0391 0.0219 0.0191 0.0073 0.0314 0.0152 0.0244 0.0146 0.0105 0.0185 0.0198 0.0123 0.0112 0.0164 0.0093 0.0241 0.0120
(0.0059) (0.0034) (0.011) (0.0309) (0.0036) (0.0268) (0.9978) (0.0113) (0.0462) (0.0853) (0.0045) (0.0109) (0.0668) (0.0133) (0.0290) (0.0149) (0.0792) (0.0136)
0.0618 0.1986 0.0621 0.0872 0.0373 0.0534 0.0422 0.0379 0.0266 0.0320 0.0236 0.0322 0.0239 0.0133 0.0175 0.0178 0.0149 0.0121
(0.5164) (0.0584) (0.5320) (0.0845) (0.0253) (0.0705) (0.3470) (0.0628) (0.1405) (0.8628) (0.1229) (0.0117) (0.1758) (0.0664) (1.3167) (0.0939) (0.1532) (0.0729)
0.0024 0.1507 0.2505 0.0525 0.0301 0.0626 0.0314 0.0142 0.0203 0.0749 0.0367 0.0141 0.0681 0.0677 0.0840 0.0675 0.0894 0.0683
(0.0143) (0.0502) (0.1208) (0.0443) (0.0165) (0.0523) (0.2489) (0.0139) (0.0748) (0.4527) (0.0648) (0.0186) (0.0853) (0.0531) (0.1972) (0.0439) (0.1073) (0.0513)
0.0244 0.1151 0.3847 0.1998 0.0248 0.1964 0.4852 0.3326 0.1234 0.0261 0.1523 0.0059 0.1405 0.0128 0.1247 0.0186 0.1761 0.0137
(0.0096) (0.1647) (0.1943) (0.1424) (0.0105) (0.1452) (0.0361) (0.0557) (0.0827) (0.0559) (0.0971) (0.0746) (0.0781) (0.0404) (0.0688) (0.0466) (0.1104) (0.0443)
0.0309 0.0291 0.0467 0.0347 0.0299 0.0426 0.05919 0.0135 0.0298 0.0169 0.0314 0.0357 0.0366 0.0178 0.0247 0.0349 0.0820 0.0179
(0.0121) (0.1007) (0.1504) (0.0813) (0.0128) (0.0863) (0.6345) (0.0382) (0.1427) (0.4565) (0.1488) (0.0775) (0.0972) (0.0289) (0.0638) (0.0282) (0.1012) (0.0298)
[Link] obs. 133 26 133 26 133 26 95 64 95 64 95 64 73 86 73 86 73 86
Threshold 21.7132 21.7132 21.7132 17.6994 17.6994 17.6994 15.2362 15.2362 15.2362

Note: Robust standard errors in parentheses. Boldface values indicate statistical significance at the 1% level.

49

Electronic copy available at: [Link]