Natural Resources Research, Vol. 28, No.

4, October 2019 ( 2019)

https://doi.org/10.1007/s11053-019-09468-7

Original Paper

Does Resource Curse Really Exist in Precious Metal

Producer Countries?

1,3 2
Melike E. Bildirici and Seyit M. Gokmenoglu

Received 21 October 2018; accepted 18 February 2019

Published online: 4 March 2019

In this study, the relationship between precious metal production and economic growth is
examined for the countries producing them the most, namely Australia, Canada, Mexico,
Philippines, Peru, South Africa and USA, for the period 1963–2016 by using Markov
switching-vector error correcting models and impulse–response functions. All economies
under examination in this study are well described by Markov switching intercept
heteroscedasticity-vector error correction models. We detect both estimated intercepts and
variances differ across regimes. According to results, the effect of precious metal production
on economic growth is various. As a whole, there is a long-run relationship between each
precious metal production and economic growth. In the short run, whereas gold production
has a negative effect on USAs and Canadas economies, which supports resource curse, it
has positive one on South Africas. The effect of silver production on economic growth of
Australia and Peru is negative, but it is positive for Canada and South Africa. While copper
production has a positive effect on the economic growth of Australia and South Africa, we
find supporting evidence for resource curse for Canada, Peru and USA. We also find
insignificant correlation between any precious metals and economic growth for Mexico and
Philippines.
KEY WORDS: Dutch disease, Economic growth, MS-VEC, Precious metals, Resource curse.

INTRODUCTION Australia, Canada and Botswana. However, it can

Is it a good chance having a resource abundance
or not? What is the influence of the natural resource
abundance on economic growth? In the literature,
several empirical studies seek an answer to this
question. At first glance, having natural resource
abundance seems a blessing and a good opportunity
for start-up. Some countries use this unique oppor-
tunity to enhance their welfare such as Norway,
1
Department of Economics, Yıldız Technical University, Davut-
pasa/Istanbul, Turkey.
2
Institute of Social Sciences, Yıldız Technical University, Davut-
pasa/Istanbul, Turkey.
3
To whom correspondence should be addressed; e-mail:
melikebildirici@gmail.com
be observed that many natural resource-rich coun-
tries, particularly in Africa, Middle East and South
America, perform worse economic performance
relative to the resource-poor countries in the same
economic development level. This phenomenon is
known as resource curse. It simply describes the
failure to benefit from natural resource revenues for
sustained growth, but there is no consensus in the
literature whether the curse really exists. In the last
two decades, while about 40% of econometric
studies support resource curse, 20% of those studies
suggest natural resources have a positive impact on
economic growth and the rest found insignificant
results (Havranek et al. 2016). The contrasting
findings of previous studies stem from the indicators

1403
1520-7439/19/1000-1403/0  2019 International Association for Mathematical Geosciences
1404
that were used for resource richness, the type of
resources, the countries analyzed, the econometric
method and the period covered.
The first empirical study supporting resource
curse was conducted by Sachs and Warner (1995).
They used the ratio of primary product exports to
GDP in the initial year as a measure of resource
richness. Many researchers adopted this method and
used this indicator for resource richness (e.g., Leite
and Weidman 1999; Sala-i-Martin and Subramanian
2003; Mehlum et al. 2006; Brunnschweiler and Bulte
2008; Kim and Lin 2017). The share of mineral
production in GDP, the share of primary product
exports in total exports, land per capita were also
firstly used by Sachs and Warner (1995), and then
some other studies utilized them. Gylfason (2001),
Stijns (2006), Cockx and Francken (2014) used the
share of natural capital in total capital as a proxy for
resource richness. Total natural capital and subsoil
wealth per capita were used by Brunnschweiler
(2008), oil income per capita was used by Wigley
(2017) and Damette and Seghir (2018), oil and gas
value per capita were used by OConnor et al.
(2018). Arezki and Ploeg (2011) employed natural
capital stock. Some studies like Atkinson and
Hamilton (2003), Collier and Hoeffler (2009),
Bhattacharyya and Hodler (2010) and De Soysa and
Gizelis (2013) employed the share of natural rents in
GDP. Arezki and Brückner (2011), Antonakakis
et al. (2017) and Arin and Braunfels (2018) used oil
rents, and Amiri et al. (2018) used natural resource
rents. While Cavalcanti et al. (2011) employed the
real value of oil production for natural resource
abundance, Smith (2015) measured it with natural
resource production data.
The outcomes of the studies on resource rich-
ness and economic growth are also diverse in terms
of economic analysis method conducted. If countries
have similar natural resources but the outcome is
not same, what makes the difference? At this point,
there are several channels used to explain the curse,
and the most emphasized one is institutional quality
in a country. However, in this article, our focus is on
the direct effect of resource abundance on economic
growth, not on the transmission channel of the ef-
fect. We therefore did not take into consideration
transmission channels in analyzing the relationship.
In this article, we aim to contribute to the lit-
erature in the following ways. Firstly, we proxy re-
source richness with production data. To the best of
our knowledge, only Stijns (2005) and Smith (2015)
have used this indicator for resource abundance.
Bildirici and Gokmenoglu
Secondly, to distinguish between types of resources,
we focused only on precious metals. The countries
with the highest production levels of precious metals
as of 2016 are included for the analysis. When
selecting countries we excluded those with inade-
quate data for the analysis. Three of the countries
under study are developed ones, and four of them
are developing ones. These properties of the coun-
tries also enable us to determine whether the effect
of resource abundance on the economic growth
differs according to the developmental level of the
countries. Finally, as the salient feature of the study,
unlike other studies on this field, we use Markov
switching-vector error correcting (MS-VEC) models
to investigate the relationship.
The second section discusses briefly the litera-
ture on natural resource richness and economic
growth. In the third section, data description and
econometric methods are presented. The fourth
section covers results of the analysis. Finally, the last
section presents conclusions and economic policy
implications.
LITERATURE AND CHANNELS
TO RESOURCE CURSE
To economize on space we introduce main
arguments and studies in this section. Readers are
referred to Frankel (2010), Deacon (2011) and Ste-
vens (2015) for detailed literature survey. The
Economist, published in 1977, drew attention to the
decline of the Netherlands manufacturing sector
after the discovery of natural gas resources in 1960s
and coined this event as Dutch Disease. Dutch
Disease claimed that a boom in natural resources
attracted foreign currency into the economy giving
rise to appreciation of the local currency, drawing
capital and labor force to resource and service sec-
tors (Corden 1984). Both of these crowd out the
manufacturing sector and make it uncompetitive,
hindering sustainable economic growth. It was firstly
formalized by Corden and Neary (1982). Gelb
(1988) analyzed the impact of oil revenues on eco-
nomic growth especially in developing countries and
found negative relationship between them and term
the negative relationship as resource curse. Auty
(1993) popularized the term in his sensational study
that found mineral-rich countries, like the oil-rich
ones, have slower growth rate compared with those
lacking in resource richness in the same develop-
ment level. Subsequently, the question whether
Does Resource Curse Really Exist
having resource abundance is a curse became very
popular in economic growth literature. Sachs and
Warner (1995) were the first to analyze the curse
empirically. They showed, by using cross-sectional
data analysis, that countries having high share of
primary product exports in GDP in 1971 had lower
economic growth performance during 1971–1989.
Sachs and Warner (1997) changed the base year to
1970 and got the same result. Likewise, studies by
Sachs and Warner (2001), Leite and Weidman
(1999), Dietz et al. (2007), Ross (2001), Sala-i-Mar-
tin and Subramanian (2003), Mehlum et al. (2006),
Moradbeigi and Law (2017) and Damette and Seghir
(2018) found supporting results for resource curse.
In contrast, studies by Davis (1995), Stijns (2006),
Brunnschweiler (2008), Alexeev and Conrad (2011),
Cavalcanti et al. (2011), Cotet and Tsui (2013),
Smith (2015), Havranek et al. (2016), Arin and
Braunfels (2018) and Dobra et al. (2018) claim
natural resources have positive effects on economic
growth.
In addition to Dutch Disease thesis, the re-
source curse literature generally suggests three other
channels leading to slow economic growth: price
volatility, rentier-state and the crowd-out effect. As
for the first channel, Prebisch (1950) and Singer
(1950) argued that mineral and agricultural com-
modity prices follow a downward trend in the long
run relative to those of manufactured and other
commodities. Ross (2003) indicated that interna-
tional prices of primary commodities have fluctuated
more compared to those of manufactured ones over
the past century. If a country mainly relies on nat-
ural resource export revenues, the volatility created
by the fluctuations in prices of these resources in the
world market and the sudden and sharp decreases in
the terms of trade can cause economic shocks. The
unpredictable fluctuations in revenues create
uncertainty and, in turn, have undoubtedly had a
negative impact on a resource-dependent countrys
economy.
DATA DESCRIPTION
AND ECONOMETRIC METHODS
Data Description
We have compiled precious metals production
data by using U.S. Geological Survey Minerals
Yearbooks issued from 1960 to the present. The
countries with the highest production levels as of
1405
2016 were included for the analysis. We excluded the
countries with inadequate data for the analysis. We
have chosen three developed (Australia, Canada,
and USA) and four developing countries (Mexico,
Philippines, Peru and South Africa), which have the
highest production levels in 2016 for analysis. We
have acquired gross domestic production data from
the World Bank database. The variables used in
analysis were economic growth, gold production,
silver production, copper production symbolized
as y, gold, slv and cop, respectively. We have
transformed all data to log ratio to minimize skew-
ness as ly ¼ logððyt  yt1 Þ=yt1 Þ, lgold ¼ log
ððgoldt  goldt1 Þ=goldt1 Þ, lslv ¼ logððslvt  slvt1 Þ=
slvt1 Þ and lcop ¼ logððcopt  copt1 Þ=copt1 Þ,
where t and t1 denote current value and the pre-
vious value of the variables, respectively.
Econometric Methods
The empirical literature on resource curse pre-
dominantly depends on cross-sectional regressions,
which suffer from various problems of endogeneity
bias as well as some limitations. They are incapable
of identifying dynamics and testing the short- and
long-run effects of natural resource abundance on
economic growth. However, the Markov switching
model used in this study relaxes the constricting
postulation that all observations in a series are
drawn from a Gaussian distribution with constant
mean and variances in the period. In this article,
empirical results were evaluated in three stages.
In the first stage, the unit root tests were applied
to decide whether variables are integrated of order
one (I(1)). If they are, we checked in the second
stage if there is any co-integration relationship be-
tween variables by using the Johansens co-integra-
tion test (Johansen 1991, 1995). If we detect co-
integration relationship, we used in the third stage a
two-step estimation method suggested by Krolzig
(1996) for MS-VEC analysis. In the first step, we
obtained the co-integration vector and equilibrium
errors by estimating linear vector error correcting
(VEC) model and then in the second step we treated
the equilibrium errors as exogenous variable in
Markov switching-vector autoregressive (MS-VAR)
model, becoming MS-VEC model. We used the
expectation maximization (EM) algorithm to esti-
mate MS-VEC models, and it yields total regime
numbers, regime durations and probabilities of
transition as well as adjustment rate of error cor-
1406
recting parameter. We defined the models for each
regime as follows.
For crisis and/or high volatility regime (regime
1):
X
n X
n any precious metal productions, we can say the
Dyt ¼ a0 ðst ¼ 1Þ þ a1;k ðst ¼ 1ÞDytk þ
k¼1 k¼1
¼ 1ÞDxtk þ by ðst ¼ 1Þectt1 þ uyt ðst ¼ 1Þ
X
n X
n
Dxt ¼ a0 ðst ¼ 1Þ þ a1;k ðst ¼ 1ÞDxtk þ
k¼1 k¼1
¼ 1ÞDytk þ bx ðst ¼ 1Þectt1 þ uxt ðst ¼ 1Þ
For the medium growth and/or the medium
volatility regime (regime 2):
X
m X
m  0
Dyt ¼ a0 ðst ¼ 2Þ þ a1;k ðst ¼ 2ÞDytk þ
k¼1 k¼1
y Xq
¼ 2ÞDxtk þ by ðst ¼ 2Þectt1 þ ut ðst ¼ 2Þ
X
n X
n
Dxt ¼ a0 ðst ¼ 2Þ þ a1;k ðst ¼ 2ÞDxtk þ
k¼1 k¼1
¼ 2ÞDytk þ bx ðst ¼ 2Þectt1 þ uxt ðst ¼ 2Þ
For the high growth and/or low volatility regime
(regime 3):
X
m X
m
Dyt ¼ a0 ðst ¼ 3Þ þ a1;k ðst ¼ 3ÞDytk þ
k¼1 k¼1
y
¼ 3ÞDxtk þ by ðst ¼ 3Þectt1 þ ut ðst ¼ 3Þ
X
n X
n ables may have influence on the relationships
Dxt ¼ a0 ðst ¼ 3Þ þ a1;k ðst ¼ 3ÞDxtk þ
k¼1 k¼1
¼ 3ÞDytk þ bx ðst ¼ 3Þectt1 þ uxt ðst ¼ 3Þ
where ectt1 is the error correction term derived
from the co-integration relationship and b is the
adjustment parameter showing how quickly vari-
ables approximate to the long-run equilibrium. If the
adjustment parameter is negative and statistically
significant, we can say that there is an error correc-
tion mechanism in the model. We also utilized
Bildirici and Gokmenoglu
orthogonalized impulse–response functions to
investigate impulse–response relation between pre-
cious metal productions and economic growth. If we
detect a response in economic growth to a change in
a2;k ðst precious metal production causes to economic
growth.
ð1Þ
MS-VEC Analysis
a2;k ðst After Hamilton (1989) introduced Markov
switching models for modeling economic time series
with a permanent and a cyclical component, Krolzig
ð2Þ (1997) extended it to vector autoregressive (VAR)
case. Krolzig (1997) also brought in MS-VEC model
with shifts in the drift dðst Þ and in the equilibrium
mean lðst Þ. The suggested MS(.)-VEC(.) model is:

a2;k ðst Dyt  dðst Þ ¼ a b xt1  lðst Þ  kðt  1Þ
þ Ck ðDxtk  dðst ÞÞ þ ut ; ut =st
ð3Þ k¼1
X 
 NID 0; ðst Þ :
a2;k ðst ð7Þ
P
symbolizes the residual variance in the re-
gimes. lðst Þ defines dependency of the mean l of the
ð4Þ K–dimensional time series vector on the regime
variable st .
In the Markov switching intercept
heteroscedasticity-vector error correction (MSIH-
a2;k ðst VEC) model, I denotes intercept and H denotes
0
heteroscedasticity in the residuals. b dðst Þ ¼ 0 is
E½Dxt jst  ¼ dðst Þ ¼ b? d ðst Þ. xt embraces economic
ð5Þ growth ðyt Þ, gold production ðgoldt Þ, silver produc-
tion ðslvt Þ and copper production ðcopt Þ. The vari-
a2;k ðst between yt ; goldt ; slvt and copt . However, the ef-
fects can be captured by shifts in the mean or
intercept or variance in MS-VEC models. Regime
ð6Þ variable is directed by a Markov chain, expressed by
the transition probabilities pij :
  1 
Pr st fst1 g1i¼1 ; fyt1 gi¼1 ¼ Pr fst jst1 ; qg ð8Þ
It indicates that situation in period t will depend
on the situation in period t  1. st is presumed to
follow an irreducible ergodic M state Markov pro-
cess with a transition matrix expressed as (Bildirici
2012a, b),
Does Resource Curse Really Exist
2 3
p11 p12  p1M
6 p21 p22  p2M 7 empirical support for long-run relationship between
6 7
P¼6 .. .. .. 7 ð9Þ
4 . .  . 5
pM1 pMM  pMM
The Markov chain is ergodic and irreducible; a
two-state Markov chain with transition probabilities
pij with unconditional distribution is showed in the
above equations. To estimate the Markov switching
(MS) models, the EM suggested by Hamilton is
employed. In these models, yt is obvious, but the
regime variable is not observable and value of the
variable could be derived by making use of the real
values of yt (Bildirici and Gökmenoğlu 2017). This
inference can be demonstrated as
nit ¼ Pr ½st ¼ ijXt ; h ð10Þ
in which i = 1,2; Xt represents information set, and h
is the parameter vector to be estimated.
The conditional log-likelihood can be given as
X symbolizes moderate growth, and regime 3 refers to
log f ðy1 ; y2 ; . . . ; yT jy0 ; hÞ ¼ log f ðyt jXt1 ; hÞ ð11Þ
EMPIRICAL RESULTS
Unit Root Tests and Co-integration Tests
We applied the augmented Dickey–Fuller
(ADF) unit root, Kwiatkowski–Phillips–Schmidt–
Shin (KPSS) stationary test and nonlinear Kapetan-
ios–Shin–Snell (KSS) (Kapetanios et al. 2003) unit
root test to check whether selected series are inte-
grated of order one or not. The ADF test is not reli-
able if the sample period has structural breaks, so it
may produce faulty results. To this end, we also em-
ploy KPSS stationary test and nonlinear KSS unit root
test. According to ADF and KSS unit root test results
in Table 1, the existence of a unit root cannot be re-
jected at the level for all countries. In contrast, first
differences of ly, lgold, lslv and lcop seem to be sta-
tionary at 5% significance level averagely. According
to the KPSS stationary test results, we can reject the
null hypothesis, indicating that the series is stationary
for all variables of all countries at the level, but we can
accept the null hypothesis when we use the first dif-
ferences of the variables. Thus, it is evident all vari-
ables are I(1) countries. Therefore, we proceeded to
examine co-integration using the Johansen test. From
the test results (Table 1), it is clear that the null
hypothesis of no co-integration can be rejected for all
1407
variables at the 0.05 level. Therefore, we found
precious metal productions and GDP. Because we
detected co-integration in the series, we applied MS-
VEC analysis in the next step.
Business Cycle Characteristics
We estimated MS-VEC models by using EM
algorithm. The business cycle dates determined by
these models and Economic Cycle Research Insti-
tute (ECRI) are given in Table 2, together with the
coincidence ratios proposed by Altuğ and Bildirici
(2010) to assess the achievement of model dating. To
calculate the ratios we compared our estimations by
ECRIs, tolerating maximum two or three diver-
gences, and calculated the average coincidence as
72%.
In the study, regime 1 denotes to crisis, regime 2
high growth. Regime 1 depicts very clearly the
recessions of 1971, 1974, 1980–1982, 1989–1991 cri-
sis, 1991–1992 and the recent 2008 crisis as a whole.
It is seen overall that the crisis (regime 1) period is
shorter than the other ones. Furthermore, the esti-
mated models show high persistency for the regimes.
Subsequently, it indicates a strong proof for business
cycle asymmetries in the countries under examina-
tion. The transition probability matrix is also ergodic
and cannot be reduced, which confirms stationarity
of the regime (Hamilton 1990).
MS-VEC Models and Impulse–Response Functions
To determine the number of regimes, linear
VEC model was tested against a two-regime MS-
VEC model. In the same way, we tested two regimes
against three regimes. All models were selected
based on the Akaike information criteria (AIC) and
likelihood ratio (LR) test statistics, and we found
that data could be well described by MSIH-VEC
models, in which intercepts and variance change
across regimes. In these models, autoregressive
parameters do not change across the regimes, but
intercepts and variance of the errors do in each re-
gime. Besides, the contemporaneous correlation
matrix of the residuals is regime switching. The
existence of contemporaneous correlation suggests
taking into consideration the orthogonalized im-
pulse–response functions (IRFs). In this sense, we
 1408

Table 1. Unit root test results for the analyzed countries

Variables Australia Canada Mexico Peru

ADF KPSS KSS ADF KPSS KSS ADF KPSS KSS ADF KPSS KSS

lyt  1.958 0.129*  0.50(C3)  0.98 0.218***  1.50 (C3)  1.917 0.240***  0.98(C3)  1.905 0.144*  2.25(C3)
dlyt  5.923*** 0.102  11.36***(C1)  4.64*** 0.068  9.5***(C1)  5.916*** 0.085  2.45**(C1)  4.549*** 0.078  19.18***(C1)
lcopt  2.137 0.179**  1.85(C1)  2.46 0.186**  1.80 (C1)  3.145 0.274***  2.18(C3)  2.046 0.190**  2.01(C2)
dlcopt  7.621*** 0.099  15.54***(C1)  7.83*** 0.098  10.58***(C1)  5.853*** 0.030  5.79***(C1)  8.194*** 0.070  8.22***(C2)
lgoldt  1.716 0.280***  2.40(C2)  2.38 0.124*  0.15 (C2)  1.749 0.239***  1.63(C3)  1.994 0.168**  1.79(C1)
dlgoldt  3.925*** 0.087  5.15***(C2)  3.218* 0.039  9.23***(C2)  6.611*** 0.051  7.44***(C1)  5.852*** 0.028  19.36***(C1)
lslvt  2.404 0.165**  2.83(C3)  0.698 0.187**  1.41 (C2)  1.732 0.135*  1.17(C1)  2.792 0.158**  2.16(C3)
dlslvt  7.892*** 0.048  7.89***(C1)  6.263*** 0.106  11.48*** (C2)  6.805*** 0.073  14.16***(C1)  9.231*** 0.074  7.489***(C1)

Philippines South Africa USA

ADF KPSS KSS ADF KPSS KSS ADF KPSS KSS

lyt  2.039 0.152**  1.17(C3)  2.280 0.153**  1.71(C3)  2.207 0.191**  1.66(C3)
dlyt  4.105*** 0.042  10.15***(C1)  4.456*** 0.040  7.25***(C1)  5.760*** 0.068  18.82***(C1)
lcopt  2.05 0.150**  1.65(C3)  2.340 0.233***  2.50(C2)  2.645 0.196**  1.39(C1)
dlcopt  5.93*** 0.061  9.99***(C1)  7.838*** 0.085  4.12***(C2)  8.130*** 0.053  11.12***(C1)
lgoldt  1.957 0.147**  1.82(C1)  2.062 0.230***  1.02(C1)  1.694 0.122*  1.12(C1)
dlgoldt  7.504*** 0.056  9.25***(C1)  5.141*** 0.079  9.78***(C1)  3.806** 0.045  15.55***(C1)
lslvt  1.838 0.215***  1.18(C2)  .161 0.173**  0.36(C1)
dlslvt  7.613*** 0.098  3.01**(C2)  6.759*** 0.086  8.44***(C1)

Johansen co-integration trace statics

AUST r = 0 44.08*r £ 1 17.13 r £ 2 4.84 PHL r = 0 26.47* r £ 1 8.07 r £ 2 1.01

CAN r = 0 56.83*r £ 1 30.84 r £ 2 13.55 SAF r = 0 69.41* r £ 1 28.55* r £ 2 13.12*
MEX r = 0 53.06*r £ 1 22.80 r £ 2 8.61 USA r = 0 53.24 *r £ 1 20.89 r £ 2 4.78
PER r = 0 48.44*r £ 1 20.76 r £ 2 5.19

ADF and KPSS tests are calculated for intercept + trend assumption. Critical values for ADF test: 1% level:  4.137279, 5% level:  3.495295, 10% level:  3.176618; for KPSS test: 1% level:
0.216, 5% level : 0.146, 10% level : 0.119; for KSS test C(1) 1% level:  2.82, 5% level:  2.22, 10% level:  1.92; C(2) 1% level:  3.48, 5% level:  2.93, 10% level:  2.66; C(3) 1% level:
 3.93, 5% level:  3.40, 10% level:  3.13. Cases 1, 2 and 3 (C1, C2 and C3) represent raw, demeaned and detrended data selections, respectively. 1%, 5% and 10% significance levels are
denoted by ***,**,*, respectively
*denotes rejection of the null hypothesis of no co-integration at the 0.05 level for Johansen co-integration trace statics
Bildirici and Gokmenoglu
Does Resource Curse Really Exist 1409

Table 2. Dating analysis (year: quarter)

Australia Canada Mexico Peru

Detected ECRI Detected ECRI Detected ECRI Detected ECRI

1980:1–1982:1 1981:2–1982:4 1963:1–1965:1 1974:2–1975:1 1964:1–1964:1 1982:1–1983:3 1967:1–1969:1 NA

1990:1–1992:1 1990:1–1992:1 1974:1-1975:1 1981:2–1983:2 1966:1–1969:1 1985:4–1986:4 1975:1–1979:1
2008:1–2010:1 2008:1–2009:3 1981:1-1984:1 1990:2–1991:4 1975:1–1975:1 1992:4–1993:4 1979:1–1982:1
2015:1–2015:1 1990:1-1992:1 1982:1–1983:1 1994:4–1995:3 1983:1–1986:1
2008:1-2008:1 1985:1–1986:1 2000:3–2000:4 1998:1–2000:1
1992:1-1995:1 2003:3–2003:4 2004:1–2007:1
2000:1–2000:1 2008:2–2009:2 2009:1–2009:1
2008:1-2009:1 2015:1–2016:1
Coin ± : 2/3 = 66.6% Coin ± : 3/3 = 100% Coin ± : 5/7 = 71.4% Coin ± : NA

Philippines South Africa USA

Detected ECRI Detected ECRI Detected ECRI

1970:1–1971:1 NA 1976:3–1977:4 1967:1–1967:1 1969:4–1970:4

1983:1–1983:1 1981:4–1983:1 1974:1–1975:1 1973:4–1975:1
1997:1–1997:1 1984:4–1986:1 1979:1–1980:1 1980:1–1980:3
1999:1–2000:1 1985:1–1985:1 1986:3–1986:4 1982:1–1982:1 1981:3–1982:4
1997:1–2001:1 1991:2–1991:3 1990:1–1991:1 1990:3–1991:1
2006:1–2008:1 1997:3–1998:2 2001:1–2002:1 2001:1–2001:4
2012:1–2015:1 2007:4–2009:2 2008:1–2010:1 2007:4–2009:2
2010:2–2010:4
Coin ± : NA Coin ± : 3/8 = 37% Coin ± : 6/7 = 85%

also generated IRFs that are different for every re-
gime. According to the AIC criteria and LR test
statics, we found that the best model for USA and
Australia is MSIH(3)-VEC(1) model; for Philip-
pines is MSIH(3)-VEC(2) model; for Canada,
Mexico, South Africa and Peru is MSIH(2)-VEC(1)
model. The estimated parameters along with the t-
statistics in the models are presented in Tables 3, 4,
5, 6, 7, 8 and 9. The error correction terms are sig-
nificant in the regressions on average; the variables
adjust toward the equilibrium.
The results of analysis for Australia (Table 3)
indicate that regime 1, regime 2 and regime 3 tend to
last 2.32, 3.53 and 2.25 years, respectively. Regime 2
is the most insistent one, and the remaining likeli-
hood in this regime is 0.71 (71%). The computed
probability (Prob(st = 2|st1=1) = 0.19) indicates a
low likelihood that a crisis could be followed by a
moderate growth regime. When the economy is in
crisis, the likelihood of defeating the crisis and
shifting to moderate growth regime is 0.2418
(24.18%), relatively higher than that of passing to
the higher growth regime. However, staying in crisis
regime is more likely with probability of 0.5693
(56.93%). In the short run, there is a statistically
significant positive correlation between GDP and
copper and negative one between GDP and silver. It
is seen that the coefficient of the error correction
term for sliver equation is not statistically significant,
indicating that silver looks like to be weakly
exogenous. In contrast, for gold and copper equa-
tions they are positive and significant, which means
there is no adjustment to long-run equilibrium. The
results also show that there are differences in the
amount of variance of the error terms across re-
gimes. They are larger in regime 1 than the ones in
other regimes, which reflects asymmetry in the
variance of the error terms. When we investigated
IRFs in Figure 1, we found that if one standard
deviation shock is given to the gold production, the
response of the economic growth of Australia in
regime 1 is always negative but gradually increases
toward zero for five periods. For copper and silver
productions, there is an instantaneous effect on
economic growth, and this effect is always positive.
In the second regime, one standard deviation shock
to gold, silver and copper production gives rise to
negative effect on economic growth, which taper off
to zero after a while. In the third regime, while the
response of economic growth to one standard devi-
ation to gold and copper production is positive, it is
negative for silver production.
The results of analysis for Canada (Table 4)
indicate that crisis regime is fully consistent with
 Table 3. Australia, MSIH(3)-VEC(1) model (Estimation sample: 1963–2015)
1410

Variables Regime 1 Regime 2 Regime 3

dlgoldt dlcopt dlslvt dlgdpt dlgoldt dlcopt dlslvt dlgdpt dlgoldt dlcopt dlslvt dlgdpt

Regime-  0.011  0.007 0.00004 0.030 0.017 0.02 0.016 0.029 0.029 0.039 0.063 0.042
dependent ( 1.14) ( 0.53) (0.003) (2.67)*** (1.36)* (2.40)*** (2.12)** (4.09)*** (4.38)*** (2.57)*** (5.88)*** (3.06)***
intercepts
Short-run dynamics
dlgoldt1 0.550  0.169  0.018  0.037
(18.39)*** ( 2.47)*** ( 0.53) ( 0.81)
dlcopt1  0.128  0.038  0.159 0.036
( 2.40)*** ( 0.35) ( 2.54)*** (2.48)***
dlslvt1  0.219  0.058  0.092  0.105
( 5.33)*** ( 2.59)*** ( 1.92)** ( 1.72)**
dlgdpt1 0.006 0.145  0.194 0.128
(0.13) (2.02)** ( 3.44)*** (1.67)**
Error correction
ecmt1 0.229 0.166  0.004  0.254
(3.88)*** (1.64)* (0.06) ( 0.58)
Standard errors
r 0.348 0.243 0.152 0.14 0.063 0.034 0.037 0.031 0.021 0.048 0.036 0.047
Transition Reg. 1 Reg. 2 Reg. 3 Durations
prob.
Reg. 1 0.5693 0.2418 0.1518 2.32
Reg. 2 0.1952 0.7170 0.0878 3.53
Reg. 3 0.2066 0.2368 0.5567 2.25

Regime 1 Regime 2 Regime 3

dlgoldt dlcopt dlslvt dlgoldt dlcopt dlslvt dlgoldt dlcopt dlslvt dlgoldt dlcopt dlslvt

Contemporaneous correlations
dlgoldt 1.000 0.483  0.380  0.885 1.000  0.346 0.055  0.254 1.000  0.041  0.95 0.815
dlcopt 0.483 1.000 0.161  0.022  0.346 1.000 0.549  0.165  0.041 1.000  0.026 0.025
dlslvt  0.380 0.161 1.000 0.524 0.055 0.549 1.000  0.372  0.950  0.026 1.000  0.914
Bildirici and Gokmenoglu
Does Resource Curse Really Exist 1411

1.000

Chi(80) = 81.7186 [0.4256], Vector normality test: Chi(8) = 20.0531 [0.0101] *, Vector hetero test: Chi(100) = 118.9703 [0.0949], F(100,205) = 1.1323 [0.2285], Vector hetero-X test:
t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 424.3146, linear system: 357.1591; AIC
criterion:  13.7044, linear system:  12.4292; HQ criterion:  12.7262, linear system:  11.9401; SC criterion:  11.1528, linear system:  11.1534; LR linearity test: 134.3110,
Chi(28) = [0.0000]**, Chi(34) = [0.0000]**, DAVIES = [0.0000]**; StdResids: Vector portmanteau(6): Chi(80) = 84.6593 [0.3395]; Vector normality test: Chi(8) = 4.4172 [0.8177]; Vector

Chi(200) = 189.5057[0.6917], F(200,175) = 0.7316[0.9838]; VAR Errors: Vector portmanteau(6): Chi(80) = 77.1958 [0.5681], Vector normality test: Chi(8) = 13.7014[0.0899], Vector hetero
hetero test: Chi(100) = 96.8913[0.5694], F(100,205) = 0.7834[0.9149]; Vector hetero-X test: Chi(200) = 182.013[0.8144], F(200,175) = 0.6447[0.9987]; PredError: Vector portmanteau(6):
dlslvt
ECRI recession dates. The possibility of staying in
crisis regime while it is in the same regime is 57%
and that of moderate growth regime is 82%, which
show the persistency of regimes. The computed
 0.914
dlcopt

probability of switching from regime 1 to regime 2

(Prob(st = 2|st1=1)) is = 0.42 which reflects a good
Regime 3

possibility that a crisis is followed by a period of

growth. The results determined by the model show
dlgoldt

that regime 1 and regime 2 tend to last 4.11 and

0.025

2.64 years, respectively. In the short run, there is a

statistically significant positive correlation between
GDP and silver and negative one between GDP and
copper and gold. As for the error correction coeffi-
0.815
dlslvt

test: Chi(100) = 118.1161[0.1043], F(100,205) = 1.0958 [0.2905], Vector hetero-X test: Chi(200) = 190.2369[0.6783], F(200,175) = 0.7096 [0.9905]

cients, it is insignificant for gold, which means it is

weakly exogenous, but for copper and silver equa-
tions they are positive and significant implying no
dlcopt

adjustment to long-run equilibrium. As for the IRFs

1.000

in Figure 2, both in the crisis regime and in growth

regimes, one standard deviation change in gold and
copper productions negatively affects economic
 0.372
dlgoldt

growth. In contrast, the response to silver produc-

tion in both regimes is positive for two periods and
Regime 2

then becomes negative.

The estimated model for Mexico (Table 5)
indicates that average duration time for crisis regime
 0.165
dlslvt

is 2.07 years and likelihood of passing to growth

regime is 0.482 (48.2%). The stickiest regime is the
second one with 3.85 years of duration. Intuitively,
the possibility of remaining in this regime is very
 0.254
dlcopt

high, 0.74. When the economy is in growth regime,

the possibility of shifting to the crisis regime is 0.26.
The possibility of staying in crisis regime while it is
in the same regime is 51.80% and that of moderate
dlgoldt

growth regime is 74%, indicating the persistence of

1.000

regimes. The results also show that there is asym-

metry in the amount of variance of the error term as
evidenced in the larger variance in regime 1 com-
pared to those of regime 2 and regime 3. Error
0.524
dlslvt

correcting parameters are positive and statistically

Regime 1

significant for gold and copper equations, meaning

there is no adjustment to long-run equilibrium. For
 0.022
dlcopt

the other equations, error correcting parameters are

negative but insignificant, meaning they are weakly
exogenous. In the short run, there is no significant
correlation between any precious metals and eco-
nomic growth. According to the IRFs in Figure 3, in
 0.885
dlgoldt

the crisis and growth regime, the effect of one

Table 3 continued

standard deviation shock to copper and silver pro-

ductions on economic growth is zero. As for the gold
production in the crisis regime, the same shock has a
dlgdpt

positive effect on economic growth for only one

period, and then it becomes negative for two periods
 1412

Table 4. Canada, MSIH(2)-VEC(1) model (Estimation sample: 1964–2016)

Variables Regime 1 Regime 2

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Regime-dependent inter- 0.003 (0.554) 0.011 (4.372)***  0.0007  0.027 0.021 0.022  0.022 0.0009
cepts (0.098) ( 2.688)*** (2.162)** (3.681)*** ( 1.69)** ( 0.055)
Short-run dynamics
dlgoldt1 0.615 (6.998)***  0.232 0.232 (1.766)** 0.085 (0.577)
( 4.988)***
dlgdpt1  0.261 0.494 (7.197)*** 0.433 (2.285)** 0.398 (1.824)**
( 2.011)**
dlcopt1 0.003 (0.033)  0.128  0.160  0.200 ( 1.183)
( 1.894)** ( 1.077)
dlslvt1 0.204 (3.262)*** 0.076 (2.368)**  0.072 0.091 (0.849)
( 0.785)
Error correction
ecmt1  0.058 ( 0.384) 0.811 (8.674)*** 0.291 (1.065) 1.457 (5.285)***
Standard errors
r 0.026 0.010 0.031 0.048 0.028 0.022 0.037 0.060
Transition prob. Reg. 1 Reg. 2 Durations
Reg. 1 0.57 0.42 4.11
Reg. 2 0.17 0.82 2.64

Regime 1 Regime 2

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Contemporaneous correlations
dlgoldt 1.000  0.032 0.086  0.314 1.000  0.374  0.420 0.654
dlgdpt  0.032 1.000 0.211 0.444  0.374 1.000 0.128  0.753
dlcopt 0.086 0.211 1.000 0.418  0.420 0.128 1.000 0.126
dlslvt  0.314 0.444 0.418 1.000 0.654  0.753 0.126 1.000

t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 465.7916, linear system:
446.5644; AIC criterion:  15.6902 linear system:  15.5685; HQ criterion:  14.9755 linear system:  15.0824; SC criterion:  13.8315 linear system:  14.3045; LR linearity test:
38.4544 Chi(14) = [0.0004]** Chi(16) = [0.0013]** DAVIES = [0.0125] *; StdResids: Vector portmanteau(6): Chi(80) = 96.5578 [0.1002]; Vector normality test: Chi(8) = 15.9365
[0.0433] *; Vector hetero test: Chi(100) = 85.2324 [0.8538] F(100,212) = 0.6631 [0.9895]; Vector hetero-X test: Chi(200) = 180.5349 [0.8348] F(200,184) = 0.6535 [0.9984];
PredError: Vector portmanteau(6): Chi(80) = 98.7076 [0.0765]; Vector normality test: Chi(8) = 9.9000 [0.2721]; Vector hetero test: Chi(100) = 111.8634 [0.1964]
F(100,212) = 0.9571 [0.5924]; Vector hetero-X test: Chi(200) = 208.0339 [0.3337] F(200,184) = 0.8795 [0.8133]; VAR Error: Vector portmanteau(6): Chi(80) = 86.5375 [0.2892];
Vector normality test: Chi(8) = 17.4060 [0.0261] *; Vector hetero test: Chi(100) = 102.0327 [0.4247] F(100,212) = 0.8213 [0.8667]; Vector hetero-X test: Chi(200) = 199.9213
[0.4883] F(200,184) = 0.8144 [0.9225]
Bildirici and Gokmenoglu
 Table 5. Mexico, MSIH(2)-VEC(1) model (Estimation sample: 1964–2016)

Variables Regime 1 Regime 2

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Regime-dependent inter-  0.051 0.012 (0.45)  0.040 ( 0.26)  0.008 ( 0.73) 0.005 0.065  0.003 0.017
cepts ( 2.29)** (0.34) (2.99)*** ( 0.11) (2.05)**
Short-run dynamics
dlgoldt1  0.231 ( 1.62)*  0.211 ( 1.06)  0.342  0.028 ( 0.36)
( 1.39)*
dlgdpt1  0.259 0.038 (0.26) 0.160 (0.83)  0.209
( 2.29)** ( 3.49)***
Does Resource Curse Really Exist

dlcopt1 0.034 (2.25)** 0.009 (0.41)  0.035 0.005 (0.69)

( 1.54)*
dlslvt1 0.502 (1.83)**  0.101 ( 0.25) 0.142 (0.31) 0.115 (0.80)
Error correction
ecmt1 0. 222 (2.45)***  0.402*** 0.0002 (1.74)**  0.0001
( 3.53) ( 1.73)**
Standard errors
r 0.060 0.068 0.641 0.033 0.041 0.058 0.062 0.021
Transition prob. Reg. 1 Reg. 2 Durations
Reg. 1 0.5180 0.4820 2.07
Reg. 2 0.2600 0.7400 3.85

Regime 1 Regime 2

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Contemporaneous correlation
dlgoldt 1.000 0.104  0.064 0.574 1.000  0.440 0.065 0.275
dlgdpt 0.104 1.000  0.043 0.253  0.440 1.000 0.109  0.151
dlcopt  0.064  0.043 1.000  0.172 0.065 0.109 1.000 0.194
dlslvt 0.574 0.253  0.172 1.000 0.275  0.151 0.194 1.000

t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 381.0563, linear system:
318.6951; AIC criterion:  12.0791, linear system:  10.5267; HQ criterion:  11.1152, linear system:  9.8794; SC criterion:  9.5650, linear system:  8.8382; LR linearity test:
124.7226, Chi(20) = [0.0000]**, Chi(22) = [0.0000]**, DAVIES = [0.0000]**; StdResids: Vector portmanteau(6): Chi(125) = 163.1337[0.0124]*; Vector normality test:
Chi(10) = 28.4616[0.0015]**, Vector hetero test: Chi(150) = 129.5747[0.8846], F(150,202) = 0.6329[0.9984], Vector hetero-X test: Chi(300) = 278.3107 [0.8107],
F(300,188) = 0.6587[0.9994]; PredError: Vector portmanteau(6): Chi(125) = 204.8597[0.0000]**, Vector normality test: Chi(10) = 239.3380[0.0000]**, Vector hetero test:
Chi(150) = 124.7546 [0.9344], F(150,202) = 0.6074 [0.9993]; Vector hetero-X test: Chi(300) = 301.0927[0.4714], F(300,188) = 0.8132[0.9443]; VAR Errors: Vector portmanteau(6):
Chi(125) = 191.6879[0.0001]**, Vector normality test: Chi(10) = 231.2960[0.0000]**, Vector hetero test: Chi(150) = 126.0395[0.9230], F(150,202) = 0.6041[0.9994], Vector hetero-
X test: Chi(300) = 320.8604[0.1950], F(300,188) = 0.9294[0.7149]
1413
 1414

Table 6. South Africa, MSIH (2)-VEC(1) model (Estimation sample: 1981–2016)

Variables Regime 1 Regime 2

dlgoldt dlgdpt dlslvt dlcopt dlgoldt dlgdpt dlslvt dlcopt

Regime-dependent  0.016 0.004 (0.697)  0.052 0.004 (0.522)  0.011 0.067  0.025  0.018
intercepts ( 2.960)*** ( 3.351)*** ( 1.728)** (2.872)*** ( 1.512)* ( 2.814)***
Short-run dynamics
dlgoldt1 0.047 (0.310) 0.210 (2.058)**  1.464 0.454
( 4.754)*** (3.000)***
dlgdpt1  0.130  0.120 ( 1.041)  0.191 0.186
( 2.000)** ( 1.498)* (2.936)***
dlslvt1 0.060 (1.667)* 0.175 (3.051)***  0.157  0.026
( 2.260)** ( 0.734)
dlcopt1 0.015 (2.118)** 0.312 (2.053)** 0.467 (1.912)**  0.057
( 0.475)
Error correction
ecmt1 0.0015 (0.023)  0.581  0.109 ( 0.854) 0.101 (1.561)*
( 5.438)***
Standard errors
r 0.021 0.021 0.067 0.034 0.017 0.077 0.051 0.017
Transition prob. Reg. 1 Reg. 2 Durations
Reg. 1 0.683 0.316 7.05
Reg. 2 0.141 0.858 3.16

Regime 1 Regime 2

dlgoldt dlgdpt dlslvt dlcopt dlgoldt dlgdpt dlslvt dlcopt

Contemporaneous correlation
dlgoldt 1.000  0.137 0.352  0.150 1.000  0.362  0.343  0.736
dlgdpt  0.137 1.000 0.286 0.422  0.362 1.000  0.625  0.043
dlslvt 0.352 0.286 1.000  0.433  0.343  0.625 1.000 0.789
dlcopt  0.150 0.422  0.433 1.000  0.736  0.043 0.790 1.000

t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 304.9442 linear system:
267.6382; AIC criterion:  14.5682 linear system:  13.3508; HQ criterion:  13.8012 linear system:  12.8292; SC criterion:  12.3463 linear system:  11.8398; LR linearity test:
74.6119 Chi(14) = [0.0000] ** Chi(16) = [0.0000]** DAVIES = [0.0000] **; StdResids: Vector portmanteau(5): Chi(64) = 73.7004 [0.1906]; Vector normality test: Chi(8) = 6.4381
[0.5983]; Vector hetero test: Chi(100) = 80.9224 [0.9189], F(100,83) = 0.4296 [1.0000]; Vector hetero-X test: Chi(200) = 176.0416 [0.8880], F(200,22) = 0.2450 [1.0000] PredErrors:
Vector portmanteau(5): Chi(64) = 81.9227 [0.0650]; Vector normality test: Chi(8) = 9.2365 [0.3227]; Vector hetero test: Chi(100) = 116.9039 [0.1189], F(100,83) = 0.9635 [0.5727];
Vector hetero-X test: Chi(200) = 218.9425 [0.1706], F(200,22) = 0.7803 [0.8130] VAR Errors: Vector portmanteau(5): Chi(64) = 71.7272 [0.2371]; Vector normality test:
Chi(8) = 19.4751 [0.0125] *; Vector hetero test: Chi(100) = 92.1586 [0.6992], F(100,83) = 0.5541 [0.9976]; Vector hetero-X test: Chi(200) = 201.3295 [0.4603], F(200,22) = 0.4071
[0.9994]
Bildirici and Gokmenoglu
Does Resource Curse Really Exist
and dies out to zero. When the economy is in the
growth regime, the response is negative but gradu-
ally rises for nearly three periods and then becomes
steady.
The results of the analysis for Peru (Table 7)
indicate that the probability of remaining in regime
1 is 0.519 (51.9%) and the probability of shifting to
growth regime is 0.481 (48.1%). The growth regime
of this economy has a tendency to last 2.3 years, and
the possibility of proceeding to regime 1 from this
regime is 0.434 (43.4%). The possibility of staying in
crisis regime while it is in the same regime is 51.90%
and that of moderate growth regime is 56.50%,
suggesting the presence of regime persistency. Error
correcting rate for copper equation is negative and
statistically significant ( 0.166), meaning there is an
adjustment toward long-run equilibrium. Besides,
the error correcting parameters for gold and silver
production are negative but are not statistically sig-
nificant, implying that they are weakly exogenous. In
the short run, there are negative correlations be-
tween silver production and economic growth and
copper production and economic growth, meaning
an increase in silver or copper production leads to a
decrease in economic growth. The correlation be-
tween gold production and economic growth is
positive but not statistically significant. When we
analyzed the IRFs in Figure 4, when one standard
deviation shock is given to the residual of copper
and silver productions in the crisis and growth re-
gimes, economic growth gives negative response.
The response of economic growth to the same shock
in the crisis regime is negative at first but converts to
positive after one period. This positive effect pro-
ceeds for one period and then disappears to zero. In
contrast, in the growth regime, the effect of a unit
shock in gold production on economic growth is
positive and dies away after four periods.
The results of the model, which well describe
the economy of Philippines (Table 8), indicate that
moderate growth regime has the most duration
(7.35 years). Probabilities of staying in the same
regime for all regimes disclose persistency of the
regimes and define existence of important asymme-
tries. If the economy is in crisis, the likelihood of
defeating the crisis and shifting to the moderate
growth phase is 0.236 (23.6%), moderately higher
than that of shifting to the higher growth phase
(12.3%). The variance of the standard errors differs
across regimes. For example, the volatility in regime
1 of the gold production (0.109) is almost two times
higher than that of regime 2 (0.05) and twenty times
1415
higher than regime 3 (0.005). The volatility for
economic growth in regime 1 is 0.046, in regime 2 is
0.025 and in regime 3 is 0.022, suggesting that it is
more volatile in regime 1 than in the other regimes.
In the short run, there is no statistically significant
correlation between GDP and copper and gold. The
adjustment speeds of gold production and economic
growth to long-run equilibrium are significantly dif-
ferent from zero with positive sign, meaning that
there is no adjustment. Regarding the IRFs in Fig-
ure 5, both in the crisis and in moderate growth re-
gimes, the effect of a unit shock in gold production
induces negative effect on economic growth, but that
of copper production is positive. However, in the
high growth regime, both precious metals have a
positive effect on economic growth.
In Table 6, the results for South Africa show
that the probability of remaining in regime 1 is 0.683
(68.3%). That of regime 2 is 0.858. These suggest
persistence of the regimes, and the most persistent
one is the second one. The computed probability of
shifting to regime 1 while economy is in second re-
gime is 14.1%, indicating how low chance of a crisis
may succeed a growth period. While the economy is
in regime 1, the probability of shift to growth regime
is 0.316 (31.6%). The results also show that regime 1
tends to last 7.05 years and regime 2 tends to last
3.16 years. In the short run, there is positive and
statistically significant correlation between GDP and
all precious metal productions under examination.
The speed of adjustment of economic growth to
long-run equilibrium is  0.581, which is signifi-
cantly different from zero. Although the coefficient
of the error correcting term for copper is statistically
significant, it has positive sign, meaning there is no
adjustment to long run. As to gold and silver pro-
duction, the error correcting terms are not signifi-
cant. There is almost no difference in the volatility
of standard errors for the gold production equation
in regimes 1 and 2. However, we cannot say the
same thing for the other equations; they differ across
regimes. In regard to the IRFs in Figure 6, in the
crisis regime and growth regime, the effect of one
standard deviation shock of copper production gives
rise to positive effect on the economic growth, but
this effect dies out to zero after two periods. The
effect of one standard deviation change in silver
production on economic growth in both regimes is
positive for nearly first two periods, then it turns into
negative for one period, and then it becomes zero.
The effect of a unit shock in gold production on
economic growth in the crisis regime fluctuates
 1416

Table 7. Peru, MSIH(2)-VEC(1) model (Estimation sample: 1964–2016)

Variables Regime 1 Regime 2

dlgoldt dlgdpt dlslvt dlcopt dlgoldt dlgdpt dlslvt dlcopt

Regime-dependent inter- 0.003 (0.12) 0.031 (5.19)***  0.0005 ( 0.05) 0.018 (2.28)** 0.052 0.032 0.036 0.049
cepts (2.731)*** (6.60)*** (13.95)*** (3.91)***
Short-run dynamics
dlgoldt1 0.105 (0.82) 0.034 (0.98) 0.030 (1.46)*  0.022 ( 0.37)
dlgdpt1  0.077 0.288 (4.51)***  0.104  0.404
( 0.30) ( 2.53)*** ( 3.76)***
dlslvt1  0.338  0.241  0.193 0.055 (0.35)
( 0.89) ( 2.54)*** ( 3.00)***
dlcopt1 0.328 (1.43)*  0.412  0.117  0.101 ( 0.91)
( 6.39)*** ( 2.97)***
Error correction
ecmt1  0.27 ( 1.27) 0.902 (17.27)***  0.015 ( 0.43)  0.166
( 1.83)**
Standard errors
r 0.085 0.023 0.041 0.027 0.068 0.020 0.009 0.053
Transition prob. Reg. 1 Reg. 2 Durations
Reg. 1 0.519 0.481 2.08
Reg. 2 0.434 0.565 2.30

Regime 1 Regime 2

dlgoldt dlgdpt dlslvt dlcopt dlgoldt dlgdpt dlslvt dlcopt

Contemporaneous correlation
dlgoldt 1.000  0.312  0.269  0.099 1.000 0.545  0.470  0.226
dlgdpt  0.312 1.000 0.285 0.277 0.545 1.000  0.454  0.015
dlslvt  0.269 0.285 1.000 0.423  0.470  0.454 1.000 0.466
dlcopt  0.099 0.277 0.423 1.000  0.226  0.015 0.466 1.000

t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 412.4801, linear system:
386.4835; AIC criterion:  13.6785, linear system:  13.3013; HQ criterion:  12.9637, linear system:  12.8152; SC criterion:  11.8197, linear system:  12.0373; LR linearity test:
51.993, Chi(14) = [0.0000]**, Chi(16) = [0.0000]**, DAVIES = [0.0001]**; StdResids: Vector portmanteau(6): Chi(80) = 108.9550 [0.0174]*; Vector normality test:
Chi(8) = 14.1917 [0.0769], Vector hetero test: Chi(100) = 97.2548[0.5591], F(100,212) = 0.8287[0.8554], Vector hetero-X test: Chi(200) = 210.1038[0.2980],
F(200,184) = 0.9611[0.6088]; PredError: Vector portmanteau(6): Chi(80) = 136.8416[0.0001]**, Vector normality test: Chi(8) = 44.6887[0.0000]**, Vector hetero test:
Chi(100) = 103.8656[0.3757], F(100,212) = 0.9585[0.5891]; Vector hetero-X test: Chi(200) = 208.7496[0.3212], F(200,184) = 1.0001[0.5005]; VAR Errors: Vector portmanteau(6):
Chi(80) = 129.8519[0.0004]**, Vector normality test: Chi(8) = 31.3909[0.0001]**, Vector hetero test: Chi(100) = 103.9338[0.3739], F(100,212) = 0.9668[0.5696], Vector hetero-X
test: Chi(200) = 205.3730[0.3823], F(200,184) = 0.9623[0.6057]
Bildirici and Gokmenoglu
 Table 8. Philippines, MSIH(3)-VEC(2) model (Estimation sample: 1964–2016)

Variables Regime 1 Regime 2 Regime 3

dlgoldt dlgdpt dlcopt dlgoldt dlgdpt dlcopt dlgoldt dlgdpt dlcopt

Regime-dependent inter-  0.016 0.0003 (0.03)  0.051 0.025 0.033 0.020 0.032 0.028 0.076
cepts ( 0.60) ( 2.77)*** (2.22)** (5.01)*** (1.83)** (12.88)*** (3.82)*** (2.43)***
Short-run dynamics
dlgoldt1  0.021 0.001 (0.02)  0.234
( 0.33) ( 2.83)***
dlgoldt2 0.021 (0.41)  0.041  0.09 ( 0.98)
( 0.90)
Does Resource Curse Really Exist

dlgdpt1  0.16 0.171 (1.91)**  0.212

( 1.79)** ( 1.33)**
dlgdpt2  0.082 0.206 (4.01)*** 0.041 (0.28)
( 1.45)*
dlcopt1 0.037 (1.60)* 0.027 (0.80) 0.065 (0.59)
dlcopt2 0.025 (1.50)* 0.047 (1.20) 0.158 (1.62)*
Error correction
ecmt1 0.143 (1.65)* 0.243 (2.42)*** 0.176 (0.95)
Standard errors
r 0.109 0.046 0.070 0.0500 0.025 0.038 0.005 0.022 0.101
Transition prob. Reg. 1 Reg. 2 Reg. 3 Durations
Reg. 1 0.640 0.236 0.123 2.77
Reg. 2 0.095 0.864 0.040 7.35
Reg. 3 0.300 0.005 0.659 2.90

Regime 1 Regime 2 Regime 3

dlgoldt dlgdpt dlcopt dlgoldt dlgdpt dlcopt dlgoldt dlgdpt dlcopt

Contemporaneous correlation
dlgoldt 1.000  0.161  0.430 1.000  0.490 0.105 1.000  0.855 0.686
dlgdpt  0.161 1.000  0.478  0.489 1.000 0.178  0.855 1.000  0.923
dlcopt  0.430  0.478 1.000 0.105 0.178 1.000 0.686  0.923 1.000

t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 279.8931, linear system:
228.6430; AIC criterion:  8.5243, linear system:  7.4960; HQ criterion:  7.7523, linear system:  7.0671; SC criterion:  6.5168, linear system:  6.3807; LR linearity test:
102.5002, Chi(18) = [0.0000]**, Chi(24) = [0.0000]**, DAVIES = [0.0000]**; StdResids: Vector portmanteau(6): Chi(36) = 35.9587[0.4706], Vector normality test:
Chi(6) = 1.0555[0.0867], Vector hetero test: Chi(84) = 62.3388 [0.9632], F(84,151) = 0.5116 [0.9996], Vector hetero-X test: Chi(210) = 206.6500[0.5524],
F(210,38) = 0.4789[0.9994]; PredError: Vector portmanteau(6): Chi(36) = 43.2576[0.1891], Vector normality test: Chi(6) = 72.4848[0.0000]**, Vector hetero test:
Chi(84) = 63.1078[0.9570], F(84,151) = 0.5727[0.9973], Vector hetero-X test: Chi(210) = 224.3910[0.2360], F(210,38) = 0.6576[0.9653]; VAR Errors: Vector portmanteau(6):
Chi(36) = 29.1710[0.7830], Vector normality test: Chi(6) = 57.7954[0.0000]**, Vector hetero test: Chi(84) = 67.2230[0.9099], F(84,151) = 0.6363[0.9883], Vector hetero-X test:
Chi(210) = 224.8563[0.2294], F(210,38) = 0.6059[0.9853]
1417
 Table 9. USA, MSIH(3)-VEC(1) model (Estimation sample: 1963–2015)
1418

Variables Regime 1 Regime 2 Regime 3

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Regime-  0.029 0.003  0.045  0.006  0.026 0.008 0.024  0.010 0.039 0.016 0.048 0.034
depen ( 2.47)*** (1.10) ( 2.31)** ( 0.23) ( 3.21)*** (4.52)*** (2.10)** ( 0.77) (2.04)** (10.43)*** (2.67)*** (1.94)**
-dent
intercepts
Short-run dynamics
dlgoldt1 0.323  0.031 0.204 (2.06)** 0.214
(4.47)*** ( 3.00)*** (1.92)**
dlgdpt1 0.666 0.683 (17.24)***  0.626  0.179
(2.67)*** ( 1.67)** ( 0.41)
dlcopt1  0.103  0.041  0.197 0.111
( 1.22) ( 3.52)*** ( 1.69)** (0.85)
dlslvt1
Error correction
ecmt1 0.028  0.039  0.384  0.009
(0.48) ( 4.02)*** ( 3.65)*** ( 0.08)
Standard errors
r 0.030 0.008 0.047 0.073 0.023 0.007 0.022 0.032 0.065 0.004 0.055 0.048
Transition Reg. 1 Reg. 2 Reg. 3 Durations
prob.
Reg. 1 0.622 0.200 0.177 2.65
Reg. 2 0.108 0.840 0.051 6.26
Reg. 3 0.089 0.047 0.863 7.32

Regime 1 Regime 2 Regime 3

dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt dlgoldt dlgdpt dlcopt dlslvt

Contemporaneous correlation
dlgoldt 1.000 0.430 0.845 0.957 1.000  0.599 0.209  0.319 1.000  0.286  0.143 0.130
dlgdpt 0.430 1.000 0.472 0.390  0.599 1.000  0.206 0.304  0.286 1.000  0.300 0.560
dlcopt 0.845 0.472 1.000 0.796 0.209  0.206 1.000 0.373  0.143  0.300 1.000 0.463
Bildirici and Gokmenoglu
 Does Resource Curse Really Exist 1419

Chi(80) = 88.8211[0.2341], F(80,205) = 1.1320[0.2430]; Vector hetero-X test: Chi(140) = 161.0089[0.1080], F(140,222) = 1.2128 [0.0999]; PredError: Vector portmanteau(6): Chi(80) = 109.9617 [0.0148] *,
1.000

Chi(24) = [0.0000]**, DAVIES = [0.0000]**; StdResids: Vector portmanteau(6): Chi(80) = 124.1004[0.0012] **; Vector normality test: Chi(8) = 8.6642 [0.3714]; Vector hetero test:

Vector normality test: Chi(8) = 27.2167[0.0006] **, Vector hetero test: Chi(80) = 112.1230[0.0104]*, F(80,205) = 1.5884[0.0049]**, Vector hetero-X test: Chi(140) = 212.2522[0.0001]**, F(140,222) = 1.9299
[0.0000]**; VAR Errors: Vector portmanteau(6): Chi(80) = 90.2778[0.2026], Vector normality test: Chi(8) = 17.7360[0.0233]*, Vector hetero test: Chi(80) = 88.8762[0.2328], F(80,205) = 1.1115[0.2751],
t-statistics are given in () parentheses. Significance at 1 percent, 5 percent and 10 percent is denoted with ***, ** and *, respectively. Log-likelihood: 405.2970, linear system: 363.0286; AIC criterion:
 13.5961, linear system:  12.9067; HQ criterion:  12.9528, linear system:  12.6065; SC criterion:  11.9232, linear system:  2.1261; LR linearity test: 84.5366, Chi(18) = [0.0000]**,
dlslvt
between negative and positive values and be-
comes zero after three periods. The same shock
in the gold production in the growth regime has a
negative effect, which tapers off to zero after
dlcopt

0.463

three periods.
The estimated model for USA (Table 9)
Regime 3

shows that the economy is well characterized by

three regimes. When the economy is in the crisis
dlgdpt

regime, the likelihood of staying in that regime is

0.560

62.2%. In contrast, the possibility of passing to

moderate growth regime and high growth regime
is 20% and 17.7%, respectively. The moderate
growth regime of the American economy has a
dlgoldt

0.130

tendency to last 6.26 years, and the possibility of

proceeding to regime 1 from this regime is low,
0.108 (10.8%). The possibility of proceeding to
the higher growth stage is the lowest one with
1.000
dlslvt

0.051 (5.1%) and that of staying in this regime is

0.84 (84%) is the highest one, making this regime
the most persistent one. Ergodic probabilities
determined dominant regime as the third one
dlcopt

0.373

with 7.32 years of duration. When the economy is

in this regime, the possibility of shifting to crisis
Regime 2
Table 9. continued

regime is 8.9%, which is higher than for shifting

to moderate growth regime (4.7%). The transition
dlgdpt

0.304

probabilities of staying in the same regime while

the economy is in a specific regime indicate per-
sistency of the regimes. In the short run, gold and
copper production is negatively correlated with
 0.319
dlgoldt

economic growth. The error correction terms are

negative and significantly different from zero for
Vector hetero-X test: Chi(140) = 176.6585 [0.0195]*, F(140,222) = 1.5170 [0.0028]**

economic growth and copper production equa-

tions, meaning they adjust toward the long-run
equilibrium. However, the error correction coef-
1.000
dlslvt

ficients for gold and silver productions are

insignificant, indicating they are weakly exoge-
nous. In respect of the IRFs in Figure 7, the re-
sponse of economic growth to gold production in
dlcopt

0.796

the crisis regime is positive but nearly zero; in the

moderate growth regime and the high growth
Regime 1

regime it is negative but nearly zero. One stan-

dard deviation shock to silver production gives
dlgdpt

0.390

rise to positive response on the economic growth

for all regimes. For the copper production, the
same shock induces negative but nearly zero ef-
fect on the economic growth for the first two
dlgoldt

0.957

regimes. In the high growth regime, its effect is

negative for two periods and tapers off to zero
after that.
dlslvt
1420 Bildirici and Gokmenoglu

Regime 1: response to orth. shock to AUS_GOLD Regime 1: response to orth. shock to AUS_COP
AUS_GOLD AUS_COP
0.04
0.025 AUS_SIL AUS_GDP
0.02
0.000
0.00
-0.025
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to AUS_SIL Regime 1: response to orth. shock to AUS_GDP
0.04 0.0010
0.02 0.0005
0.00 0.0000
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to AUS_GOLD Regime 2: response to orth. shock to AUS_COP
0.075 0.04
0.050
0.02
0.025
0.000 0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to AUS_SIL Regime 2: response to orth. shock to AUS_GDP
0.03 0.03
0.02 0.02
0.01 0.01
0.00 0.00

0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 3: response to orth. shock to AUS_GOLD Regime 3: response to orth. shock to AUS_COP

AUS_GOLD AUS_COP
0.025 AUS_SIL AUS_GDP 0.04

0.000 0.02

0.00
-0.025

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 3: response to orth. shock to AUS_SIL Regime 3: response to orth. shock to AUS_GDP
0.01
0.015

0.00 0.010

0.005
-0.01

0.000
-0.02
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Figure 1. Impulse–response functions for Australia.
Does Resource Curse Really Exist 1421

Regime 1: response to orth. shock to CAN_GOLD Regime 1: response to orth. shock to CAN_GDP
0.020
CAN_GOLD CAN_GDP
0.02 CAN_COP CAN_SIL
0.015
0.01
0.010
0.00
0.005
-0.01
0.000
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to CAN_COP Regime 1: response to orth. shock to CAN_SIL
0.03
0.03
0.02
0.02
0.01
0.01
0.00
0.00

0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 2: response to orth. shock to CAN_GOLD Regime 2: response to orth. shock to CAN_GDP
0.04
0.02
CAN_GOLD CAN_GDP
CAN_COP CAN_SIL

0.02
0.00

0.00 -0.02

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to CAN_COP Regime 2: response to orth. shock to CAN_SIL

0.03
0.015

0.02
0.010

0.01
0.005

0.00
0.000

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Figure 2. Impulse–response functions for Canada.

CONCLUSION study to investigate whether the resource curse

really exists in the precious metal producer coun-
We analyzed the relationship between eco- tries. We detected long-run relationship between
nomic growth and precious metal production in this them by employing the Johansen co-integration test.
1422 Bildirici and Gokmenoglu

Regime 1: response to orth. shock to MEX_GOLD Regime 1: response to orth. shock to MEX_GDP
MEX_GOLD MEX_GDP
MEX_COP MEX_SIL 0.00
0.04
-0.05
0.02
-0.10

0.00 -0.15

-0.20

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to MEX_COP Regime 1: response to orth. shock to MEX_SIL
0.006
1.00

0.75 0.004

0.50

0.25 0.002

0.00

0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 2: response to orth. shock to MEX_GOLD Regime 2: response to orth. shock to MEX_GDP
MEX_GOLD MEX_GDP
0.050 MEX_COP MEX_SIL

0.050

0.025

0.025

0.000

0.000
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to MEX_COP Regime 2: response to orth. shock to MEX_SIL
0.025

0.020
0.04
0.015

0.010
0.02

0.005

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Figure 3. Impulse–response functions for Mexico.

Then, we employed MS-VEC analysis to the data to VEC models with shift in the intercepts and in the
investigate short-run dynamics, error correcting error variances are well describing the data under
mechanisms and impulse–response relationships in examination. The empirical evidence for the nega-
every state of the economy. We found that MSIH- tive effect of natural resource richness on economic
Does Resource Curse Really Exist 1423

Regime 1: response to orth. shock to PER_GOLD Regime 1: response to orth. shock to PER_GDP

0.075 0.02
PER_GOLD PER_GDP
PER_SIL PER_COP

0.050 0.01

0.025
0.00

0.000

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to PER_SIL Regime 1: response to orth. shock to PER_COP
0.04
0.02

0.02 0.01

0.00
0.00

-0.01
0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 2: response to orth. shock to PER_GOLD Regime 2: response to orth. shock to PER_GDP

0.050 PER_GOLD PER_GDP

PER_SIL PER_COP
0.01

0.025

0.00
0.000

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to PER_SIL Regime 2: response to orth. shock to PER_COP

0.02 0.04

0.01 0.02

0.00
0.00

-0.01
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Figure 4. Impulse–response functions for Peru.

growth is mixed. According to the results, it is evi- the short run, generally precious metal abundance
dent that the relationship between precious metal has a negative effect or no effect on economic
productions and economic growth varies across the growth of the economies under examination except
state of the business cycle of the economy. In sum, in copper abundance for Australia, silver abundance
1424 Bildirici and Gokmenoglu

Regime 1: response to orth. shock to PHL_GOLD Regime 1: response to orth. shock to PHL_GDP
0.050
0.10 PHL_GOLD PHL_GDP
PHL_COP 0.025
0.05
0.000
0.00 -0.025

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to PHL_COP Regime 2: response to orth. shock to PHL_GOLD
0.06
0.050
0.04
0.025
0.02
0.000
0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to PHL_GDP Regime 2: response to orth. shock to PHL_COP
0.03
0.02 0.02

0.01
0.00
0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 3: response to orth. shock to PHL_GOLD

0.075
PHL_GOLD PHL_GDP
0.050 PHL_COP

0.025

0.000
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Regime 3: response to orth. shock to PHL_GDP
0.025

0.000

-0.025

-0.050
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Regime 3: response to orth. shock to PHL_COP
0.04

0.02

0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Figure 5. Impulse–response functions for Philippines.

for Canada, all precious metal abundance for South
Africa. However, the impulse–response function
analysis suggests that the response of the economic
growth to the precious metal production shocks
changes according to the countries, the business cy-
cles of the economies and the type of the precious
metals but not according to development level. This
diversity may stem from economic and political
structure of the countries. In particular, institutional
quality is determinant for economic growth of re-
Does Resource Curse Really Exist
Figure 6. Impulse–response functions for South Africa.
source-rich countries. We mean by institutions the
rule of law, democratization and income distribution
system. Under a weak institutional quality, which
generally gives rise to dictatorship, natural resource
richness has a negative effect on human capital,
development of financial institutions and competi-
tiveness of manufacturing sector. Therefore, a strong
institutional infrastructure should be established to
escape resource curse and get more benefit from
1425
natural resources. Besides, if the economy does not
depend on only natural resources, in this case, pre-
cious metal productions, the negative effects can be
easily eliminated; otherwise, the negative effects can
have harmful and lasting effects on the economy.
The precious metal producer countries should hence
diversify their economic activity and establish a
strong institutional infrastructure to minimize re-
source curse effect.
1426 Bildirici and Gokmenoglu

Regime 1: response to orth. shock to USA_GOLD Regime 1: response to orth. shock to USA_GDP
0.075 USA_GOLD USA_GDP
0.050 USA_COP USA_SIL 0.005
0.025
0.000
0.000
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 1: response to orth. shock to USA_COP Regime 1: response to orth. shock to USA_SIL
0.050 0.03
0.02
0.025
0.01
0.000 0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to USA_GOLD Regime 2: response to orth. shock to USA_GDP
0.03 USA_GOLD USA_GDP
0.010
0.02 USA_COP USA_SIL

0.01 0.005
0.00 0.000

0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 2: response to orth. shock to USA_COP Regime 2: response to orth. shock to USA_SIL
0.03
0.02
0.02
0.01 0.01
0.00 0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6

Regime 3: response to orth. shock to USA_GOLD Regime 3: response to orth. shock to USA_GDP
0.06

USA_GOLD USA_GDP 0.04

0.04 USA_COP USA_SIL

0.02
0.02

0.00
0.00
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Regime 3: response to orth. shock to USA_COP Regime 3: response to orth. shock to USA_SIL

0.04 0.004

0.02
0.002

0.00
0.000
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Figure 7. Impulse–response functions for USA.
Does Resource Curse Really Exist
REFERENCES
Alexeev, M., & Conrad, R. (2011). The natural resource curse and
economic transition. Economic Systems, 35(4), 445–461.
Altuğ, S., & Bildirici, M. (2010). Business cycles around the globe:
A regime switching approach. Koç University-TUSIAD
Economic Research Forum Working Papers 1009, Koc
University-TUSIAD Economic Research Forum.
Amiri, H., Samadian, F., Yahoo, M., & Jamali, S. J. (2018). Natural
resource abundance, institutional quality and manufacturing
development: Evidence from resource-rich countries. Resources
Policy. https://doi.org/10.1016/j.resourpol.2018.11.002.
Antonakakis, N., Cunado, J., Filis, G., & de Gracia, F. P. (2017).
Oil dependence, quality of political institutions and economic
growth: A panel VAR approach. Resources Policy, 53, 147–
163.
Arezki, R., & Brückner, M. (2011). Oil rents, corruption, and state
stability: Evidence from panel data regressions. European
Economic Review, 55, 955–963.
Arezki, R., & Ploeg, F. (2011). Do natural resources depress in-
come per capita? Review of Development Economics, 15(3),
504–521.
Arin, K. P., & Braunfels, E. (2018). The resource curse revisited:
A Bayesian model averaging approach. Energy Economics,
70, 170–178.
Atkinson, G., & Hamilton, K. (2003). Savings, growth and the
resource curse hypothesis. World Development, 31(11), 1793–
1807.
Auty, R. M. (1993). Sustaining development in mineral economies:
the resource curse. London: Routledge.
Bhattacharyya, S., & Hodler, R. (2010). Natural resources,
democracy and corruption. European Economic Review, 54,
608–621.
Bildirici, M. (2012a). The relationship between economic growth
and electricity consumption in Africa: MS-VAR and MS-
Granger causality analysis. The Journal of Energy and
Development, 37(2), 179–207.
Bildirici, M. (2012b). Economic growth and energy consumption
in G7 countries. The Journal of Energy and Development, 37,
1–30.
Bildirici, M. E., & Gökmenoğlu, S. M. (2017). Environmental
pollution, hydropower energy consumption and economic
growth: Evidence from G7 countries. Renewable and Sus-
tainable Energy Reviews, 75, 68–85.
Brunnschweiler, C. N. (2008). Cursing the blessings? Natural re-
source abundance, institutions, and economic growth. World
Development, 36(3), 399–419.
Brunnschweiler, C. N., & Bulte, E. H. (2008). The resource curse
revisited and revised: A tale of paradoxes and red herrings.
Journal of Environmental Economics and Management, 55,
248–264.
Cavalcanti, T., Mohaddes, K., & Raissi, M. (2011). Growth,
development and natural resources: New evidence using a
heterogeneous panel analysis. The Quarterly Review of
Economics and Finance, 51, 305–318.
Cockx, L., & Francken, N. (2014). Extending the concept of the
resource curse: Natural resources and public spending on
health. Ecological Economics, 108, 136–149.
Collier, P., & Hoeffler, A. (2009). Testing the neocon agenda:
Democracy in resource-rich societies. European Economic
Review, 53, 293–308.
Corden, W. M. (1984). Booming sector and Dutch disease eco-
nomics: Survey and consolidation. Oxford Economic Papers,
36, 359–380.
Corden, W. M., & Neary, J. P. (1982). Booming sector and de-
industrialisation in a small open economy. The Economic
Journal, 92(368), 825–848.
1427
Cotet, A. M., & Tsui, K. K. (2013). Oil, growth, and health: What
does the cross-country evidence really show? The Scandina-
vian Journal of Economics, 115(4), 1107–1137.
Damette, O., & Seghir, M. (2018). Natural resource curse in oil
exporting countries: A nonlinear approach. International
Economics, 156, 231–246.
Davis, G. (1995). Learning to love the Dutch disease: Evidence
from the mineral. World Development, 23, 1765–1779.
De Soysa, I., & Gizelis, I. T. (2013). The natural resource curse
and the spread of HIV/AIDS, 1990–2008. Social Science and
Medicine, 77, 90–96.
Deacon, R. T. (2011). The political economy of the natural re-
source curse: A survey of theory and evidence. Foundations
and Trends in Microeconomics, 7(2), 111–208.
Dietz, S., Neumayer, E., & de Soysa, I. (2007). Corruption, the
resource curse and genuine saving. Environment and Devel-
opment Economics, 12, 33–53.
Dobra, J., Dobra, M., & Ouedraogo, A. (2018). Does mineral
development provide a basis for sustainable economic
development? Resources Policy, 58, 71–76.
Frankel, J. A. (2010). The natural resource curse: A survey. NBER
Working Paper Series, Working Paper 15836.
Gelb, A. (1988). Oil windfalls blessing or curse?. New York:
Oxford University Press.
Gylfason, T. (2001). Natural resources, education, and economic
development. European Economic Review, 45, 847–859.
Hamilton, J. D. (1989). A new approach to the economic analysis
of nonstationary time series and the business cycle. Econo-
metrica, 57, 357–384.
Hamilton, J. D. (1990). Analysis of time series subject to changes
in regime. Journal of Econometrics, 45(1–2), 39–70.
Havranek, T., Horvath, R., & Zeynalov, A. (2016). Natural re-
sources and economic growth: A meta-analysis. World
Development, 88, 134–151.
Johansen, S. (1991). Estimation and hypothesis testing of co-in-
tegration vectors in Gaussian vector autoregressive models.
Econometrica, 59(6), 1551–1580.
Johansen, S. (1995). Likelihood-based inference in cointegrated
vector autoregressive models. New York: Oxford University
Press.
Kapetanios, G., Shin, Y., & Snell, A. (2003). Testing for a unit
root in the nonlinear STAR framework. Journal of Econo-
metrics, 112(2), 359–379.
Kim, D. H., & Lin, S. C. (2017). Human capital and natural re-
source dependence. Structural Change and Economic
Dynamics, 40, 92–102.
Krolzig, H. M. (1996). Statistical analysis of cointegrated VAR
processes with markovian regime shifts. SFB 373 Discussion
Paper 25/1996, Humboldt-Universität zu Berlin.
Krolzig, H. M. (1997). Markov-switching vector autoregressions
modelling, statistical inference, and application to business
cycle analysis. Berlin: Springer.
Leite, C., & Weidman, J. (1999). Does mother nature corrupt?
Natural resources, corruption and economic growth. IMF
Working Papers WP 99/85.
Mehlum, H., Moene, K., & Torvik, R. (2006). Institutions and the
resource curse. The Economic Journal, 116, 1–20.
Moradbeigi, M., & Law, S. H. (2017). The role of financial
development in the oil-growth nexus. Resource Policy, 53,
164–172.
OConnor, K. J., Blanco, L. R., & Nugenta, J. B. (2018). Does oil
really curse democracy? A long-run time-series analysis of
127 countries. Resources Policy, 57, 264–277.
Prebisch, R. (1950). The economic development of Latin America
and its principal problems. Lake success. New York: United
Nations.
Ross, M. (2001). Does oil hinder democracy? World Politics,
53(3), 325–361.
1428
Ross, M. (2003). How does mineral wealth affect the poor? https://
www.researchgate.net/publication/228601978_How_does_mi
neral_wealth_affect_the_poor. Accessed December 10, 2017.
Sachs, J. D., & Warner, A. M. (1995). Natural resource abundance
and economic growth. NBER Working Papers, 5398, pp. 1–54.
Sachs, J. D., & Warner, A. M. (1997). Natural resource abundance
and economic growth. Harvard Institute for International
Development Working Paper, November, pp .1–50.
Sachs, J. D., & Warner, A. M. (2001). The curse of natural re-
sources. European Economic Review, 45, 827–838.
Sala-i-Martin, X., & Subramanian, A. (2003). Addressing the
natural resource curse: An illustration from Nigeria. NBER
Working Paper Series, pp. 1–46.
Singer, H. W. (1950). The distribution of gains between investing
and borrowing countries. American Economic Review, 40,
473–485.
Bildirici and Gokmenoglu
Smith, B. (2015). The resource curse exorcised: Evidence from a
panel of countries. Journal of Development Economics, 116,
57–73.
Stevens, P. (2015). The resource curse revisited. Appendix: A lit-
erature review. N.L., Chatham House: Energy, Environment
and Resources. https://www.chathamhouse.org/sites/default/f
iles/publications/research/20150804ResourceCurseRevisitedS
tevensLahnKooroshyAppendix.pdf. Accessed December 15,
2017.
Stijns, J. P. (2005). Natural resource abundance and economic
growth revisited. Resources Policy, 30, 107–130.
Stijns, J. P. (2006). Natural resource abundance and human capital
accumulation. World Development, 34(6), 1060–1083.
Wigley, S. (2017). The resource curse and child mortality, 1961–
2011. Social Science and Medicine, 176, 142–148.