Energy Economics 80 (2019) 12–19

Contents lists available at ScienceDirect

Energy Economics

journal homepage: www.elsevier.com/locate/eneeco

Oil price collapse and challenges to economic transformation of Saudi

Arabia: A time-series analysis☆
Fredj Jawadi a,⁎, Zied Ftiti b
a
University of Lille, 104 Avenue du Peuple Belge, 59043 Lille, France
b
EDC Paris Business School, OCRE-Lab, 70 Galerie des Damiers, 92415 Courbevoie, France

a r t i c l e i n f o a b s t r a c t

Article history: This paper studies the impact of oil price changes on economic growth in Saudi Arabia to measure its dependency
Received 11 October 2017 on the crude oil sector. To this end, an On/Off threshold regression is specified to allow the oil/GDP relationship to
Received in revised form 27 November 2018 be asymmetric, nonlinear, and time-varying with regard to the business cycle phases. Further, we empirically test
Accepted 3 December 2018
the diversification hypothesis put forward by the National Transformation Program (Saudi Vision 2030) to check
Available online 18 December 2018
whether the equity-energy investment initiative could boost economic growth in Saudi Arabia. First, our findings
JEL classification:
confirm the contribution of the oil sector to economic growth in the country, but also show that the oil/Saudi
C2 economy relationship exhibits nonlinearity and threshold effects, as the impact of oil price varies per regime
G10 depending on the state of the market. Second, in line with the Vision 2030 expectations, we are in favor of
Q43 transforming the economy and opening its equity market. We also quantify a positive and significant impact of
equity investment on the Saudi Arabian economy. Further, this diversification route will stimulate a beneficial
Keywords: oil effect on the real economy.
Oil price © 2018 Elsevier B.V. All rights reserved.
Economic growth
Diversification
Saudi vision 2030
Nonlinearity
Threshold effect

1. Introduction Accordingly, Saudi Arabia's considerable oil exporting capacity has

given it two major benefits. First, as the world's leading oil producer,
With an oil extraction cost of $10 per barrel, the lowest in the world, it plays a key economic and political role in the Middle East. Second,
Saudi Arabia has an enormous comparative advantage in oil production. oil production has provided it with considerable wealth that has
This low extraction cost is because Saudi Arabia has the largest proven supported the country's public finances and economic growth for
oil reserves, estimated at around 267 billion barrels (about 24.4% of all many years. Oil production yielded $150 billion in 2015, representing
proven oil reserves in the world in 2000). Consequently, Saudi Arabia 23% of its GDP. Oil exports provide Saudi Arabia with a positive trade
has traditionally been the world's biggest oil producer, with a gigantic balance ($184 billion in 2014 and $47.3 billion in 2015). Oil production
production rate of 10.3 million barrels a day in 1980, 10.6 million barrels has sustained economic growth, not only in Saudi Arabia (about 3.5%
a day in 2006, 9.2 million barrels a day in 2008, and 9.76 million barrels a growth rate in 2014 and 3.3% in 2015)1 but also across the whole
day in 2009. Indeed, the country produced over 10% of the world's oil GCC region (5% growth rate in 2016).2 Furthermore, Saudi Arabia's
(11.8% of total crude oil production in 2001) and exported 16.5% of all oil exports have helped it to develop strong trade relationships
oil exports to OECD countries in 2000. Crude Oil Production in Saudi with the United Arab Emirates, China, India, Singapore, Egypt, and
Arabia averaged 8.101 million barrels per day from 1973 until 2018, other countries.
reaching a high of 10.7 million barrels in November of 2016 and is While oil production does have a positive effect on the real economy
about 10.6 million barrels per day in October 2018. Also, Saudi in Saudi Arabia, the flip side is that it makes Saudi very dependent on its
Arabia's Crude Oil exports are about 6.9 million barrels/day in Dec oil sector. With a lack of economic diversification in Saudi Arabia and
2017 against 7.4 million barrels/day in December 2016. the high crude oil price volatility over the last decade, the question
arises: how would Saudi Arabia's GDP react if oil prices collapsed and
☆ The authors acknowledge the financial support from the Immam University SMF
1
Chair, Saudi Arabia. The oil sector accounts for 40% of real GDP and 90% of total export earnings.
2
⁎ Corresponding author. The GCC region controls about 40% of global oil exports and possesses about 60% of the
E-mail address: [email protected] (F. Jawadi). world's proved oil reserves.

https://doi.org/10.1016/j.eneco.2018.12.003
0140-9883/© 2018 Elsevier B.V. All rights reserved.
 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19 13

Table 1
Main descriptive statistics.

DLCGDP DLRGDP DLWTI

Mean −0.001 0.037 0.076

Median 0.007 0.031 0.065
Std. dev. 0.096 0.093 0.263
Skewness −0.798 −0.408 0.429
Kurtosis 4.092 3.565 5.658
Jarque-Bera 6.863 1.811 14.309
(p-value) (0.032) (0.404) (0.000)

Note: DLWTI measures oil return. DLCGDP and DLRGDP denote the economic growth rate
using constant GDP per capita and real GDP, respectively.

It is worthwhile to make two important observations. First, the ex-

cessive volatility in crude oil price has been induced by different factors:
Fig. 1. Oil price evolution. WTI refers to the West Texas Intermediate (WTI), an oil pricing
i) the slowdown in global economic growth, ii) the slowdown in
benchmark in the US. Brent denotes the Brent Crude oil price and is considered as the
benchmark index for European-originated oil. OPEC price refers to the Organization of the Chinese oil demand (economic growth in China went from 7.4% in
Petroleum Exporting Countries (OPEC) Reference Basket. Data for WTI are obtained from 2014 to 6.9% in 2015, its lowest level since 1990), iii) the downturn in
the Federal Reserve Economic Data (FRED) service of the Federal Reserve Bank of St. Louis, the industrial sector, iv) Saudi Arabia's strategy to compete with the
while Brent data is collected from the U.S. Energy Information Administration. The
US shale price, v) the refusal of other oil exporting countries to reduce
Organization of the Petroleum Exporting Countries Data provides data for OPEC prices.
Source: Bottazzi (2016), http://cafim.sssup.it/~giulio/other/oil_price/report.html.
their production, and vi) the shale oil revolution. Second, the decline
in oil production has significantly impacted Saudi Arabian economy.
Indeed, the decrease in oil prices since 2014 has reduced the GDP by
how would Saudi Arabia's economy transform itself as it is expected to $100 billion (about 13%).4 Additionally, following the oil slump, the
do so with the on-going development program called Saudi 2030? Saudi Arabian stock market lost 10.7% by mid-2014 and public deficit
From a theoretical point of view, this transformation is required, reached 16% of GDP by 2015. According to the International Monetary
since Saudi Arabia's marked specialization in oil production also Fund, to rebalance this deficit, oil price needs to be $80 per barrel,
brings with it some severe drawbacks to its economy. The strong de- while the US shale oil production has pushed the price of a barrel
pendence on oil production and energy sector has stifled job creation toward $50.
in other sectors, leading to a high youth unemployment rate.3 For To deal with the effect of oil prices on the real economy of Saudi
instance, the oil industry (dominated by oil extraction activities) Arabia, one direct measure adopted was to reduce oil production from
represents 45% of GDP, while services represent 50% of GDP. Oil 7.2 million barrels/day in 2016 to 6.6 million barrels/day in 2017.
exports account for 90% of total exports and 80% of government in- However, there is certainly no guarantee that this measure would be
come. Consequently, the excessive oil price volatility caused by de- successful. It could also be partially successful given the uncertainty of
mand shocks and/or overall geopolitical instability and tensions reaching a consensus with the other OPEC countries on oil production
has always had a negative impact on the economy in Saudi Arabia, volumes. For this reason, Saudi Arabia has recently introduced a new,
but that has always varied over time. Oil price volatility and uncer- long-term reform and transformation program, set out in the Vision
tainty is also affected by the relative uncertainty of coordination 2030 project.
among oil producers regarding oil production quotas. The Vision 2030 program involves a series of reforms and transfor-
For illustration, from Fig. 1, it appears that oil prices have shown mations. First, the program stipulates diversification of the Saudi
episodic high volatility since the 1980s. While the oil price dynamics economy to reduce the Kingdom's dependence on the oil industry.
between 1988 and 2000 appears to be relatively stable, the oil price Accordingly, the program aims to create a more diversified and sustain-
dynamics saw more volatility in the aftermath of 2000. Indeed, from able economy that offers economic opportunities to entrepreneurs,
2000 to 2008, oil prices have shown an unprecedented increase from creates job opportunities, and fights unemployment. Second, Vision
$25 per barrel to $150 per barrel. These movements in oil prices can 2030 defines new legislation that restructures and regulates the eco-
be explained by the oil production cuts by the OPEC with a simultaneous nomic and accounting systems to improve the business environment
increase in oil demand from emerging economies. However, this spike (i.e. building an efficient and transparent accounting system) while
was stemmed by the global financial crisis in 2008 that dampened oil benefitting from the Kingdom's strategic location. It also applies the
demand pushing oil price to $40 per barrel. With the first signs of Islamic principle of moderation. Indeed, the Vision 2030 program aims
economic recovery, the crude oil price moved up to $100 per barrel at converting Saudi Arabia into a global hub, connecting Asia, Europe,
and then to $125 in 2014. However, oil prices abruptly fell in the fourth and Africa, to lead investments in the Middle East region. Third, the
quarter of 2014 because of excess oil supply. In June 2014, oil price Vision 2030 program stipulates the privatization of some government
reached a level of $112 per barrel and in December 2014, to $59. This services and the creation of a range of digital services to improve the
drop in oil price is associated with the slow Chinese economic growth service qualities and make them more efficient and responsible. The
that implied a lower oil demand from India and Brazil as well. Further, program is also expected to transform the public investment fund into
the shale oil revolution in the USA and Canada implied a sharp reduction a sovereign wealth fund. All these changes would also help improve
in their oil imports and produced a downward pressure on oil prices. the Kingdom's tourist appeal. Tourism is another sector that could be
Indeed, as suggested by James Hamilton in Jawadi (2018), the shale a major source of income for the country. At present, millions of tourists
oil revolution in the US has affected overall oil prices, and, by exten- visit Saudi Arabia annually, especially because of two mosques that are
sion, the relationship between oil price and the economy of the Saudi among the most sacred sites of Islam on Earth.
Arabia. Saudi Arabia, can endure low oil prices for a long time and These initiatives are obviously important reforms that could trans-
can possibly even push the USA and Canada to stop their costly oil form the Saudi Arabian economy. However, it is worth noting that this
production methods.
4
This is particularly important given the shale revolution and the great oil price uncer-
3
The population growth in Saudi Arabia is about 3% a year with 65% of its population tainty that might negatively affect the economic activity (see Elder and Serletis, 2010 for
under 25 years of age and 42.5% of the population under 14 years. more details).
14 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19

Table 2 1.2
Unconditional Pearson correlations.

DLCGDP DLRGDP DLWTI 0.8

DLCGDP 1.000 0.989 0.174
DLRGDP 1.000 0.187
DLWTI 1.000 0.4

Note: This table reports the values of unconditional correlations between economic
growth and the oil return series.
0.0

transformation requires other structural changes and more clearness -0.4

from the normally secretive authorities as well. For example, investors
might ask for more transparency regarding oil reserve estimates, pro-
duction cost estimates, and management practices of Saudi Aramco. -0.8
60 65 70 75 80 85 90 95 00 05 10 15
Such transparency is only feasible if the other OPEC oil producers do
the same, but in practice this is unlikely. DLCGDP DLRGDP DLWTI
Based on the National Transformation Program and the Saudi Vision
2030, this paper empirically looks at the impact of oil price change on Fig. 2. Oil return and GDP economic growth interactions.
the Saudi economy in a nonlinear framework and explores the diversi-
fication hypothesis. To this end, this study empirically investigates the
impact of oil price changes on economic growth. We analyze the Saudi pushed the government to nationalize the oil industry that led to the
GDP-oil price relationship and compare the reactions of the Saudi econ- creation of the Saudi Aramco company. Saudi Arabia also experienced
omy following an oil price shock at different stages of an economic cycle. volatility in oil-related returns during the Iranian revolution in 1979,
We allow the oil-GDP relationship to be nonlinear and time-varying the Iran-Iraq war (1980–1988) and the Gulf Wars and, more recently,
with regard to these phases, and we propose a new On/Off threshold the decline in foreign oil demand, all of which impacted the real
model that can provide a different measure of the reaction of Saudi economy negatively and significantly. Thus, the development plans
GDP to oil price change. Our model endogenously captures the contribu- launched by the Saudi authorities were designed to hedge its economy
tion of the oil sector to economic growth for both recession regime and against oil price collapse, reduce its dependency on the oil sector, and
growth regime. At the same time, we augment our model with other diversify its economy through its openness to other investments. These
financial variables to test the diversification hypothesis of the Vision actions were underpinned by certain priorities and reforms set out in
2030 program, and we double-check the impact of oil on the Saudi development plans: improvement of infrastructure (1970–1975), im-
economy when the economy is more diversified. This is particularly in- provement in societal wellbeing (1975–1980), improvement in educa-
teresting as this will enable discussion on potential economic diversifi- tion and social services (1980–1985), diversification of services and
cation solutions to improve the Saudi economy taking alternative sectors (1985–1990),5 regional defense and development, enhanced
investments into account and incorporating other economic factors. government efficiency (1990–1995), and so on. However, despite these
Our study has two interesting findings. First, we confirm the depen- efforts, the country is still highly dependent on the oil sector. This has
dency of Saudi Arabia's economy on the crude oil sector, but we also given rise to an increase in the unemployment rate, a decrease in social
show that oil price exhibits a threshold effect that varies depending services, and several budgetary cuts.
on the regime or state of the market. Identifying these states is particu- The interaction between oil price and the Saudi economy has been a
larly useful for hedging the economy against a further oil collapse. central focus of research for several economists for many years. The
Second, we show the benefit of economic transformation through the initial analyses suggested that excess oil price volatility requires a struc-
diversification option since only equity investment can boost the tural change in the Saudi economy and ongoing transformation of the
Saudi economy. However, it should be kept in mind that equity markets country that has shifted from an agricultural economy to an urbanized
can also increase the oil price effect on the real economy. This occurs one, with a large service sector dominated by the oil industry. De
when oil income is reinvested in financial markets. Overall, these Santis (2003) confirmed the country's dependency on the oil sector
findings seem to be in line with the expectations of the national and showed that the relationship between oil price and the Saudi econ-
transformation program and the Saudi Vision 2030 project. omy is largely due to the policies of oil producers and geopolitical
The remainder of the paper is as follows. Section 2 briefly shocks. Indeed, oil price fluctuations and its impact depends on quota
discusses the related literature. The econometric methodology is regimes and OPEC agreements.6 Alkhathlan et al. (2014) also showed
presented in Section 3. Section 4 discusses the main empirical results that Saudi Arabia's relations with the other OPEC countries has fluctu-
and Section 5 concludes. ated with time, and this can affect oil price and the real economy,
even if Saudi Arabia remains a price-maker.
2. Literature Samargandi et al. (2014) pointed to Saudi Arabia's efforts to diver-
sify its economy by promoting the non-oil sector, power generation,
Before moving to the analysis of the related literature survey, it is telecommunications, and natural gas exploration, and improving fi-
worthwhile to note some key policies and transformation measures nancial services (efficient banking system, well-developed financial
taken by the Saudi authorities to deal with the evolution of the oil markets, etc.).7 The authors showed that while financial develop-
price change on their economy. This provides an important backdrop ment has had a positive impact on growth in the non-oil sector in
to understand the evolution of the economic structure in Saudi Arabia the long-run, their ARDL (autoregressive distributed lag) analysis
in how it has dealt with oil price changes. Many actions and develop-
ment programs have been implemented by the Saudi rulers to boost oil
5
production profits and hedge their economy. First, in 1938–1950, oil pro- The Vision 2030 project is thus in line with this fourth development program.
6
duction increased by 400 times while oil income only increased by 18 This situation is similar in other countries such as Iran, where, according to Farzanegan
and Markwardt (2009), there is positive and significant dependency between oil price
times, leading the Saudi government to increase taxes to reduce the changes and growth in industrial output.
spread. While the 1970s was a prosperous period for the Saudi economy, 7
From Hossain and Serletis (2017), there is further evidence of substitution between
a slower oil effect was observed from the 1980s onwards. This change crude oil and gas.
 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19 15

Breakpoint Test Statistics

6

0
1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007

Fig. 3. The Andrews-Ploberger breakpoint test. Note: The test statistic is on the vertical axis.

highlighted a negative or insignificant outcome on the oil sector. Nili Teräsvirta, 1993), and threshold autoregressive models called TAR (Tong
and Rastad (2007) also noted a weaker impact of financial develop- and Lim, 1980). These models are often preferred to linear ones as they
ment on oil exporting countries (i.e. Saudi Arabia). The authors ex- offer more flexibility in modeling financial and macroeconomic variables
plained this finding by the inefficiency of financial institutions during different regimes and are, therefore, more appropriate for captur-
in oil dependent countries. Contrary to Nili and Rastad (2007), ing asymmetry in the data. However, while the transition is made by an
according to Beck (2011), even if the financial sector has a different observed state variable, the transition is expected to be smooth in
structure in oil exporting countries due to its underdeveloped finan- the STAR model (resp. abrupt for TAR models) and conducted when a
cial system, financial development significantly and positively af- known threshold variable exceeds the threshold. Both the transition vari-
fects oil exporting countries. Choudhury and Al-Sahlawi (2000) ables and the threshold were endogenously estimated and identified via
also pointed to shifts in the Saudi economy through an increase in linearity tests (Tsay, 1989; Teräsvirta, 1994). We then focused on the
the non-oil sector and the opening up of the Saudi stock market class of threshold models that appear to be the most appropriate for our
since 2008, resulting in an increased number of companies (172 research question to identify oil effects per regime while taking further
companies in 2013 against 72 companies in 2008, 22 insurance com- structural breaks and data asymmetry into account. Indeed, the peculiarity
panies in 2008 against 1 in 2004). It also yielded a higher liquidity of the TAR model is that it describes relationships that are linear per re-
ratio through increasing in the size of the Saudi banking system. gime but are nonlinear over the entire period and relationships for
Trabelsi (2017) also found a significant link between oil price which the transmission of oil effect is expected to be abrupt rather than
fluctuations and sectoral indices in Saudi Arabia. Interestingly, this smooth. Additionally, the abrupt transitions between regimes enabled
relationship appears to exhibit asymmetry. us to capture the rapid changes in dynamics between the variables follow-
Overall, the analysis of the above-mentioned studies generates two ing a shock affecting oil price. Moreover, the TAR model is preferred to the
interesting observations. First, the hypothesis of strong dependency of STAR model as the TAR model provides a pertinent specification that helps
the Saudi Arabian economy on crude oil is not rejected. Second, several to capture spikes and abrupt linkages related to a collapse in oil price.
projects involving economic initiatives and financial market openness TAR models were developed by Tong and Lim (1980) and discussed
have been launched, but related studies do not provide unanimous or in Tong (1990), while Tsay (1989) and Hansen (1996) have proposed a
conclusive results regarding the impact of these initiatives on the real complete modeling strategy for these models. Formally, a simple two-
economy. It is important to note that the econometric tools used to regime TAR model, denoted TAR (2, p, St), can be written as
assess these effects have always relied on a linear model that may not
p p
be appropriate to capture asymmetric and nonlinear effects of financial Y t ¼ α 10 þ ∑i¼1 α 1i Y t−1 þ ∑ j¼1 β1 j X t−1 þ ε1t if St ≤c
p p ð1Þ
development, except for the study by Alimi and Aflouk (2017) which Y t ¼ α 20 þ ∑i¼1 α 2i Y t−1 þ ∑ j¼1 β2 j X 1t− j þ ε 2t if St Nc:
also showed threshold effects in the relationship between oil price
and economic growth in the GCC economies. However, to our knowl- where St is the threshold variable. The coefficient c is the threshold
edge no study has yet focused on the National Transformation Program parameter, p is the maximum lag number and (α10, α1i, β1j) and (α20,
of Saudi Vision 2030. To fill this gap, the current study will investigate α2i, β2j) are the estimators in the first and second regimes respectively,
this program on its underlying assumptions regarding diversification ∀ I, j = 1, …, p. Yt is the endogenous variable (Saudi Arabia economic
and financial market openness and their effects on the real economy growth rate), while Xt is an explanatory variable (oil returns). The errors
in a nonlinear framework.8 are assumed to follow white noise processes.9
For the TAR model, the lag number (p) is specified using information
3. Econometric specification criteria and/or autocorrelation functions. Only sequential conditional
least squares can be used to estimate the TAR model. As was done in
The threshold model belongs to the general class of nonlinear models Tong and Lim (1980), implementation of the TAR modeling is carried
and includes three main classes: Markov models (Hamilton, 1994), out in three main steps: specifying p, defining the threshold parameter
smooth transition autoregressive models called STAR (Granger and c and the delay parameter d (that defines the transition variable) and es-
timating the two-equation system by the least squares (LS) method.
8
To justify this nonlinearity and threshold approach, we refer to a recent interview with However, since the values of the threshold c and the delay parameter d
James Hamilton in Jawadi (2018), in which Hamilton confirms that the relationship be- are unknown, the Tong and Lim (1980) method tends to be seldom
tween oil price and real economy is nonlinear and that it also exhibits threshold effects.
used in the literature. Accordingly, an alternative procedure based on
Indeed, the real economy reacts differently not only to a positive or a negative oil shock,
and distinctly to a small or a large oil shock, but it also appears that the economic reaction
9
may differ with regard to the state of the economy (expansionary versus recessionary). The threshold variable St might be a lagged endogenous variable (Yt−d or a lagged ex-
Further, even for the same but repetitively observed shock, the market reaction toward planatory variable (Xt−d)), where 1 ≤ d ≤ p is a delay parameter that is specified through
the first shock might differ from that of the same shock observed the second time. the linearity tests.
16 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19

Table 3 Table 4
Results of the Bai and Perron (2003) test. Results of linearity tests (p-values).

Breaks F-statistic Scaled Weighted Critical Series p d St Hansen Tsay Teräsvirta (1994) tests Model
F-statistic F-statistic value (1996) (1989)
H01 H02 H03 H12
1 3.675694 7.351387 7.351387 11.47
Dlrgdp 1 1 dlwtt−1 0.00a 0.00 0.76 0.00 0.40 0.00 TAR
2⁎ 11.74720 23.49440 27.63906 9.75
Note: H01, H02, H03 and H12 refer to the linearity tests against a STAR model as in Teräsvirta
UDMax statistic⁎ 23.49440 UDMax critical 11.70 (1994) (for more details on linearity tests, see Hansen (1996), Tsay (1989) and Teräsvirta
value⁎⁎ (1994)). H01: the fourth-order terms of the Taylor approximation are not significant. H02:
WDMax statistic⁎ 27.63906 WDMax critical 12.81 under H01, third-order terms of the Taylor approximation are not significant. H03: under
value⁎⁎ H02, second-order terms of the Taylor approximation are not significant (see Teräsvirta
(1994) for more details on these tests). St refers to the optimal threshold. d is the delay
Estimated break dates: 1: 1977; 2: 1986.
parameter and p is the autoregressive order.
⁎ Significant at the 0.05 level. a
Denotes the bootstrap p-values.
⁎⁎ Bai and Perron (2003) critical values.

linearity tests was introduced by Tsay (1989) and Hansen (1996) to a major risk since it is the main driver of the Saudi economy. This
estimate the TAR models. This approach is based on the threshold tests might explain the higher volatility of the estimated economic growth
and is conditioned by the estimated values for c and d (d is the delay rate (about 9.3%), even though it remains lower than the volatility of
parameter that specifies the threshold variable). We use this approach oil price (oil volatility is 2.82 times higher than economic growth vola-
in the current paper for three main steps: i) specification and linearity tility). Finally, the excess leptokurtic and asymmetric properties identi-
tests, ii) estimation, and iii) validation. The related linearity or threshold fied in the distribution of Saudi Arabia's economic growth might be seen
tests assess the null hypothesis of linearity against nonlinearity and help as indicators of nonlinearity in the data, even if, unlike the WTI and
to specify the threshold parameter, identify the transition variable, CDGP variables, normality is not rejected for the RDGP series. This non-
and test the presence of threshold effects and, therefore, nonlinearity linearity hypothesis will be evaluated later in the paper.
in the data.10 Third, before moving to the nonlinear estimation, we analyzed the
impact of oil price in a linear framework. To this end, we first computed
a bilateral Pearson correlation (Table 2).
4. Empirical analysis
From Table 2, we can note positive correlations between the oil
return and economic growth proxies in excess of 17%. This suggests
4.1. Data and preliminary analysis
a negative reaction and further correction of the Saudi economy
(economic growth and consumer wellbeing) following a negative oil
We used various macroeconomic and financial series from the
shock like the collapse in oil prices in 2014. This correlation is only
Federal Reserve Economic Data (FRED) service of the Federal Reserve
about 10% if we reconsider the sample till 2000, suggesting that depen-
Bank of St. Louis. We took quarterly data for 1970 to 2016 that enabled
dency on the oil industry increased in this period. This finding is illus-
us to study the impact of oil price during both calm and crisis periods, as
trated in Fig. 2. It shows that several episodes, profiles, and changes
well as to check for the Saudi Arabian economy's reaction to different oil
affected the link between oil price and Saudi Arabia's economic indica-
shocks induced by distinct causes. Our sample included four main
tors. Further, from Fig. 2, we can see that interactions between the oil
variables: constant GDP per capita as a proxy for economic growth
industry and the Saudi economy have experienced different phases.
and development, the spot WTI (West Texas Intermediate) as the
The relationship appears negative in the sub-periods 1973–1975,
benchmark proxy for oil price, the Saudi Riyal/US$ exchange rate, and
is at a weak level between 1993 and 1999, and shows positive
the Saudi stock index—Tadawul All-Shares Index—as the benchmark
co-movements for the other years in our sample. According to the
for stock price.11 We also used real GDP as a second proxy and com-
interview with James Hamilton published in Jawadi (2018), this linkage
puted Saudi Arabia's economic growth differently using real GDP and
still works even if the sources of oil shocks and their occurrences might
constant GDP per capita.
differ (supply shocks versus demand shocks).
First, we tested the null hypothesis of the unit root in the data using
Next, we look at the relationship between economic growth vari-
the augmented Dickey-Fuller test. Our findings show that oil price, GDP
ables and oil price returns in the Saudi economy in a linear framework.
(for both proxies), and stock price are I (1), and the hypothesis of
Overall, our findings point to a positive and significant (higher after
stationarity is not rejected for the exchange rate series.12
2000) correlation of the oil price with the economic growth in
Second, we computed the main descriptive statistics for the two key
Saudi Arabia. This finding is in line with our analysis above, suggesting
variables associated with Saudi Arabia's economic growth (using the
that Saudi Arabia's dependency on the oil industry increased signifi-
two proxies) and oil returns. We report the main results in Table 1.
cantly over the last decade. This result would also point to further struc-
The two proxies used for economic growth (real GDP and constant
tural changes in the oil-economic growth relationship. We check this
GDP per capita) inform us, respectively, about economic growth and de-
hypothesis next through structural and linearity tests.
velopment in Saudi Arabia. In Table 1, Saudi Arabia exhibits a growth
rate of 3.7%, while variations in the development indicator and the
wellbeing indicator are close to zero, suggesting that the country's
4.2. Structural break tests
qualitative development is lower than its quantitative development.
Oil return averages are around 7.6%, confirming the importance of oil
We checked for the presence of further structural breaks in the
income for Saudi Arabia, and the oil return/economic growth rate ratio
dynamics of Saudi Arabia's growth rate, which indirectly enabled
exceeds 2.05, which means that 48.78% of economic growth in Saudi
us to test for a threshold effect in the data. To this end, we first ap-
Arabia comes from oil production. This confirms Saudi Arabia's strong
plied Andrews-Ploberger (1994) test for structural break, which
dependence on the oil industry. However, oil prices were extremely
examines the presence of a single structural break at an unknown
volatile in the period under consideration (about 26.3%), representing
point. From the results of the Andrews-Ploberger breakpoint test sta-
10
tistics given in Fig. 3, we do not reject the presence of a significant struc-
See Tsay (1989) and Hansen (1996) for more details on these threshold tests.
11
Data for the stock index was obtained from Bloomberg.
tural break. This finding is line with the results of the linear modeling
12
We have not reported the results of the unit root tests to save space; however, the re- that point to a further shift in the oil price-Saudi Arabia economic
sults are available on request. growth relationship.
 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19 17

0.3

0.2

0.1

0.0

-0.1

-0.2

-0.3
-0.75 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00

Fig. 4. The Tsay arranged autoregression test statistic. Note: The vertical axis reports the Tsay test statistics against the threshold value placed on the horizontal axis. It provides an arranged
regression test for threshold autoregression.

Second, we applied the structural break test of Bai and Perron data and the dynamics that vary according to regimes. Accordingly, we
(2003)13 that enabled us to check for the presence of multiple breaks specify our model as follows:
and to date the breaks. Our findings point to two main significant
structural breaks, in 1977 and 1986, corresponding to the aftermath dlrgdpt ¼ β10 þ β11 dlwtit þ ε1t if st ≤c
of the first oil shock and the 1986 oil price collapse respectively 14 ð2Þ
dlrgdpt ¼ β20 þ β21 dlwtit þ ε2t if st Nc
(Table 3).
where β11 and β21 refer to coefficients that measure oil price effect in the
4.3. Linearity tests first and second regimes respectively, c represents the threshold parame-
ter, and st is the threshold variable that is identified through linearity tests.
Next, we applied different and more robust linearity tests to check ε1t and ε2t are the error-terms of the first and second regimes that are as-
for the presence of threshold effects in Saudi Arabia's economic growth sumed to be i.i.d. β10 and β20 are the two constants in regimes 1 and 2.
dynamics (Hansen (1996) test, Tsay (1989) test, and Teräsvirta (1994) We report the main results of this model in Table 5. Our findings do
test). The first two tests assess linearity against nonlinearity of the TAR not reject the threshold effect hypothesis with the one period lagged oil
type, while the Teräsvirta (1994) linearity tests check linearity against a return as a transition variable. Accordingly, we identified two distinct
STAR nonlinearity type with smooth transition between regimes.15 We regimes to characterize the Saudi Arabian economic growth-oil price
conducted these three tests and report the main results in Table 4 and relationship in which oil price has a positive and significant impact on
Fig. 4. the Saudi Arabian economy. The time variation in the oil effect in the
Our findings show further evidence of nonlinearity and threshold ef- second regime is five times lower than time variation in the first regime.
fects in the data. While both the Tsay and the Hansen tests (Table 5) re- It seems the transition occurred in 1986 in the aftermath of the 1980s oil
ject the null hypothesis of linearity, the Teräsvirta test also rejects glut caused by a collapse in global oil demand. Overall, while these
linearity but prefers an ESTAR model (exponential STAR), since H02 is results positively confirm the dependency of the Saudi Arabian econ-
rejected while H03 is accepted. Further, the high dispersion shown in omy on the oil industry, our observation of a less significant oil effect
Fig. 4 illustrates the importance of nonlinearity in the data and confirms in the second regime suggests that the Saudi rulers might benefit from
the above structural breaks hypothesis and rejects the null hypothesis of seeking additional resources for their economy, in line with the diversi-
“no threshold”. fication hypothesis that is the basis for forthcoming transformations
Accordingly, to better investigate the oil price effect on Saudi eco- (Vision 2030). However, the adjusted R2 shows weak values, and
nomic growth, we need to extend the linear model while the relation- estimated errors indicate at least a one-order autocorrelation bias.
ship varies per regime and according to the economic cycle. To this Next, we tested the diversification hypothesis by extending the
end, we propose estimating the dynamics of Saudi Arabia's economic above specification to incorporate stock returns from the Tadawul
growth using a two-regime TAR model next, in line with Tsay (1986, All-Share Index and the Riyal/US$ exchange rate among our explana-
1989) and Hansen's (1996) modeling strategies. tory variables.16 We report the main results in Table 6, which gives us
several interesting findings. First, like the previous estimate, two
regimes were identified, and a positive and significant oil effect is
4.4. Modeling the oil effect in a two-regime threshold specification
found that varies per regime, validating our specification. Second, the
diversification hypothesis suggested by the Vision 2030 program
After rejecting linearity, we model nonlinearity in the Saudi
appears to be credible as stock market investments seem to significantly
economic growth dynamics to check whether the oil price effect varies
boost Saudi Arabia's economic growth given that the elasticity of eco-
with the phases of the business cycle or not. To this end, we estimated a
nomic growth to the stock market is about 1.3%. Third, the comparison
two-regime threshold regression model. This On/Off modeling captures
of oil effects in Tables 5 and 6 indicates that this diversification has an
nonlinearity and asymmetry as well as abrupt structural changes in the
indirect and positive effect, since the oil return effect on the real econ-
omy is higher when there is a more diversified market. Finally, the
13
relative variability of the Saudi exchange rate in the first regime (before
The reader can refer to Jawadi and Sousa (2013) for more details on this test.
14 1986) seems to positively affect the dynamics of the country's economic
The 1980s oil glut corresponds to a period of considerable crude oil surplus induced by
falling demand after the 1970s energy crisis. Indeed, the global oil prices, after reaching
16
over $35 per barrel in 1980, fell from $27 to below $10 in 1986, leading to an oil price Data for the stock index has only been available since the 1990s. For this reason, this
collapse. variable was only considered for the second regime. As for the exchange rate, it was not
15
For more details on these linearity and threshold tests, see Tsay (1989), Teräsvirta introduced in the regression of the second regime because Saudi Arabia has been pegged
(1994), Hansen (1996) and Van Dijk et al. (2002). to the US$ since 1986.
18 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19

Table 5 on the dynamics of economic growth in a nonlinear framework.

Results of univariate threshold regression. We show two interesting results. First, changes in oil price have a
Regime 1 Regime 2 positive and significant effect on the dynamics of Saudi Arabia's
Constant 0.002 0.033⁎⁎⁎
economic growth, confirming the dependency of Saudi Arabia's
(0.966) (0.000) economy on the oil sector, but this oil effect enters nonlinearly and
DLWTI 0.212⁎⁎⁎ 0.037⁎⁎ appears to vary per regime. Second, the diversification hypothesis,
(0.000) (0.05) the basis of the Vision 2030 program appears credible since invest-
c −0.019⁎⁎⁎ −0.019⁎⁎⁎
ment in the Saudi stock market not only positively boosts Saudi
(0.000) (0.000)
Adjusted R2 0.12 0.07 economic growth but also stimulates the effect that oil price has on
DW 1.04 1.25 the real economy. While this supports opening up and developing
Note. DW denotes the Durbin-Watson test statistics. Values in brackets are the p-values of
the Saudi financial market, it also suggests that while rebalancing
student tests with heteroscedasticity-consistent (Eicker-White) standard errors. their portfolio to invest in financial assets and commodities, inves-
⁎⁎ Denote statistical significance at 5%. tors taking this risk would significantly and positively affect the
⁎⁎⁎ Denote statistical significance at 1%. real economy in Saudi Arabia. Accordingly, this study expects the
ongoing Vision 2030 program to provide a growth stimulus for the
growth. These findings point to a nonlinear dependency of the economy Saudi economy. Thus, our results are in favor of diversification and
of Saudi Arabia on the oil sector and supports diversifying the domestic by extension, the National Transformation Program itself. However,
economy. To check the robustness of these results, we applied an it is clear that the program's success requires more transparency
autocorrelation (LB) and heteroscedasticity (ARCH) test to show that and symmetry of information from policy makers and the Saudi
the estimated residuals have appropriate statistical properties. rulers (information on oil reserves, extraction cost of oil barrel,
Overall, in line with the findings of Alimi and Aflouk (2017), our etc.), which in itself is a significant challenge for the normally
study points to nonlinearity and threshold effects in the relationship be- secretive Kingdom. Nevertheless, this study relies on aggregate
tween oil price and economic growth. However, it differs in several data, and it would be interesting to extend the research using sec-
ways from the study of Alimi and Aflouk (2017). First, while Alimi and toral data to identify the most significant financial sector for the
Aflouk (2017) modeled nonlinearity using panel smooth transition re- Saudi Arabian economy. Finally, while the WTI is used as a proxy
gression (PSTR) models, we focused on the threshold model, a choice for oil price, it is possible to check the robustness of these findings
justified by the fact that oil shock transmission is expected to be abrupt using other oil proxies too (Brent and OPEC).
rather than smooth, as has always been the case. This is particularly true
for an economy like that of Saudi Arabia where the oil sector makes a
major contribution. Second, while Alimi and Aflouk (2017) proposed Appendix A. Supplementary data
applying a heterogeneous group related to the GCC economies, we fo-
cused on just one major oil producer (Saudi Arabia), and that enabled Supplementary data to this article can be found online at https://doi.
us to produce a more concise econometric analysis with useful eco- org/10.1016/j.eneco.2018.12.003.
nomic and political implications for policymakers, even though both
our results and those of Alimi and Aflouk (2017) point to a significant
References
nonlinear oil price effect on the real economies. Finally, unlike Alimi
and Aflouk (2017), we tested the diversification hypothesis, which is a Alimi, A., Aflouk, N., 2017. The threshold effect of oil-price shocks on economic growth.
major assumption of the Saudi Vision 2030 program. Our findings Econ. Bull. 37 (3), 1942–1958.
Alkhathlan, K., Gately, D., Javid, M., 2014. Analysis of Saudi Arabia's behavior within OPEC
indicate the value of diversification in boosting the domestic economy and the world oil market. Energy Policy 64, 209–225.
and reducing its dependence on the oil sector. Andrews-Ploberger, 1994. Testing for Serial Correlation Against an ARMA(1, 1) Process.
unpublished manuscript. Cowles Foundation, Yale University.
Bai, J., Perron, P., 2003. Computation and analysis of multiple structural change models.
5. Conclusion J. Appl. Econ. 18, 1–22.
Beck, T., 2011. Finance and oil: is there a resource curse in financial development. Center
Discussion Paper No. 2011-017. Tilburg University, Netherlands.
We studied the dependence of the Saudi Arabian economy on oil Bottazzi, G., 2016. Short Report on Oil Price History. Working paper Report December.
price through an assessment of the impact of oil price fluctuations Choudhury, M.A., Al-Sahlawi, M.A., 2000. Oil and non-oil sectors in the Saudi Arabian
economy. OPEC Rev. 24 (3), 235–250.
De Santis, R., 2003. Crude oil price fluctuations and Saudi Arabia's behaviour. Energy Econ.
Table 6
25, 155–173.
Results of augmented threshold regression. Elder, J., Serletis, A., 2010. Oil price uncertainty. J. Money Credit Bank. 42 (6), 1138–1159.
Engle, R.F., 1982. Autoregressive conditional heteroskedasticity with estimates of the var-
Regime 1 Regime 2
iance of U.K. inflation. Econometrica 50, 987–1008.
Constant −0.971⁎⁎⁎ 0.025⁎⁎⁎ Farzanegan, M.R., Markwardt, G., 2009. The effects of oil price shocks on the Iranian
(0.000) (0.000) economy. Energy Econ. 31, 134–151.
DLWTI 0.261⁎⁎⁎ 0.116⁎⁎⁎ Granger, C.W.J., Teräsvirta, T., 1993. Modelling Nonlinear Economic Relationships. Oxford
(0.000) (0.000) University Press, Oxford.
EXR 0.268⁎⁎⁎ – Hamilton, J.D., 1994. Time Series Analysis. Princeton University Press, Princeton.
Hansen, B.E., 1996. Inference when a nuisance parameter is not identified under the null
(0.000)
hypothesis. Econometrica 64 (2), 413–430.
Stock return – 0.013⁎⁎⁎
Hossain, N., Serletis, A., 2017. A century of interfuel substitution. JCM 8, 28–42.
(0.000) Jawadi, F., 2018. Understanding oil price dynamics and their effects over recent decades:
c 0.0904⁎⁎⁎ 0.0904⁎⁎⁎ an interview with James Hamilton. Energy J. (forthcoming).
(0.000) (0.000) Jawadi, F., Sousa, R., 2013. Modelling money demand: further evidence from an interna-
Adjusted R2 0.44 0.38 tional comparison. Appl. Econ. Lett. 20 (11), 1052–1055.
DW 1.90 1.99 Nili, M., Rastad, M., 2007. Addressing the growth failure of the oil economies: the role of
LB test (0.770) (0.620) financial development. Q. Rev. Econ. Finance 46, 726–740.
ARCH test (0.560) (0.340) Samargandi, N., Fidrmuc, J., Ghosh, S., 2014. Financial development and economic
growth in an oil-rich economy: the case of Saudi Arabia. Econ. Model. 43,
Note: EXR denotes the Riyal/$ exchange rate. 267–278.
LB and ARCH tests correspond to Ljung-Box autocorrelation and Engle (1982) tests, Teräsvirta, T., 1994. Specification, estimation, and evaluation of smooth transition
respectively. autoregressive models. J. Am. Stat. Assoc. 89, 208–218.
Values in brackets are the p-values of the tests. Tong, H., 1990. Non-linear Time Series: A Dynamical System Approach. Oxford University
⁎⁎⁎ Denote statistical significance at 1%. Press, Oxford.
 F. Jawadi, Z. Ftiti / Energy Economics 80 (2019) 12–19 19

Tong, H., Lim, K.S., 1980. Threshold autoregression, limit cycles and cyclical data. J. R. Stat. Tsay, R.S., 1989. Testing and modeling threshold autoregressive processes. J. Am. Stat.
Soc. 42 (B), 245–292. Assoc. 84, 231–240.
Trabelsi, N., 2017. Asymmetric tail dependence between oil price shocks and sectors of Van Dijk, D., Terasvirta, T., Franses, P.H., 2002. Smooth transition autoregressive models—
Saudi Arabia system. J. Economic Asymm. 16, 6–41. a survey of recent developments. Econ. Rev. 21, 1–47.
Tsay, R.S., 1986. Nonlinearity tests for time series. Biometrika 73 (2), 461–466.