Electricity demand modeling in Saudi Arabia

Contents lists available at ScienceDirect

The Electricity Journal

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

Electricity demand modeling in Saudi Arabia: Do regional differences matter?

Jeyhun I. Mikayilova,b,*, Fakhri J. Hasanova,c, Waheed Olagunjud, Mohammad H. Al-Shehrie

a King Abdullah Petroleum Studies and Research Center, PO Box 88550, Riyadh, 11672, Saudi Arabia b Azerbaijan State University of Economics (UNEC), Istiqlaliyyat Str., 6, Baku, Azerbaijan c The George Washington University, 2115 G Street, NW, Washington, DC, 20052, USA d Ministry of Energy, King Saud St., Al Wizarat, Riyadh, 12622, Saudi Arabia e Saudi Electricity Company, Takhassusi St., Riyadh, 12333, Saudi Arabia

A R T I C L E I N F O

Keywords: Electricity demand Price elasticity Cointegration Regions Saudi Arabia

A B S T R A C T

The paper examines the electricity demand behavior for Saudi Arabia, using annual data for the period of 1990–2016, at regional level. The study finds that income, price and population are the main drivers of elec- tricity demand at regional level. Although, the impacts vary across regions (central, eastern, southern and western), the estimated elasticities all are statistically significant, in both long and short run, and have the expected signs for all the regions. The income, price and population elasticities range from 0.10 to 0.93, from -0.63 to -0.06, and from 0.24 to 0.95, respectively, across regions in the long run. In the short run these intervals are (0.05, 0.47), (−0.27, −0.01) and (0.13, 1.49) respectively for income, price and population across the regions. The findings reported in this paper, should assist policy makers to develop insights about the potential regional impact of changes to electricity prices, income and population patterns.

1. Introduction

Since energy is a pervasive input to all business and recreational activities, total energy demand is an important indicator that helps explain the pattern of economic development within a country. Identifying and understanding the key determinants of electricity de- mand is therefore important for the economic prosperity of a country, since the availability of reliable electricity directly affects the prospects of sustainable economic development. With important mega projects already in the works and national transformation programs such as the National Industrial Development and Logistics Program (NIDLP, 2019) already being implemented, understanding the existing and projected behaviour of electricity demand is more important now than ever be- fore.

While it is relevant to conduct a study that investigates the de- terminants of aggregate electricity demand in KSA, this is a topic that has already been addressed by the existing literature to some extent (Atalla and Hunt, 2016, inter alia). However, aggregate level electricity demand has not been studied at regional level. Hence, there is a need for a research to investigate the aggregate level electricity demand accross regions.

The objective of this research is to understand how determinants of

the electricity consumption shape it over time in the Central, Eastern, Southern and Western regions of Saudi Arabia and suggest policy in- sights accordingly.

Note that this 4-region classification is adopted by the Saudi Electricity Company (SEC) and sum of these four regions’ consumption add up to the total electricity consumption of the Kingdom.

The following points motivated us to conduct electricity consump- tion analysis at the regional level. First, the regional dimension is im- portant because of the differences in weather patterns (and subse- quently the electricity demand profiles) across the country. Second, the distribution of residential, commercial, and industrial activities is dif- ferent across the regions and this may imply different relationships between electricity consumption and its drivers. For example, in the eastern region the electricity consumption is mainly industry-driven whereas in western region, where we have the holy cities of Makkah and Madinah, this is mostly population/residential driven (SAMA, 2019). Third, the implementation of the mega projects mentioned above will also have implications for electricity demand at the regional level. Finally, recent analysis show that different regions react differ- ently to the energy, of which electricity price reform, one of the key initiatives in the Fiscal Balance Program of the Saudi Vision 2030 (Alyamani et al., 2019).

https://doi.org/10.1016/j.tej.2020.106772

⁎ Corresponding author. E-mail address: [email protected] (J.I. Mikayilov).

The Electricity Journal 33 (2020) 106772

Available online 20 April 2020 1040-6190/ © 2020 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).

T

A regional approach for KSA constitutes a novel and significant contribution to the literature on electricity demand in the following ways. First, to the best of our knowledge, there are very few to none studies that examine the determinants of electricity demand at the re- gional level for KSA. Alyamani et al. (2019) considered regional aspects of the electricity consumption. However, they only considered re- sidential electricity consumption and did just a descriptive analysis. Diabi (1998) examined total electricity consumption for the five regions of Saudi Arabia over the period 1980−1992. However, he did a panel analysis and did not estimate region-specific parameters/elasticities, and hence is different from time series that we conduct in this research. Additionally, he did not address integration-cointegration and other properties of the data used such as cross-sectional dependency, which can lead a serious issue such as bias and inconsistencies and thus mis- leading policy recommendations. Second, it investigates the demand for electricity at a disaggregated regional level, which takes into account the region specific features of electricity demand behavior; Third, it takes into account the impact of demographic factors which may play a significant role in electricity demand formation; Fourth, it uses the more recent data, which partially enables us to see the impact of the ongoing energy price reforms and consequences of the low oil price environment on electricity demand.

From a policy perspective, given the current transition from a heavily subsidized electricity price environment to a market-based price environment where subsidies are gradually phased out, it is im- portant to consider the impact of this transition from a regional per- spective. The findings reported in this study would help a better un- derstanding of the regional impact (on electricity demand) of different price policy scenarios and changes in population and income. The findings will also be useful in determining which regions, and how much, to provide transitional support to in order to alleviate some of the financial hardships associated with rising electricity prices.1

This paper employs cointegration and equilibrium error correction (ECM) methodology to develop long- and short-run price, income and demographic elasticities for regional electricity demand.

The study concludes that income, price and population are the main drivers of electricity demand at regional level as theoretically expected. Although, the impacts vary across regions, the estimated elasticities all are statistically significant, in both long and short run, and have the expected signs for all the regions. The income, price and population elasticities range from 0.10 to 0.93, from -0.63 to -0.06, and from 0.24 to 0.95, respectively, across regions in the long run. In the short run these intervals are (0.05, 0.47), (-0.27, -0.01) and (0.13, 1.49) respec- tively for income, price and population across the regions. The obtained Speed of Adjustment (SoA) coefficients are significant in all cases in- dicating the short-run deviations from the long-run relationship con- verge back to the equilibrium path.

The rest of the paper is structured as follows; Section 2 reviews the literature on electricity demand modeling in the case of Saudi Arabia, Section 3 contains the theoretical framework and Section 4 briefly de- scribes the methodology used, Section 5 presents the data. Estimation results are presented in Section 6. In Section 7, we discuss our key findings, while Section 8 concludes the study and provides the policy implications where we link our findings to the current policy environ- ment.

2. Literature review

In this section, we highlight some of the key trends in the literature on residential, industrial and total electricity consumption in Saudi

Arabia given that there are only two studies, to the best of our knowledge, analyzing the regional electricity consumption in Saudi Arabia. A recent study by Alyamani et al. (2019) discussed regional aspects of electricity consumption in residential sector. However, Alyamani et al. (2019) is descriptive-type study and does not estimate elasticities. Diabi (1998) studied total electricity consumption for the five regions of Saudi Arabia, employing the data spanning from 1980 to 1992. He estimated the kingdom-wise elasticities not regional-specific ones. The study is relatively old which might not reflect the electricity demand behaviour for the current period, due to the substantial changes in terms of economic development. In addition, Diabi (1998) uses panel estimation techniques, without performing integration-co- integration analyses, did not address the potential cross-sectional de- pendency within regions. The mentioned limitations might result in biased estimation results and consequently misleading conclusions.

The studies investigating electricity consumption employed a wide range of analytical techniques that are not necessarily quantitative or econometrically sound. We found studies that focused on qualitative analysis, Granger causality analysis, simulations based on optimization models, and a range of econometric estimation methods.

For example, both Hagihara (2013) (who simply described the en- ergy outlook in KSA) and Alrashed and Asif (2014) who conducted a survey analysis of residential electricity consumption (REC) in the Eastern province of KSA can be seen as qualitative analysis.

Matar (2017) and Matar and Anwer (2017) are examples of studies that adopted a simulation-based approach when investigating elec- tricity consumption. Both studies explored the impact of electricity price changes in KSA on residential electricity consumption in 2011 and 2015, respectively. In both studies, the simulations were conducted using an optimization based partial equilibrium model. They did not explicitly consider income and demographic effects and there were no elasticities reported due to the nature of the study.

Since the focus of this review is on econometric studies in line with the nature of our research here, we continue by reviewing the existing literature with emphasis on the type of data that was used, the econometric methodology and the specifications employed, the em- pirical analysis strategy adopted (e.g., whether a study at hand ad- dressed stochastic properties of the data) for the study.

We compare the model specifications to the standard specification as dictated by the theories, which requires that we account for income, price and demographic effects (Beenstock and Dalziel, 1986; Liddle and Lung, 2010; Hasanov, 2019, inter alia). This is important because the results from studies that do not account for all factors could potentially contain some omitted variable bias.

For industrial electricity demand in KSA, there are some studies investigating this relationship. Al-sahlawi (1999) utilized aggregate time series data and Eltony and Mohammad (1993) and Liddle and Lung (2010) utilized country-level panel data but did not report KSA specific estimates. In their specifications, Al-sahlawi (1999) considered only income, Eltony and Mohammad (1993) considered income and price, and Liddle and Lung (2010) considered only urbanization rates. More recently, Hasanov (2019) studied the determinants of industrial electricity demand for Saudi Arabia by analyzing the core drivers of electricity consumption in industrial sector.

As mentioned earlier, the results from studies that do not account for income, price and demographic effects might be biased. Since first three studies mentioned above do not account for all 3 factors, they may contain some bias. Furthermore, Al-sahlawi (1999) and Eltony and Mohammad (1993) did not consider the integration-cointegration properties of the variables included in their analysis before using OLS estimation. Therefore, their results might be biased from the spurious regression perspective. In addition, both studies are quite old and the relationship between the variables of interest might demonstrate new path with the recent data.

Next, we turn our attention to key studies on residential electricity demand. In this area, there were some studies with a regional focus.

1 Ideally, a regional analysis for each electricity consumer category would provide further insights about the impact of rising electricity prices on different consumer categories and allow for a more targeted support approach, but that is beyond the scope of our analysis. Authors currently are working on this task.

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

2

Precisely speaking, they focused on the Gulf Council Countries (GCC) as a region and not the regions within KSA as we do in this study. The earliest study we found was the work of Eltony and Mohammad (1993) who examined residential electricity demand for a panel of GCC countries including KSA and found long run (LR) income and price elasticities for GCC countries of 0.20 and −0.14, respectively. The study did not explicitly account for demographic effects. Al-Sahlawi (1999) estimated the short run (long run) income and price elasticity of residential electricity demand as 0.13 (0.70) and −0.10 (−0.50), re- spectively. This study also did not explicitly account for demographic effects. Both studies used OLS and did not account for the integration- cointegration properties of the variables.

Atalla and Hunt (2016) use a structural time series model (STSM) and data from 1985 to 2012 to investigate residential electricity con- sumption for GCC countries. Unlike other studies examined thus far, they accounted for all three factors required to explain residential electricity consumption. They also used Cooling and Heating Degree Days variables as a proxy for weather conditions. For KSA, they found the following long-run elasticities: 0.48 for income -0.16 for price and 0.80 for population, which was the variable representing the demo- graphic effect. The study also concludes that the short-run (SR) price and population elasticities are-0.16 and 4.20, respectively, while in- come does not affect the demand in the short-run.

With regards to modelling total electricity consumption, there are several studies for KSA.

Al-Faris (2002) used a Vector Error Correction Model (VECM) ap- proach and data from 1970 to 1997 to find LR (SR) income and price elasticities for KSA of 0.05 (1.65) and -0.04 (-1.24), respectively. Narayan and Smyth (2009) using data from 1974 to 2002 and FMOLS method reported long run income elasticity of electricity consumption for KSA of -3.07. In addition, the paper does not explain/interpret the found unusual negative and substantially higher income elasticity.

Liddle and Lung (2010); Karanfil and Li (2015) and Mohammadi and Amin (2015) use ECM and panel data for many countries including KSA to investigate the causal relationship between total electricity consumption and urbanization. They found that GDP per capita and urbanization granger cause total electricity consumption per capita.

Hasanov et al. (2017) use an equilibrium correction model and panel data for a number of oil exporting countries including KSA to investigate the relationship between GDP, residential electricity con- sumption, foreign direct investment and employment. One of their key findings was that employment Granger-causes residential electricity consumption in the short run. Similarly, Salahuddin et al. (2015) using panel data for GCC countries found that GDP per capita granger causes total electricity consumption per capita and reported a LR income elasticity of 0.41.

Diabi (1998) analyzed regional total electricity consumption in KSA, based on panel data (1980–1992) for five KSA regions at the time (CR, WR, ER, SR and NR). The study compared the results of different esti- mation methods (OLS, CHTA, CCTA, FE, RE and MLE) and accounted for income, price and demographic effects using the urbanization rate. He reported long run elasticities for KSA of 0.09 to 0.49 for income, −0.14 to 0.00 for price and 0.93–1.30 for urbanization. Corresponding SR elasticities are income (0.05 to 0.33), price (-0.12 to 0.00) and ur- banization (0.62–1.10).

In summary, there are a number of studies examining residential, industrial and total electricity consumption in KSA, but to the best of our survey, there are no studies that examine regional total electricity consumption. Considering this fact, the current study aims to in- vestigate the determinants of electricity demand for Saudi Arabia at the regional level using different cointegration techniques.

3. Theoretical framework

We use a standard formulation suggested by the theories such as demand-side approach :

Electricity use = F (price, income, population)

Where Electricity use is the total regional electricity demand, Income is income proxy, Population is population size of the appropriate region and Price is real electricity price.

The econometric functional relationship can be formulated as follow:

= + + + +Electricity use α α Income α Population α Price e0 1 2 3

Since increases in income and population increase electricity de- mand, while price increases negatively affect the demand of electricity, the expected signs for the coefficients α α, and 1 2 are positive while α3 it expected to be negative. e

All the variables are in logarithmic form in above specification; hence, the coefficients are elasticities, which capture the percentage change in electricity use as a result of a 1% change in the variable considered.

4. Methodology

For estimating the long-run relationships between the variables of interest, the paper uses cointegration techniques such as Dynamic Ordinary Least Squares (DOLS), Canonical Cointegration Regression (CCR) and FMOLS. For estimating the short-run elasticities and Speed of Adjustment (SoA), we used ECM in the framework of General to Specific Modeling Strategy (Gets) (Campos et al., 2005; Hendry et al., 2008; Doornik and Hendry, 2009; Doornik, 2009; Doornik and Hendry, 2018, inter alia). In addition, since we are using time series data the variables should be tested for their integration properties, and the Dickey and Fuller (1981) unit root test is used for this exercise. For cointegration exercises, we used the Engle and Granger (1987) coin- tegration test. Considering that, all the aforementioned econometric methods and tests are widely used and well known to the researchers, we are not describing them here. Interested readers can refer to Saikkonen (1992); and Stock and Watson (1993) for DOLS, and Phillips and Hansen (1990) for FMOLS, and Park (1992) for CCR methods.

5. Data

In this section, we describe the data that was used in our analysis by discussing some descriptive statistics and technical properties of the data.

The paper uses annual time series data for the sample spanning from 1990 to 2016, which is chosen based on the data availability. The data used in the study are described below:

Electricity use represents regional electricity consumption in each of the four regions measured in Mwh; these data are taken from the Saudi Electricity Company (SEC).

Price is the regional (weighted based on the regional consumption types) real weighted average electricity price, in SAR per TOE (Saudi Arabian Riyals per Ton of Oil Equivalent). To convert the nominal electricity prices to real values Consumer Price Index (CPI, index 2010 = 100) and GDP deflator (PGDP, 2010 = 100) values are used2 . CPI data is taken from Saudi Arabia Monetary Authority’s (SAMA, 2017) website (SAMA annual stats: Cost of Living Index by Divisions; http://www.sama.gov.sa/en-US/EconomicReports/Pages/ YearlyStatistics.aspx). PGDP data taken from the web page of General Authority for Statistics of Saudi Arabia (https://www.stats.gov.sa/en/ 823). Nominal electricity price is the nationwide or aggregate price as Saudi Arabia does not apply different prices to different regions.

2 The nominal price values were converted to real prices using GDP deflators for the Central and Southern regions, and CPI for the Eastern and Western re- gions depending on which deflated price produce economically meaningful and statistically significant results.

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

3

Income is proxied by GDP and disposable income (DI), both in Millions SAR, at 2010 prices. GDP data is taken from the web page of General Authority for Statistics of Saudi Arabia (General Authority for Statistics (GaStat), 2018). DI data is calculated by KAPSARC re- searchers.

Population is regional population, in persons, as a proxy for the demographic factors. This data was aggregated from the General Authority for Statistics of Saudi Arabia as it is available for 13 provinces (SAMA, 2019).

Fig. 1 demonstrates the graphs of the shares of regional electricity consumption in total country level electricity consumption.

As can be seen from the figure the share of electricity consumption significantly higher in the eastern region in early 90th, while decreasing over time it approached to the consumption levels in central and wes- tern region in recent years.

Theoretically, the long-run analysis can be conducted in logs or levels; here we choose the log-log specification due to its ease of in- terpretation and compatibility with existing literature (to allow easier comparisons with previous and future studies).

6. Empirical estimation results

6.1. Unit root test results

As is standard practice in the literature, the main concern lies in dealing with stochastic properties of the variables and hence the Augmented Dickey-Fuller (ADF) Unit Root test was employed for each variable in levels and first differences. In unit-root testing, the max- imum lag length is taken to be two and optimal lag is chosen based on Schwarz information criterion (SIC). Table 1 displays the results of the Unit Root tests.

Briefly, based on the results displayed in Table 1 we can conclude that all variables are integrated of the first order [I(1)] and there is potential for cointegration relationship. Hence, we can proceed to the

next step and test variables for the common trend. The next section provides the results of cointegration analyses.

6.2. Cointegration test results

To test the existence of cointegration relationship, i.e., whether the variables share the long-run common trend, we employed the Engle- Granger (EG) cointegration test and the results are given in Table 2.

As the table demonstrates, for all regions except the EOA, we found conclusive proof of cointegration. For the three regions, all the p-values from the EG tests were below the 5% level. This confirmed the presence of a cointegration relationship at the 5% significance level. For further investigation of the existence of the long-run relationship in the case of the Eastern region, we employed the Variable Addition Test (VAT) for cointegration, proposed by Park (1990), which states existence of co- integration as a null hypothesis. The test statistic is 4.932 with the p- value of 0.085, concluding the cointegration relationship at 8.5 % significance level. Considering the results of employed tests, we con- clude that, for all the regions there is a long-run relationship.

6.3. Long-run estimation results

After concluding the cointegration relationship among the variables the long-run estimation results can be interpreted so they are not spurious. The long-run estimation results are provided in Table 3. To have an idea of the impacts in terms of magnitude ranges, we reported the 99 % confidence intervals for the estimated coefficients/elasticities. The interval estimator provides more information about the re- presentativeness quality of the estimated coefficient, in addition to delivering the limits of the impact.

Based on the estimation results presented in Table 3, one can see that for all the regions, the impacts of income, price and population are economically meaningful as they take right signs and statistically sig- nificant at 1% level. As can be seen from Table 3, the income elasticity of electricity demand varies across regions, demonstrating more stable behavior in the Western region and wider ranged in the Central region. The highest income elasticity found in Central region, while it has the smallest value in the Eastern region. Overall, the income elasticity ranges from 0.102 to 0.931 across the regions.

Price elasticity of electricity demand is found to be ranging from -0.607 to -0.060 across the regions.

The impact of population demonstrates more similar path across regions, having differences as well. Numerically the population elasti- city ranges from 0.243 to 0.947.

6.4. Short-run estimation results

For the short-run analysis, we applied Gets to ECM to determine the relationship for each region. We start with a general ECM specification, which includes the error correction term (ECT), contemporaneous va- lues of all independent variables and 2 lags of all the variables. We then exclude variables from the analysis based on the test proposed by the methodology and end up with the final short-run specification (e.g., see Campos et al., 2005 inter alia). We are not reporting the General Un- restricted Models here to save space, but they are available from the authors upon request. Table 4 documents the final ECM specification

Fig. 1. Shares of regional electricity demands in total, %.

Table 1 ADF Unit Root Test results.

Region Variable (in logs) Level First Difference

National gdp −0.948 −4.269*** di −0.138 −5.147***

Central dele_coa −0.976 −4.962*** pele_coa −1.274 −4.514*** pop_coa −0.271 −5.618***

Eastern dele_eoa −1.576 −4.403*** pele_eoa −1.782 −4.465*** pop_eoa −0.582 −3.806***

Western dele_woa −0.903 −6.994*** pele_woa −1.898 −4.438*** pop_woa −0.139 −4.668***

Southern dele_soa −0.444 −5.110*** pele_soa −1.327 −4.534*** pop_soa −0.670 −4.476***

Note: dele = demand for electricity, COA = central operating region, EOA = eastern operating region, SOA = southern operating region, WOA = western operating region, pop = population, pele = electricity price; *** =Significant at the 1% level, intercept only case is chosen based on the analyses.

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

4

and the post-estimation test results. As the table demonstrates the short-run final specifications pass all

the diagnostic and misspecification tests, hence the results are inter- pretable. Two observations are worth stating. First, all the remained regressors in the final ECM have economically meaningful and statis- tically significant impact on the electricity consumption in all regions. Second, this is true for the contemporaneous values of all the drivers, i.e., the income, price and population.

The estimated Speed of Adjustment (SOA) coefficients (coefficient of the ECT term) are statistically significant and negative in all the regions, confirming the existence of stable long-run relationship among the variables.

7. Discussion of the results

In this section, we present the key results and insights from our analysis on regional electricity demand. The key results are summarized in Tables 3 and 4. Table 3 displays the estimated long run elasticities for each region and Table 4 displays the short run elasticities.

As implied by Table 3, the long-run income elasticities range from 0.10 to 0.93 across the regions, the long-run price elasticities range from -0.61 to -0.06, and the long-run population elasticities range from 0.24 to 0.95. These results are in line with our expectation of positive elasticities for income and population, and negative elasticities for prices that we discussed in the Theoretical Framework section. In terms of magnitudes, our long-run income elasticities are smaller than finding of Al-Faris (2002), which is the only study devoted to total electricity after 2000, and seem to be bigger (1.65) than the expected magnitude. In this regard, it makes sense to find income elasticity being smaller than his finding. When it comes to long-run price elasticity, the found range is in line with the expectation for the developing country case. For example, Atalla and Hunt (2016) estimated the long-run price elasticity for residential demand to be -0.16, which is well contained in our interval. In addition, since the previous studies investigated the electricity demand modeling at overall country level, our results are not directly comparable with their results. The country level elasticities can be seen as the average representative for the whole country, while the regional ones might take into account the region-specific features.

With an income elasticity of 0.93 (the upper bound of the con- fidence interval), electricity demand in the Central region is the most sensitive to changes in income. The Eastern region is the most sensitive to changes in population as the population elasticity of electricity de- mand is the highest at 0.95 (the upper bound of the confidence in- terval).

With a price elasticity of −0.61 (The lower bound of the confidence interval. The western is chosen with the highest impact since the nar- rower confidence interval in comparison with east), the Western region displays the largest sensitivity to changes in prices. Comparing the sizes of all the elasticities for each region, income has the largest impact on electricity demand in the Central region, population has the largest impact in the Eastern region, while price has highest impact in the Western region.

Comparing the elasticities across regions, we observe that in abso- lute terms, the price elasticity is lowest in the Southern region (-0.13,

Table 2 Cointegration test results.

COA EOA SOA WOA

Value P-value Value P-value Value P-value Value P-value

EG tau-statistic −5.413 0.013** −3.598 0.226 −5.273 0.037** −4.168 0.040** EG z-statistic −561.841 0.000*** −18.136 0.189 −60.934 0.000*** −21.866 0.033**

Notes: p-values are MacKinnon (1996) probability values; ** and *** stand for the rejection of the null of no cointegration at 5% and 1% significance level, respectively.

Table 3 Long-run Estimation results.

Regressor Region

COA EOA SOA WOA

Income* (0.403, 0.931) (0.102, 0.204) (0.120, 0.360) (0.426, 0.470) Price (−0.580,

−0.362) (−0.628, −0.356)

(−0.132, −0.060)

(−0.607, −0.427)

Population (0.243, 0.661) (0.771, 0.947) (0.335, 0.645) (0.776, 0.840)

Notes: Dependent variable is dele_i and i takes COA, EOA, SOA and WOA, re- spectively; * for the Central and the Southern regions GDP used as an income measure; For the Eastern and Western regions Disposable Income is used as income proxy; Numbers in parentheses are 99 % confidence intervals for the obtained elasticities.

Table 4 Short-run Estimation results.

Independent Variables (DLOGs) Region

Central Eastern Western Southern

ECT (−0.293, −0.151) *** (−0.892, −0.532) *** (−0.984, −0.746) *** (−1.318, −0.924) *** income (0.152, 0.414) ** (0.054,0.182) * (0.075, 0.221) * (0.159, 0.471) * price (−0.079, −0.013) * (−0.234, −0.088) ** (−0.274, −0.150) *** (−0.103, −0.029) * population (0.128, 0.616) * (0.168, 0.412) ** (0.755, 1.485) *** (0.338, 0.600) *** dele (−1) (0.118, 0.386) * income (−1) (−0.212, −0.078) ** price (−1) population (−1) (−0.552, 0.134) (−1.399, −0.839) *** TESTS (p-values) Serial Correlation LM 0.344 0.062 0.067 0.400 Normality test (Jarque-Bera) 0.850 0.856 0.522 0.300 Heteroscedasticity (White) 0.636 0.870 0.484 0.873 Ramsey RESET Test 0.471 0.153 0.133 0.942

Notes: Dependent variable is DLOG(DELE_i) and i takes COA, EOA, WOA and SOA; ***, **, * stand for rejecting a null hypothesis at 1%, 5% level and 10 % significance levels, respectively. All independent variables except ECT are in DLOG.

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

5

the lower bound of the confidence interval), the population elasticity is lower in the Central and Southern regions and the income elasticity is lower in the Eastern and Southern regions. Below we rationalize these findings based on regional characteristics.

We suspect that the low price elasticity in the Southern region is a result of the relatively lower levels of income in the south (Household Income and Expenditure Survey, GaStat, 2018). Due to the relatively lower income levels, electricity consumption (which is mainly, about 62%, residential consumption in the region) is already conservative and optimized, and there are not many avenues to further reduce con- sumption as a result of price increases. This view is reinforced by the relatively low income elasticity for the southern region.

In order to explain the regional population elasticities, it is im- portant to highlight that population trends are explained by migration patterns, births and deaths. In the Southern region where emigration is prevalent due to the search for better economic opportunities in other regions by adults and population increases are mainly driven by births, a lower population elasticity is potentially explained by population growth being driven by mainly births. In comparison to adults, young children tend to consume less electricity, hence the lower population elasticity in the south is potentially due to population growth being driven by births. The Southern and Eastern regions having lower in- come elasticities is potentially explained by the economic character- istics of the regions. In the Eastern region, which is heavily in- dustrialized, the income elasticity being low makes sense because of the relatively higher levels of income in the region. Changes in income are less likely to impact the level of electricity consumption when income levels are high (Chang et al., 2016, inter alia). Not surprisingly, in the Southern region where income levels are lower, the income elasticity is also low. This is potentially explained by the conservative approach to electricity consumption (of which about 62% is resudential consump- tion in the Southern region) at relatively lower income levels. This makes consumers less sensitive to income changes. As Chang (1977, 1980) and Chang and Hsing (1991) discusses, demand for a particular thing might be luxury for some level of income. At that level of income, depending its level, the income elasticity starts to grow rapidly, and even might become bigger than unity. After a certain level of income, the additional increase in income does not contribute its consumption at the same rate. Hence, the income elasticity with respect to that item reduces and further increase does not change it significantly. Therefore, for some low level and for higher levels of income the income elasti- cities are expected to be low. While, it can be closer, even bigger than unity for income levels between the low and high ones. This point is also concluded by Chang and Hsing (1991) for residential electricity consumption in US, and by Chang et al. (2016) for electricity con- sumption in case of panel of countries. In this regard, relatively low income elasticities in cases of the Eastern and Southern regions can be explained by the above-mentioned points.

8. Conclusions and policy implications

In this study, we examined the impact of price, income and popu- lation (proxy for demographic effects) on electricity demand for four, central, eastern, southern and western, regions in Saudi Arabia. To the best of our knowledge, this is the first time series study to examine the determinants of regional electricity demand in Saudi Arabia.

The empirical results show the estimates of the regional long-run price elasticities to be around −0.5 in three regions and −0.1 in the Southern region. This leads to the conclusion that changes in the price of electricity are an effective market signal that has the potential to impact the demand for electricity.

From a policy perspective, based on the regional differences in elasticities, policy makers can develop insights about the potential re- gional impact of changes in electricity prices. Furthermore, based on the estimated regional impact, policy makers are better positioned to strike a balance between the desire to encourage efficient electricity

consumption (by transitioning to a market-based pricing scheme with higher prices) and the negative impact of the price increase. Striking a balance will require policy makers to determine the appropriate level of household and industrial support required to dampen the negative impact of price changes.

With long run elasticities that range between 0.24 and 0.95, the strong impact of population on regional electricity demand requires governments to pay special attention to policies that impact the total and regional distribution of population, which in turn affect the ag- gregate and regional distribution of electricity demand. This is espe- cially relevant for the effective planning of generation, transmission and distribution networks across the country. The recently im- plemented expat levy is a relevant example of a policy that has the potential to change the existing population dynamics.

Lastly, our results show that an increase (decrease) in income (gross domestic product and disposable income) leads to an increase (de- crease) in electricity consumption. This result proves that electricity is a normal good from an income perspective. It is important to note that there are big differences in the income elasticities across regions. The differences range from approximately 0.7 in the COA to 0.2 in the EOA. This implies that it is important to account for the regional income differences when designing policy for the electricity sector.

In summary, our findings constitute evidence on the characteristics of the regional determinants of electricity demand in Saudi Arabia. The policy implications discussed above represent our contribution to the discussion in the current policy environment in Saudi Arabia.

There is a room for future research to analyze the regional elec- tricity demand based on types of consumers. This would enable pol- icymakers to see the main types by regions driving the demand for electricity and clearer picture of demand behavior.

Author contributions

All authors contributed equally to all aspects of the research re- ported in this paper.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influ- ence the work reported in this paper.

Acknowledgments

The views expressed in this study are those of the authors and do not necessarily represent the views of their affiliated institutions. We are responsible for all error and omissions.

References

Al-Faris, A.R.F., 2002. The demand for electricity in the GCC countries. Energy Policy 30 (2), 117–124.

Alrashed, Farajallah, Asif, Muhammad Zahid, 2014. Trends in Residential Energy Consumption in Saudi Arabia with Particular Reference to the Eastern Province.

Al-Sahlawi, M.A., 1999. Electricity planning with demand estimation and forecasting in Saudi Arabia. Energy Stud. Rev. 9, 82–88.

Alyamani, Ryan, Abdulelah, Darandary, Mikayilov Jeyhun, I., Hasanov Fakhri, J., 2019. Residential Electricity Price Reforms: Are Different Income Groups and Regions Impacted Equally? KAPSARC Publications. (Accessed on December 18, 2019). https://www.kapsarc.org/research/publications/residential-electricity-price- reforms-are-different-income-groups-and-regions-impacted-equally/.

Atalla, T., Hunt, L., 2016. Modelling residential electricity demand in the GCC countries. Energy Econ. 59, 149–158.

Beenstock, M., Dalziel, A., 1986. The demand for energy in the UK: a general equilibrium analysis. Energy Econ. 8 (2), 90–98.

Campos, Julia, Ericsson, Neil R., Hendry, David F., 2005. General-to-specific modeling: an overview and selected bibliography. FRB Int. Finance Discuss. Paper 838.

Chang, H.S., 1977. Functional forms and the demand for meat in the United States. Rev. Econ. Stat. 59, 355–359.

Chang, H.S., 1980. Functional forms and the demand for meat in the United States: a

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

6

reply. Rev. Econ. Stat. 62, 148–150. Chang, H., Hsing, Y., 1991. The demand for residential electricity: new evidence on time-

varying elasticities. Appl. Econ. 23, 1251–1256. Chang, Y., Choi, Y., Kim, C.S., Miller, J.I., Park, J.Y., 2016. Disentangling temporal pat-

terns in elasticities: a functional coefficient panel analysis of electricity demand. Energy Econ. 60, 232–243.

Diabi, A., 1998. The demand for electric energy in Saudi Arabia: an empirical in- vestigation. OPEC Rev. 22 (1), 13–29.

Dickey, D., Fuller, W., 1981. Likelihood ratio statistics for autoregressive time series with a unit root. Econometrica 49, 1057–1072.

Doornik, J.A., 2009. Autometrics. Chapter 4 in In: Castle, J.L., Shephard, N. (Eds.), The Methodology and Practice of Econometrics: A Festschrift in Honour of David F. Hendry. Oxford University Press, Oxford, pp. 88–121.

Doornik, J.A., Hendry, D.F., 2009. Modelling Dynamic Systems: PcGive 13. Timberlake Consultants, LTD. London.

Doornik, J.A., Hendry, D.F., 2018. Empirical Econometric Modelling, PcGive 15. Published by Timberlake Consultants Ltd Timberlake Analytics, Inc. The Loft, 2C Blake Mews Kew Gardens Richmond, TW9 3GA, UK.

Eltony, M.N., Mohammad, Y.H., 1993. The structure of demand for electricity in the Gulf Cooperation Council Countries. J. Energy Dev. 18 (2), 213–221.

Engle, R.F., Granger, C.W.J., 1987. Co-integration and error correction: representation, estimation and testing. Econometrica 55, 251–276.

General Authority for Statistics (GaStat), 2018. Household Income and Expenditure Survey, 2018. (accessed on December 18, 2019). https://www.stats.gov.sa/en/37.

Hagihara, Jun, 2013. Saudi Arabia’s domestic energy situation and policy: focusing on its power sector. Kyoto Bull. Islamic Area Stud. 6, 107–135.

Hasanov, F.J., 2019. Theoretical framework for industrial electricity consumption re- visited: empirical analysis and projections for Saudi Arabia. KAPSARC Discussion Paper. KAPSARC, Riyadh, Saudi Arabia. https://doi.org/10.30573/KS–2019-DP66.

Hasanov, F.J., Bulut, C., Suleymanov, E., 2017. Review of energy-growth nexus: a panel analysis for ten Eurasian oil exporting countries. Renew. Sustain. Energy Rev. 73, 369–386.

Hendry, D.F., Johansen, S., Santos, C., 2008. Automatic selection of indicators in a fully saturated regression. Comput. Stat. 33, 317–335 Erratum, 337-339.

Karanfil, F., Li, Y., 2015. Electricity consumption and economic growth: exploring panel- specific differences. Energy Policy 82, 264–277.

Liddle, B., Lung, S., 2010. Age structure, urbanization, and climate change in developed countries: revisiting STIRPAT for disaggregated population and consumption-related environmental impacts. Popul. Environ. 31 (5), 317–343.

Matar, W., 2017. A look at the response of households to time-of-use electricity pricing in Saudi Arabia and its impact on the wider economy. Energy Strategy Rev. 16, 13–23.

Matar, W., Anwer, W., 2017. Jointly reforming the prices of industrial fuels and re- sidential electricity in Saudi Arabia. Energy Policy 109, 747–756.

Mohammadi, H., Amin, M.D., 2015. Long-run relation and short-run dynamics in energy consumption–output relationship: international evidence from country panels with different growth rates. Energy Econ. 52, 118–126.

Narayan, P.K., Smyth, R., 2009. Multivariate Granger causality between electricity con- sumption, exports and GDP: evidence from a panel of Middle Eastern countries. Energy Policy 37 (1), 229–236.

NIDLP, 2019. National Industrial Development and Logistics Program. Accessed on September 11. https://vision2030.gov.sa/en/programs/NIDLP.

Park, Joon Y., 1990. Testing for unit roots and cointegration by variable addition. Adv. Econometrics 8, 107–133 ed. by G. F. Rhodes and T. B. Fomby. Greenwich, CT: JAI Press.

Park, J.Y., 1992. Canonical cointegrating regressions. Econometrica 60, 119–143. Phillips, P.B., Hansen, B.E., 1990. Statistical inference in instrumental variables regres-

sion with I(1) processes. Rev. Econ. Stud. 57, 99–125. Saikkonen, Pentti., 1992. Estimation and testing of cointegrated systems by an auto-

regressive approximation. Econ. Theory 8 (1), 1–27. Salahuddin, M., Gow, J., Ozturk, I., 2015. Is the long-run relationship between economic

growth, electricity consumption, carbon dioxide emissions and financial development in Gulf Cooperation Council Countries robust? Renew. Sustain. Energy Rev. 51, 317–326.

Stock, J.H., Watson, M.W., 1993. A simple estimator of cointegrating vectors in higher

order integrated systems. Econometrica 61, 783–820.

Jeyhun I. Mikayilov Jeyhun is a research fellow at KAPSARC. His primary research interests include but are not limited to applied time series econometrics, the eco- nomics of energy and environment, and sustainable devel- opment.

Fakhri J. Hasanov Fakhri is a research fellow leading the KAPSARC Global Energy Macroeconometric Model (KGEMM) project. His research interests and experience span econometric modeling and forecasting, building and applying macroeconometric models for policy purposes, energy economics with a particular focus on natural re- source-rich countries.

Waheed Olagunju Waheed is Director of Economic Analysis in the Policy and Strategic Planning department at Ministry of Energy in Kingdom of Saudi Arabia.

Mohammad H. Al-Shehri Mohammad is head of the Modeling Department at Saudi Electricity Company. Mohammad holds MS degree in Engineering.

J.I. Mikayilov, et al. The Electricity Journal 33 (2020) 106772

7