Analysis of Price and Income Elasticities of Energy Demand in Ecuador: A Dynamic OLS Approach
AAnalysis of Price and Income Elasticities of EnergyDemand in Ecuador: A Dynamic OLS Approach
Kathia Pinz´onEscuela Polit´ecnica [email protected]
Abstract
Energy consumption (EC) in Ecuador has increased significantly during thelast decades, affecting negatively the financial position of the country since (1)large EC subsidies are provided in its internal market and (2) Ecuador is mostlya crude-oil exporter and oil-derivatives importer country. This research seeks tostate the long-run price and income elasticities of energy demand in Ecuador, byanalyzing information spanning the period 1970–2015. A co-integration analysisand an estimation by using a Dynamic Ordinary Least Squares (DOLS) approachconsidering structural breaks is carried out. Results obtained are robust and suggestthat in the long-run energy demand in Ecuador (1) is highly income elastic, (2) hasno relationship with its price and (3) has an almost unitary but inverse relationshipwith the industrial production level. Conclusions and economic policy suggestionsare also provided.
Keywords:
Ecuador, energy demand, elasticity, co-integration, DOLS, structuralbreak.
JEL classification:
Q43, Q48, O13.
Introduction
Energy consumption (EC) is implied in every economic activity: in activities related toproduction as well as in those related to consumption; so that it is one of the major factors1 a r X i v : . [ q -f i n . GN ] N ov nvolved in the economic system. Furthermore, it is demanded in such a recursive waythat, not only the enterprises demand energy for carrying out their activities, but alsothose enterprises’ performance allows people to demand more energy goods, encouragingin turn even greater production levels. Dhungel (2003), to this respect, even mentionsthat with an expansion in the economy, the production increases over time, resulting ingreater energy requirements to sustain the pace of development. Therefore, the EC in agiven country is directly related too its economic performance.According to information published in BP Statistical Review of World Energy 2016,the world EC has kept an increasing trend over the last 25 years, showing a mean annualgrowth rate of 1.9%. In 2015 it stood at 13147.3 Mtoe. (1% greater than in 2014)showing the same trend that the World GDP (At Purchasing Power Parity, measuredin U.S. dollars at constant prices of 2011) presented: this latter grew at a mean rate of3.37% during the same period according to the World Bank. Most of countries, in fact,present growing trends in EC; however, some of them—mostly in Europe and Eurasia—keep their levels relatively constant. EC is, therefore, induced differently by countriesworldwide. Moreover, as Phoumin and Kimura (2014) mentions, such differences dependon factors like GDP level, industrial structure, lifestyle of citizens, geographical locationand energy prices (especially relative energy prices).In several countries around the world, the growth in EC has led to economic growth(see, for instance Al-mulali and Binti Che Sab, 2012; Masih and Masih, 1996). Nonethe-less, it has also had negative ecological and financial effects in many countries worldwide.Firstly, its negative environmental impact has been expressed in an upward trend of CO2emissions (Ang, 2008; Soytas et al., 2007; Halicioglu, 2009; Dhakal, 2009). According toinformation of the BP Statistical Review of World Energy 2016, alike the EC, the worldCO2 emissions level has also maintained an increasing trend (with a mean growth rateof 2.2%) over the last 50 years. This has been mainly caused by the consumption of en-ergy from fossil fuels, which are the most polluting among the existent ones and whose “BP Global” is one of the world’s leading integrated oil and gas companies. One of its key reports isthe “BP Statistical Review of World Energy”, which provides information regarding global energy trendsand projections. Million tonnes oil equivalent. (Mega-toe). It is a unit of energy. A tonne of oil equivalent (toe) isthe amount of energy released by burning one tonne of crude oil; though there are several kinds of crudeoil which have different calorific values, the exact value is defined by convention so that 1 toe. equalsapproximately to 42 Gigajoules (GJ). The amount of CO2 produced when a fuel is burned is a function of the carbon content of the fuel,so that the fossil fuels such as the coal, the oil and the natural gas are those which emits more CO2 per ), in 2015 it wasof $1252.77 millions and represented the 27.1% of the total DDFA.As Aziz et al. (2013) mentions, spurred by the oil price shocks in late 1973 and duringthe period 1979 to 1980, a lot of attention was devoted to the analysis of energy demand unit of energy output or heat content In 2008, given the great gap existent between the volumes of national demand and supply of oilderivatives in the country, an account called “Deficit Derivatives Financing Account (DDFA)” was createdas part of the General State Budget, with the objective of keeping permanently the necessary funds toimport oil derivatives in order to cover the internal demand. Such account gathers (1) a transfer of theincome perceived from oil derivatives sells done by the public enterprise “PETROECUADOR”, (2) atransfer done by the Central Government and (3) a transfer regarding revenues from certain exports ofcrude oil.
3s a consequence of the dramatic events in energy markets and the increasing importanceof this sector in national economies. Thus, it was performed a great effort to estimatethe relationship between energy demand and factors such as income and energy price.Nevertheless, in Ecuador the research aimed to this respect has been lesser than in manyother countries worldwide. At the best of my knowledge, in this country no work of thisspecific kind has been carried out before, in spite of the budget problems that the currentupward trend of energy demand represents.Nowadays, the behavior of energy prices internationally is affecting again to Ecuadoras well as to several other countries worldwide. In this country, the increasing energydemand along with the high volatility of oil prices in the external market, had generatedhigh costs to the government given the existing trend to import some types of secondaryenergy and the significance of energy subsidies in its economy. This finally states thesolution of the issues related to energy usage as one of the most important topics to con-sider about economic policy. In consequence, it becomes essential to know how sensitiveenergy demand is to changes in people’s income and energy prices, in order to provideright responses of economic policy that conducts the country to a more ecologically andfinancially sustainable growth.This paper attempts to determine the price and income elasticities of energy demand inEcuador by applying a Dynamic Ordinary Least Squares (DOLS) approach. The resultswill allow us to define important conclusions in order to shed lights on the adequateeconomic policy on this issue. Obtained results are robust. The reminder of this paper isas follows. Section 1 states the theoretical and empirical context of the research. Section 2states the background of energy demand in Ecuador. Section 3 explains the methodologyapplied and data used. Section 4 details the results obtained and their interpretation.Section 5 states the conclusions and policy implications of the research.
The demand of energy plays an important role in the economic system: changes in level orstructure of EC could lead to significant changes in other macroeconomic variables in aneconomy. In analyzing energy usage and policy oriented to such issue, the whole energydemand specification represents a crucial input (Aziz et al., 2013). Price and income4lasticity of energy demand, therefore, represent important tools in searching to generateimportant changes in the economic performance of a given country. Certainly, not onlyincome and energy price are factors that define the amount of energy demanded, but alsoothers such as industrial structure, resource endowments, etc (Phoumin and Kimura,2014; Aziz et al., 2013). However, most of the studies carried out to this respect hadhighlighted their importance especially because of their usefulness as policy economictools at a macro level and, in some cases, due to the lack of data regarding other factors.Price and income elasticity of energy demand shows the level of responsiveness ofthe demanded amount of energy in respect to changes in its price and in income level,respectively. In the existent literature, energy demand is accepted to tend to be priceinelastic due to its basic nature necessity and its lack of substitutes, whereas the incomeelasticity tends to vary across countries depending on the specific conditions of each one.Additionally to this respect, Phoumin and Kimura (2014) point out that in developingcountries energy demand tends to have a higher price elasticity than in developed ones,given the higher dependence of their developing industries on EC.Given the importance of energy demand not only in terms of clearly economic aspectsbut also in environmental ones, a considerable amount of research work has been carriedout regarding this topic. However, at my knowledge, most of them have been developedregarding non-latinamerican countries. In fact, as Mitchell (2006) mentions, in the lit-erature, most of the studies have been conducted by developed countries and mentionvaguely the impact of oil prices on energy demand for developing countries. Besides,several of them are typically based on cross section analysis and therefore, offer only arepresentative measure or benchmark for price and income elasticity of energy demandin developing countries.Under such consideration, at the international level, for instance, one can quote severalauthors who had researched about this regard, in reference to developing countries aswell as to developed ones (see, for example Dhungel, 2003; Phoumin and Kimura, 2014).Several researches have looked for stating the estimates of energy demand elasticities forspecific countries (see, for example Dhungel, 2003; Mitchell, 2006), meanwhile other oneshave analyzed those estimators for groups of countries (see, for example Phoumin andKimura, 2014; Aziz et al., 2013; Madlener et al., 2011)Dhungel (2003), for instance, in his research, found the price and income elasticity of5ommercial energy demand in Nepal to be high (-1.65 and 3.04 using variables in levels,respectively); the petroleum energy demand to have a price elasticity between -0.83 and-0.55 and an income elasticity between 0.02 and 0.54; and the electric energy demand tohave a price elasticity between -1.14 and -1.25 and an income elasticity of 4.51. This, byapplying a distributed polynomial lag model with information spanning the period 1980–1999. It is worth mentioning that by the time at which such research was carried out(2003), the Nepalese economy depended at a low level on commercial energy i.e. obtainedfrom resources such as oil, coal, natural gas and the atoms’ core (nuclear energy). Thus,the use of non-commercial energy sources (which are not easily measurable) allowed ahigher sensitivity of commercial energy demand to prices.By other side, Mitchell (2006) determined that the demand for energy is fairly incomeelastic and price inelastic (50% and -20% on average, respectively) in the long run inBarbados, by applying a co-integration approach and analyzing the period from 1960 to2005.Phoumin and Kimura (2014), besides, determined the price and income elasticities,both for the short and long run, for many countries of ASEAN and some East Asiancountries by using a log-linear energy demand model. They found that the price elasticityis higher in developing countries than in developed ones and the income elasticity isrelatively high (greater than 1) in most of those countries. Moreover, they concludedthat price elasticity tends to be higher in countries where energy subsidies do not exist.Aziz et al. (2013) found, by analyzing a group of 16 developing countries for theperiod from 1978 to 2003 and employing a dynamic heterogeneous panel estimation, thatenergy price and income elasticity for those countries in the long run are 0.03 and 0.17,respectively, while the short run ones are 0.02 and 0.1, respectively. They also showedthat industrialization degree and CO2 emission levels are factors that affect significantlyenergy demand in those countries.Additionally, several investigations about this topic have been focused on specific kindsof energy. Madlener et al. (2011), for instance, estimated the price and income elasticitiesof the residential electricity demand, by applying a panel co-integration approach, for aset of 18 OECD countries spanning the period 1978–2008. They concluded that demandin most of the countries is price inelastic, suggesting that changes in prices for seekingelectrical energy conservation would fail as measures of economic policy.6n Ecuador, few investigations about this topic have been carried out, and have beenrelated only with specific kinds of EC. The research of Berrezueta and Encalada (2014),for instance, estimates the price elasticity of electric energy residential demand in the“Concession Area of the Center-South Regional Electrical Enterprise”, located in theprovinces of Azuay, Morona Santiago y Ca˜nar. They considered the period 2002–2012and applied an Error Correction Model, getting as result a short-run price elasticity of0.5383 and a long-run one of 0.2223.Rugel et al. (2009), on other side, in their research determined—for the long-run—the“Extra” gasoline price and income elasticity as 0.4 and 0.32 respectively, and the “Super”gasoline ones as 1.55 and 0.6 respectively. They used monthly data regarding the period1989–1998 and applied a Stock and Watson OLS dynamic approach. They compared itwith results of Error Correction and Johansen models.Finally, Collahuazo (2014) analyzed the gasoline demand function of Ecuador for theperiod from 1979 to 2013 by applying a co-integration analysis and Error CorrectionModel, finding that the long-run income elasticity of “Extra” and “Super” gasoline is0.202 and 0.251 respectively, suggesting that “Super” gasoline is a normal good in com-parison with the “Extra” gasoline, which is inferior; moreover, she found no relationshipbetween gasoline prices and gasoline demand.As seen above, though few researches had attempted to define price and/or incomeelasticities of the demand of specific kinds of energy in Ecuador, no one has aimed tostate such parameters regarding energy demand at a global level.
Ecuador, since its foundation, has been featured by keeping a growth model based mainlyon the exportation of raw materials; however, its macroeconomic performance have beenmodified significantly overtime, so that its industry structure—essentially aimed to thesatisfaction of its internal market—and beyond that, the behavior patterns of each eco-nomic sector within the country have also changed, affecting the evolution of the EC aswell.During the first decades after the institutionalization of Ecuador as a Republic, thecountry’s economic dynamic attempted to satisfy the demand of the international market7ith the production of goods in which the country presented “comparative advantages”so that—given the precariousness of the Ecuadorian industry at that time—the produc-tion of agricultural goods ended p being intensified and destined each time in greaterproportions to the exportation.During the period 1860–1920, the “cocoa boom” took place in the ecuadorian economiccontext. The production of such a good used to be carried out in a rudimentary waywhich was intensive in respect to workforce. Its pick, however, finished due to an adversecontext in the international market and the rise of a plague in the existent plantationsaround the 1930’s. After that, from 1943 to 1965, another agricultural product—thebanana—took the leadership among the ecuadorian exports and, beyond that, the boomof its exports developed along with the execution of a project aimed to lead the countryto a structural industrial transformation, the ISI (Imports Substitution Industrialization)model, created and proposed by ECLAC (Economic Commission for Latin America andthe Caribbean). Such model sought to eliminate the importation of goods and to eradicatethe agro-exporter model in order to encourage the modernization of the economy throughthe internal demand, so that this latter would become the creator of employment andadded value.Until the 1970’s, therefore, an agro-exporter growth model featured by two agriculturalproduct booms took place in Ecuador. However, even though the model was set up ina still colonial context based on work in “large states”, at the final of this period such acontext had changed significantly due to a deeper introduction of the capitalist systemin the economic structure and the development of different industrial sectors within thecountry. Furthermore, while at the beginning of the period people were dedicated eitherto a rudimentary agricultural production or to provide raw materials for the existentprimitive manufacture in “mitas” and “obrajes”, with the implementation of the ISImodel, a greater level of technology was applied to the production: to the agriculturalproduction as well as to the emerging industrial production related with food, beverages,tobacco and textiles.EC, consequently, also grew hastily during this period. According to the EconomicCommission for Latin America and the Caribbean, in 1937 the EC in Ecuador was ofabout 761 millions of kwh (around 0.0654 Mtoe.); made up in 77% of firewood, 16% ofoil and its derivatives and the rest of electricity (this latter used in 80% by the industrial8ector). Additionally, the oil production at that time was more than three times itsconsumption. Towards 1951, however, among other changes in EC, the oil productionbarely covered the consumption of oil and derivatives one time and a half. From 1929to 1951, in fact, the total EC increased in about 800% and the use of liquid fuels (whichmostly comes from oil) increased due to its consumption by transportation, industry andresidential sectors.In 1972, the petroleum mining and exportation at a big scale started to be developed—becoming in itself the next boom in the ecuadorian economy—along with an intensifi-cation of the industrialization process encouraged initially by the banana boom duringthe precedent decades. However, a series of events occurred during the next years suchas the so-called “debt crisis” during the 80’s and 90’s decades—wherewith the ISI modelwas removed, the dolarization of the economy in the beginning of year 2000, the worldfinancial crisis in 2008 and the current adverse international performance of oil prices hadgreatly affected such a process and, indeed, the level and structure of EC.As shown in the Figure 1, during the last 45 years the primary energy consumption(PEC) has had a general upward trend, showing a mean annual growth rate of 6%, whichwas greater than the GDP’s one (4%, considering GDP in constant prices of 2007) in thesame period. Moreover, during the last years the increasing in this variable has beeneven more accelerated. In fact, while from 1970 to 2000 the annual EC grew in 7.05516Mtoe. (it passed from 1.27 to 8.32 Mtoe.), from 2000 to 2015 it increased in 7.05502Mtoe. (passing from 8.32 to 15.38 Mtoe.), which shows that in the recent 15 years theannual EC has almost dobled, furthermore, in the period 2000–2015 it has increased inthe same measure at which it increased from 1970 to 2000.Such considerable change, in fact, has been closely related with the evolution ofecuadorian GDP, especially during the last 15 years. As shown in the Figure 2, whilebefore year 2000 the EC grew at a higher speed than GDP did, from 2000 to 2015 both ofthem have show a pretty similar trend, which agrees with the literature that establishesexistent (either uni-directional—in both directions—or bi-directional) relationships be-tween them. However, both variables have reflected the evolution of different economicsectors within the country and, indeed, the industrialization level and the consumption Primary energy is every way of energy available in the nature before being converted or transformed,e.g. oil, coal, natural gas, solar energy, among others. Likewise, the secondary energy refers to any energyobtained from the transformation of primary energy, e.g. electricity, oil derivatives.
Energy Consumption by Economic Sector 2011-2014Year Transportation Sector Industrial Sector Residential Sector2011
Additionally, oil products are the most used energy source by the transportation sectorso that while in 2010 the participation of such products in its EC was 100%, towards 2014it barely decreased (it stood at 93.48%). In general, during recent years the compositionof EC by all the economic agents in Ecuador has been modified radically in respect to thescene of 1970: As mentioned in Economic Comission for Latin America and the Caribbean(ECLAC) (1954), while in 1970 the total PEC was made up just in a marginal proportionby oil, in 2010 oil had a participation of about 75%, while other sources of energy suchas sugar cane, wood fire, hydro-energy and natural gas comprised the remaining 25%.Undoubtedly, a greater participation of consumption of oil products has taken place. Infact, as shown in Table 2, on page 13, while in 2011 the participation of such productsin total EC stood at 78%, in 2014 it stood at 76.56%, remaining relatively constant.Moreover, although from 2011 to 2014 the consumption of oil products has decreasedits participation in EC from transportation and residential sectors, it has increased itsparticipation in the EC from industrial sector even at a greater rate than the diminishesmentioned.Additionally, according to the BP Statistical of World Energy 2016, in 2015 the totalPEC was made up of 76.2% by oil sources, 3.73% by natural gas, 19.27% by hidro-electricity and 0.8% by renewable sources. Oil sources, consequently, keep being themostly used in the country. Considering that, it becomes important to analyze how oilprices have evolved.In general terms, the international oil price—considered as the “Brent Oil” price sinceit is officially the reference for international oil price—evaluated in current U.S. dollars—and thus, the prices of oil derivatives—has had an increasing trend over time. Figure 3shows the behavior of this series. It can be noticed that in 2009 (with the world financial11risis) oil price had a significant fall (it stood at $61.67). Moreover, during the recent 3years it has presented a decreasing trend: indeed, in 2015 it decreased 47.1% in respectto the price of 2014. Figure 3: Evolution of International Oil PriceEven so, if we consider i) that Ecuador is a crude oil exporter country, ii) that approx-imately the 80% of energy consumed corresponds to oil derivatives consumption, iii) thatmost of such derivatives (around 70%, according to the Ministry Coordinator of StrategicSectors of Ecuador) are imported and, iv) that the consumption of energy goods by alleconomic sectors is largely subsidized by the government, it is noticeable that the cost ofacquiring energy goods is quite high in Ecuador. Moreover, the current increasing trendof EC with such a configuration is not sustainable in time since oil—whose exports havebeen able to fund such a level of EC—is a non-renewable energy source and furthermore,its price performance does not depend only on purely market factors but also responds topolitical issues. It turns important, therefore, carrying out research work which attemptsto guide the policies established by the government in the seeking of a more sustainableperformance of the Ecuadorian economy.
As mentioned by Mitchell (2006), overtime the empirical methods used to investigate thedeterminants of energy demand—and, more specifically, its price and income elasticity—have changed substantially, but the modelling technique has generally remained the same:12 a b l e : P a rt i c i p a t i o n o f e n e r g y s o u r ce s i n E n e r g y C o n s u m p t i o nb y E c o n o m i c S ec t o r C O M P O S I T I O N O F E N E R G Y C O N S U M P T I O N E n e r g y S o u r c e s T r a n s p o r t a t i o nS e c t o r I ndu s t r i a l S e c t o r R e s i d e n t i a l S e c t o r A ll S e c t o r s * 20112014201120142011201420112014 O il - b a s e d E n e r g y G a s o li n e . % . % . % . % . % . % D i e s e l O il . % . % . % . % . % . % F u e l O il . % . % . % . % . % . % J e t F u e l . % . % G L P . % . % . % . % . % . % . % O t h e r s . % T o t a l O il - B a s e d E n e r g y . % . % . % . % . % . % . % . % N o n O il - B a s e d E n e r g y E l ec tr i c i t y . % . % . % . % . % . % . % F i r e w oo d . % . % . % . % G a s N a t u r a l . % . % . % . % P r o du c t o s d e C a˜ n a14 . % . % . % O t h e r s **6 . % . % . % . % . % . % T o t a l N o n O il - B a s e d E n e r g y . % . % . % . % . % . % . % . % T a b l e p r e p a r e db y t h e a u t h o r . I n f o r m a t i o n r e ga r d i n g2011 w a s o b t a i n e d f r o m t h e p r e s e n t a t i o n “ E n e r g e t i c M a t r i x ” , pub li s h e db y t h e M i n i s t r y C oo r d i n a t o r o f S t r a t e g i c S ec t o r s o f E c u a d o r . R e ga r d i n g2014 ,i n f o r m a t i o n w a s o b t a i n e d f r o m t h e “ N a t i o n a l E n e r g e t i c B a l a n ce , pub li s h e db y t h e s a m e M i n i s t r y . * C o m p r i s e s , b e s i d e s t h e s p ec i fi e d s ec t o r s , t h ec o mm e r ce , t h e ag r i c u l t u r e , t h e fi s h i n g , t h e m i n i n g , t h ec o n s t r u c t i o n , t h e n o n - e n e r g e t i c s ec t o r , a nd t h e o w n c o n s u m p t i o n . ** C o m p r i s e s k e r o s e n e , r e s i du a l s , n o n - e n e r g e t i c m a t e r i a l s a ndn o t s p ec i fi e d .
13n the majority of studies energy demand model comprises a price variable and an incomevariable, regressed on some measure of EC.Several investigations have considered just the GDP level and an Energy Price variablein the specification of Energy Demand equation as regressors, to estimate the income andprice elasticities of energy demand, respectively (see, for example Dhungel, 2003; Phouminand Kimura, 2014). Al-Azzam and Hawdon (1999), however, seeking to estimate suchelasticities of energy demand in Jordan, along with an income and a price variable,included in the model the variable “construction”—arguing that construction activity wasa good indicator of the development process and indeed of energy usage—and a dummyto capture the changes in the political climate in Jordan. In the research of Mitchell(2006), by the other hand, they looked for stating such elasticities including in the modelthe GDP level, energy price and a variable of energy efficiency as regressors; besides that,the authors used oil prices as energy prices arguing that energy used in Barbados wasmostly from oil sources. Additional variables, however, have also been used in researchwork in order to define the determinants and elasticities of energy demand, such as theshare of industry in GDP and the carbon dioxide emission levels (see, for instance Azizet al., 2013).Initially, the specification of energy demand model to be considered in this paper isas follows:(1) ln E t = ln Y t + ln P t + ln I t + ε t where E t represents energy demand , Y t represents the real income , P t represents energyprice , and I t represents the industrial production level. The model considers the series inlogarithms since it is a useful practice for estimating elasticities. The specific conceptu-alization of each variable included is as follows: E t is the per capita PEC, Y t is the realper capita GDP (at prices of 2007), P t , given that about the 80% of EC is comprised byoil-based EC, is the international price of crude oil (valued in dollars of 2015), and I t isthe Output of the Industrial Sector (valued in prices of 2007). Based on information of the World Bank. Industry includes manufacturing. It comprises value addedin mining, manufacturing, construction, electricity, water, and gas. Value added is the net output of asector after adding up all outputs and subtracting intermediate inputs. It is calculated without makingdeductions for depreciation of fabricated assets or depletion and degradation of natural resources. y t , takes its start in a general model given as follows:15 t = µ + βt + (1 − αL ) − C ( L ) ε t .Such model can also be expressed in the following way:∆ y t = ( α − y t − + µ (1 − α ) + αβ + (1 − α ) βt + C ( L ) ε t ,where (1 − αL ) µ t = C ( L ) ε t , with µ t such that ε t is i.i.d. and follows a normaldistribution with mean 0 and variance σ ε . C ( L ) = (cid:80) ∞ j =0 c j L j , (cid:80) ∞ j =1 j | c j | < ∞ , and c = 1. L is the lag operator. As mentioned by Haldrup et al. (2012), it is importantto note that the role of deterministic component is different in the levels and the firstdifferences representations. For instance, if α = 1, the constant term would equal to β whereas the slope would be 0. This shows the importance of carefully interpreting themeaning of deterministic terms under the null and alternative hypothesis.The definition of the AO and IO models proposed by Perron (1988) in order to testfor the existence of unit root with a structural break, embraces the definition of twodummy variables: DU t and DT t , such that DU t = 1 and DT t = t − T for t > T andzero otherwise. T represents the point of break in the series.According to the method, the following (AO) models are considered:1. y t = µ + ( µ − µ ) DU t + (1 − αL ) − C ( L ) ε t y t = µ + βt + ( µ − µ ) DU t + (1 − αL ) − C ( L ) ε t y t = µ + β t + ( β − β ) DT t + (1 − αL ) − C ( L ) ε t y t = µ + β t + ( µ − µ ) DU t + ( β − β ) DT t + (1 − αL ) − C ( L ) ε t The first model considers a non-trending specification with a break in intercept, thesecond considers a trending specification with a break in intercept, the third one considersa trending specification with a break in trend and the fourth one considers a trendingspecification with a break in intercept and trend.The (IO) models considered are the following:1. y t = µ + (1 − αL ) − C ( L )( ε t + θDU t )2. y t = µ + βt + (1 − αL ) − C ( L )( ε t + θDU t )3. y t = µ + βt + (1 − αL ) − C ( L )( ε t + θDU t + γDT t )16t is important to mention that these models consider known breakpoints. Addition-ally, as can be noticed, IO models do not embrace the case of a trending specificationwith break in trend, given that linear estimation methods cannot be used in this caseas they are in the others. An alternative technique, however, was developed by Zivotand Andrews (1992), which provided a solution for the estimation of such model, at thetime that developed a procedure for detecting break points endogenously from the data.Zivot and Andrews (1992) considered an estimated break point rather than a fixed one(an approach of endogenous break rather than one of an exogenous one). As Nilsson(2009) mentions, the procedure to test for this kind of break is the same as for exogenousbreaks: it tests for each possible break date in the sample, or some specific part of thesample, and then chooses the date with strongest evidence against the null hypothesis ofunit root, i.e. where the t-statistic from the ADF test of unit root is at a minimum . Thismethod, however, does not allow for the existence of a break under the null hypothesis.In order to test for the existence of unit root considering two structural breaks, themethodologies developed by Clemente et al. (1998) and Lumsdaine and Papell (1997) arebroadly used. Lumsdaine and Papell (1997) consider the presence of two breaks in trendvariables, while Clemente et al. (1998) consider the existence of double change in mean.Lumsdaine and Papell (1997) consider (assuming that two breaks belong to the inno-vational outlier) the following model to carry out the test:∆ y t = µ + βt + θDU t + γDT t + φDU t + Φ DT t + αy t − + (cid:80) ki =1 c i ∆ y t − i + ε t ,where DU t = 1 and DT t = t − T B t > T B
T B DU t = 1 and DT t = t − T B t > T B
T B k ) is determined based on the generalto specific approach (the t test).On the other hand, Clemente et al. (1998) consider (as Lumsdaine and Papell (1997),assuming the case of innovational outliers) the following model: y t = µ + ρy t − + δ DT B t + δ DT B t + d DU t + d DU t + (cid:80) ki =1 c i ∆ y t − i + ε t ,where DT B it is a pulse variable that takes the value 1 if t = T B i + 1 ( i = 1, 2) and0 otherwise, DU it = 1 if t < T B i ( i = 1, 2) and 0 otherwise. T B i and T B are the timeperiods when the mean is being modified. It is supposed that T B i = λ i T ( i = 1, 2),with 0 < λ i <
1, and also that λ > λ . After such estimation, the procedure consists in For a deeper explanation, see Zivot and Andrews (1992) < λ < λ , λ < − λ <
1. Therefore,it is necessary to choose some trimming value ( λ ).After the determination about whether or not unit roots exist in the time seriesanalyzed, and the nature of stationarity or non-stationarity of them, it is necessary todefine exactly the procedure to follow in order to estimate the relationships among them,granting the estimation of a non-spurious but real relationship.In 1987, Engle and Granger faced the problem of the estimation of models when unitroots exist in the series by introducing the concept of “cointegration”, establishing that,even being individually non-stationary, if the variables are I(1) and there is a stationarylinear combination of them, a correct estimation by OLS would be feasible through anError Correction Model (ECM), which consists in the regression of the first difference ofa dependent variable on its lags, lags of the first difference of regressors, and residuals ofthe OLS regression of variables in levels. This procedure, however, can potentially havea small sample bias and can examine at most one cointegrating relationship betweenvariables.The Johansen’s Vector Error Correction Model (VECM), developed in 1988, forinstance, presents several advantages over the ECM: (1) it does not assume one co-integrating relationship, (2) it does not impose any exogeneity restrictions and (3) it usesa system of equations framework to estimate the model. This methodology takes itsstarting point in the vector auto regression (VAR) of order p given by X t = Φ + Φ X t − + . . . + Φ p X t − p + ε t ,where X is a ( n ×
1) vector of variables that are integrated of order one i.e. I(1) and ε t is a ( n ×
1) vector of innovations. This VAR can also be written as∆ X t = Φ + ΠX t − + (cid:80) p − i =1 Γ i ∆ X t − i + ε t ,where it holds that Π = (cid:80) pi =1 Φ i − IΓ i = − (cid:80) pj = i +1 Φ j , j = 1, . . . , p − r is the range of the matrix Π . If r = 0, then there is no co-integration, sothat non-stationarity of I (1) type in the variables vanishes by taking differences. If r = n , this is, if Π has full rank, then the variables of X cannot be I (1) but are stationary, so18hat analysis with variables in level could be carried out. Finally, if Π has reduced rank,this is, if 0 < r < n , then at least one cointegrating relationship exists and r representsthe number of cointegrating relationships.Furthermore, Johansen proposes two different likelihood ratio tests of the significanceof the reduced rank of the Π matrix, in order to determine the number of existentcointegrating relationships between the variables: the trace test and maximum eigenvaluetest, shown in equations: J trace = − T (cid:80) ni = r +1 ln(1 − ˆ λ i ) J max = − T ln(1 − ˆ λ i )Here, T is the sample size and ˆ λ i is the i th largest canonical correlation of ∆ X t with X t − . The trace test tests the null hypothesis of r cointegrating vectors againstthe alternative hypothesis of n cointegrating vectors. The maximum eigenvalue test, onthe other hand, tests the null hypothesis of r cointegrating vectors against the alterna-tive hypothesis of r + 1 co-integrating vectors. Neither of these test statistics follows achi square distribution. Asymptotic critical values for this procedure were provided byOsterwald-Lenum (1992) and MacKinnon et al. (1999).As many authors mention, however, results of both procedures are completely reliablein large samples. For the case of small samples, therefore, an alternative procedure hasbeen developed and is widely used nowadays: the Dynamic OLS (DOLS) approach. Thismethod was proposed by Stock and Watson (1993) and basically is an improvement ofOLS estimation by dealing with small sample and dynamic sources of bias. As Al-Azzamand Hawdon (1999) mentions, it is a robust single equation approach which corrects forregressor endogeneity by the inclusion of leads and lags of first differences of the regressors,and for serially correlated errors by a GLS procedure. This method allows to estimatethe long-run equilibrium, in systems which may involve variables either integrated of thesame order or integrated of different orders but still cointegrated. In addition, it hasthe same optimality properties as the Johansen distribution. In this paper, essentiallydue to its robustness in small samples in comparison with precedent methods, the DOLSapproach is applied.The DOLS model, in general terms and for a dependent variable Y t with regressors X i , t , i = 1, 2, . . . , k , is given as follows: Y t = β + β X t + β X t + . . . + β k X k , t + (cid:80) qi = p α i ∆ X t − i + (cid:80) qi = p δ i ∆ X t − i + . . . +19 qi = p φ i ∆ X k , t − i + ε t where p and q represent the maximum orders of leads and lags (respectively) of thefirst differences of the regressors included in the model. All time series considered contain annual information spanning the period 1970–2015.The information about the real per capita GDP and the Total Output of Industry inEcuador was extracted from the website of the World Bank. The information about thePEC level and the International Crude Oil Prices (in U.S. dollars of 2015) are from theBP Statistical Review of World Energy 2016, published in the website of BP Global.
Before moving on to the estimation of equation 1, it is important to test for the existenceof unit roots in each series involved. Firstly, ADF and Phillips-Perron tests are applied tothe series. The specification of the model for testing the existence of unit root (interceptand trend, intercept, and none) is chosen considering the significance of the regressorsincluded in the model. The results obtained can be seen in the Table 3. In all cases (exceptin the case of ln P ) the regressors considered in the model of the tests are significant at95%. In fact, even the test specification embracing neither constant nor trend (shownin Table 3) is found non-significant for the case of ln P . The results suggest that ln E isstationary without drift, ln Y is non-stationary and ln I is stationary around a trend at95% level. No concrete conclusion, however, is obtained about ln P .Therefore, as the next step, unit root tests allowing for structural breaks are carriedout. The consideration of structural breaks (either in intercept or in trend) is importantparticularly in this analysis since during the period of study several facts have affectedsignificantly the economic performance of Ecuador. The behavior of the series consideredcan be appreciated in the Graph 4. Since the series suggest to present gradual changes,the type of break to be considered in all cases is “Innovational Outlier” (IO).The testing for unit roots in consideration of structural breaks starts by consideringonly one break point in each series. The specification of the model for testing the existence20able 3: Results of ADF and PP Tests TEST Model Specification Statistic Critical Values Lags Bandwidth1% 5% 10% ln Y ADF No Constant, No Trend 3.700 -2.617 -1.948 -1.612 0PP No Constant, No Trend 2.617 -2.617 -1.948 -1.612 4ln P * ADF No Constant, No Trend 0.348 -2.617 -1.948 -1.612 0PP No Constant, No Trend 0.346 -2.617 -1.948 -1.612 2ln E ADF No Constant, No Trend -4.508 -2.617 -1.948 -1.612 0PP No Constant, No Trend -4.946 -2.617 -1.948 -1.612 1ln I ADF Trend and Constant -3.835 -4.176 -3.513 -3.187 0PP Trend and Constant -3.734 -4.176 -3.513 -3.187 4Akaike Information Criterion was considered to choose the appropriate number of lags in ADFtest.Bartlett kernel spectral estimation method and Newey-West bandwidth selection were consideredin PP test. of unit root (intercept and trend, intercept, and none) as well as the specification ofthe break included in it (break in intercept, in trend, or in both) is chosen consideringthe significance of the regressors included in the model. The Table 4 shows the resultsobtained. The whole specification considered (of the deterministic component as well asof the break) regarding the series ln E , ln Y and ln I , is significant at 95% of confidence.In the case of ln P , even the specification without considering either trend or intercept(shown in the Table 4), is not significant.Table 4: Results of Unit Root Tests considering One Structural Break Variable Model Specification t-Statistic Lags BreakDeterministic Component Break ln E Constant and Trend Trend -4.207 0 1983ln Y Constant and Trend Trend -4.942 4 2004ln P * Constant Constant -3.202 0 1998ln I Constant and Trend Trend -5.308 0 2000The procedure developed in Perron (1988) and Vogelsang and Perron (1998) was consid-ered for the test of ln P . The procedure of Zivot and Andrews (1992) was considered intests of ln E , ln Y and ln I .The critical values at 95% of confidence are -4.52 for series ln E , ln Y and ln I , and -4.44for ln P .The number of lags selected in each case was chosen based on Akaike Information Crite-rion (AIC). E , ln Y and ln I , it could be concludedthat: (1) ln Y , as well as ln I are trend stationary variables, with structural (innovative)breaks in years 2004 and 2000, respectively and (2) ln E is a non-stationary variable. Inaddition, conclusive results about the existence of unit root in ln P were not found.Up to here, only ln Y and ln I series have a specifically defined stationarity conditionbased in the tests carried out, whereas ln E and ln P have been found to be non-stationarywithout any other specification. In order to obtain a better understanding about thestationarity condition of ln P and ln E , tests of unit roots considering two break pointswere performed regarding this series. The test of Lumsdaine and Papell (1997) wasperformed firstly, the results are shown in Table 5. As not all the dummies includedin each model were significant, results suggest that both series do not show a trendingbehavior even when considering 2 breaks. Therefore, the test of Clemente et al. (1998)is carried out about these series. The results can be seen in Table 6 and suggest thatln E and ln P are non stationary variables; however, ln E presents significant structuralbreaks in the years 1975 and 2004, while ln P presents them in years 1984 and 2002.What we know at this point is (1) that ln Y and ln I were found to be trend sta-tionary with structural breaks in years 2004 and 2000, respectively. (2) ln E is a non-stationary variable with significant breaks in mean in 1975 and 2004, and (3) ln P is22able 5: Results of Lumsdaine and Papell Tests Dummy Variables α t-Statistic Number of Lags Break PointsVariable Coef. Std. Error ln E DU1 0.114 0.062 -0.503 -5.050 0 1976 / 1980DT1 0.012 0.023DU2 0.147 0.048**DT2 -0.002 0.022ln P DU1 0.565 0.268** -0.330 -3.526 0 1979 / 1986DT1 -0.095 0.051DU2 -0.195 0.238DT2 0.112 0.051**Highlighted standard errors denote variables significant (at 95% of confidence).Number of lags included were determined based on the general to specific approach (the t test).For all series, the trimming parameter was established equal to 0.1.
Table 6: Results of Clemente-Monta˜nez-Reyes Tests
Dummy Variables ρ − t-Statistic Number of Lags Break PointsVariable Coef. Std. ErrorLNE DU1 0.13207 3.156*** -0.1793 -4.427 1 1975 / 2004DU2 0.08032 2.983***
LNP
DU1 -0.73789 -4.669*** -0.6147 -5.139 4 1984 / 2002DU2 0.77171 5.053***Highlighted standard errors denote variables which are significant at 95% of confidence.The critical value at 95% of confidence is -5.490.
23 non-stationary variable with significant breaks in mean in years 1984 and 2002. Thenext step is, therefore, testing for the existence of cointegration relationships between thevariables in presence of the deterministic components and structural breaks found.Through the tests applied regarding the existence of unit roots in the series, structuralbreaks in them were found for years 1975, 1984, 2000, 2002 and 2004. Year 1975 is nearto 1973, year at which officially started the oil prices crisis of 70’s. Year 1984, is nearto 1983, year at which officially started the so-called debt crisis of 80’s in Ecuador (inthis year, Ecuador signed its first Intention Letter). In addition, 2000, 2002 and 2004 areyears that refers essentially to year 2000, since in this year the debt crisis in the countrygave way officially to a dolarization process and so, to a change in Ecuadorian economicperformance. Only two possible breaks are going to be considered in the modelling energydemand, years 1983 and 2000, since a break in 1973 could represent a problem given theperiod of data considered in the study.In order to define the number of lags to include in testing for cointegration (throughthe Johansen’s method), VAR models were fitted. The exogenous variables considered insuch VAR models were: (1) dummies of the breaks found ( B and B regarding year 1983and 2000 respectively), (2) a trend variable ( T ) and (3) interactions between (1) and (2).Regarding the breaks included in VAR models, 3 different break dummy specifications areassessed: (1) 1983 (2) 2000 and (3) 1983–2000. Schwarz, Akaike and Hannan-Quinn In-formation Criterion were considered in choosing the adequate number of lags. Thereafter,Johansen’s cointegration test was performed considering the series. Asymptotic criticalvalues developed by Giles and Godwin (2012) for the test of cointegration in presence ofstructural breaks, were calculated.As illustrated by Table 7, on page 25, the results suggest that at least one cointegratingrelationship exists between the variables when considering structural breaks (1) in year1983 and (2) in years 1983 and 2000. Regarding year 2000, according to the results,2 cointegration relationships exist. However, since the retained methodology is DOLS(because of its reliability according to the sample size), this option is not considered.Thus, only the models with breaks (1) in year 1983 and (2) in years 1983 and 2000 areestimated. It has been selected only one lag and no lead for such estimations. TheDynamic OLS models to be fitted, therefore, are the following:24 a b l e : R e s u l t s o f J o h a n s e n ’ s C o i n t e g r a t i o n T e s t s i n c o n s i d e r a t i o n o f S tr u c t u r a l B r e a k s B r e a k L a g s E x og e n o u s V a r i a b l e s R a n g e o f Π T r a c e S t a t i s t i c A s y m p t o t i c C r i t i c a l V a l u e s C o n c l u s i o n % % % B , T ( B ) r = . . . . R e j ec t i o n r ≤ . . . . A cce p t a n ce r ≤ . . . . — – r ≤ . . . . — – 20001 B , T ( B ) r = . . . . R e j ec t i o n r ≤ . . . . R e j ec t i o n r ≤ . . . . A cce p t a n ce r ≤ . . . . — – 1983 , r = . . . . R e j ec t i o n B , T ( B ) r ≤ . . . . A cce p t a n ce B , T ( B ) r ≤ . . . . — – r ≤ . . . . — – N u m b e r o f l ag s i n e a c h c a s e w a s d e fin e db y r unn i n g VA R m o d e l s c o n s i d e r i n g s t r u c t u r a l b r e a k s a s e x og e n o u s v a r i a b l e s . C r i t i c a l v a l u e s w e r ec a l c u l a t e d a cc o r d i n g t o G il e s a nd G o d w i n ( ) . A cce p t a n ce a nd r e j ec t i o nd e fin e d i n a ll c a s e s a t % o f c o nfid e n ce . E t = β + β ln Y t + β ln P t + β ln I t + γ T + γ B t + γ T ( B t )+ (cid:88) i =0 α i ∆ ln Y t − i + (cid:88) i =0 δ i ∆ ln P t − i + (cid:88) i =0 φ i ∆ ln I t − i + ε t Model 2: ln E t = β + β ln Y t + β ln P t + β ln I t + γ T + γ B t + γ T ( B t ) + γ B t + γ T ( B t )+ (cid:88) i =0 α i ∆ ln Y t − i + (cid:88) i =0 δ i ∆ ln P t − i + (cid:88) i =0 φ i ∆ ln I t − i + ε t where B and B are dummy variables that represent the chosen years of structuralbreak (1983 and 2000, respectively), and T is the trend variable. Clearly, the modelsto be fitted estimate the long-run relationship between ln E and its determinants (ln Y ,ln P and ln I ), by taking into account the lagged effect of those 3 variables as well as theeffect of structural breaks in the series (either in intercept or in trend). The results aboutthe estimation of the models is shown in Table 8. As can be noticed, R , the standarderror of the regression and the long-run variances of both models suggest that Model 1and 2 are good representations of the dynamics between the variables. The Jarque-Berastatistic in both models suggests that residuals follow a normal distribution at a 5% ofsignificance. Additionally, according to the respective statistics shown in Table 8, thesignificance of ln Y , ln P and ln I as regressors in Models 1 and 2 varies across models,however, the variable ln P is not significant in any model. Since in Model 1 variablesare significant at a better level than in model 2, results shown regarding Model 1 areconsidered as a better estimation.Thereby, based on results obtained in Model 1, the following can be concluded: At thelong run, (1) the income elasticity of energy demand is 1.8, which (since it is greater than1) allows to state that energy demand is highly income elastic. This result agrees with theclose relationship observed graphically among the trends of the GDP and EC overtime.And (2) energy demand presents an almost unitary elasticity (-1.2) in relation with the26roduction level of industrial sector, however, such significant relationship is inverse.This result can be explained by the evolution of EC from the economic sectors within thecountry: Keeping constant the level of investment in the country, an increasing in the totalindustrial output might imply the movement of economic factors from other economicsectors—such as the transportation sector, which is the major energy consumer—towardsthe industrial one, causing a slight decrease rather than an increase in the total level ofEC. Table 8: Results of Estimation of Model 1 and Model 2 Variable Model 1 Model 2 ln Y P I -1.219197 -1.017343[0.222253] *** [0.277015] ** β T B T ( B ) -0.038034 -0.05584[0.00612] *** [0.01101] ** B -0.436834[0.239847] * T ( B ) 0.014229[0.00755] * Adjusted R S.E. of Regression
Long-run Variance
Jarque-Bera Statistic
Standard Deviations highlighted with * , ** and *** denote vari-ables which are significant at 90%, 95% and 99%, respectively.Variables of first differences of regressors (as well as their first lags)were not included in this results since they have only been used as atool to estimate the long-run relationship between variables. Theircoefficients do not have crucial interpretation.The long-run variance was determined using a Bartlett kernel andfixed Newey-West bandwidth.
27t is important to mention, furthermore, that no relationship between EC and energyprice was found. This results agrees with the finding of Collahuazo (2014), who did notfind relationship between gasoline prices and gasoline consumption by using data spanningthe period 1979–2013, researching that fairly were carried out about an oil-based product.Unlike Collahuazo (2014), Rugel et al. (2009) found low gasoline demand price elasticities,however, such research was developed with information regarding the period 1989–1998.Noticeably, though until 1998 gasoline (which is an oil-based enegy source) was barelyresponsive to gasoline prices, towards 2013 such responsiveness had already decreasingat the point that no relationship existed anymore between consumption and prices ofgasoline. The result of this research regarding price elasticity, therefore, can suggest thatnot just gasoline demand but also that of other energy sources suffered such decreasingin price responsiveness overtime.
EC in Ecuador has kept an accelerated upward trend over the last years and, at thesame time, has been modified significantly in its structure so that nowadays it is mainlycomprised by oil-based EC (about the 80% of the total) and it is mostly consumed by thetransportation sector (about 50% of the total). Given that Ecuador has become an oil-exporter country and an importer of oil-based products at the same time and, additionally,that large subsidies have been granted for the consumption of oil-based energy, duringthe last decades it has faced permanent problems of deficit budget. Thereby, EC issuesare important topics to analyse by policymakers nowadays. The present research worksought to state the price and income elasticities of energy demand in Ecuador.The obtained results suggest, on one hand, that in the long-run the EC is significantlyresponsive to changes in real income as well as in industrial production level. As men-tioned above, the trend shown by GDP level and EC specially during the last 15 yearssuggested the existence of a close relationship between the two series, such relationshipwas actually confirmed by the results of the fitted DOLS model. The responsiveness ofEC to changes in industrial production, however, was found to be significant but low,which is a reasonable finding considering the participation of industrial sector on EC(around 18% in 2014). Beyond that, such relationship was found to be negative, which28an be explained by the fact that given the low level of capital investment in Ecuador,an increasing in industrial production might mean the mobilization of economic factorsfrom other sectors to industry, so that EC by other sectors such as transportation woulddecrease, ending in a total negative effect on EC level.Regarding the price elasticity of energy demand, no significant relationship was foundbetween energy price and EC. To this respect, about the price elasticity of gasolinedemand, by one hand, Rugel et al. (2009) found low but significant gasoline demandprice elasticities by using information spanning the period 1989–1998, while on the otherhand, Collahuazo (2014) did not find relationship between gasoline prices and gasolineconsumption by using data spanning the period 1979–2013. Therefore, such previousresults might suggest that while until 1998 gasoline was barely responsive to gasolineprices, towards 2013 such responsiveness had already decreasing at the point that norelationship existed anymore between consumption and gasoline prices. The result ofthis research regarding price elasticity of energy demand, therefore, suggests that notonly gasoline demand but also that of other energy sources suffered such decreasing inprice responsiveness overtime, ending up in a complete independence from the evolutionof energy prices.Such a behavior regarding the relationship between EC and energy price, as Dhungel(2003) mentions, can be attributed to the substantial subsidies granted by the governmentin the acquirement of energy goods, given that subsidized energy price tends to distortenergy demand level inducing excessive consumption and waste, and also discouraging thesearching of more efficient and environmentally friendly energy sources. In Ecuador, infact, consumption of several energy goods by various economic sectors, since 70’s decadehave been and are such largely subsidized that prices in internal market has remainedrelatively constant.Important economic policy suggestions can be traced based on the results obtainedin this research. As EC is highly sensitive to changes in real income, the establishmentof income taxes as well as a better targeting of EC subsidies would (1) reduce the waste-ful use of energy by the different economic sectors and (2) help modifying the nationalbudget dynamic, which during the last decades has presented large deficits due to thepermanent trend of Ecuador—as oil-exporter country—to export of crude oil in exchangeof importing oil-derivatives and the large subsidies provided to population for the con-29umption of energy goods. More specifically, a modification in the targeting and level ofoil-based EC subsidies, might also work towards a convenient mobilization of factors fromthe transportation sector—whose development is actually strongly encouraged by largesubsidies and which does not generate any added value—to the industry sector—whichconsumes oil-based energy in lower proportions and does generate added value.Finally, even though during the last years various efforts have been made by thegovernment in order to advance in the replacement of oil-based energy sources by othermore sustainable energy sources (mainly hidro-electricity), apparently prices of oil-basedenergy products in the internal market are as lower than their prices in the internationalmarket as they continue pressuring upward the internal demand of such goods, and so,the assignment of state funds to their imports. The participation of such goods remainsaround 80% of the total EC. It is crucial for the country, however, and even more inthe current international adverse context of oil price—wherein it bounds barely the $45so that value of crude oil exports of the Ecuador have decreased dramatically—thatdecisions to modify the current patterns of EC are taken in order to achieve a morefinancially sustainable performance of ecuadorian economy in the long run. In addition,rightly structured measures of economic policy taken to this respect could contributetowards a more ecologically sustainable development of the country.
Acknowledgement
I thank Andrea Bonilla, Yasm´ın Salazar and Carolina Guevara, Professors at EscuelaPolit´ecnica Nacional, for their valuable suggestions and comments to improve this re-search. 30 eferences (2012). Budget Execution Report: Fiscal Year 2011. Technical report, Finances Ministryof Ecuador.(2013). Energetic Matrix of Ecuador. Technical report, Ministry Coordinator of StrategicSectors.(2015). National Energetic Balance - 2015. Technical report, Ministry Coordinator ofStrategic Sectors.(2016). BP Statistical Review of World Energy 2016. Technical report, BP Global.(2016). Budget Execution Report: Fiscal Year 2015. Technical report, Finances Ministryof Ecuador.Al-Azzam, A. and Hawdon, D. (1999). Estimating the Demand for Energy in Jordan:A Stock-Watson Dynamic OLS (DOLS) Approach. Surrey Energy Economics Centre(SEEC), School of Economics Discussion Papers (SEEDS) 97, Surrey Energy Eco-nomics Centre (SEEC), School of Economics, University of Surrey.Al-mulali, U. and Binti Che Sab, C. N. (2012). The impact of energy consumption andCO2 emission on the economic growth and financial development in the Sub SaharanAfrican countries.
Energy: Sustainable Energy and Environmental Protection 2010 ,19(1):180–186.Ang, J. B. (2008). Economic development, pollutant emissions and energy consumptionin Malaysia.
Journal of Policy Modeling , 30(2):271–278.Aziz, A. A., Nik Mustapha, N. H., and Ismail, R. (2013). Factors affecting energy demandin developing countries: A dynamic panel analysis.
International Journal of EnergyEconomics and Policy , 3(Special Issue):1–6.Berrezueta, J. and Encalada, A. (2014). An´alisis de los factores que afectan la demandade energ´ıa el´ectrica y su estimaci´on - Sector Residenial del ´Area de Concesi´on de laEmpresa El´ectrica Regional Centro Sur C.A., periodo 2002-2012.Clemente, J., Monta˜n´es, A., and Reyes, M. (1998). Testing for a unit root in variableswith a double change in the mean.
Economics Letters , 59(2):175–182.31ollahuazo, A. (2014). An´alisis emp´ırico de la funci´on de demanda de gasolina en Ecuadoren el per´ıodo 1979-2013.Dhakal, S. (2009). Urban energy use and carbon emissions from cities in China and policyimplications.
Energy Policy , 37(11):4208–4219.Dhungel, K. R. (2003). Income and price elasticity of the demand for energy: a macro-level empirical analysis.
Pacific and Asian Journal of Energy , 13(2):73–84.Di Bella, G., Norton, L., Ntamatungiro, J., Ogawa, S., Samake, I., and Santoro, M.(2015). Energy subsidies in Latin America and the Caribbean: Stocktaking and policychallenges. Working paper, International Monetary Fund.Economic Comission for Latin America and the Caribbean (ECLAC) (1954). El desarrolloecon´omico del Ecuador.
Publicaci´on de las Naciones Unidas , (E \ CN.12 \ Applied Economics Letters ,19(16):1561–1565.Haldrup, N., Kruse, R., Ter¨asvirta, T., and Varneskov, R. T. (2012). Unit roots, non-linearities and structural breaks. CREATES Research Papers 2012-14, Department ofEconomics and Business Economics, Aarhus University.Halicioglu, F. (2009). An econometric study of CO2 emissions, energy consumption,income and foreign trade in Turkey.
Energy Policy , 37(3):1156–1164.Lumsdaine, R. L. and Papell, D. (1997). Multiple trend breaks and the unit-root hypoth-esis.
The Review of Economics and Statistics , 79(2):212–218.MacKinnon, J. G., Haug, A. A., and Michelis, L. (1999). Numerical Distribution Func-tions of Likelihood Ratio Tests for Cointegration.
Journal of Applied Econometrics ,14(5):563–77.Madlener, R., Bernstein, R., and Alva Gonzalez, M. A. (2011). Econometric estimationof energy demand elasticities.
E.ON Energy Research Center Series , 3(8):1–62.32asih, A. M. and Masih, R. (1996). Energy consumption, real income and tempo-ral causality: results from a multi-country study based on cointegration and error-correction modelling techniques.
Energy Economics , 18(3):165–183.Mitchell, T. (2006). A co-integration analysis of the price and income elasticity of energydemand.
Central Bank of Barbados-Research Department .Nilsson, I. (2009). Unit root tests and structural breaks in the Swedish electricity price.Master’s thesis, Luea University of Technology.Osterwald-Lenum, M. (1992). A note with quantiles of the asymptotic distribution of themaximum likelihood cointegration rank test statistics.
Oxford Bulletin of Economicsand Statistics , 54(3):461–72.Perron, P. (1988). The Great Crash, The Oil Price Shock And The Unit Root Hypothesis.Papers 338, Princeton, Department of Economics - Econometric Research Program.Phoumin, H. and Kimura, S. (2014). Analysis on Price Elasticity of Energy Demandin East Asia: Empirical evidence and policy implications for ASEAN and East Asia.
ERIA Discussion Paper Series .Rugel, F., Bastidas, J., and Garc´ıa, F. (2009). Estimaci´on de las elasticidades de lademanda de gasolina en el Ecuador: Un an´alisis emp´ırico.Soytas, U., Sari, R., and Ewing, B. (2007). Energy consumption, income, and carbonemissions in the United States.
Journal of Policy Modeling , 62(3-4):482–489.Stock, J. H. and Watson, M. W. (1993). A Simple Estimator of Cointegrating Vectors inHigher Order Integrated Systems.
Econometrica , 61(4):783–820.Vogelsang, T. J. and Perron, P. (1998). Additional Tests for a Unit Root Allowing for aBreak in the Trend Function at an Unknown Time.
International Economic Review ,39(4):1073–1100.Zivot, E. and Andrews, D. (1992). Further evidence on the great crash, the oil-price shock,and the unit-root hypothesis.