Applied Energy 113 (2014) 392–403

Contents lists available at ScienceDirect

Applied Energy

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

Scenario analysis of nonresidential natural gas consumption in Italy

0306-2619/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.apenergy.2013.07.054

⇑ Corresponding author. Tel.: +39 010 353 28 72.E-mail addresses: vincenzo.bianco@unige.it, vbianco@libero.it (V. Bianco).

Vincenzo Bianco ⇑, Federico Scarpa, Luca A. TagliaficoUniversity of Genoa, DIME/TEC, Division of Thermal Energy and Environmental Conditioning, Via All’Opera Pia 15/A, 16145 Genova, Italy

h i g h l i g h t s

� The paper proposes an analysis on natural gas consumption forecasting.� A validation of the methodology is accomplished obtaining a good level of accuracy.� GDP, price and temperature elasticities are calculated.� A scenario analysis is developed by analysing twelve different scenarios.� In 2030, a natural gas consumption between 32 and 43 bcm is expected in Italy.

a r t i c l e i n f o

Article history:Received 26 March 2013Received in revised form 17 June 2013Accepted 21 July 2013

Keywords:Natural gas consumptionForecastingElasticityScenario analysisLinear regression

a b s t r a c t

Objective: The aim of the present paper is to develop a model for the long term forecasting of nonresiden-tial gas consumption in Italy. The influence of economic and climatic data, as well as the impact of reg-ulatory changes are considered.Methods: The model is developed by using a regression model and, to this scope, the necessary explainingvariables are determined. A successful validation of the model is performed, showing that it guarantees asatisfactory level of accuracy.Results: Short and long run elasticities are estimated, highlighting that Gross Domestic Product (GDP) percapita has a much greater influence on gas consumption with respect to price. Twenty-four consumptionscenarios are presented, underlining that in 2030 nonresidential gas consumption in Italy is expected tobe between about 32 and 46 bcm (billions of cubic meters).Conclusions: It can be concluded that the increase of nonresidential gas consumption is strongly linked tothe GDP evolution and according to the GDP growth scenario, nonresidential gas consumption might lar-gely change.Practice implications: The outcomes of the present analysis can be successfully utilized by energy manag-ers to design appropriate energy management policies. Particularly, the determination of the elasticitieshas practical relevance in setting up adequate pricing policies, whereas the long term forecast representsa useful support to estimate the volume of the necessary supply contracts and to plan newinfrastructures.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Italy is one of the European countries with the highest share ofnatural gas in its energy mix. Particularly, it represented the 40%and 38% [1] of the primary energy consumption, in 2010 and2011 respectively. In 2011, Italy, with its 76 bcm, resulted to bethe third consumer of natural gas in EU after Germany (78 bcm)and United Kingdom (83 bcm) [2].

Given these facts, it is of crucial importance to be able to predictgas consumption with a good degree of accuracy, in order to man-age supply contracts, indigenous production and infrastructures

planning in an optimal way. To fail one of these three targetswould cause instabilities in the Italian energy system, because ofthe unbalanced energy mix of the country, largely dependent onnatural gas.

In terms of market, it is detected that the incumbent operator isENI, the former state monopolist, with a market share of about 27%of total sells to final customers, then there is ENEL with about 12%of market share and the third operator is Edison with about 11%[2]. Therefore, the first three operators control about the 50% ofthe market, whereas the first twenty operators represent aboutthe 84% of the market and the remaining 16% is divided by 288small and very small operators [2]. In light of this, it can be saidthat the market is very concentrated.

The historical trend of natural gas consumption is reported inFig. 1, where three main contributions to the total consumption

Nomenclature

C consumption, bcm (billions of cubic meters)GDP Gross Domestic Product, €HFO price of heavy fuel oil, $/tLE long run elasticityLFO price of light fuel oil, $/tk year indexn number of observationP gas price, €/GJ and $/MbtuPOil oil price, $/bblt yearT temperature, �C

Greek lettersa regression coefficientb1–5 regression coefficients

Subscripts0 starting pointe estimatedimp importm referred to month mmin minimumnres nonresidentialPC per capitat � 1 lag of one year

V. Bianco et al. / Applied Energy 113 (2014) 392–403 393

are highlighted, namely thermo electric power plants, residentialand nonresidential [3,4].

Natural gas consumption in thermo electric power plants islinked to the electricity market, because it is directly related tothe electricity generated in gas fuelled power plants. The amountof generated electricity is set according to the variable costs com-petition, mainly influenced by fuel prices, of the different plantsavailable on the market.

Residential consumption of natural gas is determined by the de-mand for heating, sanitary water and cooking facilities in residen-tial buildings. The main consumption drivers are represented byexternal temperature of the heating season, population and build-ings characteristics (i.e. insulation, facilities, etc.).

Nonresidential natural gas consumption represents the usage ofnatural gas related to economic activities, particularly industry (i.e.manufacturing, food, construction, etc.) and services (i.e. offices,shops, healthcare, etc.). It can be assumed that natural gas con-sumed in the industrial sector is mainly used in production pro-cesses, even though a substantial share difficult to estimate isalso utilized for heating purposes (i.e. heating of very large indus-trial buildings), whereas the consumption in the service sector issubstantially due to heating demand.

By analyzing Fig. 1, it is possible to detect that, up to 1995, non-residential natural gas consumption represented the largest shareof the total, whereas after 1995 a sharp increase of thermal powerplants consumption is detected. This increase is due to the liberaliza-tion of the power generation sector, that allowed many players toenter the Italian market by building a relevant number of combinedcycle gas turbines, which boosted consumption of natural gas.

By observing the trend of the last five years, it can be roughlysaid that 35% of natural gas consumption is due to thermal power

Fig. 1. Historical trend of natural gas consump

plants, 35% to nonresidential consumption and 30% to residentialuses.

The aim of the present paper is to develop a forecasting modelfor nonresidential natural gas consumption in Italy by utilizing aregression model.

From the above mentioned data, it is detected that nonresiden-tial sector represents a relevant part (�30 bcm, billions of cubicmeters, per year) of the total consumption, therefore in order toguarantee the security of supply it is necessary to foreseen accu-rately the future consumption.

The under or over estimation of the needs can cause relevanteconomic losses to both final customers and natural gaswholesalers.

In the present case, a regression model is believed to be theoptimal choice, because it allows to analyze the impact that exter-nal explaining variables of common use (i.e. Gross Domestic Prod-uct, price, etc.) have on the consumption, leading to theformulation of a data light model (i.e. the cost of data mining isminimized) of practical interest.

Moreover, the use of a regression model has the unquestionableadvantage to allow the determination of a consumption equationwith a relatively simple structure, which can be easily imple-mented in more complex planning models (i.e. market simulators).

To the best of authors’ knowledge, this represents the first at-tempt, available in the scientific literature, to analyze in detail nat-ural gas consumption in Italy, focusing on nonresidential sector.

The first object of the present paper is to analyze nonresidentialnatural gas consumption, estimating the consumption equation byusing a regression model.

The second target is to propose an accurate forecasting of natu-ral gas consumption up to the year 2030, developing a scenario

tion in Italy for different sector of usage.

394 V. Bianco et al. / Applied Energy 113 (2014) 392–403

analysis by building different forecasts of the explaining variables,in order to study their effects on gas consumption.

It is believed that the information contained in this paper is use-ful for energy planners and policy makers, who can implement suc-cessful strategies by using accurate forecasts and analysis.

Particularly, the knowledge of price and income elasticitieshelps gas distribution companies to design suitable pricing strate-gies and allows public authorities to implement convenient de-mand side management policies. Whereas, the estimation offuture consumption helps to design optimal sourcing strategiesand to plan required infrastructures.

2. Literature review

In the last 20 years, because of the growing energy demand, anincrease of natural gas consumption is regularly detected year byyear. In the light of this, many authors decided to investigate onthe analysis and forecasting of natural gas consumptions by usingdifferent approaches and methodologies.

Erdogdu [5] investigated natural gas demand in Turkey, estimat-ing short and long-run price and income elasticities by means of lin-ear regressions; whereas the future growth of demand wasforecasted using an Autoregressive Integrated Mobile Average (AR-IMA) modeling. He showed that a relevant increase of the demandis expected in the next years. Instead, Aras and Aras [6] investigatedabout residential gas demand in a Turkish city. Their method relieson dividing a year into two seasons as heating and non-heatingperiods and estimating individual autoregressive time series mod-els for each period instead of attempting to capture the seasonalpatterns in a single model. The results revealed that to use separatemodels for each period reduces the forecast errors significantlywhen the major purpose of natural gas demand is space heating.

Sánchez-Úbeda and Berzosa [7] presented a model based on thedecomposition approaches to forecast industrial end-use naturalgas consumption in Spain, with a forecasting horizon of 1–3 yearsand a daily resolution. According to the authors, the proposed mod-el can be effectively utilized in the practical operation of gas system.

Forouzanfar et al. [8] proposed a logistic based approach toforecast the natural gas consumption for residential as well ascommercial sectors in Iran, introducing two different methods toestimate the logistic parameters. One based on the concept of thenonlinear programming and the second one based on genetic algo-rithm (GA). The two approaches showed a good predictive effi-ciency. Also Siemek et al. [9] utilized a logistic based approach toforecast natural gas consumption in Poland. The proposed modelexpresses good compatibility with the natural-gas demand forthe period 1995–2000. However, according to their results, the er-ror of prognosis may reach 20%.

Sabo et al. [10] considered the problem of hourly forecast ofnatural gas consumption on the basis of hourly movement of tem-perature and natural gas consumption in the preceding period.They proposed some mathematical models with linear and nonlin-ear model functions relating to natural gas consumption. Theytested the results on a concrete example showing an acceptabledegree of accuracy.

Azadeh et al. [11] proposed an adaptive network-based fuzzyinference system-stochastic frontier analysis (ANFIS-SFA) ap-proach for the prediction of long-term natural gas consumption.The proposed models consider GDP and population as input vari-ables and, to show the applicability and superiority of the ANFIS-SFA approach, gas consumption in four Middle Eastern countries,namely Bahrain, Saudi Arabia, Syria, and United Arab Emirates,was forecasted.

Gorucu and Gumrah [12] studied an approach to understandthe factors affecting gas demand, in order to forecast gas consump-

tion by multivariable regression analysis for the city of Ankara,Turkey. They developed a statistical model and tested it for thepast years to evaluate the degree of accuracy. The model yieldedvery satisfactory results and the projections of gas consumptionfor the years 2002 up to 2005 were obtained for the city of Ankara.Also Vondrácek et al. [13] used a statistical approach to forecastnatural gas consumption of individual residential and small com-mercial customers. Their approach was based on nonlinear regres-sion principles. Parameters were estimated using mainly real datasets from meter readings of customers.

Huntington [14] developed a statistical model of industrial USnatural gas consumption based upon historical data for the1958–2003 period. He proposed a relatively simple approach,based on linear regression, which allows to generate forecasts ina simple way. The obtained data might be used as input of largerand more complex models.

Gutierrez et al. [15] examined the possibilities of using a Gom-pertz-type innovation diffusion process as a stochastic growthmodel of natural-gas consumption in Spain. They obtained a gooddescription of the historical series and good short-medium termforecasts (1998–2000).

Very recently, Soldo [16] published a general review of studiesconcerning the forecasting of natural gas consumption. He consid-ered forecasting horizons ranging from daily to annual basis, fromindividual customer level up to total world. According to his con-clusion, in the future there will be a further development of Hubertand Grey model as main tools of forecasting, at country or multi-country level. The final remark highlighted in [16] appears to bereasonable, because Grey models were successfully utilized inthe similar problem of forecasting electricity consumption [17–20].

As for Italy, there is a lack of investigation on the topic of energydemand. In fact, apart some models on electricity forecasting [21,22]and a paper on the energy consumption and the possible substitu-tion among the different sources [23], scientific literature regardinglong term forecasting of energy consumption is quite limited.

3. Methodology

3.1. Data set

All the historical data used to develop the model are taken fromofficial sources, freely available on line [2,3,24,25]. Table 1 summa-rizes the data and the respective sources.

Population profile, reported in Fig. 2(a), highlights an interestingbehavior. It shows an absence of growth from the year 1990 up to2002, whereas in the subsequent period a growth trend is detected.As observed in [21,22] this trend is due to the decrease of the birthrate in Italy, which substantially balanced the death rate, causingstability of population [21]. On the contrary, after 2002 populationstarted to increase due to immigration from other countries [21].

Historical trend of nominal GDP is reported in Fig. 2(b). Datahighlights a substantial linear trend of growth for GDP up to theyear 2008. After 2008, because of the global economic downturnof 2009 and the crisis of sovereign debt of 2010, a decrease ofthe GDP is detected, whereas in 2011 a slight increase is observed.

Fig. 2(c) reports nominal gas price, including taxes, for nondo-mestic customers, showing a substantial increasing trend with afluctuating behavior, mainly dependent on oil price fluctuationson the spot market. In fact, gas price is usually determined bymeans of ‘‘oil-linked’’ price formulas, having the following struc-ture [26]:

Pm ¼ P0 þ c1ðLFOm � LFOoÞ þ c2ðHFOm �HFOoÞ ð1Þ

where LFOo and HFOo are the starting price of light fuel oil and heavyfuel oil, LFOm and HFOm represent the price for the month m, which

Table 1Sources of historical data used in the analysis.

Variable Dates Source

Natural gas consumption 1990–2010

European Statistical Office (Eurostat)[3]

Natural gas consumption 2011 Italian Ministry of EconomicDevelopment [4]

Natural gas price 1990–2011

Eurostat [3]

Gross Domestic Product(GDP)

1990–2011

Italian Statistical Office (ISTAT) [24]

Population 1990–2011

ISTAT [24]

Temperature data 1990–2011

‘‘Il Meteo’’ [25]

Fig. 3. Series of historical nonresidential gas consumption in Italy.

V. Bianco et al. / Applied Energy 113 (2014) 392–403 395

generally takes the average value of the previous six to nine months[26]. c1 and c2 are two coefficients taking into account the naturalgas market segments competing with HFO and LFO, factors to sharerisks or rewards between sellers and buyers and technical convert-ing factors to have homogeneous units of measures.

Finally, Pm is the price paid by the buyers in the month ‘‘m’’,whereas P0 depends on the starting price of natural gas when thesupply agreement is signed and by the private negotiations be-tween buyers and sellers.

Therefore, the formula can be split into two contributions: P0,substantially determined by market conditions at the time of theagreement plus private negotiations, and the remaining part,linked to the oil spot market.

Fig. 2(d) reports the average annual minimum temperature inItaly, showing a random pattern, typical of meteorological data.Average minimum temperature in Italy from 1990 until 2011 re-sults to be equal to 10.5 �C.

Fig. 3 reports nonresidential gas consumption in Italy from 1990up to 2011.

An increasing trend is substantially detected from 1990 up to2007, with different fluctuations due to average annual minimum

(a)

(c)

Fig. 2. Historical trend of explaining variables: (a) population; (b) Nominal Gross Domestprice, including taxes, for nonresidential consumers; (d) yearly average minimum temp

temperature and GDP variations. For example in 2009, a sharp de-crease of GDP and a relevant increase of temperature caused astrong decrease of gas consumption. Similarly, in 2002 it can be no-ticed the effect of the temperature, higher with respect to the aver-age, causing a decrease in consumption.

3.2. Consumption model estimation

In this subsection an equation linking nonresidential gas con-sumption in Italy to three explaining variables, namely average an-nual minimum temperature, gas price and GDP per capita, isproposed.

The model is expressed as a linear logarithmic function and itassumes the form of a standard dynamic constant elasticity func-tion of the consumption [27,28]:

LogðCnres;eÞ ¼ aþ b1 � LogðTminÞ þ b2 � LogðPnresÞþ b3 � LogðGDPPCÞ þ b4 � LogðCnres;t�1Þþ b5 � LogðGDPPC;t�1Þ ð2Þ

(b)

(d)

ic Product and Nominal Gross Domestic Product per Capita; (c) Average nominal gaserature in Italy.

396 V. Bianco et al. / Applied Energy 113 (2014) 392–403

where Cnres represents the nondomestic gas consumption in bcm,Tmin is the annual average minimum temperature in �C, Pnres isthe average gas price for nonresidential customers in €/GJ HHV(high heating value) and GDPPC represents the GDP per capita in €

per inhabitant, a and bi are the regression coefficients and the sub-script ‘‘t � i’’ refers to the lag term (i.e. lag 1 in the present case).

The coefficients b1, b2 and b3 are very important, because theyrespectively indicate temperature, price and GDP per capita shortrun elasticities of nonresidential gas consumption.

As for the expected signs of the coefficients, one expects b1 lessthan zero, because at the increase of the temperature gas con-sumption tends to decrease (i.e. less demand for space heating,heating process in industrial applications, etc.), b2 less than zerofor usual economic reasons [29] and b3 higher than zero, becausea greater level of economic activity is expected to accelerate gasconsumption, particularly in the industrial processes.

Long run elasticities of price, GDP per capita and temperatureare calculated by dividing short run elasticities by (1 � b4) [27,28]:

LE Pnres ¼ b2=ð1� b4Þ ð3Þ

LE GDPPC ¼ b3=ð1� b4Þ ð4Þ

LE Tmin ¼ b1=ð1� b4Þ ð5Þ

where LE_Pnres, LE_GDPPC and LE_Tmin are the long run price, GDPper capita and temperature elasticities of nonresidential gas con-sumption in Italy.

Microsoft Excel is used to create the model and to develop allthe calculations and the results of these estimates are summarizedin Table 2.

There is the possibility that the ordinary least square (OLS) re-sults may be misleading due to inappropriate standard errors be-cause of the presence of heteroskedasticity [5,27]. Therefore, totest for the presence of heteroskedasticity, White heteroskedastic-ity test is performed. Since White’s test statistic value of 18.6 issmaller than 95% critical v2 value of 25, it is possible to confirmthe null hypothesis of no heteroskedasticity.

To account for the presence of serial correlation, the Breusch–Godfrey Serial Correlation LM test [30] is applied to the model.Since Breusch–Godfrey’s test statistic value of 7.59 is smaller thanthe 95% critical v2 value of 7.81, it is possible to confirm the nullhypothesis of no serial correlation in the residuals.

Finally, the Augmented Dickey Fuller (ADF) [30] test is used totest for the presence of unit roots and to establish the order of inte-gration of the considered variables (i.e. natural logarithm of Cnres,Tmin, Pnres and GDPPC). On the basis of ADF statistics, reported in Ta-ble 3, the null hypothesis of a unit root cannot be rejected at 10%level of significance for the series Cnres and Pnres.

Stationarity is obtained by running the ADF test on the first dif-ference of the variables, indicating that the series Cnres and Pnres areintegrated of order 1, I(1) in nature. As given in [27,30–32] if, afterrunning the ADF test on the first difference of the considered vari-ables, stationarity is obtained, then Eq. (2) may be regarded as a va-lid long run equilibrium relation, if the resulting residuals arestationary, I(0). As reported in Table 4, ADF test on the first differ-

Table 2Summary of coefficients, statistics (t statistics are reported in parenthesis) andestimation of price and income elasticities over the period 1990–2011 for Eq. (2).

a �0.8485 (�1.52) LE_Tmin �0.8419b1 �0.3414 (�1.90) LE_Pnres �0.2772b2 �0.1124 (�2.82) LE_GDPPC 2.7100b3 1.0989 (3.15) R2 0.915b4 0.5945 (4.06) F 32.2b5 �0.7784 (�2.34)

ences of all the variables shows that they are stationary and theADF test performed on the residuals, known as AEG (AugmentedEngle Grenger) [30], of Eq. (2) confirms that they are I(0). There-fore, it can be concluded that the variables are co-integrated andthe estimated equation may be considered as a valid expressionto forecast gas consumption [33].

3.3. Error analysis and model validation

The estimation of the prediction accuracy of the proposed mod-el is of fundamental importance, in order to guarantee forecastingvalidity. For such reason an error analysis based on three statisticalmeasures, i.e. mean absolute percentage error (MAPE), mean abso-lute deviation (MAD) and mean square error (MSE), is employed toestimate model performance and reliability [17,34,35].

MAPE is a general accepted percentage measure of predictionaccuracy. MAD and MSE are two indicators of the average magni-tude of forecasted errors, but the latter imposes a greater penaltyon a large error rather than several small deviations [34,35]. Thethree measures are defined as follows:

MAPEð%Þ ¼ 1n

Xn

k¼1

jCnres;eðkÞ � CnresðkÞjCnresðkÞ

ð6Þ

MAD ¼ 1n

Xn

k¼1

jCnres;eðkÞ � CnresðkÞj ð7Þ

MSE ¼ 1n

Xn

k¼1

ðCnres;eðkÞ � CnresðkÞÞ2 ð8Þ

For comparison purposes, a very simple model, represented by astraight line fitting over time (i.e. naïve forecasting), is considered[22]. This approach allows to confirm the necessity of developing amore complex model, such as the one represented by Eq. (2).

Fig. 4 shows the high level of accuracy guaranteed by Eq. (2), infact the figure shows that the estimated values fit in a band of ±3%with respect to the historical values. This demonstrates the abilityof Eq. (2) to represent nonresidential gas consumption in Italy. Onthe contrary, Fig. 4 also proves that a straight line fitting is totallyinadequate to describe the complex pattern of nonresidential gasconsumption.

Data reported in Table 5 illustrates the higher degree of accu-racy of Eq. (2) with respect to a naïve forecasting. In fact, MAPEof the proposed model is less than a half of a simple straight line.Similar conclusions can be derived by analyzing values of MADand MSE (less than a half and less than one sixth with respect tothe naïve forecasting, respectively).

A validation test of Eq. (2) is also proposed, estimating the fore-casting equation on data ranging from 1990 up to 2007, such thatthe remaining four years are reserved for model validation on newdata. In this way it is possible to assess equation robustness on newdata. It should be noted that this testing procedure results inslightly different coefficients values from the ones given in Table 2,since years 2008–2011 are now excluded from the estimation.

Table 3Augmented Dickey Fuller (ADF) unit root test on the considered variables. It can bedetected that for Cnres and Pnres the null hypothesis of a unit root cannot be rejected at10% level of significance.

Variable ADF test statistic Critical value 90% Test equation

Log (Cnres) �2.0923 �2.6461 ConstantLog (GDPPC) �4.8979 �2.6461 ConstantLog (Pnres) �0.6043 �2.6461 ConstantLog (Tmin) �2.7999 �2.6461 Constant

Table 4Augmented Dickey Fuller (ADF) unit root test on the first difference of the consideredvariables. It can be detected that for all the variables null hypothesis of a unit root canbe rejected at least at 10% level of significance.

Variable ADF test statistic Critical value 90% Test equation

Log (Cnres) �5.3997 �2.6504 ConstantLog (GDPPC) �2.7907 �2.6504 ConstantLog (Pnres) �5.0322 �2.6504 ConstantLog (Tmin) �8.8141 �2.6504 Constant

Table 5Comparative analysis of forecasting errors.

Error measure Eq. (2) Naïve forecasting

MAPE 1.6% 4.6%MAD 0.5 1.3MSE 0.4 2.5

V. Bianco et al. / Applied Energy 113 (2014) 392–403 397

In order to perform a comparison, the same procedure is alsoapplied to the naïve forecasting equation and the results are re-ported in Table 6.

The validation highlights the capacity of the proposed model togive a future outlook with a sufficient degree of accuracy, showingrelative errors always less than 10%.

It should be taken into account that the years considered formodel validation (i.e. 2008–2011) represent a tough test, becausestrong changes in the trend are present, but the proposed modeldemonstrates its capability to capture the directions (i.e. in-crease/decrease) of the consumption variations correctly, provingits strong prediction ability.

Instead, straight fitting forecast shows its inadequacy to capturesuch a complex consumption pattern, resulting in high relative er-rors (i.e. equal or higher than 15%) versus observed values.

4. Analysis and outlook of consumption drivers

In this section, it is discussed the possible evolution of theexplaining variables of the model and the expected impact of reg-ulations and policies about gas markets. All these variables have adirect effect on the final consumption, therefore a careful assess-ment is necessary.

4.1. Future outlook for explaining variables

In order to perform the forecasting of nonresidential natural gasconsumption, it is necessary to have an outlook on the futuredevelopment of the explaining variables. The variables consideredin the present paper are of complex estimation, therefore, ratherthan using simplified methodologies as proposed in [21,22], thevalues are taken or estimated from specific and detailed studiesavailable in the literature.

4.1.1. GDP per capita outlookTwo scenarios of GDP growth are proposed: a baseline and a

high growth outlook. Growth rate are taken from [36], where a realgrowth rate of 0.2% is foreseen for Italy from 2012 up to 2017,

Fig. 4. Comparison of data generated by means of Eq. (2) and straight lineforecasting (i.e. naïve forecasting) with respect to the real historical consumptionpattern.

whereas a growth rate of 0.7% is expected for the period 2018–2030.

As for high growth scenario, it is assumed that Italian GDP willgrow with the average rate of Euro area, which means at a pace of2.0% from 2012 up to 2017 and 2.2% for the period 2018–2030.

The above mentioned growth rates are expressed in real terms,therefore it is necessary to consider the inflation, in order to applythem to the nominal historical series of GDP reported in Fig. 2(b).As suggested in [37], an inflation rate of 2% is assumed all alongthe period of study (i.e. 2012–2030). Finally, to get the GDP per ca-pita, a projection of Italian population until 2030 is taken from[24]. All the projections are reported in Fig. 5.

4.1.2. Gas price projectionsDuring the last three/four years natural gas prices have shown a

different trend with respect to the previous period (i.e. previous10–15 years), highlighting a ‘‘de-coupling’’ with respect to oil pricetrend, resulting in a decrease of prices. The causes which deter-mined the price ‘‘de-coupling’’ are different:

� Economic downturn caused a reduction in natural gasdemand, with the consequence of increasing volumes onthe market.

� Such as natural gas, also electricity demand decreased andelectricity generators try to sell on the gas market part oftheir ‘‘take or pay’’ (a clause in the contract according towhich the gas must be bought even if it is not needed)quota of their contracts, in order to reduce financial losses.

� Because of the strong development of unconventional gasextraction in USA, a huge quantity of liquefied natural gas(LNG), originally directed in USA, is diverted towards Euro-pean and Asian markets.

All the above mentioned facts have the effect to increase theliquidity on the spot markets, which causes a decrease of price,determining a de-coupling from oil linked contracts, because, onthe contrary, oil price on the spot market increases, as shown in[38].

In order to develop a robust outlook of gas price, it is necessaryto assess if the de-coupling will be maintained in the future, or if itis only a transient condition determined by contingent events.According to [38], the price de-coupling from oil-linked contractsis supposed to be a transient condition and in the long term a re-coupling of prices to oil linked contracts is expected.

To the scope of the present paper, the hypothesis of a re-cou-pling of gas prices to oil linked gas contracts appears to be reason-able. The focus of the present work is on nonresidential gasconsumption and it is expected that most of the companies not in-volved in the energy business prefers to sign long term agreementswith suppliers, rather than to invest resources to develop a know-how to operate on the gas market. For energy related industries,such as electricity generators, the topic is much more complicated,because re-coupling or de-coupling of prices means to pursue twodifferent long term strategies in terms of gas supply and capacityinvestments.

In order to provide a reliable outlook of gas prices, the historicalcorrelation between oil and gas prices is analyzed. Fig. 6(a) shows

Table 6Observed and forecasted gas consumption values for Eq. (2) and straight line fitting. Relative errors (RE) are reported in parenthesis.

Year Observed value (bcm) Eq. (2) forecasting (bcm) RE (%) Naïve forecasting (bcm) RE (%)

2008 29.8 29.6 �0.8 31.9 6.32009 26.6 29.0 8.3 32.3 17.62010 27.9 30.0 7.1 32.7 14.82011 27.1 29.9 9.3 33.1 18.2

(a) (b)

Fig. 5. Future outlook of GDP and population (a) and GDP per capita (b).

398 V. Bianco et al. / Applied Energy 113 (2014) 392–403

that a high correlation is detected, with a R2 of 0.96, confirmingthat the two prices were strongly related. Therefore, a simple linearregression of gas and oil prices is proposed to estimate the futuredevelopment:

Pimp ¼ 0:768þ 0:099 � POil ð9Þ

where Pimp represents the average price of natural gas on the borderof the considered country and POil is the spot price of oil.

To perform the forecasting, a projection of oil prices is neededand, to this end, different outlooks provided by internationally rec-ognized institutions [39–41] are considered and compared. Thecomparison, reported in Fig. 7, shows that [39] is aligned withthe low price scenario proposed in [41], whereas the outlook of[40], in long term, is similar to the base cases of [41]. Scenarios re-ported in [41] are the result of a supply and demand model, such asthe scenario proposed in [40], whereas [39] does not provide infor-mation on the methodology used. Considering that scenarios sug-gested by [40,41] are quite close, data proposed in [41] areconsidered here, because different cases (i.e. low, base and highgrowth) are proposed, giving the possibility to develop a scenarioanalysis. Using Eq. (9) and oil projections given in [41], adjusted

(a)

Fig. 6. Gas price analysis: (a) correlation between oil spot price and natural gas importprice for final users [4].

considering the inflation rate suggested in [37], natural gas priceoutlook is obtained (Fig. 7(b)).

Price series obtained in this way are to be considered at the na-tional border, therefore, to get prices for final users, other compo-nents (i.e. logistic, commercial margin, taxes, etc.) might beconsidered. These components are estimated from the ratio ofgas price to final users and gas price on the border for the last fiveyears (Fig. 6(b)), accounting for about 30%. Only the last five yearsare taken into account, because a decreasing trend of the differencebetween final user and import gas prices is observed in the last tenyears in Italy (see Fig. 6(b)). Therefore, it can be assumed that, atmoment, a price surplus of 30% is charged to final users with re-spect to the import price.

4.1.3. Assumptions on temperatureAs reported in Eq. (2), gas consumption depends also on mini-

mum temperature, therefore assumptions on this parameter are re-quired to develop the analysis. To this scope, three simple scenariosare considered in the present paper, namely average value,maximum value and minimum value. As for the average value, theaverage temperature from 1990 until 2011 (i.e. 10.5 �C, highlighted

(b)

price [1]; (b) trend of oil spot price [1], natural gas import price [1] and natural gas

(a) (b)

Fig. 7. Oil prices scenarios (a) and natural gas import prices scenarios (b).

V. Bianco et al. / Applied Energy 113 (2014) 392–403 399

in Fig. 2(d)) is considered for all the forecasting horizon; whereas formaximum and minimum value the highest (11.3 �C) and lowest (i.e.9.3 �C) values of temperature are respectively considered all theyears.

Moreover, to get a more realistic consumption profile, a random ser-ies of temperature is generated, with values contained between themaximum and minimum values recorded in the period 1990–2011.

4.2. Regulatory framework

From 1998, European Union started to reform the gas market,with the objective to increase the competitiveness of the sector.The main target is to establish a single European market, wherethe price is ideally set by the interaction of supply and demand(i.e. clearing price).

In order to do this, EU launched policies of liberalization andintegration of the different national markets. The third gas direc-tive, 2009/73/EC [42], represents the European gas legislation inforce at present. It focuses on three main issues [43]:

� Unbundling of transport and other activities.� Regulated third party access.� Concept of ‘‘eligible customer’’.

The first point lead to the separation of vertical integrated com-panies, usually former state monopolists (ENI in Italy); in fact, thedirective imposes the separation of companies dealing with theraw material (producers, importers, wholesalers and retailers)and companies providing the system with infrastructure and ser-vices (transporters, LNG plant operators, and storage).

The second point regards the definition of a regulation to guar-antee the access to the transportation network for all the operators,in order to create a fair competitive environment.

Whereas the third issue aims at giving customers the freedomto choose their energy suppliers.

The directive 2009/73/EC is of fundamental relevance especiallyfor the issue of the unbundling. In fact, it proposes three possiblemodels to manage this issue, namely ownership unbundling(OU), independent system operator (ISO) and independent trans-mission operator (ITO); more details are reported in Table 7. InItaly the ITO model is adopted.

In terms of quantitative modeling, it is very difficult to take intoaccount the impact of policies, because their effect needs to bemonitored in the course of the years and the prescriptions of thedirective 2009/73/EC started to be mandatory from March 2012,therefore an extremely limited amount of data are available to de-tect a significant trend.

Anyway, some important consideration can be drawn. Particu-larly, as reported in Fig. 5(b) and previously discussed, gas price

for final nonresidential consumers is about 30% higher than the le-vel of the import price.

The aim of the EU directives is to minimize, by enhancing theconcurrency among the operators, price surplus with respect tothe level of the import price, in order to guarantee a competitivesupply for the EU customers.

According to this, it is possible to imagine two extreme scenar-ios: one where the EU directive completely fulfills its target, there-fore gas price for final customer is assumed to be equal to theimport price, and a second scenario, where it is supposed thatthe directive fails to reach its scope and the extra price with re-spect to the import remains at the current level (i.e. 30%). All theother possible situations are in between the two hypothesizedscenarios.

5. Results and discussion

5.1. Price, income and temperature elasticities

To investigate the consumption structure, Eq. (2) is built tounderstand the influence of economic variables, price and GDPper capita, and the impact of climatic factor (i.e. average minimumtemperature).

It is found that long run price elasticity, LE_Pind, GDP per capitaelasticity, LE_GDPPC, and minimum temperature elasticity, LE_Tmin,have the following values: 2.7100, �0.2772 and�0.8419. As for theexpected signs, they seem consistent. In fact, LE_GDPPC has a posi-tive sign, in agreement with the fact that a higher economic activ-ity results in higher degree of consumption of primary resources,among which natural gas.

On the contrary, LE_Pind and LE_Tmin have negative signs, whichis once again consistent with the economic fundamentals. In fact,higher price will cause a decrease in consumption and, in the caseof natural gas, higher temperature reduces the consumption of nat-ural gas (i.e. there is a lower need for heating).

To the best of authors’ knowledge, it is the first time that price,GDP per capita and minimum temperature elasticities of Italiannonresidential natural gas consumption are determined, thereforethere are no other values in the open literature which can be usedto do a significant comparison.

Observing the estimated values of long run elasticities, it can bedetected that LE_GDPPC has the highest value, showing that ItalianGDP per capita is closely linked to natural gas consumption, there-fore any recession or any shocks that have a negative influence onGDP per capita will result in a negative impact on nonresidentialnatural gas consumption, as noticed in the last years.

On the contrary, among all the estimated elasticities, LE_Pind hasthe lowest impact, showing that nonresidential gas consumptionhas a limited reaction to price signals, therefore pricing policies

Table 7Description of unbundling models according to the Directive 2009/73/EC [42].

Definition Description

Ownership unbundling (OU) A new company which owns and manages the transport network is created. This company results to be totally independent by thevertically integrated companies operating in the exploration, production and retail businessIn [42], OU is indicated as the most effective way to promote investments in infrastructure in a non-discriminatory way, fair access tothe network for new entrants and transparency in the market

Independent system operator(ISO)

Vertically integrated company maintains the ownership of the transport network, but its management is in charge of an independentcompany

Independent transmissionoperator (ITO)

Vertically integrated company maintains the ownership of the transport network and the control of the company in charge of itsmanagement, but it must guarantee its independence. The independence of the transmission operator is assessed by controls of thenational authorities

Table 8Scenario Grid: all the scenarios considered are reported in the table below. Bold rows show unlikely scenarios, which are not taken into account. A total of twelve scenario isexamined.

Scenario GDP base GDP high Gas price base Gas price high Gas price low Tmin average Tmin low Tmin high 2009/73/EC success 2009/73/EC failure

1 x x X x2 x x X x3 x x X x4 x x x x5 x x x x6 x x x x7 x x x x8 x x x x9 x x x x

10 x x X x11 x x X x12 x x X x13 x x x x14 x x x x15 x x x x16 x x x x17 x x x x18 x x x x19 x x X x20 x x X x21 x x X x22 x x x x23 x x x x24 x x x x25 x x x x26 x x x x27 x x x x28 x x X x29 x x X x30 x x X x31 x x x x32 x x x x33 x x x x34 x x x x35 x x x x36 x x x x

400 V. Bianco et al. / Applied Energy 113 (2014) 392–403

cannot be used to stimulate consumption; unless, price variation isrelevant.

The low price elasticity highlights that for nonresidential cus-tomers the possibility to switch to another primary energy sourceis quite complex. In fact, if a production process or a heating sys-tem are designed to be fuelled with natural gas, it becomes extre-mely complicated and expensive to substitute or convert thesystems in order to use other sources of primary energy.

For example, as for heating systems, it might be possible the sub-stitution of plants based on natural gas boiler with electrical heatpumps, but the price increase of natural gas should be as high asto justify this investment, which results to be quite relevant in thecases of large office buildings or production plants. These kinds ofinvestments are very difficult to be evaluated by companies, becausethe price of gas is very fluctuating, therefore it is complicated tounderstand if a variation of price is due to a random phenomenon

or it represents a structural change that will last in the time and,thus, it might justify an investment.

LE_Tmin also has a substantial impact on nonresidential gas con-sumption, determining fluctuations in the consumption profile.LE_Tmin is of fundamental importance, because Tmin, being a ran-dom value, is very difficult to estimate. Therefore to avoid shortageof gas supply, storages are necessary to manage unpredictable sit-uations, among which severe meteorological conditions. This con-firms that to know how consumption react to temperature changesis crucial.

5.2. Scenario analysis

Forecasts are obtained by using Eq. (2) and a scenario analysis isperformed, in order to present different cases determined by vari-ous projections of the explaining variables. The impact of economic

(a) (b)

(c) (d)

Fig. 8. Scenarios analysis of nonresidential natural gas consumption in Italy in the case of failure in the implementation of the Directive 2009/73/EC. GDP base, price base anddifferent temperatures (a). GDP base, price low and different temperatures (b). GDP high, price base and different temperatures (c). GDP high, price high and differenttemperatures (d).

V. Bianco et al. / Applied Energy 113 (2014) 392–403 401

and climatic data, as well as the effect of regulatory changes isconsidered.

Table 8 reports all the investigated scenarios, twenty-four in to-tal, even though by the mere combination of all the variablesthirty-six scenarios are determined, but twelve of them are consid-ered to be not significant.

Particularly, it is unlikely to have a base GDP growth and a highgas price, or, on the contrary, a high GDP growth and a low gasprice. In the present analysis, gas price is supposed to be oil-in-dexed, therefore it does not seem consistent to have a high GDPgrowth and a low oil price, or vice versa.

Fig. 8 reports the forecasting of natural gas consumption for allthe considered scenarios, in the case that the directive 2009/73/ECfails to reach its target (i.e. the creation of a competitive market)and a price surplus (i.e. 30% in Italy) with respect to the level of im-port price is maintained.

Fig. 8(a) considers the base GDP and gas price scenarios, high-lighting the impact of average minimum temperature. In 2030,consumption in ‘‘Scenario 4’’ is about 5.5 bcm higher than ‘‘Sce-nario 5’’, therefore the impact of about �2 �C (i.e. 9.3 �C in Scenario4 vs. 11.3 �C in Scenario 5) on the minimum temperature is quiterelevant, representing more than 10% of the total consumption.Average minimum temperature scenario (i.e. 10.5 �C in Scenario1) presents consumption 2 bcm higher than ‘‘Scenario 5’’ and3.5 bcm lower than ‘‘Scenario 4’’ in 2030.

Fig. 8(b) takes into account base GDP growth and low pricesscenarios, showing the impact of the different temperatureassumptions. In 2030, average minimum temperature scenario(i.e. 10.5 �C in Scenario 3) presents consumption 2.3 bcm higherthan ‘‘Scenario 9’’ (i.e. 9.3 �C) and 4.0 bcm lower than ‘‘Scenario8’’ (i.e. 9.3 �C), similarly to the data reported in Fig. 8(a).

It can be noticed that consumptions reported in Fig. 8(b) aregenerally higher with respect to those of Fig. 8(a), because lowprice scenario tends to encourage natural gas usage. Even thoughprice elasticity is low, differences in the prices of the proposed sce-narios (Fig. 7(b)) are quite large, therefore they have a substantialimpact on the level of consumption.

From a practical point of view, it happens that a low price leveldiscourages investments in energy efficiency, because their profit-

ability decreases or it even becomes negative and companies preferto consume larger quantities of natural gas.

Fig. 8(c and d) examine the cases where high GDP growth to-gether with base and high gas prices scenarios are employed fordifferent minimum temperature levels (i.e. average, low, and high).

By comparing Fig. 8(a and c), the strong effect of GDP can be de-tected. For example consumption of ‘‘Scenario 13’’ is 5.2 bcm high-er than that of ‘‘Scenario 4’’. Instead, the comparison of Fig. 8(c andd) shows that a sustained price scenario determines a strong de-crease of consumption, even if GDP growth rate is high.

When high GDP growth and high gas price are combined to-gether, they determine a positive combination to pursue energyefficiency. In fact, a high GDP growth means that, in average, com-panies are in a positive cycle and they have possibilities to sustaininvestments. Therefore, if natural gas price is high, they can investin energy efficiency actions or to convert their facilities to usecheaper primary energy sources.

Finally, Fig. 9(a) reports the two extreme (Scenarios 5 and 13)and the average scenarios (Scenario 1), showing that the averageconsumption is closer to the lowest level, rather than to thehighest.

This is due to the fact that the only difference between scenario1 and 5 is represented by temperature (i.e. average vs. high),whereas in scenario 13, also different GDP and price levels are con-sidered. It can be said that in 2030 a consumption of 32 bcm is ex-pected, recording an average growth rate (i.e. CAGR) of 0.9% peryear in the horizon 2011–2030; whereas the maximum foreseenconsumption is estimated to be about 43 bcm, with a pace ofgrowth of 2.4% per year from 2011 up to 2030 and 34 bcm are fore-casted in an ‘‘average’’ condition, meaning a growth rate of 1.2%per year in the period 2011–2030.

These estimations are of relevant importance for the infrastruc-ture planning, because it is possible to analyze if new facilities areneeded to satisfy the increase of demand in the nonresidentialsector.

At moment Italy has a capacity to import a maximum of114 bcm/year [2], about 100 bcm via pipelines and 14 bcm viaLNG terminals, the current level of saturation of the infrastructuresis 86% [2]. Considering a maximum load factor of 95% for safety

(a) (b)

Fig. 9. Assuming the failure of the Directive 2009/73/EC: extreme and average consumption scenarios (a) and comparison of the impact of average and ‘‘random’’ yearlyaverage minimum temperature on the average scenario, i.e. ‘‘Scenario 1’’ (b).

Table 9Comparison of scenarios in the case of successful implementation or failure of Directive 2009/73/EC.

Scenario 2009/73/EC Cnres 2011 (bcm) Cnres 2030 (bcm) D (2030 vs. 2011) (bcm) DScenario (bcm)

1 vs. 19 Failure 27.1 34.0 6.9 �2.8Success 36.8 9.7

3 vs. 21 Failure 27.1 38.7 11.6 �3.2Success 41.8 14.7

4 vs. 22 Failure 27.1 37.5 10.5 �3.1Success 40.6 13.5

5 vs. 23 Failure 27.1 32.0 4.9 �2.6Success 34.6 7.6

8 vs. 26 Failure 27.1 42.7 15.6 �3.5Success 46.2 19.1

9 vs. 27 Failure 27.1 36.4 9.3 �3.0Success 39.4 12.3

10 vs. 28 Failure 27.1 38.7 11.6 �3.2Success 41.9 14.8

11 vs. 29 Failure 27.1 35.5 8.4 �2.9Success 38.4 11.3

13 vs. 31 Failure 27.1 42.7 15.6 �3.5Success 46.2 19.1

14 vs. 32 Failure 27.1 36.4 9.4 �3.0Success 39.4 12.3

15 vs. 33 Failure 27.1 39.2 12.1 �3.2Success 42.4 15.3

16 vs. 34 Failure 27.1 33.4 6.4 �2.7Success 35.2 9.1

402 V. Bianco et al. / Applied Energy 113 (2014) 392–403

reason, it is possible to increase the import of about 10 bcm, thusnew infrastructures are needed to guarantee the supply in the non-residential sector, unless the demand of natural gas in the othersectors decreases.

Fig. 9(b) shows consumption in scenario 1, when a random tem-perature profile is applied, in fact, different fluctuations can be ob-served. Cumulated consumption from 2012 up to 2030 for scenario1 is equal to about 582 bcm, whereas scenario 1 with a randomtemperature profile has a cumulated consumption of about592 bcm. The average difference is about 0.5 bcm per year, whichis not relevant (i.e. about 2% per year), but on a specific year,according to the temperature value, it can be also relevant, forexample 1.5 bcm in 2016 (i.e. three times the average).

In light of this, it can be said that to estimate average consump-tion, the average temperature can be used, but when there is thenecessity to investigate on the consumption profile, a series of ran-dom temperature furnishes more detailed results.

For example, the use of Eq. (2) with a random temperature pro-file represents a powerful tool for the physical design of gas storag-es which guarantee supply, when there are rigid climaticconditions.

As for scenarios 19–36, they assume that the directive 2009/73/EC is successfully implemented, thus the level of natural gas pricefor final users converges to that of the import price, resulting in adiscount of 30% with respect to prices assumed in Scenarios 1–18.

Trends for scenarios 19–36 are similar to those of scenarios 1–18, therefore they are omitted for sake of brevity, but the level ofthe consumption is higher. Table 9 reports a comparison for allthe considered scenarios.

Differences at the year 2030 are analyzed for couples of scenar-ios whose unique difference is represented by success or failure inthe implementation of the directive 2009/73/EC.

Table 9 highlights that a successful implementation of thedirective should determine an average increase of the consumptionequal to about 3 bcm in 2030.

The two extreme scenarios on the implementation of the direc-tive 2009/73/EC considered in the present paper lead to a mini-mum and maximum increase of consumption respectively equalto 0 and 3 bcm, therefore it seems consistent to expect that a likelyimpact of the directive implementation (i.e. average scenario)should determine an increase of consumption between 1 and2 bcm in 2030.

6. Conclusion

The present paper introduces a model to forecast nonresidentialconsumption of natural gas in Italy. Consumption drivers are iden-tified and discussed, therefore a single equation demand model ispresented. The model takes the classical form of a standard

V. Bianco et al. / Applied Energy 113 (2014) 392–403 403

dynamic constant elasticity function of consumption. The esti-mated equation is validated on historical data, demonstrating agood prediction ability, even though the consumption patternhas a complex trend.

Results revealed that long run price elasticity is quite low, i.e.about �0.28, with respect to the one of GDP per capita, i.e. about2.71; whereas average minimum temperature is detected to havea significant impact on nonresidential gas consumption, with ashort run elasticity of about �0.34 and a long run elasticity of�0.84.

The low elasticity to the price is probably due to the limitedamount of options and the relative high cost that companies haveto switch towards other primary sources of energy.

In order to perform a scenario analysis of expected consump-tion, different evolutions of explaining variables and regulatoryframework are proposed, leading to the determination of twenty-four significant scenarios.

By analyzing the obtained scenarios, it is concluded that in 2030a natural gas consumption between about 32 and 46 bcm shouldbe expected, highlighting a growth rate, from 2011 until 2030,ranging from 0.9% to 2.9% respectively.

Estimation of elasticities, consumption forecasts and commentscontained in this paper are expected to be helpful to energy plan-ners and policy makers in building future scenarios about Italiannatural gas sector. Particularly, the proposed estimations mightbe of help in the planning of new infrastructures and in the phys-ical design of storages. Furthermore, the estimates of GDP andprice elasticities have particular relevance for designing appropri-ate pricing policies in the energy sector.

Acknowledgements

The authors want to acknowledge Felice Daniele Russo for theprecious help for the download of meteorological data. The authorsalso want to express their gratitude to two anonymous reviewersfor the helpful comments provided.

The present work was supported by Genuenese Atheneum(PRA2012 Grant No. CUP D31J13000000005).

References

[1] BP Statistical Review of World Energy, June 2012. <http://www.bp.com/statisticalreview> [accessed 12.18.12].

[2] Relazione Annuale sullo Stato dei Servizi e sull’Attività Svolta, Autorità perl’Energia Elettrica e il gas, March 2012. <http://www.aeeg.it> [accessed06.07.13].

[3] EUROSTAT. European Statistical Office, December 2012. <http://epp.eurostat.ec.europa.eu/portal/page/portal/energy/data/database> [accessed 12.18.12].

[4] MISE. Italian Ministry of Economic Development, December 2012 [in Italian].<http://dgerm.sviluppoeconomico.gov.it/dgerm/consumigasannuali.asp>[accessed 12.18.12].

[5] Erdogdu E. Natural gas demand in Turkey. Appl Energy 2010;87(1):211–9.[6] Aras H, Aras N. Forecasting residential natural gas demand. Energy Sources

2004;26(5):463–72.[7] Sánchez-Úbeda EF, Berzosa A. Modeling and forecasting industrial end-use

natural gas consumption. Energy Econ 2007;29(4):710–42.[8] Forouzanfar M, Doustmohammadi A, Menhaj MB, Hasanzadeh S. Modeling and

estimation of the natural gas consumption for residential and commercialsectors in Iran. Appl Energy 2010;87(1):268–74.

[9] Siemek J, Nagy S, Rychlicki S. Estimation of natural-gas consumption in Polandbased on the logistic-curve interpretation. Appl Energy 2003;75(1–2):1–7.

[10] Sabo K, Scitovski R, Vazler I, Zekic-Sušac M. Mathematical models of naturalgas consumption. Energy Convers Manage 2011;52(3):1721–7.

[11] Azadeh A, Asadzadeh SM, Saberi M, Nadimi V, Tajvidi A, Sheikalishahi M. Aneuro-fuzzy-stochastic frontier analysis approach for long-term natural gasconsumption forecasting and behavior analysis: the cases of Bahrain, SaudiArabia, Syria, and UAE. Appl Energy 2011;88(11):3850–9.

[12] Gorucu FB, Gumrah F. Evaluation and forecasting of gas consumption bystatistical analysis. Energy Sources 2004;26(3):267–76.

[13] Vondrácek J, Pelikán E, Konár O, Cermáková J, Eben K, Maly M, et al. Astatistical model for the estimation of natural gas consumption. Appl Energy2008;85(5):362–70.

[14] Huntington HG. Industrial natural gas consumption in the United States: anempirical model for evaluating future trends. Energy Econ 2007;29:743–59.

[15] Gutiérrez R, Nafidi A, Sánchez RG. Forecasting total natural-gas consumptionin Spain by using the stochastic Gompertz innovation diffusion model. ApplEnergy 2005;80(2):115–24.

[16] Soldo B. Forecasting natural gas consumption. Appl Energy 2012;92:26–37.[17] Bianco V, Manca O, Nardini S, Minea AA. Analysis and forecasting of

nonresidential electricity consumption in Romania. Appl Energy2010;87(11):3584–90.

[18] Kayacan E, Ulutas B, Kaynak O. Grey system theory-based models in timeseries prediction. Expert Syst Appl 2010;37(2):1784–9.

[19] Akay D, Atak M. Grey prediction with rolling mechanism for electricitydemand forecasting of Turkey. Energy 2007;32(9):1670–5.

[20] Zhou P, Ang BW, Poh KL. A trigonometric grey prediction approach toforecasting electricity demand. Energy 2006;31(14):2503–11.

[21] Bianco V, Manca O, Nardini S. Linear regression models to forecast electricityconsumption in Italy. Energy Sources Part B: Econ Plan Pol 2012;8(1):86–93.

[22] Bianco V, Manca O, Nardini S. Electricity consumption forecasting in Italyusing linear regression models. Energy 2009;34(9):1413–21.

[23] Gori F, Takanen C. Forecast of energy consumption of industry and household& services in Italy. Int J Heat Technol 2004;22(2):115–21.

[24] ISTAT. Italian Institute of Statistics [in Italian]. <http://dati.istat.it/> [accessed12.18.12].

[25] Il Meteo [in Italian]. <http://www.ilmeteo.it/portale/archivio-meteo/>[accessed 12.18.12].

[26] Kanai M. Decoupling the oil and gas prices. Paris: IFRI; 2011.[27] Erdogdu E. Electricity demand analysis using cointegration and ARIMA

modelling: a case study of Turkey. Energy Pol 2007;35(2):1129–46.[28] Haas R, Schipper L. Residential energy demand in OECD-countries and the role

of irreversible efficiency improvements. Energy Econ 1998;20(4):421–42.[29] Halicioglu F. Residential electricity demand dynamics in Turkey. Energy Econ

2007;29(2):199–210.[30] Gujarati DN. Basic econometrics. New York: The Mc Graw Hill Companies;

2004.[31] Amarawickrama HA, Hunt LC. Electricity demand for Sri Lanka: a time series

analysis. Energy 2008;33(5):724–39.[32] Wooldridge JM. Introductory econometrics: a modern approach. New

York: South Western, Division of Thomson Learning; 2005.[33] Engle RF, Granger CWJ. Co-integration and error correction: representation,

estimation and testing. Econometrica 1987;55(2):251–76.[34] Zhou P, Ang BW, Poh KL. A trigonometric grey prediction approach to

forecasting electricity demand. Energy 2006;31:2839–47.[35] Wilson JH, Keating B. Business forecasting. New York: McGraw-Hill; 2009.[36] OECD Economic Outlook; 2012. doi: 10.1787/16097408 [accessed 12.18.12][37] PWC. Global Economy Watch. Projections. <http://www.pwc.co.uk/economic-

services/global-economy-watch/gew-projections.jhtml> [accessed 12.18.12][38] Asche F, Oglend A, Osmundsen P. Gas versus oil prices the impact of shale gas.

Energy Pol 2012;47:117–24.[39] World Bank. Commodity Price Forecast Update. <http://econ.worldbank.org/

WBSITE/EXTERNAL/EXTDEC/EXTDECPROSPECTS/0,,contentMDK :21574907~menuPK:7859231�pagePK:64165401�piPK:64165026�theSitePK:476883,00.html> [accessed 12.18.12].

[40] Energy Information Administration. Annual Energy Outlook 2013 EarlyRelease. <http://www.eia.gov/forecasts/aeo/er/> [accessed 12.18.12].

[41] Department of Energy and Climate Change. Oil Price Projections. 2012 Update<http://www.decc.gov.uk/en/content/cms/about/ec_social_res/analytic_projs/ff_prices/ff_prices.aspx> [accessed 12.18.12].

[42] European Parliament and Council. Directive 2009/73/EC concerning commonrules for the internal market in natural gas and repealing Directive 2003/55/EC, Bruxelles, Belgium; 2009. <http://www.europarl.europa.eu/>.

[43] Capece G, Cricelli L, Di Pillo F, Levialdi N. New regulatory policies in Italy:Impact on financial results, on liquidity and profitability of natural gas retailcompanies. Util Pol 2012;23:90–8.