A Regional Demand Forecasting Study for Transportation Fuels in Turkey

Transcription

1 A al Demand Forecasting Study for Transportation Fuels in Turkey by Özlem Atalay a, Gürkan Kumbaroğlu Bogazici University, Department of Industrial Engineering, 34342, Bebek, Istanbul, Turkey, Phone : , ozlematalayy@gmail.com Abstract Fuel demand in transportation sector has received considerable attention for the last decades. Population growth and economic development in the past few decades have caused a growing demand for fuel in transportation sector in Turkey. The objective of this study is to project the fuel consumption of each region in Turkey for transportation sector by applying different forecasting methods for the year This study is based on Multiple Linear Regression, Moving Average, Double Moving Average, Simple Exponential Smoothing, Double Exponential Smoothing, Holt Winters model using the as independent variables such as: gross value added, population, the numbers of registered vehicles, and fuel prices. The models are based on 11-year historical data between the years 2001 and 2012 which have been provided by various governmental institutions, and are used to project the trends in future transport fuel consumption for year the In most of each regions, Multiple Regression Analysis gives the best result for explaining the fuel demand. In developing/ recently developed countries, economic growth has been followed by increase in road transportation fuel consumption. For this reason, it was expected that gasoline consumption would increase. However, gasoline consumption has been constantly decrease in Turkey between the years 2006 and 2013 which is a developing/ recently developed country. Whereas diesel consumption reached to record levels in the same period. There is not a dramatic change in the consumption of LPG. Model projections indicate in all regions a shift away from gasoline-fueled vehicles towards diesel-fueled ones. 1. INTRODUCTION The prediction of fuel consumption has been an important tool for energy planning, with the following purposes which cited as: to (i) develop right pricing and taxation systems, (ii) help decide future investments on fuels, (iii) aid address emission and pollution issues, and (iv) plan of future energy needs, identify national infrastructure and research and development requirements. The fuel demand is critical for decision makers so that they can implement corresponding policies and regulations to ensure sustainable development [1]. People have become reliant on private vehicles, and have been using them more frequently. When the economic prosperity level of a country increases, people are able to afford a vehicle to go places, any time, instead of taking public transport which is not as comfortable as a private vehicle. In good economic conditions, travel demand increases. These increases also cause an increase in the amount of fuel consumption from personal travel. A similar situation can be seen in

2 freight transport. The growing economy has stimulated more goods transport activities from one location to other parts of the country, as well as to the rest of the world. Therefore, the increase in the number of freight trips means an increase in the amount of fuel consumption [2]. The main objectives of this study are to analyze past Gasoline, Diesel and Liquefied Petroleum Gas (LPG) consumption in Turkey s transportation sector and to identify the main factors affecting future demand. To achieve these objectives, the study develops six models using different independent variables based on 11-year historical data between years 2001 and 2012 refined from scattered data sources. In general, established models can be handled in six main categories. Multi Linear Regression Models (region-based and yearly) Moving Average (region-based and yearly) Double Moving Average (region-based and yearly) Simple Exponential Smoothing (region-based and yearly) Double Exponential Smoothing (region-based and yearly) Time series models Holt Winters model (region-based and quarterly) With the help of these models, it is possible to perform sales forecasts in terms of different fuel types, sectors and geographical regions. Figure 1. shows that these regions in Turkey can be categorized in twelve categories according to TUIK; Figure 1. TUIK s Used in the Analyses After many trial and error procedures, 36 econometric models have been developed revealing the effect of different socio-economic and technological factors on annual fuel demand in each region. The validity of each model is tested by means of statistical and diagnostic tests. Results provide interesting insight into the underlying drivers of fuel demand in different regions. In addition to the annual econometric models based on linear regression, various quarterly time series models were developed including several autoregressive integrated moving average (ARIMA) models with different exponential smoothing methods. The validity of the models has been checked and results have been compared based on mean absolute deviation as well as percentage and squared error values yielding the most accurate set of forecasting models. It is found that econometric models generally outperformed the time series ones and provided effective alternatives for forecasting annual fuel demand. Model projections indicate in all regions a shift away from gasoline-

3 million tonnes fueled vehicles towards diesel-fueled ones. The main reason for the attractivity of diesel cars is fuel efficiency as diesel engines are 20 percent to 40 percent more fuel efficient than equivalent gasoline engines. However, diesel fuel contains about 15% more carbon per litre reducing the CO 2 emission advantage by favourable fuel efficiency. It is expected that increasingly stringent emissions regulations and the high cost of new anti-pollution technology will make diesel engines much more expensive [3]. Criticism on diesel vehicles has recently increased and expectations reversed such that diesel automobiles will see a downward trend (e.g. [4], [5], [6]). Results of this study indicate that this will not be the case in Turkey in the short term unless there is a new environmental tax policy or standard to discourage the use of diesel fuel. According to the International Energy Agency, tax as a percentage of total price for gasoline and diesel was %60 and %49, respectively. The sustainability of diesel s tax advantage in some European countries including Turkey, however, is questionable. Since Turkey is a developing/ recently developed country, it is expected that the fuel demand grows fast in the short and the long term. The results show that in each region fuel demand increase dramatically. To overcome this demand increase, the studied scenarios, can be implemented Road Transportation Fuel Demand in Turkey Forecasting fuel demand becomes more difficult because of the unprecedented increase in fuel consumption. Petrobased road transportation fuels, namely; gasoline, diesel, and liquefied petroleum gas (LPG) have a market share of more than 99% in road transportation sector. As more than 90% passengers and goods are transported by roads in Turkey. In Figure 2., road transportation fuel demand is shown according to fuel types. Gasoline demand decreased linearly between the years 2006 and 2012 in Turkey. Whereas, diesel consumption reached to record levels in the same period. There is not a dramatic change in the consumption of LPG. In this study, all regions have been analyzed. The result of Istanbul region is given as the most fuel consumption in this region. In Figure 3., the gasoline consumption of Istanbul in 2006 is 738,894 tonnes; however, the result shows that the gasoline consumption of Istanbul will be 399,807 tonnes in Both diesel and LPG consumption reached record levels in the same period in Turkey. The diesel consumption of Istanbul in 2006 is 1,602,673 tonnes. It will reach 2,655,771 tonnes in In the same manner LPG consumption in 2006 is 233,471 tonnes and it will reach 287,572 tonnes in The gasoline consumption in 2006 is 17% larger than its counterpart in The difference in diesel consumption between the years 2006 and 2012 is significant, about 17%. As a result, Turkey s dependence on these fuels could reach to precarious levels in the next decade Turkey Gasoline Diesel LPG 14,689,966 tonnes 3,024,773 tonnes 1,638,667 tonnes Figure 2. Road Transportation Fuel Demand [7]

4 Millions of Vehicles It can be seen from the Figure 3., 5.7 million number of vehicles used gasoline, 7.5 million number of vehicle used diesel, and 3.6 million number of vehicle used liquefied petroleum gas (LPG) in Turkey ,549,806 5,722,940 3,649,739 Figure 3. The Number of Vehicles According to Fuel Type [8] Gasoline Diesel Lpg 1.2. Dynamics of the Vehicle Stock for Road Transportation In Figure 4., the percentage of vehicle number in Turkey and Istanbul are displayed for the years 2006 and The number of vehicles which use gasoline in 2006 is 23% larger than the number of vehicles in The differences in diesel and LPG between the years 2006 and 2012 are significant, about 11% and 14%, respectively. 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 8% 22% 34% 45% 57% 34% % 90% 6% 10% 80% 70% 27% LPG 60% 53% 50% diesel 40% gasoline 30% 67% 20% 37% 10% 0% LPG diesel gasoline Figure 4. The Percentage Distribution of Vehicles According to Fuel Type in Turkey and Istanbul [8] In Figure 5. the percentage of car numbers according to fuel type in Turkey and Istanbul are displayed for the years 2004 and The differences in gasoline, diesel and LPG between the years 2004 and 2012 are 30%, 13% and 17% respectively. The number of cars which used gasoline in 2004 is 30% larger than the number of cars in However, both diesel and LPG consumption reached record levels in the same period in Turkey and Istanbul.

5 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 21% 38% 14% 27% 65% 35% LPG Diesel Gasoline 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 8% 15% 7% 41% 85% 45% LPG Diesel Gasoline Figure 5. The Percentage Distribution of Cars According to Fuel Type in Turkey and Istanbul [8] 2. METHODS Various time series and econometric models have been developed for the short-term forecasting of regional gasoline, diesel and, liquefied petroleum gas (LPG) demand in road transportation in Turkey. After many trial and error procedures, 36 econometric models have been developed revealing the effect of different socio-economic and technological factors on annual fuel demand in each region. This enabled the analysis of the whole road transport sector. The employed validity and diagnostic tests included estimation of the coefficient of determination, colinearity test of independent variables, statistical test of the F-statistics, statistical test of the standard error, model function form test, first degree self-correlation test to residuals, residuals correlation tests, heteroscedasticity and normality tests, model stability test and forecasting ability of the models through Theil s U statistic. In addition to the forecasting of fuel demand, several independent variables of the regression models were projected including population, gross value added of the transport sector, fuel prices, vehicle stock and utilization rates etc. All projections have been calculated with different methods so as to make use of the most reliable forecasts. Moving Average, for example, turned out to be the best method for estimating the prices of gasoline and diesel whereas Multiple Regression Analysis was used for LPG prices. TUIK (Turkish Statistical Institute) is the responsible authority to collect data from institutions and publish them as the official statistics for Turkey. Although, the transport sector plays an important role on the economic and social growth of Turkey, limited statistics exist regarding this sector. Based on all collected data, a series of assumptions are made to provide the relevant data required by the model. Multi Linear Regression analysis in this study is analyzed with focus on vehicle fuel usage indicator, national gross domestic product, fuel prices, crude gasoline imports, consumer price index, population, and are responsible for nearly all of the national fuel demand. The final model includes the vehicle fuel usage indicator, national gross domestic product, population, and fuel prices variables. These independent variables based on 11-year historical data between years 2001 and 2012 refined from scattered data sources. Transport sector energy modeling studies heavily depend on social and economic factors such as, registered vehicles, selling price of fuel, national gross domestic product, consumer price index, population. Since all of these parameters require much effort to be analyzed in detail, incorporating these factors into models are difficult.

6 2.1. Population The population is an independent variable for the Multiple Regression Analysis. The model needs population as an input. For that reason, this chapter covers the calculation of population by region. The data belongs to past years and is not available in terms of the population by region. The region-based population data is only available for the year 2000, and between the years 2007 and The region-based population data has not been recorded between the years 2000 and For this reason, we estimated the population as the average of the difference between the years 2000 and 2007 as the increment of the each year. There is an important point that needs to be stressed which is, that there is a negligible increase or decrease for each and every year. The models are based on population data between the years 2007 and 2013 which have been provided by TUIK, and are used to project the trends in future transport fuel consumption for year the Population plays a vital role on the evolution and development of fuel demand Gross Value Added Gross Value Added (GVA) is the key economic metric used to measure all economic activity within a geographical area over a given period. Understanding GVA and its calculation is essential when trying to measure economic impact. Since GVA data have not been recorded in specific years, this is a main problem for data collection in this study. The region-based GVA data are only available from the year 2004 to Since unavailable data of GVA for the years 2002 and 2003, the missing data in these years were calculated by regression analysis. However, region-based GVA are available between the years 2002 and based GVA data are need after the year For that reason, region-based GVA is calculated according to the GDP data. In this study, real values of region-based GVA are calculated for each region with the help of the consumer price index forecast value for GVA is calculated with the help of Moving Average method. The objective is to use past to data to develop a forecasting model for future periods. Gross Domestic Product data indicates a slow improvement within consequtive years. Therefore, taking average for small amount of periods is expected to give better results for forecasting. Two different methods have been compared. Results of MA=2 & MA=3 are given in Table 1. MA=2 gives the best result for all regions when we look at the MSE, MAD, and MAPE values forecast is available as shown in figures for all regions. The region of Istanbul is selected as an example to show the result of Moving Average method to present 2013 forecast value for GVA. The trend of GVA between the years 2000 and 2013 for Istanbul region can be seen from Figure 6. Table 1. Forecasts and Accuracy Results of Istanbul for MA=2 and MA=3 Forecast 2013 MSE MAD MAPE MA=2 177,921,777 TL 182,197,788,884,410 11,606, % MA=3 171,083,829 TL 282,149,287,971,249 15,117, %

7 Selling Price (Turkish Lira) GVA (millions) 200 TL 180 TL 160 TL 140 TL 120 TL 100 TL 80 TL 60 TL 40 TL 20 TL 0 TL 177,921,777 TL Istanbul/MA=2 Actual Istanbul/MA=2 Fitted Istanbul/MA=2 Forecast Figure 6. MA=2 Results of GVA for Istanbul 2.3. Fuel Prices For the fuel prices there is a small improvement within consecutive years. Therefore taking the average of small periods is expected to give more reliable results as a consequence. Here again the two different periods are compared. Results of MA=2 & MA=3 is given in Table 2. for all fuel types. In addition to that, the selling price of LPG has an upward trend. For that reason regression analysis is considered to be an alternative method for MA=2. MA=2 gives the best result when MSE, MAD, and MAPE values are considered for gasoline and diesel fuel types. The results of the Moving Average method are shown below, which are built up by relying on the selling price data provided by EMRA (Energy Market Regulatory Authority). In addition to that, trends of selling price for all fuel types are displayed in Figure 7., Figure 8. and Figure 9. Table 2. Forecasts and Accuracy Results of Istanbul for Gasoline, Diesel and LPG Selling Prices Forecast 2013 MSE MAD MAPE Gasoline MA = TL % MA = TL % Diesel MA = TL % MA = TL % LPG Regression TL % MA = TL % TL2.40 TL2.20 TL2.00 TL1.80 TL TL Istanbul(MA=2) Actual Istanbul(MA=2) Fitted Istanbul(MA=2) Forecast Figure 7. MA=2 Result of Gasoline Selling Price for Istanbul

8 Selling Price (Turkish Lira) Selling Price (Turkish Lira) 2.00 TL 1.80 TL 1.60 TL 1.40 TL 1.20 TL 1.00 TL TL Istanbul(MA=2) Actual Istanbul(MA=2) Fitted Istanbul(MA=2) Forecast Figure 8. MA=2 Result of Diesel Selling Price for Istanbul It can be seen from the Figure 9. the selling price of LPG has an upward trend. For that reason Regression Analysis is considered to be an alternative method for MA=2. When Regression Analysis and Moving Average=2 is compared for the LPG selling price, Regression Analysis gives the best result for all regions TL 1.25 TL 1.00 TL 0.75 TL 0.50 TL 0.25 TL 0.00 TL TL Istanbul(Regression) Actual Istanbul(Regression) Fitted Istanbul(Regression) Forecast Figure 9. Regression Analysis of LPG Selling Price for Istanbul 2.4. Vehicle Fuel Usage Indicator The Vehicle Fuel Usage Indicator is an independent variable for the multiple regression analysis. During the creation of the Vehicle Fuel Usage Indicator; - Population, - GVA, - The Vehicle Fuel Usage Indicator which belongs to past years independent variables have been tested. In general, the rate of GVA and Population, and the Vehicle Fuel Usage Indicator from the previous years have been effective in these Multiple Regression Model. However, the rate of GVA and Population is not an independent variable for some regions. For example, the regression analysis for West Marmara, East Black Sea, and Northeast Anatolia shows that the rate of GVA and Population do not affect the Vehicle Fuel Usage Indicator. Vehicle fuel usage indicator defines the number of road motor vehicles by kind of fuel used. Since vehicle fuel consumption and activity rates are different, coefficients have been identified for each vehicle. The information about coefficients does not come from any sources. It is an expert guess in this study. Vehicle types are multiplied by these coefficients. They can be seen from the Table 3.

9 Table 3. Equivalent Coefficients for the Types of Vehicles Coefficient Car 1 Minibus 4.5 Bus 16.7 Van 3.3 Truck 17.5 Motorcycle 0.15 Special Purpose Vehicle 0.8 Tractor 1.6 for each region; Vehicle fuel usage indicator for Istanbul region in a specific year can be calculated as follows and it is calculated VFUI (it) = ( number of gasoline-powered cars (t) * ( 1000/P Istanbul(t) ) ) * 1 + ( number of gasoline-powered minibuses (t) * ( 1000/P Istanbul(t) ) ) * ( number of gasoline-powered buses (t) * ( 1000/P Istanbul(t) ) ) * ( number of gasoline-powered vans (t) * ( 1000/P Istanbul(t) ) ) * ( number of gasoline-powered trucks (t) * ( 1000/P Istanbul(t) ) )* 17.5 (1) + ( number of gasoline-powered motorcycles (t) *( 1000/P Istanbul(t) ) )* (number of gasoline-powered special purpose vehicles (t) * ( 1000/P Istanbul(t) ) )* ( number of gasoline-powered tractors (t) * ( 1000/P Istanbul(t) ) )* 1.6 VFUI (it) is the vehicle fuel usage indicator of fuel type i in time t, P (t) is the population in time t. The data belongs to past years for which the number of tractors is not available. For this reason, like the estimation of population, the number of tractors between the years 2000 and 2004 is estimated. For example, since the number of tractors in 2004 is known Turkey-wide, the number of regional tractors in 2004 is estimated according to the number of regional tractors in For the Vehicle Fuel Usage Indicator the best result can be derived by regression analysis. In this regression model there are various explanatory variables are tried, but the most reliable conclusion could be reached by using the rate of GVA and population, vehicle fuel usage indicator from the previous years.

10 2.5. Multiple Regression Analysis The general formula for the multiple regression models for the Istanbul is shown in Equation 1; FD (it) = b 0 + b 1 * GVA (t) / Population (t) + b 2 * FD (i,t-1) +b 3 * VFUI (it) + b 4* SP (it) (2) where, i is the fuel type index, t is the year index, FD (it) is the fuel demand of fuel type i in t, GVA (t) is the gross value added in t, Population (t) is the population in t, SP (it) is the selling price of fuel type i in t, VFUI (it) is the vehicle fuel usage indicator of fuel type i in t, b 0 is the fixed sales value, b 1 is the increment of the rate of the GVA and Population which causes the fuel demand, b 2 is the increment of the fuel demand from previous year which causes the fuel demand, b 3 is the increment of vehicle fuel usage indicator which causes the fuel demand, b 4 is the increment of the selling price which causes the fuel demand, Regarding to the developed model here is the corresponding fitted values and accuracy of model: Fuel Demand 2013 (Istanbul) = -1,051, ,648.76* GVA 2013 / Population * Fuel Demand ,717.83* Selling Price ,743.36* Vehicle Fuel Usage Indicator 2013 (Gasoline) (3) When we take the calculated value of GVA/Population, Selling Price, and Vehicle Fuel Usage Indicator for the year 2013 into consideration, the formulation will be as follows; Fuel Demand 2013 (Istanbul) = -1,051, ,648.76* 177,921,777 /14,107, * 504, ,717.83* ,743.36* Fuel Demand (Istanbul) = 399,807 tonnes (4) As expected; the rate of GVA and population, the fuel demand from previous year and vehicle fuel usage indicator has a positive effect on the fuel demand. However, Selling Price has a negative effect on the fuel demand. Figure 10. shows the overlap performance of fitted values vs. actual data. In addition to that forecasted value of the year 2013 is presented on the graph. Corresponding accuracy values are shown on the table.

11 Sales(Gasoline) x1000 tonne Istanbul Actual Fitted Forecast ,807 tonnes Gasoline Demand (Istanbul) = -1,051, ,648.76* GVA/Population * The Gasoline Demand from previous year - 545,717.83* Selling Price + 7,743.36* Vehicle Fuel Usage Indicator (Gasoline) Forecast MSE MAD MAPE 399, E % Figure 10. Regression Results of the Amount of Yearly Gasoline Sales for Istanbul 3. RESULTS AND DISCUSSION In this study mostly the aim is expected to build a model regarding to fuel type and region changes. As fuel demand is highly sensitive to political and economic fluctuations and new records could highly affect the fuel demand. In this study six forecasting methods namely, Moving Average= 2, Moving Average= 3, Double Moving Average= 2, Regression Analysis, Simple Exponential Smoothing, Double Exponential Smoothing have been considered to present fuel demand for the year The growth of fuel consumption creates trending data, therefore the most reliable forecast methods would be the ones which takes trend into account. After many trial and error procedures, 36 econometric models have been developed revealing the effect of different socio-economic and technological factors on annual fuel demand in each region. The validity of each model is tested by means of statistical and diagnostic tests. Results provide interesting insight into the underlying drivers of fuel demand in different regions. The validity of the models has been checked and results have been compared based on mean absolute deviation as well as percentage and squared error values yielding the most accurate set of forecasting models. Results of twelve regions have been analyzed. As a result the Istanbul region is listed as the highest fuel consuming region in Turkey. Regression analysis gives the best results for the gasoline and LPG demand forecast for the Istanbul. However, MA=3 gives the optimum result for diesel demand forecast. It can be said that there is not any upward or downward trend in past data for diesel sales. Therefore, taking an average of small amount of periods is expected to give better results for forecasting. For that reason, the Moving Average method gives the most reliable result for the calculation of diesel demand forecast. The gasoline consumption of Istanbul in 2006 was tonnes; however, the results shows that the gasoline consumption of Istanbul will be tons in The diesel consumption of Istanbul in 2006 was tons. It will reach to tonnes in In the same manner LPG consumption in 2006 was tonnes and it will reach to tons in Gasoline demand decreased linearly between the years 2006 and 2012 in Turkey. However, both diesel and LPG consumption reached to record levels in the same period of time. As a result, Turkey s dependence on these fuels could reach precarious levels in the next decade. All methods are investigated in details and each model created a result consideration of validity. In general, it is very clear that Multiple Regression Analysis gives the best solution for explaining the fuel demand. The growth of fuel

12 demand creates a trending data; therefore the most reliable forecast methods would be the ones which take trend into account. So this is not a surprise that Regression and DES performs better than the others. To sum up, the forecast methods for 2013 are as follows: Multiple regression analysis is the best method to forecast gasoline demand in Istanbul, West Marmara, Aegean, East Marmara, West Anatolia, Central Anatolia, West Black Sea, North East Anatolia, and East Black Sea. Simple Exponential Smoothing gives the best result for South East Anatolia and Mediterranean. Gasoline demand in Central East Anatolia can be calculated with Moving Average=3 method. Multiple regression analysis is the best method to forecast diesel demand in North East Anatolia, Central East Anatolia, South East Anatolia, West Marmara, Aegean, West Anatolia, Mediterranean, Central Anatolia, West Black Sea, and East Black Sea. Moving Average=3 and Double Moving Average=2 is the best method for Istanbul and East Marmara to calculate the diesel demand for the year 2013, respectively. To forecast LPG demand for the year 2013 for North East Anatolia, Central East Anatolia, South East Anatolia, Istanbul, West Marmara, East Marmara, Central Anatolia Multiple Regression Analysis will gives the best forecasted value. For the other seven regions namely, East Black Sea, West Black Sea, Mediterranean, West Anatolia, Aegean Double Exponential Smoothing gives the best result for the year CONCLUSION Road Transport takes an important role in fuel sales. In the past few decades, population growth and economic development have caused a growing demand for fuel in transportation sector in Turkey. For that reason, sales of gasoline, diesel and, liquefied petroleum gas (LPG) are modeled as region-based in terms of significant time intervals (quarterly and yearly). Gasoline, diesel and LPG sales are affected by different socio-economic and technological factors such as population, the vehicle fuel usage indicator, gross domestic product (GDP), and fuel prices. Various time series and econometric models have been developed for the short-term forecasting of regional gasoline, diesel and, liquefied petroleum gas (LPG) demand in road transportation in Turkey. After many trial and error procedures, 36 econometric models have been developed revealing the effect of different socio-economic and technological factors on annual fuel demand in each region. This enabled the analysis of the whole road transport sector. The employed validity and diagnostic tests included estimation of the coefficient of determination, co linearity test of independent variables, statistical test of the F-statistics, statistical test of the standard error, model function form test, first degree self-correlation test to residuals, residuals correlation tests, heteroscedasticity and normality tests, model stability test and forecasting ability of the models through Theil s U statistic. Results provide interesting insight into the underlying drivers of fuel demand in different regions. In addition to the annual econometric models based on linear regression, various quarterly time series models were developed including several autoregressive integrated moving average (ARIMA) models with different exponential smoothing methods. The validity of the models has been checked and results have been compared based on mean absolute deviation as well as

13 percentage and squared error values yielding the most accurate set of forecasting models. It is found that econometric models generally outperformed the time series ones and provided effective alternatives for forecasting annual fuel demand. Model projections indicate in all regions a shift away from gasoline-fueled vehicles towards diesel-fueled ones. Table 4. shows changes in the gasoline demand with the best performing method between 2012 and The table indicates that multiple regression analysis is the best performing method for Istanbul region. A closer look at the data reveals that there is dramatic decrease in the gasoline demand in Istanbul. Table 4. Best Performing Methods and Projections for Gasoline Demand The Best Performing Method Gasoline(tonnes) 2012 Actual 2012 Forecasted 2013 Actual 2013 Forecasted 2014 Forecasted Regression Istanbul 504, , , , ,037 Regression Regression Regression Regression SES Regression Regression Regression West Marmara Aegean East Marmara West Anatolia Mediterranean Central Anatolia West Black Sea East Black Sea 115, , , , , , , , , , , , , , , , , , , , , , , , ,489 59,624 62,952 58,795 56,785 56,194 76,509 80,032 75,795 77,712 75,300 38,072 39,221 38,795 37,711 36,832 Regression Northeast Anatolia MA=3 SES Central East Anatolia South East Anatolia 21,851 25,946 21,479 20,932 20,086 31,824 31,696 31,305 31,901 31,671 72,880 72,879 75,564 81,607 82,479 In addition to the forecasting of fuel demand, several independent variables of the regression models were projected including population, gross value added of the transport sector, fuel prices, vehicle stock and utilization rates etc. All projections have been calculated with different methods so as to make use of the most reliable forecasts. Moving average, for example, turned out to be the best method for estimating the prices of gasoline and diesel whereas multiple regression analysis was used for LPG prices.

14 In developing/ recently developed countries, economic growth has been followed by increase in road transportation fuel consumption. For this reason, it was expected that gasoline consumption would increase. However, gasoline consumption has been constantly decreasing in Turkey which is a developing/ recently developed country. The main reason for this reduction is gasoline s steep price. Turkey has the highest gasoline price among all the OECD member states due to the high taxes that are reflected at the level of retail price. As gasoline prices increase dramatically, consumers are shifting to diesel and LPG in Turkey [9]. Since the late 1990s, the European diesel car market boomed whereas diesel vehicles were phased out of the Japanese market and remained at a negligibly small level in the United States while gaining popularity recently. Registrations for diesel cars and sport utility vehicles rose 24 percent in the United States from 2010 through 2012 [3]. The main reason for the attractiveness of diesel cars is fuel efficiency, as diesel engines are 20 percent to 40 percent more fuel efficient than equivalent gasoline engines. However, diesel fuel contains about 15% more carbon per litre reducing the CO 2 emission advantage by favourable fuel efficiency. It is expected that increasingly stringent emissions regulations and the high cost of new anti-pollution technology will make diesel engines much more expensive [4]. Criticism of diesel vehicles has recently increased and expectations changed such that diesel automobiles will see a downward trend (e.g. [5], [6], [7]). Results of this study indicate that this will not be the case in Turkey in the short term unless there is a new environmental tax policy or standard to discourage the use of diesel fuel. According to the International Energy Agency, tax as a percentage of total price for gasoline and diesel was %60 and %49, respectively. The sustainability of diesel s tax advantage in some European countries including Turkey, however, is questionable. Around 50% of Turkish total oil demand was consumed in the transport sector. This means that crude oil import is very much dependent on the transport sector. Turkey fulfills most of its crude oil need by import. The crude oil is processed at the refineries in Turkey. When 100 tons of crude oil are processed at the refinery, 48 tons gasoline, 18 tons diesel, and 11 tons LPG are obtained. However, diesel and LPG are in more demand than gasoline. For this reason, Turkey is increasingly dependent on diesel and LPG imports. Instead of processing crude oil to obtain diesel and LPG, importing them could be much more profitable. That s why oil-rich countries should take into the consideration the export of diesel and LPG. It can be concluded that the decrease in gasoline consumption leads to an increase in import of diesel and LPG. Since Turkey is a developing/ recently developed country, it is expected that the fuel demand grows fast in the short and the long term. The results show that in each region fuel demand increased dramatically. To overcome this demand increase, the studied scenarios, can be implemented. In conclusion, all models need an improvement and update of the data. The model data in this study can be used as basis for the road transport sector, but as time passes, the data for the future years should be updated. As a result, more precise and realistic analysis can be conducted. The collection of relevant and accurate data is the main part in this study.

15 REFERENCES [1] Li, Z., Rose. J.M., Hensher, D.A., Forecasting automobile petrol demand in Australia: An evaluation of empirical models, Transportation Research, Part A 44, pp.16 38, [2] Limanond, T., Jomnonkwao, S., Srikaew, A., Projection of future transport energy demand of Thailand, Energy Policy, 39, pp , [3] Hilton Holloway, Diesel dominance threatened by EU emissions rules, 2013, October [4] Zachary Shahan, Diesel Cars Finally Given The Axe In Europe?, 2013, October [5] Cames and Helmers, Critical evaluation of the European diesel car boom - global comparison, environmental effects and various national strategies, 2013, [6] Charlie Dunmore, Wood fires and diesel cars pose pollution threat: EU watchdog, 2013, idindee99e03s , October [7] Energy Market Regulatory Authority, Road Transportation Fuel Demand 2012, Ankara, [8] Turkish Statistical Institute, Road Motor Vehicle Statistics 2012, Ankara, [9] Melikoglu, M., Demand forecast for road transportation fuels including gasoline, diesel, LPG, bioethanol and biodiesel for Turkey between 2013 and 2023, Renewable Energy, pp , 2013.