Académique Documents
Professionnel Documents
Culture Documents
abstract
In recent years, Thailand Government has made ambitious targets for biofuels and the demand
for biofuels has increased. This study estimates the long-term factors that affect the supply and
demand for biofuels (bioethanol and biodiesel) in Thailand by developing a theoretical model
based on simultaneous equations, using Two stage least square and Three stage least square
approaches. Quarterly data were collected for the years 2007 e 2013:2. Amongst many factors,
we found three major factors affecting the long-term demand for biofuels: the price of biofuels,
the real gross domestic product and the number of vehicles. On the other hand, the long-run
supply of biodiesel and bioethanol was influenced by the gross domestic product, the cost of
substitution using biofuel (i.e., price of gasoline 95 compared to blended gasoline) and the price
of biodiesel. These findings have important policy implications to the government plans for
biofuel development, suggesting that maintaining the price difference between gasoline and
gasohol (blended gasoline) and a greater penetration of biofuel vehicles in fleet would be
important policies for biofuel consumption. On biofuel supply side, results indicate that policies
to compensate the biofuel producers for prices and reducing cost of bio-fuel production are
important.
Keywords:
Biofuels demand
Biofuels supply
Two stage least square (2SLS)
Three stage least square (3SLS)
Introduction
Energy crises have been observed in many countries over the past decade due to higher
demand and prices of energy. Global rise in crude oil prices and environmental concerns drove
the Thai government to institute the national development plan in the past. The Renewable and
Alternative Energy Development Plan of Thailand (AEDP) calls for 25% of the country's energy
to be produced from renewable resources by 2021. The plan, entitled as AEDP 2012e2021, is
aimed of using alternative energies such as solar and wind energy, aswell as using biodiesel and
bioethanol [1]. One of the key aspects of AEDP is to replace imported oil and increase the
energy security of the country [2]. According to the plan, Thailand will replace imported oil by
44% with biofuel and encourage ethanol and biodiesel (B100) consumption to reach nine and six
million liters per day respectively by 2021. Governmental policies promoting biofuels during the
years 2005e2010, resulted in a more than 10-fold increased production of bioethanol and
biodiesel. The country's share of biofuel production in the Asia- Pacific region increased from
6% in 2005 to19% in 2009 [3]. Nevertheless, the government's target for biofuel consumption
has not been achieved despite the presence of subsidy and taxation policies to raise biofuels
demand [4].
Energy demand management is becoming an important issues since the future world is
dependent on today's decision. Managing the energy resources in an optimal manner has become
imperative among energy planner and policy makers [5]. However, lack of sufficient and reliable
historical data of biofuel especially in developing countries is one of the main challenges in
developing accurate forecasting models [5]. In the context of biofuel management, it is necessary
to understand the major factors (e.g. socioeconomic, demographic factors) that affect biofuel
production and consumption and how they affect the biofuel demand and
supply.
Suganthi and Samuel [4] reviewed energy forecasting models and divided them into
twelve categories namely, time series models, regression models, econometric models,
decomposition models, unit root test and integration models, ARIMA models, expert systems
and ANN models, grey models, input-output models, genetic algorithm/fuzzy logic/neuro fuzzy,
integrated model and bottom up models. Jebaraj and Iniyan [5] also reviewed models for
forecasting energy quantities into six categories of energy planning models, energy supply-
demand models, forecasting models, Optimization models, energy models based on neural
networks and emission reduction models. Among these models, regression analysis was widely
used to forecast biomass and bioenergy consumption because of mature theory and simply
algorithms [6]. However, limitations of regression analysis is that it requires large sets of data for
forecasting [7]. The Artificial neural network (ANNs) also widely chosen as the energy
forecasting tool such as industrial sector [8], residential sector [9] and transport sector [10].
ANNs can forecast the future values of many variables simultaneously and nonlinear relation in
the data structure [8] but cannot find out the coefficient of each independent variables [11].
Simultaneous equations, especially two stage least square (2SLS) and three stage least
square (3SLS), have been developed as a less complex models that require minimal data and are
effective in long term forecasting. The application of 2SLS and 3SLS solves the problem of
heteroscedasticity, multicollinearity [12], autocorrelation [13] and endogeneity of price and
quantity of goods [14]. The results form 2SLS and 3SLS are more accurate than time series [15]
and ordinary least square (OLS) [12]. Simultaneous equations are two-way relationship between
y and (some of) the x's. This relationship makes the distinction between the dependent and
explanatory variables of dubious value [16]. Generally, simultaneous equations are composed of
2 x's and y equations. Each simultaneous equation is used for expressing the mutually or jointly,
dependent or endogenous variables. Unlike the single equation model, simultaneous equation
models take into account information provided by other equations in the system [17]. Two stage
least square is a method of extending regression to cover models which violate ordinary least
square (OLS) assumption of exogeneity so that the disturbance term of the dependent variable is
correlated with the cause(s) of the independent variable(s) [18]. Three stage least square is a
combination of multivariate regression (SUR estimation) and two stage least square. It obtains
estimated instrumental variable, taking into account the covariances across equation disturbances
as well [19]. The objective function for three stage least square is the sum of squared
transformed fitted residuals [20]. The system equations estimated by 2SLS and 3SLS, are more
efficient to capture the interrelation between equations and causal and feedback effect between
the core variables of the system [17].
Although a number of studies have been conducted to estimate biofuels demand and
supply (i.e. demand and supply for ethanol in U.S. [14,21], ethanol demand in Brazil [22,23],
energy demand in EU [24,25], energy demand in Turkey [26] and African countries [27,28]),
only a few studies have specifically examined biofuel demand in Thailand e.g. Limanong et al.
[6]. Previous studies employed different methods for forecasting the supply and demand of bio
fuels such as regression, panel data, trend analysis, time series analysis, simultaneous equation
(OLS, 2SLS and 3SLS). From literature reviews, only few studies compared demand and supply
models by using simultaneous equation such as OLS compared with 2SLS [29] and OLS
compared with 3SLS [30].
The literature reviews found that the key factors which affect demand of biofuels are
GDP [23], number of vehicle [21], biofuel price [21,29], and previous year biofuel price [23].
None of past literature attempted to identify five groups of factors (i.e. socioeconomic factors,
cost and substitution factors, lagged value of the dependent variable, information factors and
environment factors). In fact, Limanong et al. [6] developed the demand for gasohol (91 and 95)
in Thailand (focus on Bangkok and Nakhon Ratchasima) by using quarterly data from 2005 to
2008 and analyzed by OLS. In terms of factors, this study is using three groups of factors such as
socio-economic factors (i.e. GDP, population, number of vehicle and number of gas station), cost
and substitution factors (i.e. gasoline price) and information factors (i.e. government subsidizes).
Therefore, to fill this important knowledge gap in the literature, this study has two main
key objectives. First, this study attempts to identify, quantify and select the suitable factors
affecting biodiesel and bioethanol in Thailand using quarterly data from 2007 to 2013:2 which
are the most update data. Secondly, this study also develops the long run supply and demand
alternative models (i.e. 2SLS and 3SLS) for biodiesel and bioethanol in Thailand with four group
of factors (i.e. socio-economic factors, cost and substitution factors, lagged value of the
dependent variable and information factors) and selects the most accurate one to forecast biofuel.
This study used 2SLS and 3SLS for more accuracy compared to other models and it is ideal to
the model under limitation of data. Finally, we compared the best fit model with ANNs to show
the robustness of the results. This study aims to reflect on those factors which come up with
useful policy messages for biofuel development in Thailand.
This paper is organized into five sections. Section 2 provides concise literature review on
demand and supply of biofuels. The methodology used in this analysis including data collection
and model analysis is presented in Section 3. The empirical findings from these two modelling
approaches are presented in Section 4. Finally, Section 5 presents discussions, conclusions and
policy implications.
2. Literature review
The most important influencing factors identified in previous studies that affect biofuel
demand and supply as summarized in Table 1. According to Limanong et al. [6], and 26
literature review, these factors can be categorized into five groups such as socioeconomic
factors, cost and substitution factors, lagged value of the dependent variable, information factors
and environment factors as described below.
Socio-economic factors consist of population, gross domestic product (GDP),
income/income per capita, technology, number of vehicles and number of station. Among these
factors GDP, income and number of vehicles are the most widely used and significant factors in
biofuel forecasting, followed by population and number of stations, respectively.
Cost and substitution factors include the cost of biofuel and substitution of biofuel
products that affect the decision choices of what modes or which type of biofuel to be used.
These factors comprise of price of biofuel, gasoline, diesel, crude oil and natural gas, as well as
feedstock price and substitution feedstock price. Biofuel price and gasoline/diesel price are
significant factors in biofuel forecasting.
The lagged variable is a crucial factor to examine the long-term effect on various aspects
of travel demand or biofuel consumption with the consideration that the external factor is
dynamic, and the effect could gradually change. Biofuel price in previous year is also significant
factor in biofuel forecasting.
The fourth group includes other important information that is unique to issue of interest
such as government campaign and seasonal dummy variable. Environment factors is the last
group which is usually found in biofuel demand and supply forecasting. The main factors are
CO2 emission and CO2 emission per capita.
We considered variables commonly used in previous studies to forecast biofuel demand
and supply. For demand function, four groups of factors considered are socio-economic factors
(i.e. real GDP, employment, number of vehicles and number of biofuel stations), cost and
substitution factors (i.e. price of oil product and price of biofuel), lagged of the dependent
variable (i.e. price of feedstock in previous year) and information factors (i.e. campaign for
compensation),. Whereas environment factors were not included in the model because of the
insufficient amount of data.
3. Methodology
In order to analyze the factors affecting long-run biofuel demand and supply, a set of
methodologies were implemented that had been widely used in time-series forecasting.
Simultaneous equations were used and analyzed by 2SLS and 3SLS methods and results were
compared with Artificial Neural Networks (ANNs).
Where 𝑃 = 𝑋(𝑋 ′ 𝑋)−1 𝑋 ′ is the projection matrix onto the linear space spanned by the exogenous
regressors X. The estimated Rsquare from the reduced form of equations gives some idea of the
adequacy of the instrument.
3.1.2. Three stage least square
Three stage least square method combines two-stage least squares (2SLS) with seemingly
unrelated regressions (SUR) [16,41]. SUR is a generalization of a linear regression model that
consists of several regression equations. Each equation has its own dependent variable and
potentially different sets of exogenous explanatory variables. Each equation is a valid linear
regression on its own and can be estimated separately. The model can be estimated equation by-
equation using standard ordinary least squares (OLS). Such estimates are consistent, however
generally not as efficient as the SUR method, to feasible generalized least squares with a specific
form of the variance-covariance matrix. Two important cases when SUR is in fact equivalent to
OLS, are: either when the error terms are in fact not correlated between the equations (so that
they are truly unrelated), or when each equation contains exactly the same set of regressors on
the right-hand-side [16].
where Qd bioethanol is the demand quantity for bioethanol for all sectors in Thailand during
quarter i and year t. It is measured in millions of liters. The variable, P gasoline is the gasoline 95
retail price in the Bangkok market (Baht per liter), P gasohol is the gasohol 95 retail price in
Bangkok market (Baht per liter). The variable, ve is number of gasoline engine vehicle in
Thailand, sta is the number of gasohol station in Thailand, and real GDP is real gross domestic
product of Thailand (millions of Baht) (Eq. (7)). P cassava is the price of cassava, (Baht/kg). P
sugar cane is price of sugar cane (Baht/kg), and GDP per capita is real gross domestic product
per capita of Thailand (Baht). These are used as exogenous variables in some equations. The
signs for 𝛽2, 𝛽3, 𝛽4 and 𝛽5 were expected to be positive, while the sign for 𝛽1 was expected to be
negative.
Supply : Qs bioethanol = f(P gasoline 95; P gasohol 95; labor employment; real GDP;
compensation)
(8)
The exogenous variables in these equations were the price of gasoline 95 (P gasoline 95),
number of persons employed (# labor employment), real GDP and compensation. The price of
gasohol 95 (P gasohol 95) and supply quantity of bioethanol (Qs bioethanol) were always treated
as endogenous (Eq. (8) and (9)). Meanwhile, price of gasohol were first treated as exogenous to
the model and then as endogenous. If the price of gasohol 95 was treated as endogenous,
exogenous variables in the system of equation (9) would be the price of gasohol 95 in the
previous year, the price of cassava, and the price of sugar cane. Thus, the bioethanol supply
function in double-log form for two stage least square (2SLS) and three stage least square
(3SLS) can be expressed as:
ln 𝑄 𝑠 𝑏𝑖𝑜𝑒𝑡ℎ𝑎𝑛𝑜𝑙𝑖𝑡
= 𝛽0
+ 𝛽1 ln 𝑃 𝑔𝑎𝑠𝑜𝑙𝑖𝑛𝑒 95𝑖𝑡
+ 𝛽2 ln 𝑃 𝑔𝑎𝑠𝑜ℎ𝑜𝑙 95𝑖𝑡 + 𝛽3 ln 𝑙𝑎𝑏𝑜𝑟𝑖𝑡 + 𝛽4 ln 𝑟𝑒𝑎𝑙 𝐺𝐷𝑃𝑖𝑡
+ 𝛽5 ln 𝑐𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛𝑖𝑡 + 𝑒
(10)
𝛽2 , 𝛽3 dan 𝛽4
where Qs bioethanol is the supply quantity of bioethanol for all sectors in Thailand during quarter
i and year t (millions of liters), P gasoline 95 is the retail price of gasoline 95 in the Bangkok
market (Baht per liter), P gasohol 95 is the retail price of gasohol 95 in the Bangkok market
(Baht per liter), labor is all of labor employment in Thailand (thousands of persons), real GDP is
real gross domestic product in Thailand (millions of Baht), and compensation is the value which
the government subsidizes the product (Baht per liter). Moreover, P cassava is the price of
cassava (Baht per kilogram), and P sugar cane is price of sugar cane (Baht per kilogram). The
signs for 𝛽1 , 𝛽3 , 𝛽4 𝑑𝑎𝑛 𝛽5 were expected to be positive, while the sign for 𝛽2 was expected to be
negative.
The exogenous variables in these equations were the number of vehicles (# vehicles),
number of biodiesel stations (# biodiesel station) and real GDP. The price of biodiesel (P
biodiesel) and demand quantity of biodiesel (Qd biodiesel) were always treated as endogenous
variables (Eq. (11) and (12)). The price of biodiesel was treated first as an exogenous variable in
the model (11) and then as an endogenous variable as shown in (Eq. (12)). If price of biodiesel
was treated as an endogenous variable, the price of biodiesel in previous four quarters was used
as an exogenous variable in the system as shown in (Eq. (12)). The biodiesel demand function in
double-log form of two stage least square (2SLS) and three stages least square (3SLS) can thus
be written as:
This analysis starts with the unit root tests, bioethanol demand function, bioethanol
supply function, biodiesel demand function, and biodiesel supply function.
Table 2
Summary of statistical analysis of biofuel fuel markets in Thailand (Quarterly 2006e2013:2)
4.4. Modelling the long-run biodiesel demand
The long-run estimation of biodiesel demand showed relationship between biodiesel
consumption and determining factors including the price of biodiesel, number of vehicles,
number of biodiesel stations and real GDP. Table 6 summarizes the parameters in the long-run
estimation. Two models were developed, C-1 and C-2. C-1 used a 3SLS approach, while C-2
used a 2SLS approach.
The 3SLS equation obtained from Model C-1 had an Adj. R2 = 0.872 (R2 = 0.896), S.E. =
0.111 and DW = 1.838, which indicated the absence of serial correlation. The 2SLS estimation
from Model C-2 had an Adj. R2 = 0.872 (R2 = 0.896), S.E. = 0.111, and DW = 1.838 which
indicated the absence of serial correlation. The most important variable to have significant effect
were stability, robustness and presence of the expected signs. Model C-1 and C-2 had the same
elasticity, R2 and adj. R2, but the t-test statistic for each factor was different. The elasticity of
price of biodiesel with a lag of 3 quarters was a positive value of 0.5881, while the elasticity of
number of vehicles with a lag 4 of quarters and real GDP with a lag of 4 quarters were 0.588 and
1.807, respectively. The number of biodiesel stations was not significant.
This elasticity suggests that there is a direct effects on biodiesel demand. This mean, an
increase in the price of biodiesel of 1 Baht per liter 3 quarters earlier will lead to increase
biodiesel consumption by 0.5881 million liters. The number of vehicles had positive effect to
biodiesel consumption, as a result, an increase in biodiesel consumption of 2.134 million liters
resulted from increase in the number of vehicles. The real GDP of 4 quarter earlier had a positive
effect to biodiesel consumption. Biodiesel consumption increases 1.807 million liters because the
GDP 4 quarters earlier increased by 1 million Baht. This positive price elasticity of biodiesel
shows the effect of Thailand's promotion of biofuels in many
of its policies. One of the most important policies was to subsidize biodiesel market price so as to
keep it lower than 30 Baht per liter. Biodiesel contribute 46% of total petroleum product
consumption in 2012 [63]. When the price of biodiesel increases, it tends to increase
consumption after 3 quarters. Moreover, increase in number of vehicles and GDP per capita
during this period affects the demand of biodiesel after 3 and 4 quarters, respectively. The main
reason is that biodiesel is used in some sectors which do not tend to increase GDP directly such
as agriculture and transportation. A previous study also showed that an increase in GDP does not
affect biodiesel demand immediately and the impact of GDP on the renewable energy
consumption is significant for the countries [36,66]. Another study [6] also shows the biodiesel
demand regarding lags in the price of biodiesel. The dependent variables affecting biodiesel
demand had significantly high correlation such as number of biodiesel stations and vehicles.
Acknowledgments
The authors would like to thank the Asian Institute of Technology and The HM King's
Scholarship project (King HRD), Energy Policy and Planning Office (EPPO), Ministry of
Energy and Suratthani Rajabhat University for supporting this research.