Natural Resources, 2011, 2, 102-106
doi: 10.4236/nr.2011.22014 Published Online June 2011 (
Copyright © 2011 SciRes. NR
Leading Indicators of Heating Coal Pricing in
Turkey: A Coal Pricing Model (2003-2009)
Mehmet Mithat Uner1, Nezir Kose2, Soner Gokten3
1Department of Business Administration, Gazi University, Ankara, Turkey; 2Department of Econometrics, Gazi University, Ankara,
Turkey; 3(Corresponding Author) Department of Accounting and Finance, Gazi University, Ankara, Turkey.
Received January 11th, 2011; revised March 22nd, 2011; April, 1st, 2011.
In this study, a coal pricing model for Turkey is developed employing Granger causality and cointegration analysis by
using monthly data between January 2003 and April 2009. Empirical results based on Granger causality tests indicate
that foreign coal futures prices and domestic consumer price index for energy sector can be used as the leading indica-
tors for domestic coal prices for Turkey. An error correction model for Turkish coal pricing is specified by taking into
account the results of Granger causality. The forecast of the coal prices based on error correction model is giving very
successful results. It is observed that the coal prices and forecasted coal prices values are almost moving together or
very close to each other.
Keywords: Pricing, Granger Causality, Heating Coal
1. Introduction
Leading indicators of Turkish coal prices can be con-
sidered as foreign energy prices, domestic inflation,
cost factors, and economic growth of Turkey. The list
of leading indicators [1] used in this study is given in
Table 1.
Our analysis is based on monthly data over the pe-
riod January 2003-April 2009. Turkish coal prices (TL
per tone) were taken from the Turkish Statistics Insti-
tution. The code for this data is 0454001 according to
Classification of Individual Consumption by Purpose
(COICOP). We use US West Texas Intermediate (WTI)
for foreign crude oil spot prices and Cushing, Okla-
homa Crude Oil Future Contract 1 for foreign crude oil
futures prices. Henry Hub Natural Gas Spot Prices and
Natural Gas Futures Contract 1 are considered for spot
and futures prices of foreign natural gas. Spot and fu-
tures prices for foreign coal are Coal Commodity Spot
Prices Central Appalachia (CAPP) 12,500 Btu, 1.2 SO2,
and CAPP Coal Futures, respectively.
The source for the spot prices is the Energy Informa-
tion Administration (EIA), while futures prices were
taken from the New York Mercantile Exchange (NY-
MEX). Foreign energy prices are converted to Turkish
Lira from US Dollar to consider the effect of exchange
rate on Turkish coal prices.
Table 1. The indicators for Turkish coal prices.
Foreign Energy Prices
Crude Oil Spot Prices
Crude Oil Futures Prices
Natural Gas Spot Prices
Natural Gas Futures Prices
Coal Spot Prices
Coal Futures Prices
Domestic Inflation
Producer Price Index
Producer Price Index for Mining Sector
Consumer Price Index
Consumer Price Index for Energy Sector
Cost Factors
Unit Labour Cost
Other Inputs Cost
Real Sector
Industrial Production
Four indicators are used for domestic inflation;
Leading Indicators of Heating Coal Pricing in Turkey: A Coal Pricing Model (2003-2009)
namely, producer price index, producer price index for
mining and stone quarrying industry, consumer price
index, and consumer price index for electricity, gas and
other fuels. Base year of the price indexes is 2003. Data
were taken from the electronic data delivery system of
Central Bank of the Republic of Turkey (CBRT). Index
of Wages per Production Hour Worked in Manufacturing
of Cook and Refined Petroleum, whose base year is 1997,
is used as a proxy for the unit labour costs. Monthly data
are obtained through an interpolation by linear method as
the wage index is available only at a quarterly frequency
in the CBRT Electronic Data Delivery System. The elec-
tricity price is used as a proxy variable to capture the
effect of prices of other cost inputs on coal price levels.
The data were taken from the Turkish Statistics Institu-
tion (TL per KWh, The COICOP code is 0451001).
Industrial Production Index, whose base year is 2005,
is used as a proxy to measure real income at a monthly
frequency. Data source is the Electronic Data Delivery
System of the CBRT. To account for the seasonal ef-
fects, the data are seasonally adjusted by using the
Tramo/Seats method. All data in this study are in loga-
rithmic form.
2. Methods
2.1. Granger Causality Test for the Leading Indi-
Granger [2], [3], [4], proposed a time-series data based
approach in order to determine causality relationships
among variables. According to Granger [2], the defini-
tion of causality is based entirely on the predictability of
some series, say. If some other series contains informa-
tion in past terms that helps in the prediction of and if
this information is contained in no other series used in
the predictor, then is said to Granger cause. Granger
causality has been used in the context of rational expec-
tations, definition of strong exogeneity, and econometric
modelling strategy. A better term for Granger causality is
precedence [5]. Therefore, this test can be used for de-
termining leading indicators of any variable. We may
also use the results of Granger causality for evaluating
forecasting performance since it is concerned with one-
ahead forecast accuracy.
Park and Philips [6], Sims, Stock and Watson [7] and
Toda and Philips [8] have shown that the standard as-
ymptotic theory is not applicable to hypothesis testing in
level VAR model if the variables are integrated or coin-
tegrated. Therefore, the usual Wald test statistics for
Granger non-causality based on level VAR not only has
nonstandard asymptotic distribution but depends on nui-
sance parameters in general if variables are nonstation-
In this study, we examine Granger causal relationships
between Turkish coal prices and other variables using
Toda-Yamamoto [9] approach to determine the leading
indicators for Turkish coal price level. Toda-Yamamoto
procedure considers a lag augmented or modified Wald
(M-Wald) test which has conventional asymptotic chi-
square (2
) distribution when a VAR (p + dmax) is es-
timated where dmax is the maximal order of integration
suspected to occur in the system. In other words, this lag
augmentation procedure provides standard asymptotic
although the time series have integration/cointegration
properties, and therefore, can be applied without a priori
information about the presence (absence) and location of
unit roots. The results of Granger causality test based on
M-Wald statistic are given in Table 2.
Foreign coal future prices and consumer price index
for energy sector are Granger causes of Turkish coal
prices at 5% level. There is also Granger causality from
domestic electricity prices and industrial production to
Turkish coal prices at 10% level. These results imply that
foreign coal futures prices and consumer price index for
energy sector can be used as the leading indicators for
domestic coal prices for Turkey. Therefore, these vari-
ables will be used the next steps of our analysis.
2.2. Unit Root Test with one Structural Break
Figure 1 shows a time plot of the data set over the sam-
Table 2. The Results of Pairwise Granger Causality Tests
Null hypothesis:
X does not Granger cause of Turkish coal
Lag M-Wald
Statistic p-value
Foreign oil spot prices 3 2.97 0.3961
Foreign oil future prices 2 1.28 0.5273
Foreign natural gas spot prices 3 2.30 0.5122
Foreign natural gas future prices 3 0.68 0.8774
Foreign coal spot prices 11 15.94 0.1432
Foreign coal future prices 6 1454 0.0241
Consumer price index 2 3.59 0.2741
Consumer price index for electricity, gas
and other fuels 12 34.55 0.0006
Producer price index 2 0.48 0.7872
Producer price index for mining and stone
quarrying 2 1.53 0.4655
Unit labour cost 6 4.37 0.6256
Electricity Prices 10 17.58 0.0624
Industrial Production 8 14.11 0.0790
While maximum lag is 12, optimal lag length is determined by using two
types of information criteria (Schwarz and Akaike). If the two selection
criteria determine different lag order, Modified-Wald test, developed by
Toda and Yamamoto [9] is performed to eliminate lags from a general to
more specific model.
Maximum order of integration in the system is equal to 1.
Copyright © 2011 SciRes. NR
Leading Indicators of Heating Coal Pricing in Turkey: A Coal Pricing Model (2003-2009)
Copyright © 2011 SciRes. NR
ple period that displays to have an upward trend in levels
with a non-deterministic structure. Domestic coal prices
for Turkey and consumer price index for energy sector
both exhibit similar shapes while foreign coal futures
prices represent different shape from other series. For-
eign coal futures prices have been very volatile, changing
their trajectories and behaviour with respect to the eco-
nomic situation. Moreover, all variables include struc-
tural breaks in 2008. The visual analysis therefore tenta-
tively suggests that all the variables are not stationary.
The next step is to verify this conclusion using unit root
Model C). Model A allows for a one-time change in the
intercept of the trend function. It is known as the “Crash
Model”. Model B allows only a change in the slope of
the trend function at the time of the break. Model C in-
cludes a one time change in both level and trend. As
suggested in Figure 1, we use Model A for all series due
to there is only a change in the inter- cept of the trend
function. The results for the Perron [13] unit root test are
reported in Table 3. The unit root null hypothesis cannot
be rejected for all variables at 5% sig- nificance level.
These results indicate that the order of integration for all
series is equal to one.
To test for a unit root in time series data, the Aug-
mented Dickey-Fuller [10] procedure is commonly used
in empirical studies. Perron [11] was the first to point out
that power to reject the unit root null declines if the data
contains a structural break that is ignored. Per- ron [11]
incorporated an exogenous structural break into an ADF
test. More recently, Zivot Andrews [12] and Perron [13]
proposed unit root tests that allow for a structural break
to be determined endogenously from the data. In order to
check whether a unit root is present in the data or not, we
used Perron [13] test because of structural breaks in the
series. Perron [11] defined three types of models for
one-time break in the trend function (Model A, Model B,
2.3. Cointegration Analysis
The long-run relationship is investigated using the coin-
tegration analysis of Johansen and Juselius [14]. It is
well known that the results of cointegration tests using
this technique depend on the deterministic components
included in the VAR and on the chosen lag length. The
appropriate lag length is selected by using two types of
information criteria (Schwarz and Akaike). The VAR
order in logarithm level is estimated as 7 by Schwarz
information criteria while it is estimated as 12 by Akaike
information criteria. Due to the two selection criteria
were determined different lag order as 7 and 12, respect-
tively, Modified-Wald test [9], [15] was performed to
eliminate lags, and the appropriate lag length is estimated
as 7. We carried out both the trace and maximum eigen-
value type cointegration tests of Johansen and Juselius
[14]. Owing to the trace statistic and the maximum ei-
genvalue statistic may yield conflicting results. Number
of cointegrating equations by the deterministic compo-
nents in model is summarized in Table 4. Both the trace
and maximum eigenvalue (except test type 1) test statis-
tics indicate that there are two cointegrating equations
for all deterministic trend assumption.
Table 3. The Results of Unit Root Test
Series Estimated Break
Point: TB Lagt-Statistic Methods
2004: M03 44.5746 Mint
Turkish Coal
Prices 2004:M03 44.5746 Max ˆ
2008:M09 14.8556 Min t
Consumer Price
Index for Energy
Sector 2008:M09 14.8556 Max ˆ
2006:M06 54.7004 Min t
Foreign Coal
Futures Prices 2006:M06 54.7004 Max ˆˆ
*At 5% for T = 80.
The appropriate lag length is determined through general to specific
testing which is suggested by Perron [11].
The long-run analysis results are based on determines-
tic trend assumptions that both the time series and the
cointegrating equation have linear trends which corre-
spond to assumption 3 since all series have an upward
Figure 1. Graphic representation of the variables in logarithmic levels.
Leading Indicators of Heating Coal Pricing in Turkey: A Coal Pricing Model (2003-2009)
trend in levels. When the error correction term is nor-
malized with respect to coal prices for Turkey, the results
of VEC model are given Table 5. The VEC model is
very robust as all the diagnostic tests are insignificant,
indicating that the residuals are normally distributed,
homoskedastic and not serially correlated.
The coefficient of error correction term in coal prices
equation is estimated as 0.1350 and it is statistically
significant. It shows that the adjustment speed toward
long-run equilibrium will be 0.1350 and the full adjust-
ment of deviation takes about 7 months. The coefficients
of consumer price index for energy sector and foreign
coal future prices in cointegrating equation are estimated
as 1.54 and 0.44 respectively and they are statistically
significant at the 5% level.
The positive sign of coefficients is consistent with e-
conomic theory. The results indicate that an increase in
consumer price index for energy sector and foreign coal
future price of 1 percent will increase coal price for Tur-
key by approximately 1.54 percent and 0.44 percent re-
Table 4. Number of cointegrating relations by the determi-
nistic components in model (5% level).
Trend None None Linear Linear Quad-
Intercept Intercept Intercept Intercept Intercept
Test Type No
Trend Trend Trend
Trace 2 2 2 2 2
Eigen 1 2 2 2 2
Table 5. VECM results (cointegrating equation).
Turkish Coal Prices 1.0000
Consumer Price Index for Energy Sector 1.5351
Foreign Coal Futures Prices 0.4369
Intercept term 3.7425
Error Correction 0.1350
diagnostic test Test Test Statistic
Normality test Lutkepohl (Jar-
que-Bera) 3.36 0.1862
Serial correlation Breusch-Godfrey
5.92 for lag 1
7.39 for lag 2
9.82 for lag 3
5.53 for lag 4
Heteroskedasticity White 203.60
Values in parentheses are t-statistics.
3. Results
Actual values and forecasts for Turkish coal prices ob-
tained from VEC model are graphed in Figure 2. The
forecasting level of domestic coal prices for Turkey is
remarkably close to actual level as is seen by examina-
tion of Figure 2.
To assess the forecast performance of VEC model, we
compare their Root Mean Square Error (RMSE) and
Theil Inequality Coefficient with a naive model, where
the naive model is a first order autoregressive model for
logarithmic domestic coal prices. RMSE and Theil Ine-
quality Coefficient are computed for the forecasted value
of domestic coal prices which enables us to make com-
parisons across different models.
Table 6 shows that the VEC model has a much lower
RMSE and Theil Inequality Coefficient than the naive
model. These results imply that we have a significant
power in predicting Turkish coal prices using consumer
price index for energy sector and foreign coal futures
prices as the leading indicators in Turkey.
4. Conclusions
This study tries to determine the important factors that
affect the heating coal market of Turkey in the frame of
pricing decisions. A coal pricing model for Turkey was
developed by using Granger causality and cointegration
analysis. Empirical results suggest that settlement prices
of coal futures and domestic consumer price index for
energy sector can be used as the leading indicators in
order to determine and forecast the domestic heating coal
Figure 2. Actual and forecast values of Turkish coal prices.
Table 6. Forecast error statistics.
RMSE Theil Inequality Coefficient
VEC Model2.4506 0.0042
Naive Model10.0176 0.0176
Copyright © 2011 SciRes. NR
106 Leading Indicators of Heating Coal Pricing in Turkey: A Coal Pricing Model (2003-2009)
prices for Turkey. An error correction model for Turkish
coal pricing is specified by taking into account the results
of Granger causality. The forecast of the coal prices
based on error correction model is giving very successful
results. It is observed that the coal prices and forecasted
coal prices values are almost moving together or very
close to each other.
[1] M. M. Uner, N. Kose, S. Gokten, and P. Okan, “Financial
and Economic Factors Affecting the Lignite Prices in
Turkey: An Analysis of Soma and Can Lignites,” Re-
sources Policy, Vol. 33, No. 4, 2008, pp. 230-239.
[2] C. W. J. Granger, “Investigating Causal Relations by
Econometric Models: Cross Spectral Methods,” Econo-
metrica, Vol. 37, No. 3, 1969, pp. 424–438.
[3] J. R. McCrorie and M. J. Chambers, “Granger Causality
and the Sampling of Economic Processes,” Journal of
Econometrics, Vol. 132, No. 2, 2006, pp. 311-336.
[4] D. TjØstheim, “Granger-causality in multiple time se-
ries,” Journal of Econometrics, Vol. 17, No. 2, 1981, pp.
157-176. doi:10.1016/0304-4076(81)90024-5
[5] G. S. Maddala and M. Kim, “Unit Roots, Cointegration
and Structural Change,” Cambridge University Press,
[6] J. Y. Park and P. C. B. Philips, “Statistical Inference in
Regressions with Integrated Process: Part 2,” Economet-
ric Theory, Vol. 5, 1989, pp. 95-132.
[7] C. A. Sims, J. H. Stock and M. W. Watson, “Inference in
Linear Time Series Models With Some Unit Roots,”
Econometrica, Vol. 58, 1990, pp. 113-144.
[8] H. Y. Toda and P. C. B. Phillips, “Vector Autoregression
and Causality,” Econometrica, Vol. 59, 1993, pp.
229-255. doi:10.1016/0304-4076(93)90024-Y
[9] H. Toda and T. Yamamato, “Statistical Inference in Vec-
tor Autoregressions with Possibly Integrated Processes,”
Journal of Econometrics, Vol. 66, 1995, pp. 225-250.
[10] D. A. Dickey and W. A. Fuller, “Likelihood Ratio Statis-
tics for Autoregressive Time Series with a Unit Root,”
Econometrica, Vol. 49, 1981, pp. 1057-1072.
[11] P. Perron, “The Great Crash, the Oil Price Shock, and the
Unit Root Hypothesis,” Econometrica, Vol. 57, 1989, pp.
1361-1401. doi:10.2307/1913712
[12] E. Zivot and D. Andrews, “Further Evidence on the Great
Crash, the Oil-Price Shocks, and the Unit Root hy-
pothesis,” Journal of Business and Economic Statistics,
Vol. 10, 1992, pp. 251-272. doi:10.2307/1391541
[13] P. Perron, “Further Evidence on Breaking Trend Func-
tions in Macroeconomic Variables,” Journal of Econo-
metrics, Vol. 80, 1997, pp. 355-385.
[14] S. Johansen and K. Juselius, “Maximum Likelihood Es-
timation and Inference on Cointegration with Applica-
tions to the Demand for Money,” Oxford Bulletin of
Economics and Statistics, Vol. 52, 1990, pp. 169-210.
[15] H. Lütkepohl and M. M. Burda, “Modified Wald
Tests under Nonregular Conditions,” Journal of
Econometrics, Vol. 78, No. 1, 1997, pp. 315-332.
doi:10.1016 /S0304-407 6(97)80015 -2
opyright © 2011 SciRes. NR