The Invisible and Asymmetric Tax Policy in Gasoline Prices:

The Case of Turkey

Ozgur Bor

(Corresponding Author)

(Atilim University, Department of Economics, Kizilcasar Mah., Incek-Golbasi, 06680, Ankara, Turkey;

Tel. +90 (312) 586 86 36;Fax. +90 (312) 586 8091; e-mail: obor@atilim.edu.tr

Mustafa Ismihan

(Atilim University, Department of Economics, Kizilcasar Mah., Incek-Golbasi, 06680, Ankara, Turkey)

1

Abstract

This study analyzes the role of tax policy in gasoline prices in Turkey by utilizing time series

techniques. It provides and compares empirical results by using daily gasoline prices between

January 2005 and December 2013, with and without the effect of taxation. Our results, based on the

threshold cointegration analysis, indicate asymmetric adjustment between retail and crude oil

prices over the long-run. These results are in line with the “rockets and feathers” phenomenon; that

is, retail prices tend to rise faster than they fall in response to crude oil price changes. Additionally,

our results, with retail prices including taxes, indicate a negative threshold value, which implies that

the Turkish governments follow asymmetric taxation of gasoline prices through adjusting excise

taxes. Nevertheless, one can miss the big picture in gasoline pricing by concentrating only on the

adjustment dynamics. Therefore, we also analyzed and compared the long-run relationships between

crude oil and gasoline prices with and without taxes. The results indicate that Turkish governments

succeeded at implicitly imposing an exceptionally high tax burden on gasoline (about 62%) over the

longer term by adjusting non-salient excise tax amounts on gasoline and benefited from the resultant

tax revenues as means of public finance.

2

1. Introduction

In many oil importing economies retail prices of gasoline depend largely on changes in crude

oil prices and exchange rate movements. However, the role of taxation on the level of the retail

prices of gasoline is an important but somehow neglected issue in developing countries. In some

emerging market economies the fiscal authorities resort to taxes on petroleum products excessively

to create resources for financing the budget. Therefore, the amounts of taxes contribute both to the

adjustment as well as the level of the retail prices of gasoline.

An interesting feature of the oil products is the asymmetric price adjustment in retail prices.

In response to international crude oil prices retail oil prices tend to rise faster than they fall; a

phenomenon, Bacon (1991) calls “rockets and feathers”. Several theoretical explanations; such as

oligopolistic pricing, asymmetric consumer searching behavior, and the role of inventories, are used

to explain this phenomenon (Bacon and Kojima, 2010). Price asymmetries are important because the

existence of such asymmetries imply that consumers are not benefiting from price reductions as in

the case of symmetric price adjustment. Therefore, price asymmetries have important welfare

implications (Meyer and Cramon-Taubadel, 2004:582).

Bacon and Kojima (2010) note that asymmetric pricing in oil products is observed in all

studies conducted on eight developing countries. i However, for developed countries, the empirical

evidence on rockets and feathers pricing is mixed as a result of differences in econometric

specification, data frequency (daily, weekly, biweekly, monthly), the time period, the scope of the

transmission chosen and whether the prices include taxes or not (Grasso and Manera 2007). Two

seminal studies on the US economy exemplify the importance of these differences. Borenstein et al.

(1997) used a non-standard error correction model and weekly data over the 1986-1998 period to

test the asymmetry of the US gasoline prices. They found that retail gasoline prices rose quickly

following an increase in the price of crude oil, but fell slowly following a decrease. Contrarily,

Bachmeier and Griffin (2003: 772) noted that “daily data may provide more reliable estimates than

3

weekly data … [since] aggregation over time can create a type of omitted variables bias problem”.

They used a standard Engle-Granger error correction model with daily data over the 1995-1998

period, and found no evidence of asymmetry in the wholesale gasoline prices. They found similar

results when they employed Borenstein et al.’s (1997) non-standard specification on daily data, and

reached the conclusion that Borenstein et al.’s (1997) results were fragile.ii Their studies demonstrate

that econometric choices are highly influential on empirical outcomes, and confirm Geweke’s (1978)

conviction that daily data mitigate estimation bias of aggregated (e.g. weekly, monthly) data, when

available. Lastly, it is also important to mention that more recent studies employed more flexible and

efficient methods like threshold cointegration techniques, compared to the standard cointegration

(Engle-Granger) and asymmetric error-correction models, for analyzing asymmetric adjustment

between retail prices and crude oil prices (see Mann, 2012 and the references cited therein for

an overview of the threshold cointegration techniques and the related literature).

In Turkey, prices of oil products have been high for a long time and this has been affecting

the welfare of the society seriously.iii Despite the importance of this topic, Alper and Torul (2009)

appear to be the only relevant research on Turkey. By using monthly data for the 1991-2007 period,

and employing a structural VAR model iv with first differenced data, they found that retail gasoline

prices in Turkey responded to increases in international crude oil prices more than the decreases.

The authors attributed the asymmetry mainly to the government’s price setting policies via taxation

and pointed to a common belief among economists and public that Turkish fiscal authorities used

high gasoline taxes to create resources to finance government budget. Nevertheless, Alper and Torul

(2009) focused on the rockets and feathers pricing in Turkish economy with first differenced data and

hence they lost valuable information contained in the level of price variables, which would otherwise

be utilized in cointegation analysis for understanding the existence of high oil prices over the long

run. In other words, asymmetric price adjustments in the short run may not explain the existence of

high level of prices in the long run, so one can miss the big picture in gasoline pricing by

4

concentrating only on the short term price adjustment dynamics. Therefore, for a more thorough

analysis long run (level) relationships with and without taxes should be analyzed and compared.

In this study, we investigate the role of tax policy in gasoline pricing in Turkey with daily data

from 4th of January 2005 to 31st of December 2013. In order to identify the role of taxation in gasoline

pricing, we provide empirical results, by employing the popular threshold cointegration analysis

developed by Enders and Siklos (2001), on gasoline price data with and without taxes. Additionally,

we give a particular emphasis to the long-run (level) relationships between crude oil prices and

gasoline prices with and without taxes, which is a neglected issue in the existing literature on

gasoline pricing.

The rest of the paper is organized as follows. Section 2 provides an overview of the role of oil

and its taxation in the Turkish economy. The model, data and empirical results are provided in

Section 3. Finally, Section 4 provides the concluding remarks.

2. The Role of Oil and Its Taxation in the Turkish Economy: An Overview

Turkey is an emerging market economy and a net importer of oil. Turkey produced only

around 7-8% of its total petroleum consumption between 2005 and 2013.v As a consequence of high

dependency on oil imports, crude oil price movements in international energy markets have a direct

impact on domestic retail prices of oil products. This dependency also contributes to the high current

account deficit of the Turkish economy (about 10% of GDP in 2011).

In Turkey, until 2005 the government determined retail gasoline prices.vi With the Petroleum

Market Law No:5015, that entered into force in 2003, Energy Market Regulatory Authority (EMRA)

became responsible from guiding, monitoring and surveilling the energy market. From January 2005,

the refinery prices are determined freely in the market which is regulated by the EMRA to ensure

that it reflects the developments in international oil markets and exchange rate movements.

Licensed refineries and distributors are required to notify their ceiling prices to EMRA.

5

Nevertheless, as shown in Figure 1, the total share of refineries, distributors and retailers was

less than the share of taxes in retail (final) gasoline vii prices in 2011.viii Recent annual EMRA reports on

the petroleum market make it clear that this situation is not unique to 2011. ix In sum, tax is the single

largest component of retail prices in gasoline and other oil products (EMRA, 2012).

Figure 1. The Share of Final Price Components of Unleaded 95 octane Gasoline in 2011

Refinery (Tax Free) Distributors’ Share Retailers’ Share Total Tax (VAT+SCT)

30.5

4.8 4.3

60.4

Note: SCT=special consumption tax and VAT=value added tax.

Source: EMRA (2012: 140).

Considering the high tax burden on gasoline, the retail prices of oil products are still

regulated by the government. However, it is important to underline that total tax burden comprises

of two different taxes; the special consumption tax (SCT) and the value added tax (VAT). In order to

harmonize Turkey’s indirect tax system to the European Union acquis communautaire, in 2002 SCT

on gasoline was introduced. SCT is an excise or specific tax that is charged only once. It is imposed on

specific goods and the amount imposed varies among products. The main purpose of SCT in the

European Union is to maximize the social welfare. Therefore, it is imposed on luxury, unhealthy and

polluting goods.x On the other hand, VAT is levied at each stage of the production and the

distribution process.xi In case of oil products, SCT is imposed first on importers and/or producers

6

(including refineries) and then VAT is levied on the refinery price, distributor and vendors’ share as

well as on special consumption tax. This is referred to as the taxation of an already taxed item.

Moreover, while the VAT is an ad valorem tax and its rate is fixed (18%), at least over the sample

period, SCT is levied on “per unit” basis and adjusted over time. The fact that VAT rates are fixed but

SCT amounts are adjustable leaves SCT as the only way that government could affect gasoline prices.

Predictably, the SCT amounts (on gasoline) have been changed five times since 2005 (Figure 2).

Figure 2. Special Consumption Tax Adjustments from 2005 to 2013*

1.3625 TRY1.4915 TRY

1.6915 TRY1.8915 TRY 1.8765 TRY

2.1765 TRY

01.01.2005 01.07.2008 15.07.2009 01.01.2010 18.05.2012 22.10.2012

* The special consumption tax is remained at 2.1765 TRY from 22.10.2012 to 31.12.2013

Note: TRY=Turkish Lira.Source: Turkish Tax Authority

Table 1 provides strong evidence that the government can use SCT as an “instrument” to

affect retail gasoline prices through effective tax rate, which is the total tax over retail price. It is clear

from Table 1 that the effective tax rate on gasoline is very high in Turkey, fluctuating between 60%

and 75% with an average of 66.4% since 2005. However, as explained in the following sections, the

effective tax rate is not clearly perceptible to the final users since only VAT rate (18%) is visible on the

gasoline bills of the consumers.

7

Table 1. Effective Tax Rates on Gasoline

Year Retail Price After Tax (TRY)a Total Tax (TRY)a EffectiveTax Rate (%)b

2005 2.30 1.71 74.492006 2.58 1.76 68.092007 2.78 1.79 64.332008 3.09 1.95 63.072009 2.79 1.92 68.772010 3.74 2.46 65.942011 4.19 2.53 60.402012 4.48 2.78 61.972013 4.80 2.92 60.85

aAnnual average.bEffective tax rate = (Total taxes paid /retail price including taxes)*100, where total tax=VAT+SCT.Source: EMRA (2012, 2013 and 2014)

The main motivation behind imposing excessive tax on gasoline is more obvious when one

considers the role of indirect taxes on oil products in public finance in Turkey. In fact, the two indirect

taxes (SCT and VAT) on these products constitute 19.34% and 13.50% of the total tax revenues in

2005 and 2013, respectively (EMRA, 2014: Table 3.27, p.182). However, the share of these taxes in

GDP is a better indicator for the importance of indirect taxes in the total budget and this share has

remained stable, fluctuating between 3% and 3.6%, over this period (EMRA, 2014: Table 3.27, p.182).

In sum, these figures clearly indicates that the taxes levied on oil products are an important source of

tax revenue for Turkish governments.

3. Model, Data and Empirical Results

3.1. Model and Methodology

Initially, the long run equilibrium relationship between the retail price of gasoline (RGP) and

crude oil (COP) is specified by the following equation:

RGPt=β0+β1COPt +μt, (1)

where RGP and COP are retail (output) and crude oil (input) prices (see Section 3.2 for more detail on the

definitions and sources of the data) and μ is the error term.

8

In line with the existing literature, the standard Engle and Granger (EG) approach is used in

the estimation of Equation (1) due to the possible non-stationarity in the data (see Bacon and Kojima,

2010; and Bachmeier and Griffin, 2002). Since Equation (1) relates the output or final price ( RGP) to

the input price (COP), β1 is expected to be 1, that is, the input costs are passed fully to the final

(retail) prices (see, for example, Bacon and Kojima, 2010; and Bachmeier and Griffin, 2002). However,

as will be explained in more detail in the next sub-section, β1 can exceed unity when sales taxes such

as VAT and SCT are levied. Thus, it is important to provide empirical results by using retail prices with

and without the taxes.

Engle and Granger (1987) show that cointegration exists if t is stationary [t I(0)]. Therefore,

residuals from equation (1) are used to estimate the following relationship:

t =t-1 + t (2)

where is the speed of adjustment coefficient.

Rejection of the null hypothesis of no cointegration ( 0) implies that the residuals in equation (1)

are stationary.

Enders and Siklos (2001) argue that if the adjustment is asymmetric, the standard tests for

cointegration and its extensions are mis-specified and they consider an alternative error correction

specification called the threshold autoregressive (TAR) model. When we incorporate this argument

into equation (2) we obtain,

t = It1t-1+(1- It)2t-1 + t (3)

where It is the Heaviside indicator function such that:

It=¿ (4)

and is the estimated threshold value. Note that equations (1), (3) and (4) jointly represent a TAR

model.

9

Here, 1 and 2 gives the speed of adjustment coefficients for different regimes [i.e. the adjustment

is represented by 1t-1 (2t-1) when t-1 is above (below) the threshold, ]. If 1 = 2, then the

adjustment is symmetric and the TAR model reduces to the standard Engle-Granger cointegration

model. Here, the threshold has particular importance because it implies that movements toward long

run equilibrium do not take place at all points in time but only when the divergence from equilibrium

exceeds the threshold (Ghoshrey 2011). However, as noted by Enders and Siklos (2001:167), "in a

number of economic applications it is natural to set =0 so that cointegrating vector coincides with

attractor". However, in general, is unknown and it is consistently estimated (see Enders and Siklos

2001) and we call such model(s) as Consistent TAR model(s).

In equation (4), the Heaviside indicator depends on the level of t-1 (Enders and Siklos 2001). In an

alternative suggested by Enders and Granger (1998) and Enders and Siklos (2001), the threshold

depends on the previous periods change in t-1 (t) and t series exhibit more momentum in one

direction and such model is called momentum-threshold autoregressive (M-TAR) model. Here, the

Heaviside indicator is set by using lagged changes in t.

It=¿ (5)

Considering the above issues, there are four alternative models; namely, TAR, M-TAR, Consistent TAR

and M-TAR models. In empirical applications, however, the "best fitted" model is generally

determined by utilizing the information criteria such as Akaike Information Criterion and Scwarz

Information Criterion.

It should be also noted the empirical finding of cointegration justifies the existence of an error-

correction representation (via Granger Representation Theorem):

∆ RGP t=θ+φ+¿ ECT t−1+¿+φ

−¿ECT t−1

−¿+ ∑i=1

nα

+ ¿∆ RGPt−i+¿ + ∑

i=1

nα−¿∆ RGPt−i

−¿+ ∑i=1

nβ

+ ¿∆COPt−i+¿ +¿+ ∑

i=1

nβ−¿∆ COPt−i

−¿+ϑt( 6)¿¿ ¿

¿ ¿

¿¿ ¿

¿ ¿

¿¿ ¿

10

All the lagged prices (RGP and COP) are split into positive and negative components represented by

“-”and “+” superscripts. The error correction terms “ECT” are constructed from the threshold

cointegration regressions in equations (3), (4) and (5).

As mentioned above, in order to capture the asymmetries in the short run α+¿∆ RGPt− i+¿ ¿¿ and α−¿ ∆RGPt− i

−¿ ¿¿

(the lagged gasoline price increases and decreases, respectively) β+¿∆ COP t− i+¿ ¿¿ and β−¿ ∆COPt− i

−¿ ¿¿ (the

lagged crude oil price increases and decreases, respectively) are used. Additionally, the asymmetry in

the adjustment speed is also checked by defining disequilibrium terms using φ+¿ECT t−1+¿¿ ¿ and φ−¿ ECTt−1

−¿¿¿.

We can check for the presence of asymmetry by performing a standard Wald test both on the speed

and magnitude of the adjustment by using an appropriate null hypothesis.

3.2. Data

Our data span ranges from January 4, 2005 to December 31, 2013 and the number of total

observations for each variable is 3,282.xii It should be noted that this study uses daily data since

aggregated data can create substantial econometric problems (Geweke 1978), and also daily data

provides more information for separating effects from lagged changes.

Definitions and sources of the variables are given below.xiii

COP: Crude oil prices of gasoline, in USD per liter (Fob Europe Brent spot price). Source: The US

Energy Information Administration (http://www.eia.gov/dnav/pet/pet_pri_spt_s1_d.htm)xiv

RGP (excluding taxes): Net retail price of unleaded 95-octane gasoline price, in USD per liter

(vendor/sale station price without taxes). Source: EMRAxv

RGP (including taxes): Gross retail price of unleaded 95-octane gasoline price, in USD per liter

(vendor/sale station price with taxes). Source: EMRAxvi

Figure 3 shows the time plot of the three variables. As expected, these variables are non-

stationary,xvii and they tend to move closely.

11

Figure 3. Time Plot of COP and RGP (Including and Excluding Taxes)

2005 2006 2007 2008 2009 2010 2011 2012 20130.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

3.3 Empirical Results

Initially, we estimate the long run equilibrium relation between retail gasoline price (excluding the

taxes) and crude oil price, as specified in Equation (1), by OLS. The results are provided in Equation

(7).

RGPt (excluding taxes) = 0.289 + 1.074 COPt (7)

(75.596) (155.839)

where t-values are provided in parentheses.

In table 2, the results for the standard Engle-Granger, TAR, M-TAR, consistent TAR and M-TAR models

are provided based on the residuals of Equation (7).xviii In all models, including Engle-Granger, two

price series are cointegrated.xix According to the results of Akaike Information Criterion (AIC) and

Scwarz Information Criterion (SIC), the Consistent M-TAR model fits the data better than the other

the models. Thus, we focus on the results of the Consistent M-TAR model.

Table 2. Estimates of the cointegration models (With retail prices excluding taxes)

12

RGP including taxes taxes

RGP excluding taxes

COP

Item Engle-Granger TAR Consistent TAR M-TAR Consistent M-TAR

Threshold ()a NA 0 +0.0631 0 -0.0001

1 -0.026***d -0.028*** -0.034*** -0.020*** -0.020*** (-6.488)e (-5.067) (-5.519) (-3.912) (-3.848)2 NA -0.025*** -0.021*** -0.037*** -0.037*** (-4.069) (-3.761) (-5.534) (-5.623)

1 -0.016 -0.016 -0.016 -0.018 -0.018 (-0.941) (-0.938) (-0.903) (-1.049) (-1.054)2 -0.032* -0.032* -0.032* -0.031* 0.031* (-1.833) (-1.830) (-1.811) (-1.799) (-1.795)AIC -1315.44 -1313.58 -1315.92 -1317.33 -1317.83SIC -1297.15 -1289.20 -1291.54 -1292.95 -1293.45Φb NA 20.850 22.032 22.748 23.000

1= 2c NA 0.140 2.474 3.889 4.385

[0.708] [0.116] [0.049] [0.036]a Estimated threshold value.b Φ represents the sample value of F statistic with the null hypothesis of 1=2= 0 (Enders and Siklos, 2001). c In this row the sample values of F statistic, for the null hypothesis 1= 2 , are provided. P-values are provided in square brackets.d Three, two and one asterisks (*) denote that the estimated coefficient is statistically significant at, or below, one, five, and ten percent level, respectively.e t-values are provided in parentheses.

The values of the adjustment parameters (1 and 2) have the correct signs and plausible values in

the Consistent M-TAR model. They also have significant p-values, and hence we can safely say that

our results suggest convergence toward the equilibrium relation. However, the estimates suggest

that the gaps arising from the long-term equilibrium due to the increase in crude oil price are

eliminated more quickly (2=-0.037) compared to the decline in crude oil prices (1=-0.020). The null

hypothesis that the adjustment coefficients are equal (1=2) is also rejected (p-value=0.036). In

sum, our results suggest that the retail and crude oil prices are cointegrated and the adjustment is

asymmetric. Thus, these results indicate asymmetric adjustment between retail prices and crude

oil prices over the long-run. In other words, our results are in line with the “rockets and feathers”

phenomenon; that is, retail prices tend to rise faster than they fall in response to crude oil price

changes.

13

It should be noted in passing that the threshold value is almost zero (-0.0001) and hence it is not

surprising that the results from the M-TAR model is very similar to those from the Consistent M-TAR

model. This confirms the robustness of our results.

The findings of cointegration with M-TAR consistent adjustment justifies estimation of the error-

correction model as specified in Equation (6). The results are given in Table 3.xx

Table 3. Results of the asymmetric error correction model (M-TAR and Consistent M-TAR) Threshold 0 -0.0001

0.000(1.181) a,b 0.000(1.030) α 1

+¿¿ 0.054(2.636)*** 0.059(2.891)*** α 2

+¿¿ -0.032(-1.735)* -0.016(-1.009) α 1

−¿¿ 0.043(1.532) 0.034(1.280) α 2

−¿¿ 0.044(2.036)** 0.044(1.533) β1

+¿¿ 0.172(4.117)*** 0.160(3.847)*** β2

+¿¿ 0.081(2.108)** 0.083 (2.486)** β1

−¿¿ 0.082 (2.414)** 0.075 (2.230) ** β2

−¿¿ 0.018(0.522) 0.054(1.370) φ+¿¿ -0.013(-2.376) ** -0.020(-4.360)*** φ−¿¿ -0.017 (-3.197)*** -0.029 (-4.360)*** R2 0.031 0.036 AIC -3091.16 -3109.68SIC -3024.11 -3042.64

a t-values are provided in parentheses. b Three, two and one asterisks (*) denote that the estimated coefficient is statistically significant at, or below, the one, five, and ten percent level, respectively (the results are based on Newey-West standard errors).

As seen from Table 3, estimates of parameters that represent short-run adjustment dynamics are

quite similar. In other words, the short run coefficients (* and *) suggest the presence of price

symmetries (for both models). Wald tests also confirm our expectations.xxi Nevertheless, it should be

noted that the speed of adjustment terms (φ+¿¿ and φ−¿¿ ) are usually sensitive to the sample period

and have poor small-sample properties (See, Enders and Siklos, 2001 and references cited therein for

more detail). Thus, probably for this reason the respective Wald test result has not shown asymmetry

14

for the speed of adjustment terms for both error-correction models based on M-TAR and Consistent

M-TAR models.xxii

In line with the aim of this study, we continue our analysis by estimating the long run relation

between retail gasoline price (including taxes) and crude oil price.

The OLS estimates of the long-term relationship between the two oil prices are provided in Equation

(8).

RGPt (including taxes) = 1.467 + 1.620 COPt (8)

(166.841) (102.273)

where t-values are provided in parentheses.

Table 4 presents the results for the standard Engle-Granger, TAR, M-TAR, consistent TAR and M-TAR

models based on the residuals of Equation (8).xxiii In all models, including Engle-Granger, two price

series are cointegrated. However, based on the AIC and SIC, Consistent TAR model fits the data better

than the rest of the models.

Table 4. Estimates of the cointegration models (With retail prices including taxes)

Item Engle-Granger TAR Consistent TAR M-TAR Consistent M-TAR

Threshold ()a NA 0 -0.1314 0 -0.0118

1 -0.014***d -0.012*** -0.010** -0.013*** -0.012*** (-4.522)e (-2.978) (-2.516) (-3.328) (-3.312)2 NA -0.016*** -0.021*** -0.016*** -0.022*** (-3.471) (-4.207) (-3.131) (-3.410)

1 -0.044** -0.044** -0.043** -0.043** -0.042** (-2.392) (-2.505) (-2.462) (-2.499) (-2.433)2 -0.042** -0.042** -0.041** -0.042** -0.042** (-2.392) (-2.382) (-2.358) (-2.393) (-2.386)AIC 2324.48 2326.17 2323.08 2326.20 2324.51SIC 2342.77 2350.55 2347.46 2350.58 2348.90Φ b NA 10.379 11.930 10.360 11.210

1= 2c NA 0.314 3.397 0.278 1.965

[0.575] [0.065] [0.598] [0.161]a Estimated threshold value.b Φ represents the sample value of F statistic with the null hypothesis of 1=2= 0 (Enders and Siklos, 2001).

15

c In this row the sample values of F statistic, for the null hypothesis of 1= 2 , are provided. P-values are provided in square brackets.d Three, two and one asterisks (*) denote that the estimated coefficient is statistically significant at, or below, one, five, and ten percent level, respectively.e t-values are provided in parentheses.

In the Consistent M-TAR model, the values of the adjustment parameters (1 and 2) have the correct

signs and plausible values. Again, they also have significant p-values, and hence we can safely say

that our results suggest convergence toward the equilibrium relation. In this case, the null hypothesis

of symmetric adjustment (1=2) is rejected at 6.5% significance level (p-value=0.065). Thus, there is

some evidence of asymmetric adjustment between retail prices and crude oil prices over the long-

run.

Additionally, our results, with retail prices including taxes, indicate a negative threshold value, =-

0.1314, (compared to almost zero threshold value with retail prices excluding taxes) which may imply

that the Turkish governments follow the asymmetric taxation of gasoline prices via adjusting excise

taxes. That is, the retail prices with taxes are adjusted more quickly (2=-0.021), in response to the

rise in crude oil prices, beyond the threshold value since the government may temporarily delay the

adjustment of excise tax possibly with the aim of preserving political popularity. Thus, reduction in

the profit margin up to a certain point is bearable for firms for a while due to a delay in the

adjustment of excise taxes, but beyond the threshold value a further reduction in the margin is a

threat to profitability and hence different behavior is adopted by the firms.

The findings of cointegration with TAR consistent adjustment justifies estimation of the error

correction model (the results are given in Table 5). In line with the previous case (without taxes), as

can be seen from Table 5, estimates of parameters that represent short-run adjustment dynamics are

quite similar, and hence short run coefficients (* and *) suggest the presence of price symmetries.

Again, Wald tests confirm our expectations.xxiv However, now the Wald test result has shown some

evidence of asymmetry for the speed of adjustment terms (p-value=0.066).

16

Table 5. Results of the asymmetric error correction model with consistent TAR cointegration.Item Estimate

θ -0.000 (-0.669) a,b

α 1+¿¿ 0.017 (0.690)

α 2+¿¿ - 0.002 (-0.078)

α 1−¿¿ 0.026 (0.715)

α 2−¿¿ 0.035 (1.216)

β1+¿¿ 0.331 (6.297)***

β2+¿¿ 0.157 (3.151)***

β1−¿¿ 0.144 (0.858)

β2−¿¿ 0.265 (1.265)

φ+ ¿¿ - 0.008 (-2.469)**φ−¿¿ - 0.020 (-3.860)***

R2 0.032AICSIC

1107.301174.35

a t-values are provided in parentheses. b Three, two and one asterisks (*) denote that the estimated coefficient is statistically significant at, or below, the one, five, and ten percent level, respectively (the results are based on Newey-West standard errors).

As we mentioned before, one can miss the big picture in gasoline pricing by concentrating only on

the adjustment dynamics. Therefore, now, we analyze the long-run relations in more detail.

As noted in Section 3.1, the retail price of gasoline (RGP excluding taxes) is expected to fully

adjust to the changes in crude oil prices (COP), and accordingly, parameter of COP (β 1) is expected to

be 1. As is seen from Equation (7), the estimated value of this parameter is 1.074 (close to 1), and

therefore, it follows the theoretical expectation. However, as can be seen from Equation (8), the

response of gasoline prices (including taxes) to crude oil price changes is much higher (1.62), and it

exceeds unity. This means that one dollar increase in the price of crude oil reflects as an increase of

1.62 dollars on the retail price, 62 cents of which go to the tax authority. Ironically, Turkish

government succeeded at implicitly imposing an exceptionally high tax rate of 62%, [(1.62 – 1)*100]xxv

on gasoline over the longer term by adjusting the SCT amounts on gasoline as explained in Section 2.

17

This, in turn, implies that SCT is the main policy instrument in the price setting process of gasoline. It

should be recalled that VAT is an ad valorem tax and its rate is fixed at 18% in Turkey but SCT is levied

on “per unit” basis and it is not prominent (visible) to final users. Consequently, the SCT amounts

have been altered five times over the sample period (2005-2013), which confirms that the ruling

governments have preferred to adjust SCT over the long run.

In sum, the above results provide explanation to the persistence of high gasoline prices in

Turkey, which is arising from the reliance of Turkish government on indirect taxes as an instrument of

public finance. Therefore, Turkish governments benefited from crude oil price increases in the long-

run by adjusting the SCT amounts in such a way that the tax burden on gasoline is implicitly fixed at

around 60%. This result is consistent with the recent empirical evidence that individuals under react

to taxes when they are not visible and this limited attention to taxes may have serious welfare

consequences (Chetty et al., 2009).

4. Conclusion

This study investigated the role of taxation in gasoline prices in Turkey by using daily data

between 4th of January 2005 and 31st of December 2013. Our results, based on the threshold

cointegration analysis, indicate asymmetric adjustment between retail prices and crude oil prices

over the long-run. These results are in line with the “rockets and feathers” phenomenon; that is,

retail prices tend to rise faster than they fall in response to crude oil prices. Additionally, our results,

with retail prices (including taxes) indicates a negative threshold value which implies that the Turkish

governments follow asymmetric taxation on gasoline prices through adjusting excise (indirect) taxes.

However, one can miss the big picture in gasoline pricing by concentrating only on the price

adjustment dynamics, which is a commonly followed perspective in this literature. Deviating from the

scope of existing studies, we also analyzed the long-run relationships between crude oil prices and

gasoline prices with and without taxes. This is important because Turkish Tax Authority seems to

18

employ indirect taxes (especially, SCT) as a policy instrument with an aim to maximize its tax revenue

from oil products including gasoline. As a result, the nature of the taxation of gasoline is complicated

and its effects on the retail prices are not straight forward. Therefore, long-run analysis of price

dynamics with and without the effect of taxation on gasoline is a valuable empirical exercise.

Our results indicate that one-dollar increase in the price of crude oil leads to an increase of

1.62 dollars on the gross retail price of gasoline over the long-run. This, in turn, implies that 62 cents

(or 62 %) goes to the tax authority. Nevertheless, it is important to stress that 62 cents consists of the

tax revenues arising from the imposition of SCT and VAT (including the VAT on SCT amounts)

However, only VAT amount is visible to the final users and the amounts of SCT and VAT part of SCT

(although included in VAT) are hidden within the retail price of gasoline.

To sum up, the Turkish government implicitly imposes an exceptionally high tax rate on

gasoline over the long-run, by frequently adjusting the excise tax amounts which are not visible to

final consumers (only VAT amounts, 18%, are visible on bills). The main motivation behind this

behavior is the over-dependence of Turkey on the indirect taxes as a means of public finance. Thus,

this study indicates that the indirect tax revenue obtained from gasoline sales contributes

substantially to the Turkish governments' budget at the expense of consumer welfare. It is,

therefore, not unfair to say that the Turkish governments affect the gasoline pricing via “Invisible

Tax” Policy.

The findings of this paper suggest the following policy implications: First, the gasoline prices

should be monitored more carefully by Energy Market Regulatory Authority, considering the

evidence on “rockets and feathers” phenomenon, since consumers are not benefiting from price

reductions in a timely manner and this, in turn, results in welfare losses. Second, the government

must avoid the “Invisible Tax” Policy applied on gasoline as well as on other oil products since such

policies also creates consumer welfare losses. In other words, there is a need for more transparent

tax policy in a way that tax amounts on oil products become more visible to final consumers.

19

20

References

Al Gudhea, S., Kenc, T., and Dibooglu, S., (2007) Do retail gasoline prices rise more readily than they

fall? A threshold cointegration approach. Journal of Economics and Business 59:560-574.

Alper, C.E.,andTorul, O., (2009) Asymmetric adjustment of retail gasoline prices in Turkey to world

crude oil price changes: the role of taxes. Economics Bulletin 29(2):775-787.

Bacon, R. W.,(1991) Rockets and feathers: The asymmetric speed of adjustment of U.K. retail gasoline

prices to cost changes. Energy Economics13:211-218.

Bacon R.W.,and Kojima M., (2010) Rockets and feathers: Asymmetric petroleum product pricing in

developing countries. Extractive Industries for Development Series #18,June 2010, The World Bank.

Bachmeier, L.J., and Griffin,J.M., (2003) New evidence on asymmetric gasoline price responses. The

Review of Economics and Statistics 85(3):772-776

Borenstein, S., Cameron, A.C., and Gilbert,R., (1997)Do gasoline prices respond asymmetrically to

crude oil prices? Quarterly Journal of Economics 112(Feb):305-39.

Chetty, R., Looney, A., and Kroft, K. (2009) Salience and taxation: theory and evidence. American

Economic Review 99(4):1145-1177.

Enders, W. and Siklos P.L. (2001) Cointegration and Threshold Adjustment. Journal of Business and economic Statistics 19(2): 166-176.

Energy Market Regulatory Authority (EMRA), (2012) Petroleum Market Sector Report 2011.

Petroleum Market Department, Ankara.

Energy Market Regulatory Authority (EMRA), (2013) Petroleum Market Sector Report 2012.

Petroleum Market Department, Ankara.

Energy Market Regulatory Authority (EMRA), (2014) Petroleum Market Sector Report 2013.

Petroleum Market Department, Ankara.

Engle, R.F., and Granger C.W.J. (1987) Cointegration and Error Correction: Representation, Estimation

and Testing Econometrica, 55:251-276.

Geweke,J.,(1978) Temporal aggregation in the multiple regression model. Econometrica 46(3):643-

61.

Ghoshrey A. (2011) Underlying Trends and International Price Transmission of Agricultural

Commodities, ADB Economics Working Paper Series 257, Asian Development Bank.

21

Granger, C.W.J., and Lee, T.H., (1989) Investigation of production, sales and inventory relationships

using multicointegration and non-symmetric error correction models. Journal of Applied

Econometrics 4:145–159.

Grasso, M., and Manera,M., (2007) Asymmetric error correction models for the oil-gasoline price

relationship. Energy Policy35:156-177.

Mann, J. (2012). Threshold cointegration with applications to the oil and gasoline industry. PhD

Thesis, Queen's University. https://qspace.library.queensu.ca/bitstream/1974/7281/1/Mann_Janelle

_M_ 201206_PhD.pdf

Meyer, J., and Cramon-Taubadel, S.V., (2004) Asymmetric price transmission: a survey. Journal of

Agricultural Economics 55(3):581-611.

22

Endnotes

23

i See Bacon and Kojima (2010) for more detail on these eight studies.iiA more recent study by Al-Gudhea et al. (2007) investigated crude, spot, wholesale and retail gasoline adjustments in the US. They have used daily data for the period from December 1998 to January 2004 and tested for asymmetries by using a set of cointegration and error correction models with non-linear adjustment. They have found that prices were cointegrated and long run equilibrium adjustments were asymmetric.iiiAs of the end of January 2013, Turkey has the highest retail gasoline prices in the world. In August 2012, Turkey was the second in the world –closely behind Norway whose real average income is more than three times that of Turkey. (http://www.bloomberg.com/slideshow/2012-08-13/highest-cheapest-gas-prices-by-country.html #slide1)ivSpecifically, they stated that the span of time period in their study is 17 years and it cannot justify long-term relation properly (Alper and Torul, 2009: footnote 10). Even though this concern seems to be right, when one considers the other studies in the literature (i.e. those cited in Grasso and Manera, 2007), including monthly data, the span of time period is less than 20 years and usually not more than ten years in the case of weekly and daily data. Therefore, from this point of view, Alper and Torul’s justification is not valid. v Source: U.S Energy Information International Energy Statistics (www.eia.gov).viDuring the 1998-2004 period, Automatic Pricing Mechanism was used. This system determines the ceiling prices for oil products based on CIF Mediterranean Market spot prices.viiUnless stated otherwise, gasoline refers to the unleaded 95 Octane gasoline throughout this study.viiiThe retail price is calculated by adding the refinery price, the distributors’ profit, retailers’ (sale stations) share, the special consumption tax and the value added tax.

ixDuring 2009-2011 period, the distributors and the vendors’ total share is about 9.5%, the refinery price is around 27.5% and the remaining lion’s share of 63% belongs to the taxes.xHowever, SCT is imposed even on basic consumption goods in Turkey.xiIt is a general consumption tax that covers all goods and services and applied to all stages from producer and consumer and calculated from the transaction value.xiiFor the missing data concerning the weekends and the national holidays, the latest figure (i.e. before the weekend/holiday) is used, considering the fact that gasoline is also used during holidays.xiiiConsidering the purpose of the study and the role of taxation in oil products, we have used two different data sets. The first set consists of gasoline retail prices (excluding taxes) and the second one includes taxes (gross retail prices). We also converted the TRY-denominated prices to US Dollar in order to eliminate the role of exchange rate changes on pricing (see Alper and Torul, 2009). The exchange rate data, for the conversion, is taken from Turkish Central Bank.xivFor the sake of consistency and comparison, we have converted the unit of the original series from barrel to liter.xvThe daily prices are the average prices of eight distribution companies with the highest market share in İstanbul (the European Side) and are obtained from EMRA’s annual reports. xviSee the previous footnote.xvii These variables have unit roots. Formal (ADF) tests are available upon request from the authors.xviii Akaike Information Criterion (AIC) suggested that two-lagged changes, as shown in Table 2, are adequate.xix As shown in the first column of Table 2, the t statistic for the coefficient of equals to -6.488. Therefore, the Engle-Granger test indicates that the two price series are cointegrated at 1 % level. Φ (F) statistics, Enders and Siklos (2001), indicate that two series are cointegrated in all other four models [Note that the critical values from Enders and Siklos (2001) are used]. xx According to the Breusch-Godfrey tests, the residuals are not serially correlated in all the error-correction models reported in this paper. Nevertheless, the Newey-West standard errors are used for the t-tests.xxi Wald test results (not reported), as explained in Section 3.1, do not provide empirical evidence on asymmetric pricing in the short-run. Test results can be provided upon request from the authors.xxii The null hypothesis of symmetric speed of adjustment is not rejected in both error-correction models based on M-TAR and Consistent M-TAR models (p-values are 0.67 and 0.30, respectively).xxiii As in the previous case (without taxes) two-lagged changes are adequate based on the AIC.xxiv Again, upon request, test results can be provided from the authors.xxvIn this case, β1 is expected to be 1+effective tax rate over the long-run. In fact, 62% is very close to the arithmetic average of effective tax rate on gasoline, 65.3%, from 2005 to 2013 (Table 1).