Price Adjustment in Taiwan Retail Gasoline Market

This paper uses weekly data over a sample research period of 2002M4 2011M11 to estimate the impact of crude oil price on pre-tax retail gasoline price in Taiwan. We found that there is a significant, long-run equilibrium relationship between crude oil price and retail gasoline price. In the asymmetric ECM framework, this paper finds that there was distributed lag effect symmetry (DLES) between oil price and retail gasoline price. By the cumulated adjustment function, we show that retail gasoline price in Taiwan respond more quickly to reductions in crude oil price.


Introduction
A high degree of fluctuation in crude oil prices during 2007-2009 had a powerful impact on gasoline prices worldwide.For example, in the period spanning Jan 2007 to July 2008, the pre-tax retail price of 95 octane unleaded gasoline in Taiwan increased from NT3990 to NT5740 per barrel.By December 2008, the price had plummeted to NT3362/barrel, before sharply increasing to NT4928/barrel (December 2009).The public is extremely sensitive to changes in the retail price of gasoline; therefore, these fluctuations sparked considerable controversy regarding the pricing policies of petroleum companies.Galeotti, Lanza, & Manera (2003) described the reliance of modern society on the flexibility and mobility afforded by motor vehicles, which makes the demand for gasoline relatively inelastic.While the public reacts favorably to any reduction in gasoline prices, increases are met with strong displeasure.Whether in Taiwan or abroad, public opinion is generally of the mind that an increase in oil prices leads directly to an increase in gasoline prices, while reduction in oil prices correspond to a slower decline in the prices of gasoline.(Note 1) Price fluctuations in the gasoline market described above represent price asymmetry, which has long been a topic of significance to both economists and the general public (see, for example, Manning, 1991;Borenstein, Cameron, & Gilbert, 1997;Eltony, 1998;Reilly & Witt, 1998;Godby, Lintner, & Wandschneider, 2000;Bachmeier & Griffin, 2003;Radchenko, 2005;Radchenko & Shapiro, 2011).
Figure 1 shows the price of crude oil (per barrel) in NTD.Dubai Crude and Brent Crude comprise the majority of crude oil imported in Taiwan.As Taiwan's main petroleum company, CPC Corporation bases its adjustment of gasoline prices on changes in marker crude price.The marker crude price is Dubai and Brent Crude prices calculated at weights of 70% and 30%, respectively.The price shown in Fig. 1 was obtained through a conversion of the marker crude price described above from USD to NTD.The trend of gasoline price in Taiwan is largely similar to that of crude oil price; however, it does not show immediate adjustment to oil price fluctuations, but appears to exhibit asymmetry and lag (Bettendorf, van der Geest, & Varkevisser, 2003).Additionally, the relatively large fluctuations in crude oil prices after 2007 caused more frequent adjustments to gasoline prices in Taiwan.These phenomena appear consistent with the viewpoint of Radchenko (2005).
Price asymmetry reveals differing degrees of adjustment to output price in response to cost impact.It also shows lag and rigidity in price adjustment.Theoretically, the causes of this type of input-output price asymmetry are categorized as follows: the trigger or focal point pricing strategies of oligopolistic sellers (increase in oil prices diminishes profit on retail gasoline, which immediately drives gasoline prices up; however, a reduction in oil prices does not produce the same inhibitory effect); adjustment to storage and production costs; menu costs, and consumer search costs (Reagan & Weitzman, 1982;Thurman, 1998;Borenstein, 1991;Pindyck, 1993Pindyck, , 1994;;Ball & Mankiw, 1994;Borenstein & Sheperd, 1996;Damania & Yang, 1998).However, Borenstein & Shepard (2002) 1) shows the equilibrium relationship between variables of output price and costs.If there were a stable industry structure, changes in costs would not affect this equilibrium relationship (Johnson, 2002).A super-consistent coefficient estimator ( , ) can be obtained using OLS.
The retail price of gasoline in Taiwan is substituted for by the pre-tax retail price of 95 octane unleaded gasoline (average announced price).Crude oil price is indicated by marker crude price (in NTD).All prices are expressed as unit price per barrel, obtained from the website of the Bureau of Energy of Taiwan (http://www.moeaboe.gov.tw/oil102/).Information on the Taiwan-U.S. exchange rate was obtained from the Central Bank of Taiwan.Weekly data was collected over a sample research period of 2002M4 -2011M11 (sample size = 504).
If the price series in Equation (1) were I(1) series and showed a cointegrating relationship, the short-run dynamic model expressed in error correction form would be as follows: where indicates the first difference; is the error term; measures the short-run impact of oil price fluctuation, and indicates the immediate effect of variation in oil price., ∀ 1, ⋯ , denotes the distributed lag effects of oil price variation; measures the short-run impact of lagged gasoline prices; is the error correction term, and is the adjustment coefficient of long-run equilibrium.
The ECM tells us that if crude oil price were unchanged and long-run equilibrium between gasoline and oil prices was attained, then there would be no further change to gasoline price.measures the long-run equilibrium relationship of permanent change to the price of oil.Even if asymmetric adjustment responses are plausible, the long-run cointegrating relationship between gasoline and oil prices must be identical for price increases or decreases (Bachmeier & Griffin, 2003).
To explore the asymmetric short-run response to price changes, we must now extend the basic ECM to an asymmetric ECM (Granger & Lee, 1989), as shown below: The above equation differentiates changes in oil prices and the error correction terms as positive and negative variations.∆ is defined as ∆ , 0 and ∆ as ∆ , 0 ; and are also similar definitions.Equation (3) retains the basic concept of ECM but allows for more flexible adjustment of gasoline price in response to oil price.
Table 1 shows the annual number of adjustments to retail gasoline prices in Taiwan.During the sample period of this study, retail gasoline prices were adjusted a total of 224 times (123 of these adjustments were increases in price (55%) while 101 were reductions (45%)).Retail gasoline prices were not adjusted on a weekly basis (for example, prices were adjusted roughly once every 3.55 weeks at 2002 on average); particularly during the pre-2006 period, prices were adjusted roughly once every 5.16 weeks.Following 2007, however, price adjustment occurred approximately once every 1.45 weeks.Table 2 shows the distribution of adjustments (by value) to retail gasoline prices in Taiwan.It is evident that most price adjustments were on a relatively small scale ( 22).

Empirical Findings
To analyze the symmetrical (Equation 2) and asymmetrical (Equation 3) pass-through effects of retail gasoline price caused by oil price shocks, we first determined whether there was a stationary equilibrium relationship between and .If and were integrated of order one and exhibited a cointegrating relationship, this would imply an equilibrium relationship between and .This facilitated the construction of an ECM (Engle & Granger, 1987).The results of ADF (augmented Dicky-Fuller) and PP (Phillips-Perron) unit root tests (including both the constant term and the time trend) showed that the null hypothesis with a unit root was not rejected for either or .After obtaining the first-order difference for the variables, we applied the same test process and found that the null hypothesis with a unit root was significantly rejected for both and .The test results are as shown in Table 3, which indicate that and are I(1) series..956***The autoregression models include both constant term and time trend, and the optimal lags are determined using AIC (maximum lags = 12).The ADF test and PP test are based on the null hypothesis of a unit root.***, **, and * indicate that the null hypothesis is rejected at the 1 %, 5 %, and 10 % significance levels.
Next, we used OLS to estimate Equation (1) and applied ADF and PP unit root tests to the residual.The results strongly rejected the null hypothesis with a unit root, indicating a cointegrating relationship between and .The results of the Johansen cointegrating tests also showed that, with regard to both trace eigenvalue and maximum eigenvalue statistics, the null hypothesis of no conintegrating relationship was rejected (see Table 4).This outcome supports the results of the unit roots to the residual.Equation (1) implies that fluctuation in crude oil price lead to changes in the retail gasoline price and not vice versa.The results of Granger causality tests suggest the null hypothesis that the crude oil price does not Granger cause the retail gasoline price was rejected.However, the null hypothesis of that the retail gasoline price does not Granger cause the crude oil price was not rejected.2.043 1.The optimal lags of VAR system are determined using AIC (maximum lags = 12).2.
refers to the trace eigenvalue statistics; refers to the maximum eigenvalue statistics; and r in the null hypothesis refers to the number of cointegration relationships in the VAR system; and ***, **, and * indicate that the null hypothesis is rejected at the 1 %, 5 %, and 10 % significance levels.
Under the premise that and are I(1) series and have a cointegrating relationship, we were able to construct the ECM for Equation (2) or Equation ( 3).We utilized Akaike information criterion (AIC) to determine the optimal lags of m and n in Equation ( 3), and used Schwarz criterion (SC) as the basis to identify minimum lags.The purpose was to avoid an overly short lag phase, which would prevent full expression of the data form.Under the premise of maximum lags = 13 (a quarter), we set m=6 and n=9.
Table 5 shows the estimation results of the cointegrating relationship, and demonstrates that there is a significant long-run equilibrium relationship between retail gasoline price and crude oil price.The long-run pass through effect of oil price into retail price is 0.563.When oil price increased by 1%, which will be passed 0.563% into retail gasoline price.Table 6 shows the coefficient estimation results for Equations ( 2) and (3).The standard deviation of the coefficients was calculated using Newey-West HAC covariance matrix estimation.With regard to both symmetric and asymmetric ECM, it is evident that the immediate effect of oil price shocks is not significant.In the symmetric ECM, oil price shocks showed significant distributed lag effects (lags of 1, 2, 7, and 9).In asymmetric ECM, positive oil price shocks showed significant distributed lag effects in lags of 1 and 7; negative oil price shocks showed significant distributed lag effects in lags of 1, 2, and 7.Although oil price shocks appear to have an asymmetric pass-through effect, positive shocks do not necessarily produce a greater response.The adjustment coefficients of error correction were estimated to have negative values, which imply that the system is converging to equilibrium.Negative disequilibrium, however, was responded to with more rapid adjustment.

Gasoline Price Asymmetries
Frey & Manera (2007) proposed a clear definition and categorization of price asymmetries, which we have interpreted using Equation (3) as follows: 1. and measure the contemporaneous impact of ∆ and ∆ on .Therefore, if test results reject : , this is defined as contemporaneous impact asymmetry (COIA); the opposite is defined as contemporaneous impact symmetry (COIS).

If test results reject :
, ∀ 1, ⋯ , , then this indicates the existence of distributed lag effect asymmetry (DLEA); if the null hypothesis is not rejected, then this indicates distributed lag effect symmetry (DLES).
3. The third type of price asymmetry is the cumulated impact asymmetry (CUIA) of ∆ and ∆ with regard to (in the past n periods).If the test results reject : ∑ ∑ , this type of price asymmetry is known as CUIA; if the null hypothesis is not rejected, the phenomenon is termed cumulated impact symmetry (CUIS).Interestingly, COIS and DLES were established as sufficient but non-required conditions for CUIS.The simultaneous establishment of COLA and DELA does not necessarily imply CUIA or CUIS.

Because
and respectively measure the adjustment speed when 0 and 0, if the test results reject : λ λ , this is defined as equilibrium adjustment path asymmetry (EAPA); the reverse is defined as equilibrium adjustment path symmetry (EAPS).
Finally, although the coefficients of the error correction term are labeled adjustment speeds, the actual paths of adjustment are also determined by other coefficients in the model.In other words, when ∆ or ∆ occurs, the cumulative adjustment function must be used to calculate the scale of cumulated adjustment to gasoline price.When crude price is assumed to increase by 1% at time t, then indicates the cumulated adjustment process of the retail gasoline price at time t+i.The cumulative adjustment function can be expressed as follows:

∑
The cumulated adjustment process involved in the reduction of oil prices is similar to Equation (4).The cumulated adjustment function measures the persistent influence of increases (or decreases) in oil price on gasoline price.0.051 1.The optimal lags of asymmetric ECM are determined using AIC (maximum lags = 12).***, **, and *, which indicate significance at the 1 %, 5 %, and 10 % levels.Since it is impossible to eliminate the presence of correlated and heterogeneous variable in the residuals, the standard derivation is calculated using the Newey-West HAC covariance matrix estimation.2. Data are weekly, spanning from 2002M4 to 2011M11 (sample size = 504).The data have been obtained from Bureau of Energy and Central Bank of Taiwan.
Table 7 shows the test results of price asymmetries (COIA, DLEA, CUIA, and EAPA).Regarding the adjustments of gasoline price in response to oil price changes, the null hypothesis of DLES was rejected at a 10% significance level.The other null hypotheses relating to the existence of COIS, CUIS, and EAPS were not rejected, even at a 10% significance level.In other words, the gasoline price only shows DLEA.
Table 8 shows the process involved in the cumulated adjustment of gasoline price in response to changes in crude oil price.Oil price shocks were categorized as symmetric, positive, and negative.The effects of symmetric oil price shocks on retail gasoline price are calculated from symmetric ECM (equation ( 2)).It is interesting to note that in the first two weeks, gasoline price was adjusted more rapidly in response to positive oil price shock.After the third week, gasoline price was adjusted more rapidly in response to negative oil price shocks.Maximum asymmetry (difference of 0.22%) was evident after the sixth week.When oil price increased by 1%, gasoline price increased by approximately 0.61% after a quarter.Conversely, when oil price declined by 1%, gasoline price decreased by approximately 0.69% after a quarter.This shows that gasoline price in Taiwan respond more quickly to reductions in crude oil price.We have also illustrated , , and in Figure 2, demonstrating this unique phenomenon.

Concluding Remarks
This paper collects weekly data over a sample research period of 2002M4 -2011M11 to estimate the impact of oil price on pre-tax retail gasoline price in Taiwan.In this study, the crude oil price is substituted by marker crude price, which is Dubai and Brent Crude prices calculated at weights of 70% and 30%, respectively.We found that there is a significant, long-run equilibrium relationship between crude oil price and retail gasoline price.In the asymmetric ECM framework, the test results rejected the null hypothesis of distributed lag effect symmetry (DLES) between oil price and retail gasoline price in the short-run.However, there is a lack of clear evidence to prove the existence of contemporaneous impact asymmetry (COIA), cumulated impact asymmetry (CUIA), and equilibrium adjustment path asymmetry (EAPA).On the cumulated adjustment to gasoline price, it is interesting to note that in the first two weeks, gasoline price was adjusted more rapidly in response to positive oil price shock.After the third week, gasoline price was adjusted more rapidly in response to negative oil price shocks.Maximum asymmetry (difference of 0.22%) was evident after the sixth week.When oil price increased by 1%, gasoline price increased by approximately 0.61% after a quarter.Conversely, when oil price declined by 1%, gasoline price decreased by approximately 0.69% after a quarter.This shows that gasoline price in Taiwan respond more quickly to reductions in crude oil price.In Taiwan, CPC Corporation is the market leader with a market share of 75%.But CPC Corporation is also attached to the Ministry of Economic Affairs in Taiwan.This would explain why the adjustment of Taiwan retail gasoline prices in response of shocks to oil price is "politico-economic asymmetry".
felt that the international crude oil market is an efficient open market, without factors such as menu costs or incomplete information.(Note 2)

Table 1 .
Number of Taiwan Retail GasolinePrice Adjustments There are 39 weeks and 48 weeks in 2002 and 2011, respectively.Retail gasoline price is average commended retail price (before taxes).The data have been obtained from Bureau of Energy of Taiwan and this study.

Table 2 .
Distribution of the Size of Taiwan Retail Gasoline Price Adjustments (x%) Data are weekly, spanning from 2002M4 to 2011M11.There are 39 weeks and 48 weeks in 2002 and 2011, respectively.Retail gasoline price is average commended retail price (before taxes).The data have been obtained from Bureau of Energy of Taiwan and this study.

Table 3 .
Unit Root Tests

Table 4 .
Johansen Cointegrating Tests on Oil Price and Retail Gasoline Price

Table 6 .
Error Correction Models