- Email: [email protected]
Volatility spillover from world oil spot markets to aggregate and electricity stock index returns in Turkey
Applied Energy 88 (2011) 354–360
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
Volatility spillover from world oil spot markets to aggregate and electricity stock index returns in Turkey Ugur Soytas *, Adil Oran 1 METU, Dept. of BA, 06531 Ankara, Turkey
a r t i c l e
i n f o
Article history: Received 8 April 2010 Received in revised form 8 July 2010 Accepted 16 July 2010 Available online 13 August 2010 Keywords: Oil price Stock market returns Volatility spillover Electricity index returns Emerging market Cheung–Ng procedure
a b s t r a c t This study examines the inter-temporal links between world oil prices, ISE 100 and ISE electricity index returns unadjusted and adjusted for market effects. The traditional approaches could not detect a causal relationship running from oil returns to any of the stock returns. However, when we examine the causality using Cheung–Ng approach we discover that world oil prices Granger cause electricity index and adjusted electricity index returns in variance, but not the aggregate market index returns. Hence, our results show that the Cheung–Ng procedure with the use of disaggregated stock index returns can uncover new information that went unnoticed with the traditional causality tests using aggregated market indices. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction The recent rise and subsequent fall in world oil prices has renewed the interest on impacts of oil shocks on the economy and financial markets as also stated by Chen et al. [2]. Policy makers are mostly concerned about the short run and the long run effects on the macroeconomy. Whereas, investors try to understand how the financial asset returns will respond to these shocks and whether the impacts are permanent or transitory. Several studies have examined the relationship between oil prices and macroeconomic and financial variables for both developed and developing countries. Oil price changes are generally found to have significant effects both on the economy and financial markets. Some studies focus on the impact of oil shocks on stock markets, where academicians and practitioners alike have been trying to understand the dynamic link between world oil prices and stock returns. The ‘‘traditional” view holds that as oil prices rise they will cause input prices to increase, driving down profits and returns. Obviously, these effects may be drastically different for firms (or industries or even countries) that are in businesses that may benefit from higher oil prices. However, as the majority of firms in most markets tend to be oil consumers, the effects of increasing oil prices are expected to be negative for the majority. Although there is support for the negative impact of oil price increases on stock market per* Corresponding author. Tel.: +90 312 2102048; fax: +90 312 2107962. E-mail addresses: [email protected] (U. Soytas), [email protected] (A. Oran). 1 Tel.: +90 312 2102041; fax: +90 312 2107962. 0306-2619/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.apenergy.2010.07.018
formance in general, the results are not unanimous and differ across countries. The fluctuations in world oil prices naturally bring forth a number of questions regarding the stock market returns. For example, are oil price shocks transmitted to the stock markets of open economies? One expects to observe a positive shock in world oil prices providing a downward pressure on financial asset returns. If a transmission occurs, then how long it lasts is the natural extension of the question in concern. Another interesting outcome would be the lack of transmission, which would then be followed by a search for the reasons for the neutrality of stock returns to oil price changes. One may be tempted to question the country specific characteristics such as the degree of openness in the presence of this neutrality. However, this explanation does not seem satisfactory in a globalized world, at least for most countries. A following question is: Is there volatility spill over between financial markets and energy markets? As markets become more integrated [6] and global investors view alternative markets as possible arenas of speculative behavior, one expects significant volatility transmissions across these markets. Another question is: Do different industry stocks (for example utilities vs. oil and gas stocks) respond differently to oil price shocks? Industries that are closely related to energy markets may be more sensitive to fluctuations in world oil markets than other industries. Hence, the responses of different industries may have an offsetting effect in the stock index return behavior if one focuses on the aggregate stock market index returns. Another plausible explanation may be the inadequacy of the analytic approach used in uncovering the effect of an oil shock.
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
The introduction of powerful time series techniques allowed examination of temporal relationships via creation of large models with many lags. Since the interpretation of lagged variable coefficients poses a problem, in the literature many have referred to the Granger causality tests as well as impulse response and variance decomposition analyses (which are viewed as out of sample Granger causality tests) in order to interpret the results. If lags of a variable X improve the forecasts of another variable Y, then it is said that X Granger causes Y. In the context of the oil price-stock market prices literature, one expects to observe that oil price changes lead the prices of financial instruments, but not vice versa, especially in small markets relative to the world markets. The empirical results do not always support this expectation. There may be several reasons ranging from the problems related to the methodologies used to the characteristics of the financial market under investigation. The lack of a causal link running from oil price changes to financial assets may also arise because the dynamic link may be between the variances of the variables rather than between the variables themselves. Furthermore, the lack of Granger causality in mean does not necessarily imply that there is also non-Granger causality in variance. Hence, in order to achieve a full examination of the dynamic links between oil and stock returns, one should not limit his/her attention to mean spillovers. This paper fits the literature on the impact of oil price changes on stock returns; however, distinguishes itself from the rest of the literature by examining the impact of oil shocks on specific electricity index returns in an emerging economy. To that respect, we first investigate the temporal relationship between oil prices and Istanbul Stock Exchange 100 index (ISE100) returns. Then, we follow a recent trend in the literature and extend our analysis to a disaggregate level, specifically to the link between world oil prices and the electricity index returns in ISE. The stocks of companies operating in energy markets are expected to be more sensitive to world oil price changes. Furthermore, another issue that we explore in this study is that the information transmission may not be directly between the returns themselves but may be in terms of volatility spillovers. We show that world oil markets have a significant contemporaneous impact on both the ISE100 and ISE electricity returns. However, the test results based on traditional tests failed to detect any Granger causality relationships between world oil prices and stock returns in Turkey. The generalized impulse responses confirm these results, since no significant impact of oil price shocks is observed. This came as a surprise because Turkey is a net oil importer and one expects to see the impact of oil price shocks on the ISE or at least on the electricity index. When we proceed to the Granger causality in variance tests proposed by Cheung and Ng [3], we find that there is volatility spillover from world oil spot markets to the electricity stock index returns in Turkey. Hence, use of disaggregated index returns and looking for volatility spillovers seem to uncover the link between oil prices and financial asset returns which has been undetected using more traditional methods. Our results provide important implications for investors and interesting insights for further research. The next section introduces the literature review on the relationship between stock returns and oil price. Then in Section 3, we discuss data characteristics and methodology followed by empirical results in Section 4. Section 5 concludes and provides implications for investors, policy makers and further research.
2. Literature review The early literature on the macroeconomic effects of oil price shocks has been led by Hamilton [8] who shows that ‘‘oil shocks were a contributing factor in at least some of the US recessions
355
prior to 1972.” Furthermore, he states that almost all US recessions after World War II were preceded by large increases in oil prices. The following literature has extended into examining the impact of oil shocks on other economies and other macroeconomic variables (see for example [32,16,22,30,31]). Within this vast literature, there are also a quite a number of studies on how oil price changes influence the stock market returns. Along this line, most of the studies are on developed countries [7,26,27,24], but there is an increasing interest on developing countries [23,10,11,1,29] as well. Furthermore, most of the earlier work on stock indexes has usually focused on how a level shift in oil prices influences the aggregate stock market returns. Recently, however, the literature is extending into a disaggregate analysis of oil price and industry and/or individual stock returns [27,9,11,28,4,21]. Another line that the literature takes is how oil price volatility influences the volatilities in stock markets. For example, Sadorsky [26] employs monthly data for 1947– 1996 to investigate the interaction between oil prices and economic activity. He finds that oil prices and oil price volatility affect real stock returns, though the effects may not be symmetric. Additionally, the dynamics of the relation seem to have changed after 1986 with the explanatory power of oil price shocks with regard to real stock returns rising and exceeding that of interest rates. In a disaggregated study, Hammoudeh et al. [9] use daily data for 1995–2001 to study the dynamic relationship between five US oil sector indices and five different oil prices. They find that oil prices are cointegrated amongst themselves but oil sector indices are not. For mixed systems (oil sector indices plus one oil price) only one cointegrating equation is found. Their results show that oil prices can be used to predict changes in some oil sector indices. Furthermore, they also report significant volatility spillovers from oil prices to the stock indices. However, the effect is to increase the volatility in some indices, while decreasing it in others. Another disaggregated analysis is due to Sadorsky [27] where he uses a multifactor model to estimate expected returns for the Canadian oil and gas industry on a monthly basis between 1983 and 1999. He finds that exchange rates, oil prices and interest rates, in addition to the usual market factor, all affect Canadian oil and gas industry stock returns. Increases in oil prices lead to increases in Canadian oil and gas industry stock returns. This study focuses on energy industries in Canada only, but there are other works that examine the link between oil markets and other related industries. For example, Sadorsky and Henriques [28] use weekly data for 2001–2007 in a LA-VAR model to examine the relationship between alternative energy company stock prices, technology company stock prices, oil prices and interest rates in the US. They find that oil prices, technology stock prices and interest rates Granger cause alternative energy stock prices, with technology stock price shocks having a larger impact. This is interpreted as alternative energy stocks behaving more like technology stocks. The literature also includes studies on several other countries including emerging markets. Basher and Sadorsky [1], for example, employ daily and monthly data to study 21 emerging markets (including Turkey, which contains the highest market risk among all) in an international CAPM model. Their results show a significant negative impact of oil prices on stock market returns. They argue that the results are sensitive to the frequency of the data and the world CAPM model may not hold for all countries. In their work there is also evidence of a nonlinear and asymmetric conditional relationship between oil price risk and stock returns. In another study, taking into account economic activity and employment, Papapetrou [23] finds that oil price shocks affect stock returns in Greece. Narayan and Narayan [18] report the theoretically unexpected positive impact of oil price shocks on Vietnamese stock prices for the 2000–2008 period, using daily data and accounting for the nominal exchange rate.
356
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
Park and Ratti [24] use monthly data for 1986–2005 in a multivariate VAR analysis to examine the effects of oil price shocks and volatility on the stock returns of the US and 13 European countries. They find significant impact of oil price shocks on real stock returns in all countries. They report that in many European markets, but not in the US, higher volatility has a tendency to depress real returns. Finally, the results show that many effects differ across countries. Hammoudeh et al. [9] use daily data for 1994–2001 to examine the relationship between oil prices and five oil exporting Gulf Coast Country stock markets. The study finds that only the Saudi Arabian index seems to have a significant relationship with oil prices and the relationship is bi-directional. Hammoudeh and Li [11] use daily data for 1986–2003 and find that oil prices lead the Norwegian and Mexican stock markets as well as US oil sensitive stock indices they examine. Unfortunately, there are only a small number of studies that investigate the impact of oil prices on the real economy as well as the stock market in Turkey. Among these, Kibritcioglu and Kibritcioglu [13] examine the oil price and inflation link and suggest that innovations in oil prices do not affect the price level in Turkey. Sari and Soytas [29] study the relationship between stock returns, crude oil prices, interest rates and output in Turkey using monthly data for the period 1987:01–2004:03. They find that oil prices do not have any significant effect on stock returns. To the extent of our knowledge, there are no studies that examine the Granger causality relationship between world oil prices and ISE returns as well as the electricity index returns using daily data. Furthermore, this study is probably the first to examine the volatility spillover from the world oil spot markets to the electricity stock returns in Turkey. In these respects, the paper aims to contribute to the literature on the effects of oil price shocks and volatilities on disaggregated stock returns in emerging markets. 3. Data and methodology We use daily data on ISE100 index, ISE electricity index, and spot oil price for the period 05/02/2003–03/01/2007. The data range is limited by both the availability of the electricity index and the need to avoid the impacts of the 2001 crisis. The ISE100 and electricity index are retrieved from the ISE. The electricity index is composed of four electricity generating firms. The daily exchange rate between US Dollar and Turkish Lira is obtained from the Central Bank of the Republic of Turkey. The Bloomberg European Dated Brent Forties Oseberg Ekofisk (BFOE) Spot Price reflects the average value of Brent contracts for differences over the 10–21 day period. The natural logarithms of all data are arranged in 5 day weeks and all holidays are removed. ELIN, ISE100, and BRENT represent the first differences of the logged data and hence represent continuously compounded growth rates/returns. Therefore, ELIN is the electricity index return, ISE100 is the Istanbul Stock Exchange 100 index return, and BRENT is the oil return. We extract the impact of the stock market from the electricity index using a simple market model.
ELINt ¼ a þ bISE100t þ et
ð1Þ
where et is then the return on the electricity index with market effects removed (AELIN). AELIN therefore represents the market adjusted electricity index returns. This provides us with returns that are peculiar to the electricity index, where the effects of factors affecting the general market have been removed.We first investigate the time series properties of the series in concern. Table 1 summarizes the data characteristics. Table 1 presents evidence on non-normality of the series both via the Jarque–Bera statistics and excess kurtosis. Narayan and
Wong [19] argue that although conventional unit root tests are frequently conducted in energy economics literature, the results are not unanimous regarding several energy variables, including oil prices. Furthermore, Maddala and Kim [17] argue that conventional unit root tests have low power. Therefore we use GLS detrended and Ng and Perron’s [20] Z alpha tests that exhibit better power properties. It is also clear from the unit root tests in Table 1 that all returns are stationary in levels. Then we specify the VAR systems for both ELIN and AELIN separately, but including the USD-TRL exchange rate and ISE100 in both systems, as follows:
V t ¼ av þ b1 V t1 þ b2 V t2 þ þ bk V t k þ ev t
ð2Þ
where k is the lag length determined via examining five different information criteria (final prediction error (FPE), Akaike information criterion (AIC), Schwarz information criterion (SIC), Hannan-Quinn (HQ) information criterion, and likelihood ratio test), Vt = (ELINt or AELINt, ISE100t, BRENTt, ERt), ER is the exchange rate, and et are the white noise residuals. In addition to the block exogeneity tests we also examine the generalized responses of the adjusted and unadjusted electricity index returns to impulses in oil prices. Note that the generalized approach developed by Koop et al. [14] and Pesaran and Shin [25] overcomes the orthogonality problem associated with the standard impulse response analysis. Hence, ordering of variables in the VAR system does not change the results. Next we consider the Granger causality in mean and variance approach developed by Cheung and Ng [3]. They propose estimating univariate GARCH models for the stationary variables in order to get the conditional means (lt) as well as the conditional variances (r2t ). The strength of the procedure is that it does not necessitate the use of complicated MGARCH models that suffer from the dimensionality problem (Engle and Sheppard [5]). In order to overcome problems with using GARCH models for financial data, we employ the exponential GARCH (EGARCH) model. Hence, there is no need to impose restrictions to ensure positive variances. Furthermore, the asymmetric effects of positive and negative shocks to returns on volatility are also accounted for. The EGARCH model’s mean and variance equations are as follows:
Z t ¼ lt þ ut ; ut Nð0; r2t Þ
ð3Þ 2
logðr2t Þ ¼ x þ b log
r2t1
ut1 6 jut1 j þ c qffiffiffiffiffiffiffiffiffi þ a4qffiffiffiffiffiffiffiffiffi
r2t1
r2t1
3 rffiffiffiffi 27 5
p
ð4Þ
where zt denotes the stationary series, lt is a constant, ut are normally distributed error terms whose conditional variances are given by Eq. (2). The model is exponential since the log of the conditional variance is modeled in the variance equation. This ensures that the forecasts of the variance are non-negative. There is asymmetric impact (leverage effect) if c is not zero. Note that the choice of the mean function in (1) is made based on some information criteria via comparison of alternative models with AR and/or MA terms. ^ it ; are obtained ^ it Þ=h The standardized residuals, ^eit ¼ ðzit l from the EGARCH models, where zit represent the stationary series ^ it are the estimates of r . Then the sample residual cross-corand h it relation functions between the two standardized residuals ^ u1 u2 ðkÞ) are derived. The Granger causality in mean test is then (q pffiffiffi ^ u1 u2 ðkÞ that asymptotically follows the based on the statistic T q standard normal distribution, where T is the sample size. For the Granger causality in variance test the squared standardized resid^2 are obtained and the sample residual cross^ it Þ2 =h uals ^e2it ¼ ðzit l it correlation functions between the squares of the two standardized pffiffiffi ^ v 1 v 2 ðkÞ) are derived. Again the test statistic T q ^ v 1 v 2 ðkÞ residuals (q follows the normal distribution asymptotically. For the derivation of the sample cross-correlation functions see Cheung and Ng [3] and Inagaki [12].
357
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360 Table 1 Data summary statistics for log returns.
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque–Bera DF-GLS1 DF-GLS2 NP-Za1 NP-Za2
ISE100
ELIN
AELIN
BRENT
ER
0.001307 0.001905 0.100954 0.086708 0.018023 0.048697 5.209295 195.4151a 2.820423a (9) 4.204168a (9) 9.25116b (9) 15.1791c (9)
0.00017 0.000263 0.103023 0.10216 0.020735 0.04854 6.150185 396.9097a 7.193349a (6) 26.98017a (0) 59.5143a (6) 470.244a (0)
7.87E19 0.001143 0.087719 0.072479 0.016578 0.874222 7.487222 926.7223a 29.00875a (0) 23.50248a (1) 477.025a (0) 583.009a (1)
0.000991 0.001121 0.074127 0.085546 0.02004 0.015212 3.547021 11.99381a 31.66426a (0) 31.67747a (0) 478.740a (0) 478.731a (0)
9.57E05 0.000907 0.047744 0.02775 0.007785 0.951178 7.292198 880.7583a 1.143224 (8) 6.834642a (4) 11.1635b (8) 194.139a (4)
Lag lengths are determined via SIC and are in parentheses. DF-GLS1 and 2 represent the Dickey–Fuller GLS detrended unit root test statistics with intercept and with both intercept and trend respectively. NP-Za1 and NP-Za2 are the Ng–Perron Z alpha unit root test statistics with intercept and with both intercept and trend respectively. For a detailed description of DF-GLS see Maddala and Kim [17], and for NP see Ng and Perron [20]. a Significance at 1%. b Significance at 5%. c Significance at 10%.
4. Results and discussions Having established the stationarity of returns and selected the optimal lag length we check the stability of the VAR systems. The VAR systems both satisfy the stability condition in that all roots are within the unit circle. In order to overcome the serial correlation and heteroscedasticity problems observed in diagnostic analyses, we use the Newey–West adjusted standard errors for further analysis. All diagnostic results and Newey–West adjusted standard errors are available upon request. In Table 2, Panel A gives the test results for the VAR system with ELIN and Panel B with AELIN. When we follow the traditional block exogeneity tests based on the stable VAR systems, we find that there is no Granger causality running from oil spot returns to either stock index or to electricity index returns. When we examine both panels in Table 2 we see that the lack of Granger causality from oil returns to other indices persists when we use the adjusted electricity index. The block exogeneity tests reveal Granger causality running from electricity index returns to the oil returns. This causality also persists (and is more prominent) when we use the adjusted electricity index. This result may appear as unexpected at first because the Turkish economy is not expected to influence the global energy markets. However, the explanation that stock returns can improve the forecasts of oil returns may not be too far fetched. World oil prices uni-directionally Granger cause the exchange rate. This is
as expected since world oil trade is conducted in US dollars, and increases in oil prices raise the demand for dollars changing the dynamics in the foreign exchange market. An oil price shock may depress oil demand and therefore the demand for dollars (we thank an anonymous referee for pointing this out). Furthermore, the unadjusted electricity returns lead the exchange rate while ISE100 does not (Panel A), but in Panel B the adjusted electricity returns do not have any impact while ISE100 is driving the ER. The block exogeneity test results are confirmed via the generalized impulse response functions. The generalized impulse response
Table 2 Block exogeneity test results.
Panel A ISE100 ELIN BRENT ER
ISE100
ELIN
BRENT
ER
– 7.070401a 7.649358b 0.462923
2.511115 – 5.010741c 150.6405a
2.672842 1.929749 – 7.911088b
1.003266 0.686409 0.5621 –
ISE100 – 3.403173c 0.31924 208.5967a
AELIN 1.976251 – 7.968952a 0.417049
BRENT 1.973536 0.028871 – 7.035089a
ER 0.354177 3.226693c 0.062962 –
Panel B ISE100 AELIN BRENT ER
Significance implies the column variable Granger causes the row variable. a Significance at 1%. b Significance at 5%. c Significance at 10%.
Fig. 1. Generalized impulse responses of ISE100 and ELIN returns to oil shocks.
358
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
graphs depicting responses of ISE100 returns along with the unadjusted electricity index and the adjusted electricity index are presented in Figs. 1 and 2, respectively. According to Fig. 1, neither ISE100 nor ELIN significantly responds to an innovation in spot oil price growth. The responses of ISE100 and AELIN are also insignificant as shown in Fig. 2. The results of the traditional tests are not in line with Sadorsky [27] results for Canada and Hammoudeh et al. [9] results for the US. Although the different results may be attributed to the fact that this study focuses on an emerging market instead of a developed market, the Basher and Sadorsky [1] results for emerging markets (including Turkey) do not comply with our preliminary findings. Therefore, in order to check the robustness of our results and to investigate any volatility spillover links, we next examine Granger causality in mean and in variance using the Cheung and Ng [3] methodology. We first identify the EGARCH univariate models for ISE100, AELIN, and BRENT. Based on the SIC, we select models with no autoregressive or moving average terms for each variable. The asymmetric natures of the conditional variance equations are verified for all series indicating that bad and good news trigger different market reactions. Hence a simple mean equation with an asymmetric EGARCH specification is applied. The standardized residuals for all equations are subjected to the BDS test and no evidence of remaining time dependence is observed. Table 3 gives estimates of the variance equations. According to the results in Table 3, the leverage effects are apparent due to significance of c. Since |b| is less than 1, the model is stationary. The extent of a shock on the conditional variance is captured by a and x is a constant term. Then following the Cheung and Ng [3] procedure we compute the test statistics for Granger causality in mean, which are summarized in Table 4.
Fig. 2. Generalized impulse responses of ISE100 and AELIN returns to oil shocks.
Table 3 EGARCH variance equations. Variance equations
x a c b
ISE100
AELIN
BRENT
0.673075a 0.175643a 0.066147a 0.9338a
1.493140a 0.300242a 0.083243a 0.845829a
5.716711a 0.006904 0.180920a 0.269753
logðr2t Þ ¼ x þ b log
jut1 j t1 ffiffiffiffiffiffiffi r2t1 þ c puffiffiffiffiffiffiffi þa p r2 r2 t1
a
t1
qffiffiffi 2 p .
Significance at 1%.
Table 4 shows that there is no Granger causality in mean linkages between the ISE100 index and oil prices; however, there is significant contemporaneous correlation between them. According to Table 4, oil returns do not Granger cause the electricity index returns, but electricity returns negatively Granger cause oil returns at first lag. This result persists when we examine the link using the adjusted electricity index returns. There is also weak evidence that oil returns positively Granger cause adjusted returns at lag 6. However, as we have no reason to expect an effect only at lag 6, it seems likely that this finding may be spurious. Further evidence or a good explanation may be required. These findings indicate that using disaggregated indices while studying the impacts of oil price shocks, rather than dealing with aggregate market indices, may reveal new information. When we examine the Granger causality in variance tests, whose results are presented in Table 5, we observe that the spot oil returns do not Granger cause the ISE100 returns in variance either. There is bi-directional Granger causality in variance from oil returns to the electricity index returns and vice versa. The volatility in world oil markets negatively Granger cause the volatility in unadjusted electricity returns at lag 2, with a positive feedback effect from unadjusted electricity index volatility to oil volatility at lag 4. The structure of the bi-directional causality in variance changes when adjusted electricity returns are used. Oil and electricity volatilities positively Granger cause each other at lag 4. The spillovers occurring at higher lags are probably due to delays in information processing and portfolio and risk position adjustments based on updated information by investors. The volatility spillover results confirm the mean spillover results in that using industry returns instead of aggregate returns helps uncover new information regarding the link between world oil prices and financial returns. Furthermore, in examining the information transmission between market prices, volatility spillover is a dimension that should not be overlooked. Therefore, our results show that Turkish stock market is not immune to oil price shocks as indicated by Sari and Soytas [29] who focus on returns only. We show that volatility in the world oil market has an impact on the energy stock volatility in Turkey. This means that investors cannot effectively diversify risk away by including Turkish energy stocks in their portfolios. The overall results of the study provide answers to questions posed in the introduction section. First of all, we show that focusing on the general stock index may not be adequate in identifying the link between world oil market and stock returns. The Turkish energy stocks are not unresponsive to developments in the world oil market. Additionally we also show that the link may not be limited to returns but volatility linkages must also be taken into account. There is volatility spillover from world oil market to Turkish electricity index returns. Hence, a more disaggregated approach including volatility spillover analyses uncovers new information that is useful to both policy makers and investors.
359
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360 Table 4 Granger causality in mean test statistics. i
GC in MEAN ISE100 and BRENT
0 1 2 3 4 5 6 7
ELIN and BRENT
AELIN and BRENT
Lag
Lead
Lag
Lead
Lag
Lead
1.787769b 1.12162 0.570103 0.45546 0.625874 0.189 0.71573 0.514332
1.787769b 0.433774 0.67855 0.529824 0.1921 0.06507 0.65066 0.619677
1.195977 0.79629 0.597989 0.26336 0.991484 0.610382 0.771498 0.340823
1.195977 2.08212b 0.34082 0.20759 0.13323 0.75291 0.17041 0.11464
0.297445 0.071263 0.439971 0.01859 0.886139 0.830368 1.341601c 0.07746
0.297445 2.85361a 0.399692 0.83656 0.3935 1.3354 0.567005 0.05267
BRENT Granger causes the first variable in mean if the test statistic is significant for some lags; vice versa if the test statistic is significant for some leads. a Significance at 1%. b Significance at 5%. c Significance at 10%.
Table 5 Granger causality in variance test statistics. i
GC in VARIANCE ISE100 and BRENT
0 1 2 3 4 5 6 7
ELIN and BRENT
AELIN and BRENT
Lag
Lead
Lag
Lead
Lag
Lead
0.065066 0.827269 0.52982 0.65066 0.41518 1.12471 0.84586 0.390397
0.065066 0.356314 0.170411 0.312937 0.678547 0.48335 0.622776 0.35322
0.31604 0.21069 1.31372b 0.03408 1.022468 1.48413 0.05887 1.264142
0.31604 0.05887 1.013172 0.32843 2.320692a 0.632071 0.42138 1.35709
0.20759 0.3873 1.01007 0.709531 2.094509a 1.7351 0 0.455463
0.20759 0.997681 0.848958 0.89543 1.648342a 0.309839 0.216887 1.3509
BRENT Granger causes the first variable in variance if the test statistic is significant for some lags; vice versa if the test statistic is significant for some leads. a Significance at 5%. b Significance at 10%.
5. Conclusions This paper investigates the Granger causality relationship between Istanbul Stock Exchange 100 index returns, ISE electricity index returns and world spot oil market returns. The block exogeneity results from the VAR systems imply that a unidirectional Granger causality exists running from Turkish electricity index returns to oil returns. This unexpected result may be due to the fact that the stock returns of energy companies in Turkey may be following the same dynamics with the stock returns of global energy companies that do have an impact on the world oil markets or possibly that it is anticipating future movements. This poses a new question, which can be answered in an analysis that includes both the Turkish and global energy companies’ stock returns. A similar explanation can be used for the bi-directional Granger causality in variance between oil and electricity stock returns. In this emerging economy, Turkey, we find evidence that world spot oil returns Granger cause the electricity returns in variance and vice versa, but not the ISE100 returns. Hence, the volatility spillover is limited to the electricity index. Our results show that the lack of Granger causality in mean, as documented in the literature by several papers as well as by this paper, does not necessarily imply that changes in world oil prices have no impact on financial returns in Turkey. Hence, our results emphasize that the use of industry (or maybe even firm level) returns to examine the influence of oil price changes may provide new insights. Furthermore, our results also show that investigating Granger causality in variance may reveal transmission mechanisms that went unnoticed by tests of Granger causality in mean. The new line of
literature on the effects of oil price changes on returns with disaggregated indices (or even with individual stocks) and variance causality will provide essential information to global investors and a promising venue for further research. A natural extension of our results would be to evaluate the impact of different energy prices (Lee and Lee [15]) as they can be viewed as separate markets with arbitrage opportunities.
References [1] Basher SA, Sadorsky P. Oil price risk and emerging stock markets. Glob Finan J 2006;17:224–51. [2] Chen S-T, Kuo H-I, Chen C-C. Modeling the relationship between the oil price and global food prices. Appl Energy, forthcoming. doi:10.1016/ j.apenergy.2010.02.020. [3] Cheung Y, Ng LK. A causality-in-variance test and its application to financial market prices. J Econometrics 1996;72:33–48. [4] Cong R-G, Wei Y-M, Jiao J-L, Fan Y. Relationships between oil price shocks and stock market: an empirical analysis from China. Energy Policy 2008;36:3544–53. [5] Engle RF, Sheppard K. Theoretical and empirical properties of dynamic conditional correlation multivariate GARCH. Working paper No. 8554, NBER; 2001. [6] Ewing BT, Malik F, Ozfidan O. Volatility transmission in the oil and natural gas markets. Energy Econ 2002;24:525–38. [7] Ewing BT, Thompson MA. Dynamic cyclical comovements of oil prices with industrial production, consumer prices, unemployment, and stock prices. Energy Polym 2007;35:5535–40. [8] Hamilton JD. Oil and the macroeconomy since world war II. J Polit Econ 1983;91:228–48. [9] Hammoudeh S, Dibooglu S, Aleisa E. Relationships among US oil prices and oil industry equity indices. Int Rev Econ Finan 2004;13:427–53. [10] Hammoudeh S, Aleisa E. Dynamic relationships among GCC stock markets and NYMEX oil futures. Contemp Econ Policy 2004;22:250–69.
360
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
[11] Hammoudeh S, Li H. Oil sensitivity and systematic risk in oil-sensitive stock indices. J Econ Bus 2005;57:1–21. [12] Inagaki K. Testing for volatility spillover between the British pound and the euro. Res Int Bus Finan 2007;21:161–74. [13] Kibritcioglu A, Kibritcioglu B. Inflationary effects of crude oil prices (in Turkish). General directorate of research working paper. Department of the Treasury; 1999. [14] Koop G, Pesaran MH, Potter SM. Impulse response analysis in nonlinear multivariate models. J Econometrics 1996;74:119–47. [15] Lee C-C, Lee J-D. Energy prices, multiple structural Breaks, and efficient market hypothesis. Appl Energy 2009;86:466–79. [16] Lorde T, Jackman M, Thomas C. The macroeconomic effects of oil price fluctuations on a small open oil-producing country: the case of Trinidad and Tobago. Energy Policy 2009;37:2708–16. [17] Maddala GS, Kim I. Unit roots, cointegration, and structural change. Cambridge: Cambridge University Press; 1998. [18] Narayan PK, Narayan S. Modelling the impact of oil prices on Vietnam’s stock prices. Appl Energy 2010;87:356–61. [19] Narayan PK, Wong P. A panel data analysis of the determinants of oil consumption: the case of Australia. Appl Energy 2009;86:2771–5. [20] Ng S, Perron P. Lag length selection and the construction of unit root tests with good size and power. Econometrica 2001;69:1519–54. [21] Oberndorfer U. Energy prices, volatility, and the stock market: evidence from the Eurozone. Energy Polym 2009;37:5787–95.
[22] Oladosu G. Identifying the oil price–macroeconomy relationship: an empirical mode decomposition analysis of US data. Energy Polym 2009;37:5417–26. [23] Papapetrou E. Oil price shocks, stock market, economic activity, and employment in Greece. Energy Econ 2001;23:511–32. [24] Park J, Ratti RA. Oil price shocks and stock markets in the US and 13 European countries. Energy Econ 2008;30:2587–608. [25] Pesaran MH, Shin Y. Generalized impulse response analysis in linear multivariate models. Econ Lett 1998;58:17–29. [26] Sadorsky P. Oil price shocks and stock market activity. Energy Econ 1999;21:449–69. [27] Sadorsky P. Risk factors in stock returns of Canadian oil and gas companies. Energy Econ 2001;23:17–28. [28] Sadorsky P, Henriques I. Oil prices and the stock prices of alternative energy companies. Energy Econ 2008;30:998–1010. [29] Sari R, Soytas U. The relationship between stock returns, crude oil prices, interest rates, and output: evidence from a developing economy. Emp Econ Lett 2006;5:205–20. [30] Sari R, Hammoudeh S, Soytas U. Dynamics of oil price, precious metal prices, and exchange rate. Energy Econ 2010;32:351–62. [31] Soytas U, Sari R, Hammoudeh S, Hacihasanoglu E. World oil prices, precious metal prices, and macroeconomy in Turkey 37 (2009) 5557–66. [32] Uri ND. Crude-oil price volatility and agricultural employment in the USA. Appl Energy 1996;54:355–73.
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
Volatility spillover from world oil spot markets to aggregate and electricity stock index returns in Turkey Ugur Soytas *, Adil Oran 1 METU, Dept. of BA, 06531 Ankara, Turkey
a r t i c l e
i n f o
Article history: Received 8 April 2010 Received in revised form 8 July 2010 Accepted 16 July 2010 Available online 13 August 2010 Keywords: Oil price Stock market returns Volatility spillover Electricity index returns Emerging market Cheung–Ng procedure
a b s t r a c t This study examines the inter-temporal links between world oil prices, ISE 100 and ISE electricity index returns unadjusted and adjusted for market effects. The traditional approaches could not detect a causal relationship running from oil returns to any of the stock returns. However, when we examine the causality using Cheung–Ng approach we discover that world oil prices Granger cause electricity index and adjusted electricity index returns in variance, but not the aggregate market index returns. Hence, our results show that the Cheung–Ng procedure with the use of disaggregated stock index returns can uncover new information that went unnoticed with the traditional causality tests using aggregated market indices. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction The recent rise and subsequent fall in world oil prices has renewed the interest on impacts of oil shocks on the economy and financial markets as also stated by Chen et al. [2]. Policy makers are mostly concerned about the short run and the long run effects on the macroeconomy. Whereas, investors try to understand how the financial asset returns will respond to these shocks and whether the impacts are permanent or transitory. Several studies have examined the relationship between oil prices and macroeconomic and financial variables for both developed and developing countries. Oil price changes are generally found to have significant effects both on the economy and financial markets. Some studies focus on the impact of oil shocks on stock markets, where academicians and practitioners alike have been trying to understand the dynamic link between world oil prices and stock returns. The ‘‘traditional” view holds that as oil prices rise they will cause input prices to increase, driving down profits and returns. Obviously, these effects may be drastically different for firms (or industries or even countries) that are in businesses that may benefit from higher oil prices. However, as the majority of firms in most markets tend to be oil consumers, the effects of increasing oil prices are expected to be negative for the majority. Although there is support for the negative impact of oil price increases on stock market per* Corresponding author. Tel.: +90 312 2102048; fax: +90 312 2107962. E-mail addresses: [email protected] (U. Soytas), [email protected] (A. Oran). 1 Tel.: +90 312 2102041; fax: +90 312 2107962. 0306-2619/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.apenergy.2010.07.018
formance in general, the results are not unanimous and differ across countries. The fluctuations in world oil prices naturally bring forth a number of questions regarding the stock market returns. For example, are oil price shocks transmitted to the stock markets of open economies? One expects to observe a positive shock in world oil prices providing a downward pressure on financial asset returns. If a transmission occurs, then how long it lasts is the natural extension of the question in concern. Another interesting outcome would be the lack of transmission, which would then be followed by a search for the reasons for the neutrality of stock returns to oil price changes. One may be tempted to question the country specific characteristics such as the degree of openness in the presence of this neutrality. However, this explanation does not seem satisfactory in a globalized world, at least for most countries. A following question is: Is there volatility spill over between financial markets and energy markets? As markets become more integrated [6] and global investors view alternative markets as possible arenas of speculative behavior, one expects significant volatility transmissions across these markets. Another question is: Do different industry stocks (for example utilities vs. oil and gas stocks) respond differently to oil price shocks? Industries that are closely related to energy markets may be more sensitive to fluctuations in world oil markets than other industries. Hence, the responses of different industries may have an offsetting effect in the stock index return behavior if one focuses on the aggregate stock market index returns. Another plausible explanation may be the inadequacy of the analytic approach used in uncovering the effect of an oil shock.
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
The introduction of powerful time series techniques allowed examination of temporal relationships via creation of large models with many lags. Since the interpretation of lagged variable coefficients poses a problem, in the literature many have referred to the Granger causality tests as well as impulse response and variance decomposition analyses (which are viewed as out of sample Granger causality tests) in order to interpret the results. If lags of a variable X improve the forecasts of another variable Y, then it is said that X Granger causes Y. In the context of the oil price-stock market prices literature, one expects to observe that oil price changes lead the prices of financial instruments, but not vice versa, especially in small markets relative to the world markets. The empirical results do not always support this expectation. There may be several reasons ranging from the problems related to the methodologies used to the characteristics of the financial market under investigation. The lack of a causal link running from oil price changes to financial assets may also arise because the dynamic link may be between the variances of the variables rather than between the variables themselves. Furthermore, the lack of Granger causality in mean does not necessarily imply that there is also non-Granger causality in variance. Hence, in order to achieve a full examination of the dynamic links between oil and stock returns, one should not limit his/her attention to mean spillovers. This paper fits the literature on the impact of oil price changes on stock returns; however, distinguishes itself from the rest of the literature by examining the impact of oil shocks on specific electricity index returns in an emerging economy. To that respect, we first investigate the temporal relationship between oil prices and Istanbul Stock Exchange 100 index (ISE100) returns. Then, we follow a recent trend in the literature and extend our analysis to a disaggregate level, specifically to the link between world oil prices and the electricity index returns in ISE. The stocks of companies operating in energy markets are expected to be more sensitive to world oil price changes. Furthermore, another issue that we explore in this study is that the information transmission may not be directly between the returns themselves but may be in terms of volatility spillovers. We show that world oil markets have a significant contemporaneous impact on both the ISE100 and ISE electricity returns. However, the test results based on traditional tests failed to detect any Granger causality relationships between world oil prices and stock returns in Turkey. The generalized impulse responses confirm these results, since no significant impact of oil price shocks is observed. This came as a surprise because Turkey is a net oil importer and one expects to see the impact of oil price shocks on the ISE or at least on the electricity index. When we proceed to the Granger causality in variance tests proposed by Cheung and Ng [3], we find that there is volatility spillover from world oil spot markets to the electricity stock index returns in Turkey. Hence, use of disaggregated index returns and looking for volatility spillovers seem to uncover the link between oil prices and financial asset returns which has been undetected using more traditional methods. Our results provide important implications for investors and interesting insights for further research. The next section introduces the literature review on the relationship between stock returns and oil price. Then in Section 3, we discuss data characteristics and methodology followed by empirical results in Section 4. Section 5 concludes and provides implications for investors, policy makers and further research.
2. Literature review The early literature on the macroeconomic effects of oil price shocks has been led by Hamilton [8] who shows that ‘‘oil shocks were a contributing factor in at least some of the US recessions
355
prior to 1972.” Furthermore, he states that almost all US recessions after World War II were preceded by large increases in oil prices. The following literature has extended into examining the impact of oil shocks on other economies and other macroeconomic variables (see for example [32,16,22,30,31]). Within this vast literature, there are also a quite a number of studies on how oil price changes influence the stock market returns. Along this line, most of the studies are on developed countries [7,26,27,24], but there is an increasing interest on developing countries [23,10,11,1,29] as well. Furthermore, most of the earlier work on stock indexes has usually focused on how a level shift in oil prices influences the aggregate stock market returns. Recently, however, the literature is extending into a disaggregate analysis of oil price and industry and/or individual stock returns [27,9,11,28,4,21]. Another line that the literature takes is how oil price volatility influences the volatilities in stock markets. For example, Sadorsky [26] employs monthly data for 1947– 1996 to investigate the interaction between oil prices and economic activity. He finds that oil prices and oil price volatility affect real stock returns, though the effects may not be symmetric. Additionally, the dynamics of the relation seem to have changed after 1986 with the explanatory power of oil price shocks with regard to real stock returns rising and exceeding that of interest rates. In a disaggregated study, Hammoudeh et al. [9] use daily data for 1995–2001 to study the dynamic relationship between five US oil sector indices and five different oil prices. They find that oil prices are cointegrated amongst themselves but oil sector indices are not. For mixed systems (oil sector indices plus one oil price) only one cointegrating equation is found. Their results show that oil prices can be used to predict changes in some oil sector indices. Furthermore, they also report significant volatility spillovers from oil prices to the stock indices. However, the effect is to increase the volatility in some indices, while decreasing it in others. Another disaggregated analysis is due to Sadorsky [27] where he uses a multifactor model to estimate expected returns for the Canadian oil and gas industry on a monthly basis between 1983 and 1999. He finds that exchange rates, oil prices and interest rates, in addition to the usual market factor, all affect Canadian oil and gas industry stock returns. Increases in oil prices lead to increases in Canadian oil and gas industry stock returns. This study focuses on energy industries in Canada only, but there are other works that examine the link between oil markets and other related industries. For example, Sadorsky and Henriques [28] use weekly data for 2001–2007 in a LA-VAR model to examine the relationship between alternative energy company stock prices, technology company stock prices, oil prices and interest rates in the US. They find that oil prices, technology stock prices and interest rates Granger cause alternative energy stock prices, with technology stock price shocks having a larger impact. This is interpreted as alternative energy stocks behaving more like technology stocks. The literature also includes studies on several other countries including emerging markets. Basher and Sadorsky [1], for example, employ daily and monthly data to study 21 emerging markets (including Turkey, which contains the highest market risk among all) in an international CAPM model. Their results show a significant negative impact of oil prices on stock market returns. They argue that the results are sensitive to the frequency of the data and the world CAPM model may not hold for all countries. In their work there is also evidence of a nonlinear and asymmetric conditional relationship between oil price risk and stock returns. In another study, taking into account economic activity and employment, Papapetrou [23] finds that oil price shocks affect stock returns in Greece. Narayan and Narayan [18] report the theoretically unexpected positive impact of oil price shocks on Vietnamese stock prices for the 2000–2008 period, using daily data and accounting for the nominal exchange rate.
356
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
Park and Ratti [24] use monthly data for 1986–2005 in a multivariate VAR analysis to examine the effects of oil price shocks and volatility on the stock returns of the US and 13 European countries. They find significant impact of oil price shocks on real stock returns in all countries. They report that in many European markets, but not in the US, higher volatility has a tendency to depress real returns. Finally, the results show that many effects differ across countries. Hammoudeh et al. [9] use daily data for 1994–2001 to examine the relationship between oil prices and five oil exporting Gulf Coast Country stock markets. The study finds that only the Saudi Arabian index seems to have a significant relationship with oil prices and the relationship is bi-directional. Hammoudeh and Li [11] use daily data for 1986–2003 and find that oil prices lead the Norwegian and Mexican stock markets as well as US oil sensitive stock indices they examine. Unfortunately, there are only a small number of studies that investigate the impact of oil prices on the real economy as well as the stock market in Turkey. Among these, Kibritcioglu and Kibritcioglu [13] examine the oil price and inflation link and suggest that innovations in oil prices do not affect the price level in Turkey. Sari and Soytas [29] study the relationship between stock returns, crude oil prices, interest rates and output in Turkey using monthly data for the period 1987:01–2004:03. They find that oil prices do not have any significant effect on stock returns. To the extent of our knowledge, there are no studies that examine the Granger causality relationship between world oil prices and ISE returns as well as the electricity index returns using daily data. Furthermore, this study is probably the first to examine the volatility spillover from the world oil spot markets to the electricity stock returns in Turkey. In these respects, the paper aims to contribute to the literature on the effects of oil price shocks and volatilities on disaggregated stock returns in emerging markets. 3. Data and methodology We use daily data on ISE100 index, ISE electricity index, and spot oil price for the period 05/02/2003–03/01/2007. The data range is limited by both the availability of the electricity index and the need to avoid the impacts of the 2001 crisis. The ISE100 and electricity index are retrieved from the ISE. The electricity index is composed of four electricity generating firms. The daily exchange rate between US Dollar and Turkish Lira is obtained from the Central Bank of the Republic of Turkey. The Bloomberg European Dated Brent Forties Oseberg Ekofisk (BFOE) Spot Price reflects the average value of Brent contracts for differences over the 10–21 day period. The natural logarithms of all data are arranged in 5 day weeks and all holidays are removed. ELIN, ISE100, and BRENT represent the first differences of the logged data and hence represent continuously compounded growth rates/returns. Therefore, ELIN is the electricity index return, ISE100 is the Istanbul Stock Exchange 100 index return, and BRENT is the oil return. We extract the impact of the stock market from the electricity index using a simple market model.
ELINt ¼ a þ bISE100t þ et
ð1Þ
where et is then the return on the electricity index with market effects removed (AELIN). AELIN therefore represents the market adjusted electricity index returns. This provides us with returns that are peculiar to the electricity index, where the effects of factors affecting the general market have been removed.We first investigate the time series properties of the series in concern. Table 1 summarizes the data characteristics. Table 1 presents evidence on non-normality of the series both via the Jarque–Bera statistics and excess kurtosis. Narayan and
Wong [19] argue that although conventional unit root tests are frequently conducted in energy economics literature, the results are not unanimous regarding several energy variables, including oil prices. Furthermore, Maddala and Kim [17] argue that conventional unit root tests have low power. Therefore we use GLS detrended and Ng and Perron’s [20] Z alpha tests that exhibit better power properties. It is also clear from the unit root tests in Table 1 that all returns are stationary in levels. Then we specify the VAR systems for both ELIN and AELIN separately, but including the USD-TRL exchange rate and ISE100 in both systems, as follows:
V t ¼ av þ b1 V t1 þ b2 V t2 þ þ bk V t k þ ev t
ð2Þ
where k is the lag length determined via examining five different information criteria (final prediction error (FPE), Akaike information criterion (AIC), Schwarz information criterion (SIC), Hannan-Quinn (HQ) information criterion, and likelihood ratio test), Vt = (ELINt or AELINt, ISE100t, BRENTt, ERt), ER is the exchange rate, and et are the white noise residuals. In addition to the block exogeneity tests we also examine the generalized responses of the adjusted and unadjusted electricity index returns to impulses in oil prices. Note that the generalized approach developed by Koop et al. [14] and Pesaran and Shin [25] overcomes the orthogonality problem associated with the standard impulse response analysis. Hence, ordering of variables in the VAR system does not change the results. Next we consider the Granger causality in mean and variance approach developed by Cheung and Ng [3]. They propose estimating univariate GARCH models for the stationary variables in order to get the conditional means (lt) as well as the conditional variances (r2t ). The strength of the procedure is that it does not necessitate the use of complicated MGARCH models that suffer from the dimensionality problem (Engle and Sheppard [5]). In order to overcome problems with using GARCH models for financial data, we employ the exponential GARCH (EGARCH) model. Hence, there is no need to impose restrictions to ensure positive variances. Furthermore, the asymmetric effects of positive and negative shocks to returns on volatility are also accounted for. The EGARCH model’s mean and variance equations are as follows:
Z t ¼ lt þ ut ; ut Nð0; r2t Þ
ð3Þ 2
logðr2t Þ ¼ x þ b log
r2t1
ut1 6 jut1 j þ c qffiffiffiffiffiffiffiffiffi þ a4qffiffiffiffiffiffiffiffiffi
r2t1
r2t1
3 rffiffiffiffi 27 5
p
ð4Þ
where zt denotes the stationary series, lt is a constant, ut are normally distributed error terms whose conditional variances are given by Eq. (2). The model is exponential since the log of the conditional variance is modeled in the variance equation. This ensures that the forecasts of the variance are non-negative. There is asymmetric impact (leverage effect) if c is not zero. Note that the choice of the mean function in (1) is made based on some information criteria via comparison of alternative models with AR and/or MA terms. ^ it ; are obtained ^ it Þ=h The standardized residuals, ^eit ¼ ðzit l from the EGARCH models, where zit represent the stationary series ^ it are the estimates of r . Then the sample residual cross-corand h it relation functions between the two standardized residuals ^ u1 u2 ðkÞ) are derived. The Granger causality in mean test is then (q pffiffiffi ^ u1 u2 ðkÞ that asymptotically follows the based on the statistic T q standard normal distribution, where T is the sample size. For the Granger causality in variance test the squared standardized resid^2 are obtained and the sample residual cross^ it Þ2 =h uals ^e2it ¼ ðzit l it correlation functions between the squares of the two standardized pffiffiffi ^ v 1 v 2 ðkÞ) are derived. Again the test statistic T q ^ v 1 v 2 ðkÞ residuals (q follows the normal distribution asymptotically. For the derivation of the sample cross-correlation functions see Cheung and Ng [3] and Inagaki [12].
357
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360 Table 1 Data summary statistics for log returns.
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque–Bera DF-GLS1 DF-GLS2 NP-Za1 NP-Za2
ISE100
ELIN
AELIN
BRENT
ER
0.001307 0.001905 0.100954 0.086708 0.018023 0.048697 5.209295 195.4151a 2.820423a (9) 4.204168a (9) 9.25116b (9) 15.1791c (9)
0.00017 0.000263 0.103023 0.10216 0.020735 0.04854 6.150185 396.9097a 7.193349a (6) 26.98017a (0) 59.5143a (6) 470.244a (0)
7.87E19 0.001143 0.087719 0.072479 0.016578 0.874222 7.487222 926.7223a 29.00875a (0) 23.50248a (1) 477.025a (0) 583.009a (1)
0.000991 0.001121 0.074127 0.085546 0.02004 0.015212 3.547021 11.99381a 31.66426a (0) 31.67747a (0) 478.740a (0) 478.731a (0)
9.57E05 0.000907 0.047744 0.02775 0.007785 0.951178 7.292198 880.7583a 1.143224 (8) 6.834642a (4) 11.1635b (8) 194.139a (4)
Lag lengths are determined via SIC and are in parentheses. DF-GLS1 and 2 represent the Dickey–Fuller GLS detrended unit root test statistics with intercept and with both intercept and trend respectively. NP-Za1 and NP-Za2 are the Ng–Perron Z alpha unit root test statistics with intercept and with both intercept and trend respectively. For a detailed description of DF-GLS see Maddala and Kim [17], and for NP see Ng and Perron [20]. a Significance at 1%. b Significance at 5%. c Significance at 10%.
4. Results and discussions Having established the stationarity of returns and selected the optimal lag length we check the stability of the VAR systems. The VAR systems both satisfy the stability condition in that all roots are within the unit circle. In order to overcome the serial correlation and heteroscedasticity problems observed in diagnostic analyses, we use the Newey–West adjusted standard errors for further analysis. All diagnostic results and Newey–West adjusted standard errors are available upon request. In Table 2, Panel A gives the test results for the VAR system with ELIN and Panel B with AELIN. When we follow the traditional block exogeneity tests based on the stable VAR systems, we find that there is no Granger causality running from oil spot returns to either stock index or to electricity index returns. When we examine both panels in Table 2 we see that the lack of Granger causality from oil returns to other indices persists when we use the adjusted electricity index. The block exogeneity tests reveal Granger causality running from electricity index returns to the oil returns. This causality also persists (and is more prominent) when we use the adjusted electricity index. This result may appear as unexpected at first because the Turkish economy is not expected to influence the global energy markets. However, the explanation that stock returns can improve the forecasts of oil returns may not be too far fetched. World oil prices uni-directionally Granger cause the exchange rate. This is
as expected since world oil trade is conducted in US dollars, and increases in oil prices raise the demand for dollars changing the dynamics in the foreign exchange market. An oil price shock may depress oil demand and therefore the demand for dollars (we thank an anonymous referee for pointing this out). Furthermore, the unadjusted electricity returns lead the exchange rate while ISE100 does not (Panel A), but in Panel B the adjusted electricity returns do not have any impact while ISE100 is driving the ER. The block exogeneity test results are confirmed via the generalized impulse response functions. The generalized impulse response
Table 2 Block exogeneity test results.
Panel A ISE100 ELIN BRENT ER
ISE100
ELIN
BRENT
ER
– 7.070401a 7.649358b 0.462923
2.511115 – 5.010741c 150.6405a
2.672842 1.929749 – 7.911088b
1.003266 0.686409 0.5621 –
ISE100 – 3.403173c 0.31924 208.5967a
AELIN 1.976251 – 7.968952a 0.417049
BRENT 1.973536 0.028871 – 7.035089a
ER 0.354177 3.226693c 0.062962 –
Panel B ISE100 AELIN BRENT ER
Significance implies the column variable Granger causes the row variable. a Significance at 1%. b Significance at 5%. c Significance at 10%.
Fig. 1. Generalized impulse responses of ISE100 and ELIN returns to oil shocks.
358
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
graphs depicting responses of ISE100 returns along with the unadjusted electricity index and the adjusted electricity index are presented in Figs. 1 and 2, respectively. According to Fig. 1, neither ISE100 nor ELIN significantly responds to an innovation in spot oil price growth. The responses of ISE100 and AELIN are also insignificant as shown in Fig. 2. The results of the traditional tests are not in line with Sadorsky [27] results for Canada and Hammoudeh et al. [9] results for the US. Although the different results may be attributed to the fact that this study focuses on an emerging market instead of a developed market, the Basher and Sadorsky [1] results for emerging markets (including Turkey) do not comply with our preliminary findings. Therefore, in order to check the robustness of our results and to investigate any volatility spillover links, we next examine Granger causality in mean and in variance using the Cheung and Ng [3] methodology. We first identify the EGARCH univariate models for ISE100, AELIN, and BRENT. Based on the SIC, we select models with no autoregressive or moving average terms for each variable. The asymmetric natures of the conditional variance equations are verified for all series indicating that bad and good news trigger different market reactions. Hence a simple mean equation with an asymmetric EGARCH specification is applied. The standardized residuals for all equations are subjected to the BDS test and no evidence of remaining time dependence is observed. Table 3 gives estimates of the variance equations. According to the results in Table 3, the leverage effects are apparent due to significance of c. Since |b| is less than 1, the model is stationary. The extent of a shock on the conditional variance is captured by a and x is a constant term. Then following the Cheung and Ng [3] procedure we compute the test statistics for Granger causality in mean, which are summarized in Table 4.
Fig. 2. Generalized impulse responses of ISE100 and AELIN returns to oil shocks.
Table 3 EGARCH variance equations. Variance equations
x a c b
ISE100
AELIN
BRENT
0.673075a 0.175643a 0.066147a 0.9338a
1.493140a 0.300242a 0.083243a 0.845829a
5.716711a 0.006904 0.180920a 0.269753
logðr2t Þ ¼ x þ b log
jut1 j t1 ffiffiffiffiffiffiffi r2t1 þ c puffiffiffiffiffiffiffi þa p r2 r2 t1
a
t1
qffiffiffi 2 p .
Significance at 1%.
Table 4 shows that there is no Granger causality in mean linkages between the ISE100 index and oil prices; however, there is significant contemporaneous correlation between them. According to Table 4, oil returns do not Granger cause the electricity index returns, but electricity returns negatively Granger cause oil returns at first lag. This result persists when we examine the link using the adjusted electricity index returns. There is also weak evidence that oil returns positively Granger cause adjusted returns at lag 6. However, as we have no reason to expect an effect only at lag 6, it seems likely that this finding may be spurious. Further evidence or a good explanation may be required. These findings indicate that using disaggregated indices while studying the impacts of oil price shocks, rather than dealing with aggregate market indices, may reveal new information. When we examine the Granger causality in variance tests, whose results are presented in Table 5, we observe that the spot oil returns do not Granger cause the ISE100 returns in variance either. There is bi-directional Granger causality in variance from oil returns to the electricity index returns and vice versa. The volatility in world oil markets negatively Granger cause the volatility in unadjusted electricity returns at lag 2, with a positive feedback effect from unadjusted electricity index volatility to oil volatility at lag 4. The structure of the bi-directional causality in variance changes when adjusted electricity returns are used. Oil and electricity volatilities positively Granger cause each other at lag 4. The spillovers occurring at higher lags are probably due to delays in information processing and portfolio and risk position adjustments based on updated information by investors. The volatility spillover results confirm the mean spillover results in that using industry returns instead of aggregate returns helps uncover new information regarding the link between world oil prices and financial returns. Furthermore, in examining the information transmission between market prices, volatility spillover is a dimension that should not be overlooked. Therefore, our results show that Turkish stock market is not immune to oil price shocks as indicated by Sari and Soytas [29] who focus on returns only. We show that volatility in the world oil market has an impact on the energy stock volatility in Turkey. This means that investors cannot effectively diversify risk away by including Turkish energy stocks in their portfolios. The overall results of the study provide answers to questions posed in the introduction section. First of all, we show that focusing on the general stock index may not be adequate in identifying the link between world oil market and stock returns. The Turkish energy stocks are not unresponsive to developments in the world oil market. Additionally we also show that the link may not be limited to returns but volatility linkages must also be taken into account. There is volatility spillover from world oil market to Turkish electricity index returns. Hence, a more disaggregated approach including volatility spillover analyses uncovers new information that is useful to both policy makers and investors.
359
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360 Table 4 Granger causality in mean test statistics. i
GC in MEAN ISE100 and BRENT
0 1 2 3 4 5 6 7
ELIN and BRENT
AELIN and BRENT
Lag
Lead
Lag
Lead
Lag
Lead
1.787769b 1.12162 0.570103 0.45546 0.625874 0.189 0.71573 0.514332
1.787769b 0.433774 0.67855 0.529824 0.1921 0.06507 0.65066 0.619677
1.195977 0.79629 0.597989 0.26336 0.991484 0.610382 0.771498 0.340823
1.195977 2.08212b 0.34082 0.20759 0.13323 0.75291 0.17041 0.11464
0.297445 0.071263 0.439971 0.01859 0.886139 0.830368 1.341601c 0.07746
0.297445 2.85361a 0.399692 0.83656 0.3935 1.3354 0.567005 0.05267
BRENT Granger causes the first variable in mean if the test statistic is significant for some lags; vice versa if the test statistic is significant for some leads. a Significance at 1%. b Significance at 5%. c Significance at 10%.
Table 5 Granger causality in variance test statistics. i
GC in VARIANCE ISE100 and BRENT
0 1 2 3 4 5 6 7
ELIN and BRENT
AELIN and BRENT
Lag
Lead
Lag
Lead
Lag
Lead
0.065066 0.827269 0.52982 0.65066 0.41518 1.12471 0.84586 0.390397
0.065066 0.356314 0.170411 0.312937 0.678547 0.48335 0.622776 0.35322
0.31604 0.21069 1.31372b 0.03408 1.022468 1.48413 0.05887 1.264142
0.31604 0.05887 1.013172 0.32843 2.320692a 0.632071 0.42138 1.35709
0.20759 0.3873 1.01007 0.709531 2.094509a 1.7351 0 0.455463
0.20759 0.997681 0.848958 0.89543 1.648342a 0.309839 0.216887 1.3509
BRENT Granger causes the first variable in variance if the test statistic is significant for some lags; vice versa if the test statistic is significant for some leads. a Significance at 5%. b Significance at 10%.
5. Conclusions This paper investigates the Granger causality relationship between Istanbul Stock Exchange 100 index returns, ISE electricity index returns and world spot oil market returns. The block exogeneity results from the VAR systems imply that a unidirectional Granger causality exists running from Turkish electricity index returns to oil returns. This unexpected result may be due to the fact that the stock returns of energy companies in Turkey may be following the same dynamics with the stock returns of global energy companies that do have an impact on the world oil markets or possibly that it is anticipating future movements. This poses a new question, which can be answered in an analysis that includes both the Turkish and global energy companies’ stock returns. A similar explanation can be used for the bi-directional Granger causality in variance between oil and electricity stock returns. In this emerging economy, Turkey, we find evidence that world spot oil returns Granger cause the electricity returns in variance and vice versa, but not the ISE100 returns. Hence, the volatility spillover is limited to the electricity index. Our results show that the lack of Granger causality in mean, as documented in the literature by several papers as well as by this paper, does not necessarily imply that changes in world oil prices have no impact on financial returns in Turkey. Hence, our results emphasize that the use of industry (or maybe even firm level) returns to examine the influence of oil price changes may provide new insights. Furthermore, our results also show that investigating Granger causality in variance may reveal transmission mechanisms that went unnoticed by tests of Granger causality in mean. The new line of
literature on the effects of oil price changes on returns with disaggregated indices (or even with individual stocks) and variance causality will provide essential information to global investors and a promising venue for further research. A natural extension of our results would be to evaluate the impact of different energy prices (Lee and Lee [15]) as they can be viewed as separate markets with arbitrage opportunities.
References [1] Basher SA, Sadorsky P. Oil price risk and emerging stock markets. Glob Finan J 2006;17:224–51. [2] Chen S-T, Kuo H-I, Chen C-C. Modeling the relationship between the oil price and global food prices. Appl Energy, forthcoming. doi:10.1016/ j.apenergy.2010.02.020. [3] Cheung Y, Ng LK. A causality-in-variance test and its application to financial market prices. J Econometrics 1996;72:33–48. [4] Cong R-G, Wei Y-M, Jiao J-L, Fan Y. Relationships between oil price shocks and stock market: an empirical analysis from China. Energy Policy 2008;36:3544–53. [5] Engle RF, Sheppard K. Theoretical and empirical properties of dynamic conditional correlation multivariate GARCH. Working paper No. 8554, NBER; 2001. [6] Ewing BT, Malik F, Ozfidan O. Volatility transmission in the oil and natural gas markets. Energy Econ 2002;24:525–38. [7] Ewing BT, Thompson MA. Dynamic cyclical comovements of oil prices with industrial production, consumer prices, unemployment, and stock prices. Energy Polym 2007;35:5535–40. [8] Hamilton JD. Oil and the macroeconomy since world war II. J Polit Econ 1983;91:228–48. [9] Hammoudeh S, Dibooglu S, Aleisa E. Relationships among US oil prices and oil industry equity indices. Int Rev Econ Finan 2004;13:427–53. [10] Hammoudeh S, Aleisa E. Dynamic relationships among GCC stock markets and NYMEX oil futures. Contemp Econ Policy 2004;22:250–69.
360
U. Soytas, A. Oran / Applied Energy 88 (2011) 354–360
[11] Hammoudeh S, Li H. Oil sensitivity and systematic risk in oil-sensitive stock indices. J Econ Bus 2005;57:1–21. [12] Inagaki K. Testing for volatility spillover between the British pound and the euro. Res Int Bus Finan 2007;21:161–74. [13] Kibritcioglu A, Kibritcioglu B. Inflationary effects of crude oil prices (in Turkish). General directorate of research working paper. Department of the Treasury; 1999. [14] Koop G, Pesaran MH, Potter SM. Impulse response analysis in nonlinear multivariate models. J Econometrics 1996;74:119–47. [15] Lee C-C, Lee J-D. Energy prices, multiple structural Breaks, and efficient market hypothesis. Appl Energy 2009;86:466–79. [16] Lorde T, Jackman M, Thomas C. The macroeconomic effects of oil price fluctuations on a small open oil-producing country: the case of Trinidad and Tobago. Energy Policy 2009;37:2708–16. [17] Maddala GS, Kim I. Unit roots, cointegration, and structural change. Cambridge: Cambridge University Press; 1998. [18] Narayan PK, Narayan S. Modelling the impact of oil prices on Vietnam’s stock prices. Appl Energy 2010;87:356–61. [19] Narayan PK, Wong P. A panel data analysis of the determinants of oil consumption: the case of Australia. Appl Energy 2009;86:2771–5. [20] Ng S, Perron P. Lag length selection and the construction of unit root tests with good size and power. Econometrica 2001;69:1519–54. [21] Oberndorfer U. Energy prices, volatility, and the stock market: evidence from the Eurozone. Energy Polym 2009;37:5787–95.
[22] Oladosu G. Identifying the oil price–macroeconomy relationship: an empirical mode decomposition analysis of US data. Energy Polym 2009;37:5417–26. [23] Papapetrou E. Oil price shocks, stock market, economic activity, and employment in Greece. Energy Econ 2001;23:511–32. [24] Park J, Ratti RA. Oil price shocks and stock markets in the US and 13 European countries. Energy Econ 2008;30:2587–608. [25] Pesaran MH, Shin Y. Generalized impulse response analysis in linear multivariate models. Econ Lett 1998;58:17–29. [26] Sadorsky P. Oil price shocks and stock market activity. Energy Econ 1999;21:449–69. [27] Sadorsky P. Risk factors in stock returns of Canadian oil and gas companies. Energy Econ 2001;23:17–28. [28] Sadorsky P, Henriques I. Oil prices and the stock prices of alternative energy companies. Energy Econ 2008;30:998–1010. [29] Sari R, Soytas U. The relationship between stock returns, crude oil prices, interest rates, and output: evidence from a developing economy. Emp Econ Lett 2006;5:205–20. [30] Sari R, Hammoudeh S, Soytas U. Dynamics of oil price, precious metal prices, and exchange rate. Energy Econ 2010;32:351–62. [31] Soytas U, Sari R, Hammoudeh S, Hacihasanoglu E. World oil prices, precious metal prices, and macroeconomy in Turkey 37 (2009) 5557–66. [32] Uri ND. Crude-oil price volatility and agricultural employment in the USA. Appl Energy 1996;54:355–73.