Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
Master’s Thesis
Oil Risk and Oil Stocks
An Empirical Research on NYSE Listed Oil & Gas Firms
Lei Wang
MSc in Business Administration－Specialization Finance
Faculty of Economics, University of Groningen
Oil Risk and Oil Stocks
An Empirical Research on NYSE Listed Oil & Gas Firms
Lei Wang
Program: MSc in Business Administration－Specialization Finance
Faculty of Economics, University of Groningen
Supervisor: Prof. Dr. L.J.R. Scholtens
First Draft: July 20
This Draft: August 30
Key Words: oil risk, NYSE, oil &gas firm
Corresponding Author: Lei Wang, Faculty of Economics, University of Groningen,
9700 AV, Groningen, the Netherlands.
E-mail: ryanwang1978@hotmail.com
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
Abstract
In this paper, weassess the oil price sensitivities and oil risk premiums of the NYSE listed Oil & Gas firms’ returns by using a two-step cross-sectional regression analysis under two different APT models. We find that under both models the return of NYSE oil stocks is positively associated with the return of the market, the increase of the spot crude oil price, and negatively with the firm’s book-to-market ratio. Moreover, we find that the oil firms’ sensitivities to the market, the oil price and the book-to-market ratio are positively priced by the market under the integrated APT model. However, both the size and significance of the oil risk premium are rather variable than stable in the sample period. This evidence suggests that the increasing oil price will raise investors’ expectation of the oil stocks’ future return. This positive oil risk premium may disappear as the investors change their perception of the effect of oil price movement on the oil stock returns. Hence, the research results provide a cautionary story for the investors and managers who either want to take advantage of or to diversify the oil price risk.
Table of Contents
Table of contents
3
1. Introduction
4
2. Literature review
7
2.1 The effects of oil price changes on equity return 8
2.2 The oil price as an explaining factor in asset pricing 10
3. Methodology 13
3.1 Data
13
3.2
Methodology
18
3.2.1 Existing methods used in previous studies 18
3.2.2 A macroeconomic factor APT model 19
3.2.3 An integrated multi-factor APT model 20
3.2.4 Two-step cross-sectional regression 21
3.2.5 Other issues in model selection 21
3.2.6 Robustness Tests 22
3.3 Data and Model Summary 23
4. Results 24
4.1 Main Test
Results 24
4.2 Robustness Test
Results
32
5. Conclusion 36
Appendix
38
References
41
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
1. Introduction
Over the past five years, the oil prices have increased very sharply, with the WTI (West Texas Intermediate) spot crude oil price rising from a $25.56 per barrel in January 4th 2000 to a peak of $71.79 per barrel on the August 18th 20061(see Figure1). Although the oil price is soaring, the world oil consumption has surged from 77.66 million barrels per day in 2001 to 82.47 million barrels per day in 20042. Thus, it is not unreasonable to expect that the oil companies will benefit from high oil prices as oil prices levels determine their revenues. Indeed, according to the data published by the New York Stock Exchange (NYSE)3, the stock prices of most Integrated Oil & Gas companies have risen dramatically during the last year. For example, the stock price of A.O. Tatneft, a Russian oil company, skyrocketed from 26.80 U.S. dollars at December 10th 2004 to 74.00 U.S. dollars at October 4th 2005.
Figure 1
Crude Oil Price (Dollar per Barrel)
0 10 20 30 40 50 60 70 80 J an-02 A pr-02 J ul-02 O ct-02 J an-03 A pr-03 J ul-03 O ct-03 J an-04 A pr-04 J ul-04 O ct-04 J an-05 A pr-05 J ul-05 O ct-05 J an-06 A pr-06 J ul-06
However,the relation between crude oil price changes and the return of oil companies is still a controversial one. While a large number of studies focus on the effects of the oil price changes on the equity returns at country, industry and individual company levels (Chen et al’s, 1986; Al-Mudhaf & Goodwin, 1993; Jones and Kaul, 1996; Faff & Brailsford, 1999; Boyer & Filion, 2004), none of them examined the relationship between the stock returns of Oil & Gas industry and the oil price changes by introducing the macroeconomic and the fundamental factors into a multivariate estimating model. Our research introduces the firm’s fundamental
1. Data Resource: U.S. Department of Energy 2. Data Resource: U.S. Department of Energy
http://www.eia.doe.gov/emeu/international/petroleu.html#ConsumptionA
factors namely Earning-to-Price Ratio (E/P) and Book-to-Market Ratio (BE/ME) into a multi-factor APT model in respect that Fama & French (1992) addressed the importance of these factors in explaining the asset return. Comparing to previous works (Al-Mudhaf & Goodwin 1993, Faff & Brailsford 1999, Sadorsky 2001, Boyer & Filion 2004), our research has its relative advantage. The previous studies only capture the “systematic” factor risk that determines asset return while ours captures both the “systematic” and the “unsystematic” or “firm specific” factor risks and hence provides valuable information about firm’s oil risk sensitivities and market premium of oil risk. By including the “firm specific” factors, we can examine if the oil price sensitivity as well as oil risk premium are linearly related to firm’s fundamental factors, such as BE/ME and E/P, besides the relation between oil price and market as whole.
The ability of accurately pricing the oil stocks is important for both individual and institutional investors, especially for the portfolio managers. Cavaglia, et al (2004) provide evidence that suggests the home country of a company is becoming less important and the industry classification is becoming more as an important determinant of security returns. Moreover they found that the relative comparison of stock prices of companies in global industries (but across countries) is becoming a more prominent feature of active portfolio management. Focusing on single industry, our paper may provide an insight into the effect of the change of oil price on the return of stocks within the Oil & Gas Industry. This insight is useful for those investors who either concentrate on a single industry, such as sector fund managers or investors who seek to hold an industrially diversified portfolio. This paper introduces the oil price as an additional explaining variable to find the determinants of the oil stock returns under a multi-factor APT model. Our research can be summarized in the following main research questions:
1. Do oil price changes affect the stock return of the Oil & Gas firms under a
macroeconomic APT model which includes market return, default premium and term
premium?
2. Do oil price changes affect the stock return of the Oil & Gas firms when two fundamental
factors, namely the Book/Market ratio and the E/P ratio, are taken into consideration?
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
Moreover, we develop two different sets of hypotheses to make the research themes testable. The first set of hypotheses concerns the relationship between oil price risk as well as other macroeconomic factors and the returns of oil stocks. Since the oil price determines the level of revenue of the oil companies, we expect a positive relation between the oil price and the returns of oil stocks. The second hypothesis concerns if these factor sensitivities are priced by the market. Due to the same reason we addressed for the first hypotheses, we expect a positive market premium for this oil risk. The other hypotheses set (in terms of null hypotheses) teststhe second research theme.
H11: There is no significant statistic relationship between the oil price changes and the oil
stock return under the macroeconomic factor APT model.
H12: The oil price risk of the oil stock returns is not priced by the market under the
macroeconomic factor APT model.
H21: There is no significant statistic relationship between the oil price changes and the oil
stock return under an integrated multi-factor APT model- i.e. the additional firm’s fundamental
factors are introduced.
H22: The oil price risk of the oil stock returns is not priced by the market under an
integrated multi-factor APT model
2. Literature Review
Asset pricing is a subject that has been investigated by many researchers. Although some scholars have tried to find common factors to a majority of stocks or to a specific class of stocks (Fama & French, 1989, 1993, Chen, 1991), the construction of an accurate and feasible asset pricing model is still a controversial and complex issue. The influence of macroeconomic factors or firm specific factors may change as measure, dataset or testing period changes. Moreover, even the same factor can generate different influences across different industries. Oil price changes, for instance, may have a positive association with the stock return of Oil & Gas industry, but a negative association with Transportation industry (Hammoudeh & Li, 2005).
During the last decades, researchers have been suspicious about the ability of the capital asset pricing model (CAPM) to explain the expected return of risky assets, for example in the work of Gibbons (1982) and Fama & French (1992). Therefore, many studies have been done
under thearbitrage pricing theory (APT) model in terms of asset pricing (Bower et al 1984, Burmeister & Wall 1986, Chen et al 1986). A set of macroeconomic influences, such as inflation, market return and interest rate change, have been included in the model to estimate the factor sensitivities. However, only a few of these studies have focused on analyzing whether the oil price change is a determinant of the equity returns. Due to the soaring oil price over last few years, researchers have become increasingly interested in examining the sensitivity of equity return to the energy price changes and how market prices those sensitivities. Taking into consideration different return generating factors, most of the researchers have focused on how the energy price, especially the oil price, affects equity returns under a framework of APT (Al-Mudhaf & Goodwin 1993, Faff & Brailsford 1999, Sadorsky 2001, Boyer & Filion 2004).
A review of these previous works can be classified into two categories, namely the researches concerning oil price sensitivity and those concerning oil risk premium. The oil price sensitivity is the ex post sensitivity that is in oil firm stock returns to changes in the price of crude oil while the oil risk premium is the ex ante return that the oil firms arerequired to pay
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
investors for bearing this sensitivity to oil price changes (Al-Mudhaf & Goodwin, 1993). In fact, the former one is the ex ante testing stage of the latter one as the oil price sensitivities gained from the first step regression become the objects of second step regression in the Two-step Cross-sectional Regression Analysis. The rest of the chapter will be organised as follows. Subsection 2.1 introduces previous works with respect to the oil price sensitivity of equity return. Subsection 2.2 presents a further review of the literature concerning the oil risk premium.
2.1 The effects of oil price changes on equity return
Much research has been done on the way oil price change influences the financial market and stock prices of companies from different dimensions. On the country level, Driesprong,
Jacobsen & Maat (2005) found that the changes in oil prices predict stock returns by using
stock market data of 48 countries, a world market index and price series of several types of oil. Furthermore, they found that investors tend to underreact to the information in the oil price i.e. a rise in the oil price lowers the future stock market return. This can be attributed to the gradual information diffusion hypothesis, namely the stocks with slower information diffusion tend to exhibit more pronounced momentum (Hong & Stein, 1999). However,
Driesprong, Jacobsen & Maat (2005) also argued that this predictability effect is weaker for
oil related sectors. Moreover, Jones and Kaul (1996) studied oil price impact across Canada, Japan, the U.K. and the U.S., and explained that the different oil price sensitivities depends on different concentrations of particular natural resources and industrial sectors. In addition,
Hammoudeh & Li (2005) have made a comparison between the oil sensitivity of the
On the industry level, Faff & Brailsford (1999) found evidence that industries are indeed heterogeneous with respect to the equity return sensitivities to the oil price factor in their research on oil sensitivities across industries in the Australian market. Sadorsky (2001)
analyzed the oil price sensitivity of Canadian Oil & Gas industry by using an APT model where the TSE Oil and Gas index is explained by market return, crude oil price, exchange rate and interest rate. He observed that crude oil prices and market return have a positive effect on industry return. An evidence from Hammoudeh & Li (2005) indicates a negative relationship between the U.S. transportation industry and the oil price under the APT model.
On the firm level, Al-Mudhaf & Goodwin (1993) used a multi-factor APT model including a market and an oil price change factor to explain return differences in 29 U.S. oil companies in a period surrounding the oil shock of 1973. They found that Oil price shocks unambiguously drove up ex post returns for oil firms. In addition, Rajgopal & Venkatachalam (1999) studied 25 petroleum refining firms listed on the COMPUSTAT tapes in a period of ten years to examine whether the earnings sensitivity measures are risk-relevant. They concluded that the simulated earnings sensitivity measure exhibits strong contemporaneous association with the firms’ oil betas. Moreover, Boyer & Filion (2004) employed an APT model, which include both macroeconomic and firm fundamental factors, to investigate the determinants of stock return of a sample of 105 Canadian Oil and Gas companies. Their results reveal the same significant relationship between oil price change and stock returns.
However, many researchersargued that the impact of crude oil prices on the return of equity is still ambiguous. For example, Chen et al (1986) explored possible state variables influencing the equity returns on the U.S. market. They found no evidence of the oil price as such a state variable. Hamao (1989) used an extension of Chen et al’s (1986) model on the Japanese market and found the same insignificant oil price factor. Contrary to that, Kaneko &
Lee’s (1995) evidences proved that the oil price is a significant factor in determining equity
return in Japan. They explained this difference in conclusion by differences in the length of the sample period and empirical methodology. To be specific, Hamao (1989) applies the multi-factor APT model to a sample period from 1975 to 1984 while Kaneko & Lee (1995) adopt the Vector Auto-regression (VAR) model in a period from 1975 to 1993.
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
2.2 The oil price as an explaining factor in asset pricing
Beside the relation between the oil risk and equity return, some researches have focused on the question whether the oil price sensitivity can be seen as an explanatory factor in asset pricing. According to the semi-strong form of the market efficiency theory, investors should not be able to trade profitably on the basis of publicly available information. Since the oil price is publicly available for every investor, it indicates that investors can not earn an extra return from bearing the oil price risk. In other words, the oil price risk is not priced by the market. As mentioned before, Al-Mudhaf & Goodwin (1993) tested if oil price risk is priced in the sense of the APT. Their results revealed that domestic oil producers and the large multinationals had to pay investors ex ante risk premium in the period immediately following the oil shock, although this risk premium doesn’t exist before the oil shock. Moreover, Aleisa et al (2003) investigated the dynamic relationship among five S&P oil sector indices and five
different oil prices for the U.S. oil markets using daily data for the available period from 1995 to 2001. Their results show that the change of the WTI spot oil price and the 1-month to 4-month NYMEX future oil price explain the stock price movement of exploration, refinery and marketing oil companies. Using Multivariate Co-integration Techniques and a Vector Error Correction model, Lanza et al (2003) examined the long-run financial determinants of the stock prices of six major oil companies and found a significant oil risk premium on a weekly basis. Contrarily, some researchers addressed opposite results. Chen et al (1986) examined the impact of an index of oil price changes on asset pricing and found no overall effect.
Table 1
Author and
Year Study Object Area
Number of Observation / Firms
Period Model Independent variables Dependent varaibles Main Results
Chen et al 1986
Identify the relationship between economic variables and the asset
pricing
U.S. 1 Portfolio 1958- 1984 APT/ two-path Multi-factor regression
Growth rate in Oil price, term structure, risk premium,
inflation, etc.
Stock Portfolio Return Oil risk is not priced Al-Mudhaf &
Goodwin 1993
If the oil risk of 29 oil companies is priced by
market
U.S. 29 firms Surrounding 1973 APT/FIML method Multi-factor Market return, Production, Crude oil price, Refining value
Individual Stock return
Oil risk is priced (after the shock)
Oil risk is not priced (before the shock) Faff &
Brailsford 1999
Test the sensitivity of Industry equity return to
the oil price factor
Australia 24 industries 1983- 1996 Multi-factor APT/OLS Market return, exchange rate, oil price Industry portfolio returns
Oil price has positive sensitivity to the Oil & Gas
industry, but has negative sensitivity to other four
industries. Rajgopal &
Venkatachalam 1999
Examine whether the earnings sensitivity measures are risk-relevant
U.S. 25 firms 1987- 1996 Regression/ARCH Single-factor Earning sensitivity Oil Beta
Earnings sensitivity measures exhibit strong
contemporaneous association with firms’ oil
betas Sadorsky Perry
2001
Estimate the return of Canadian Oil & Gas
industry stocks Canada 1 index
1983- 1999
Multi-factor APT/OLS
Market return, Interest rate, Exchange rate, Oil price
Stock price return of Oil & Gas stock index
Positive relation: market return, oil price Negative relation: interest
rate, weakness of local currency
Author and
Year Study Object Area
Number of Observation / Firms
Period Model Independent variables Dependent varaibles Main Results
Aleisa et al 2003
Examine the relationship among five
S&P oil sector indices and five oil prices
U.S. 5 indices 1995-2001
Multivariate cointegration techniques & vector error correction
model
Five different spot/future oil price
Five different S&P oil sector stock
indices
The WTI spot and 1- to 4- month NYMEX future oil price change explains stock
price movement of exploration, refinery and marketing oil companies Lanza Alessandro et al 2003 Long-run financial determinants of the
stock prices of six major oil companies
Relevant
countries 6 firms 1998-2003
Multivariate cointegration techniques & vector error correction
model
Market return, exchange rate,
spot & future oil price Stock price return of oil Companies
Positive relation: market return, spot & future oil price
Negative relation: weakness of local currency.
Oil risk is priced M. Martin
Boyer Didier Filion
2004
The determinants of Canadian oil and gas
stocks returns
Canada 105 firms 1995-2002 Multi-factor APT
Market return, Interest rate, Exchange rate, Debt, Production, Cash flows Proven reserves, Drilling success, Crude oil price
Natural gas price
Stock price return of Companies within Oil & Gas Industry
Positive relation: market return, crude oil and natural
gas prices, cash flows, proven reserves. Negative relation: interest rate, production, weakness of
local currency Hammoudeh
Shawkat, Li Huuimin
2005
Compare oil sensitivity between oil-based market and world capital market U.S. Mexico Norway World 5 indices 1986-2003 Multivariate cointegration techniques & vector error correction
model
Oil future price, trading day dummy, lagged dependent
variable, event dummy
First difference of vector (World index,
related countries’ market indices)
There is a negative bi-directional dynamic relationship between oil
future price and world market return. il Risk and Oil Stocks L. Wang
pirical Research on NYSE listed Oil & Gas Firms
Table 1 continued
3. Methodology
In this section, a description of the characteristics of the dataset used in our study will be provided first. Then, we will discuss the existing methods used in the previous researches. In addition, we will motivate the choice of the methods for this research and explain the procedure more specifically. Finally, some important issues concerning the main testing methods and robustness testing methods will be clarified.
3.1 Data
In this sub-section, we will depict the characteristics of the dataset and the formation of the variables employed in the research. The sample contains weekly stock returns of all the companies listed on the Oil & Gasindustrial sector (a company list can be seen in appendix 2) of the NYSE (New York Stock Exchange). The reason why we don’t use monthly data is that the availability of the firm’s fundamentals such as E/P ratio and BE/ME ratio are limited (only last five years’ data is available). If monthly data is used, the number of observations would be quite small and therefore result in biases in the results. However the weekly data involves some noise and extreme observations. Therefore, the monthly data will be employed to test the robustness of the main results. According to the notation of the NYSE, the classification of industries is based on the Industry Classification Benchmark4 (ICB) created by FTSE International Limited and Dow Jones & Company Inc5. To meet the selection criteria, all companies must have an unbroken series of historical price and fundamental data within the test period. The sample does not contain income trusts –i.e. trustsformed for the purpose of owning and administering the Oil & Gas companies’ interest- and foreign companies - i.e. companies located outside the United States but traded on NYSE (Boyer & Filion, 2004). In addition, the companies with an unmatched fiscal year and calendar year will also be excluded.
The selection procedure yields a sample of 96 NYSE listed Oil & Gas firms. Among these 96 companies, 49 are pure Oil & Gas Producers6 and the others are firms operating in Oil Equipment, Services & Distribution7.This is the final sample we will employ in the second path cross-section regression. The sample size in our study is larger than that of any previous
4Ferson & Harvey (1991) have used different industrial classification called (SIC) for NYSE listed stocks. As SIC is not
available and the ICB is the official classification standard of the NYSE, we used ICB in our study.
5Dow Jones Indexes and FTSE have created a definitive classification system called the Industry Classification Benchmark
(ICB). The system is supported by the ICB Universe Database, which contains over 40,000 companies and 45,000 securities worldwide from the FTSE and Dow Jones universes.
6 According to NYSE’s description Oil & Gas Producers are companies engaged in the exploration for and drilling,
production, refining and supply of oil and gas products.
7 According to NYSE’s description Oil Equipment Services & Distribution includes suppliers of equipment and
services to oil fields and offshore platforms, such as drilling, exploration, seismic-information services and platform construction as well as operators of pipelines carrying oil, gas or other forms of fuel (excludes pipeline operators that derive the majority of their revenues from direct sales to end users).
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
works –i.e. 29 in Al-Mudhaf & Goodwin’s (1993), 6 in Lanza et al ‘s (2003)- except for
Boyer & Filion’s (2004) which consist of 105 firms. Though Boyer & Filion’s (2004)
sample size is slightly larger than ours, the quarterly based data makes their number of observations considerably lower than that of ours. More importantly, Boyer & Filion’s (2004) only aim at the relationship between oil price changes and oil stock returns while ignoring the oil risk premium. Therefore, our research is the one that has the largest sample size with respect to examining the oil risk premium.
The sample period is from January 1st 2002 to December 31st 2005. Due to the fact that the companies’ fundamental data, namely the BE/ME, Earning and Market Value, are available only for the most recent five years, the number of observations in the time series regressionis limited. Furthermore, in order to construct Fama & French (1992)’s control portfolio, stocks must be ranked by their E/P and BE/ME of year t-1. Therefore, only four year’s data can be used in our study. Compared to previous studies, Al-Mudhaf & Goodwin (1993) tested only 29 NYSE listed oil companies and they didn’t test the robustness of the results with respect to different sub-sectors, as their sample composition largely falls into one sub-sector. Our sample composition allows us to supply this gap by applying the model to different sub-sectors namely oil producer and integrated oil companies. Moreover, all returns in their study are either measured on quarterly basis within 7 years or on monthly basis within 8 years. Although our sample period covers only 4 years, the weekly based data make the number of observation in time series regression comparably larger than that of their study. Appendix 1 lists all the companies included in our study.
All individual stocks’ historical prices are adjusted prices. In accordance with DataStream’s definition, the adjusted prices are closing prices adjusted for capital gains, which is the most relevant price measure for calculating the stock return. In respect of the fact that we study NYSE listed oil companies which are assumed to have close connection with the main index of the market, the weekly NYSE Composite Index8 return serves as the proxy of the market portfolio return. The risk-free rate is represented by the rate on the 1-month U.S. Treasury bill,
(insert comma) as the 1-month Treasury bill rate was also employed by many previous studies (Ferson & Harvey 1991, Sadorsky 2001, Boyer & Filion 2004). Since our Treasury bill rate is measured on annual basis, the weekly Treasury bill rate is derived by using the formula 1-month rate/ square root of 52 weeks. The oil price return is the weekly Crude Oil WTI (West Texas Intermediate) Spot Cushing Price (Dollars per Barrel). We use the spot oil price in the main tests (Al-Mudhaf, & Goodwin, 1993), although an argument that future oil price
8The NYSE Composite is a value-weighted index and it closely reflects the broader market, as it represents 77% of the total
should be used has been run among the researchers. Sadorsky (2001) argues that the future oil price should be used because the spot prices are more affected by short-run price fluctuations due to temporary shortages or surpluses. To take his argument into consideration, we will use the WTI crude oil future price as an alternative oil price measure to test robustness of the main test results. The interest rate measure has been used as astate variable to capture the effect of default risk (Chen, et al, 1986). We introduce the difference between the weekly returns of corporate bonds rated Baa by Moody and the return of long-term U.S. government bond (in our case a 7-year Treasury9 bond)(Ferson & Harvey, 1991) as the state variable representing the default risk. Moreover, the difference between the average yield to maturity of a 7-year Treasury bond and a 1 month Treasury bill served as a state variable capturing the change in Treasury yield curve (Ferson & Harvey, 1991). Chen et al (1986) also suggest that the change in expected inflation functions as a state variable influencing the asset pricing. As a proxy of this variable, the rolling average of the Consumer Price Index (CPI) is measured on a monthly basis, whereas, our data is measured on a weekly basis. So, the inflation variable is not contained in both the macroeconomic factor and multi-factor APT model. Considering the fact that the previous study (Chen et al 1986) has included inflation rate changes as controlling variable, we will include it as additional state variable in the robustness test.
All historical price data, the firm’s fundamental data and macro-economic data are collected from DataStream (Boyer & Filion, 2004). Table 2 below lists the variables and their definition.
We use a one-sample T test to test the difference between the sample mean and a known or hypothesized value (zero in this case)(Boyer & Filion, 2004; Sadorsky, 2001).The results in
Table 3 indicate that the means of all variables do not significantly differ from zero. This
result meets the condition of standard normal distribution. With respect to the location and variability of the data set, the statistics of skewness and kurtosis reveal that none of the variables meets the standard of normal distribution (skewness=0; kurtosis=3), which implies a natural log of the data set should be used. Moreover, the summary statistics show that the average market return in our study is negative (-0.138) which is consistent with Boyer &
Filion (2004) although the size is relatively small. Sadorsky (2001) has an equivalent size of
average market return except for a positive sign. The positive oil price return is in accordance with both Boyer & Filion (2004) and Sadorsky (2001) while the size is again smaller than previous studies.
9 The historical price of 10-year Treasury bond is not available on the DataStream.
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
Table 2
Variable Symbol Definition
Market Return Rm weekly NYSE Composite Index return- 1 month U.S. Treasury Bill rate Individual Stock Return Ri weekly individual stock return- 1 month U.S. Treasury Bill rate
Default Premium DP Weekly return of U.S. corporate bonds rated Baa by Moody- the Yield of 7 year U.S. Treasury bond
Term Structure TS Weekly return of 7 year U.S. Treasury bond – 1 month U.S. Treasury Bill rate Crude Oil Price Return Roil Roil,t = ln(price of the WTI barrel in $US at t)/ (price of the WTI barrel in $US at t-1)
E/P Factor Return Re/p The difference between return on the small and big E/P portfolio BE/ME Factor Return Rb/m The difference between return on the high and low BE/ME portfolio
Data Source: DataStream
Table 3 provides the summary statistics of the final sample.
Variable Mean (%) Median (%) Stand. Dev (%) t-statistic Skewness Kurtosis
Ri 16.2400 0.2600 3.1900 0.7360* -0.8392 5.2667 Rm -13.8300 0.1025 1.9152 -1.0440 -0.4274 4.6026 Roil 0.5223 1.0288 4.6724 1.6160 -1.0682 6.6548 DP -0.2480 -0.3070 0.0271 -1.1760 -0.4015 2.6244 TS -1.0510 -1.5080 0.1079 -1.3660 0.6109 3.9578 Re/p -4.4900 -2.5200 1.7846 -0.3630 0.4183 3.5095 Rb/m 12.0300 23.1100 1.7369 1.0010 -0.8611 5.4222 Note:*significant at 1%.
Data Source: DataStream
Table 4 presents the average weekly return of the Fama & French control portfolios.
Highest BE/ME 2 3 Lowest BE/ME
Lowest Earning 0.4282 0.0123 -0.0338 0.2227
2 0.1174 0.1280 0.1160 0.1205
3 0.4344 0.1971 0.1649 0.1017
Highest Earning 0.3032 0.1251 0.0232 0.3574
Table 5 presents the correlation matrix between the different variables. Rm Roil DP TS Re/p Rb/m Ri 0.2930* 0.3200* -0.1920* 0.1410* 0.4140* 0.2400* Rm -0.1280 -0.3110* 0.2030* 0.2260* 0.0990 Roil 0.0330 -0.0410 -0.0650 -0.0020 DP -0.3520* -0.1630* -0.0830 TS -0.0200 0.0740 Re/p 0.2620* Note:*significant at 1%.
Data Source: DataStream
The sign of the term structure premium is consistent with Boyer & Filion (2004) but has a smaller size. It should be noticed that the mean return of the SMB and HML control portfolio factor suggests that firms with alow earning-price ratio have higher mean returns than those with a high earning-price ratio and firms with ahigh book-to-market ratio have higher mean returns than those with a low book-to-market ratio according to our sample and test period. This is consistent with the theory documented by Fama & French (1992). They argue that E/P and BE/ME are factors that affect the return of equity. More explanation about how these two factors affect the return of equity and why they are useful in our study will be given in the next section.
Table 5 provides the bivariate correlation coefficient between different variables. All the coefficients are relatively small in size except for correlation between the return of Oil & Gas stocks and E/P factor return. This is in accordance with Jones & Kaul (1996) and Boyer &
Filion (2004), who find that U.S. market reacts negatively to the oil price increase. Jones & Kaul (1996) argue that this negative reaction of the U.S. stock market can be completely
accounted for by the impact of the oil price shocks on current and expected future real cash flows. The argument implies that a high oil price lowers the investors’ expectation of firm’s current and future cash flow and accordingly lowers the expected stock return. Consequently, the return of the market will be depressed. However, our result is inconsistent with Sadorsky
(2001) in which a weak yet positive correlation is found. This difference may result from the
fact that a future oil price is used in Sadorsky (2001)’s study while a spot price is used in ours.Moreover, both the market excess returns and oil price returns are positively correlated with the returns of Oil & Gas stocks. This is consistent with Sadorsky (2001). Furthermore, the significant correlation between Oil & Gas stock returns and two fundamental factor returns indicates the influences of E/P and BE/ME on equity return as suggested by Fama &
French (1992). The positive and significant correlation between the two factor returns
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
themselves implies that firms with high E/P tend to have high BE/ME (Fama & French, 1992). Since none of the correlation coefficients exceed 0.5, the multicollinearity should not be a problem in our study.
3.2 Methodology
3.2.1 Existing methods used in previous studies
Depending on the purpose of the research, the characteristics of the sample and the time horizon, two different models have been used by previous studies (see Appendix 1). In this section, these two methods will be discussed in a nutshell and our choice of method will be explained in detail.
The first category, multi-factor APT model, uses oil price risk as a return generating factor in addition to other economic and fundamental risk factors, such as market return, default risk, interest rate risk and exchange rate risk, in explaining assets return (Chen, et al 1986, Al-Mudhaf & Goodwin 1993, Faff & Brailsford 1999, Sadorsky 2001, Boyer & Filion 2004). Based on the law of one price, this model assumes that the arbitrage opportunity should be eliminated as all the return generating factors are added into the model. It has been widely used in measuring the exposure of stock return to different risk factors and whether this exposure is priced by market. For example, Chen et al. (1986) employ a two-path cross-sectional regression to estimate oil risk premium in the sense of the multi-factor APT model. In the first pass, they estimate the factor loadings or the factors. Then, in the second pass, the regression of the market price on the estimated loadings or the factors is estimated. However this two-path regression is subject to an “error in the variable” problem, as the estimated rather than actual factor loading or factors are used in the second path regression (Geweke & Zhou, 1996). The ignorance of the variation in estimates will lead to potential biased inference.
The second category, Multivariate Cointegration Techniques & Vector Error Correction Model (VECM), is a restricted vector auto-regression model (VAR) designed to focus on the long-run determinants of the market value of individual firms (Lanza et al. 2003, Aleisa et al. 2003,
Hammoudeh & Li 2005). By including several error-correction terms that represent deviations
from the long-run equilibrium, this model can avoid the misspecification of an unrestricted VAR when a set of non-stationary variables is cointegrated (Hammoudeh & Li, 2005). For example, if an oil-sensitive stock market is well integrated with the world stock market and the returns on both markets are included as independent variables, the VECM should be employed.
a multi-factor APT model. In order to specify a set of return generating factors and then to estimate the factor sensitivity and market price of the factors in the multi-factor APT model, two approaches are introduced in our study. One approach is to determine a set of macroeconomic influences that might affect return and then to use regression analysis to estimate the factor sensitivity. Another is to specify a set of portfolios that might capture the relevant influences affecting stock return as factors. For either approach, a second-step cross-sectional regression analysis is used to estimate the factor risk premium (Elton et al, 2003). We use the first approach to investigate whether the oil is priced by market in the first place. Then a combination of two approaches is adopted to test the same hypothesis, namely the market premium of the oil risk. By comparing the results, a comment can be made with regard to the difference in significance as well as magnitude of the oil risk premium between the two methods.
3.2.2 A macroeconomic factor APT model
Before the specific discussion of this model construction, a few words should be mentioned concerning how the factor set is specified. There are two approaches that can determine a set of factors affecting the stock return, namely the factor-analysis approach and the macroeconomic-factor approach. Burmeister & McElroy (1988) claim that the
macroeconomic-factor has two primary advantages: (1) the factors and their APT prices in principle can be given economic interpretation, while with a factor-analysis it is unknown what factors are being priced; and (2) rather than using only asset prices to explain asset prices, macroeconomic-factor introduce additional information, linking asset-price behavior to macroeconomic events. Therefore, to test the oil price exposure of the companies in the sense of macroeconomic factor APT, we apply Faff & Brailsford’s (1999) model, a so-called multi-factor market model, to each company. This model has also been used by many other researchers- i.e. (Chen et al. 1986, Al-Mudhaf & Goodwin 1993, Sadorsky 2001, Boyer & Filion 2004) as the return-generating process in the APT. The first step is to use a set of macroeconomic factors as explanatory variables for each firm’s stock return. The model can be structured as follows:
Rit = αi +βoil Roil,t+βm Rm,t+βdp DPt+βts TSt+eit (1)
Where, Rit is the return of the ith company for time t. The return is based on the change in the closing price from the last trading day of the week to the next. Rmt is the return of the market index for time t, DPt is the default premium for time t, TSt is the premium of changes in term structure and Roil,t is the oil price return for time t, denominated in U.S. dollar. The Greek letter βoil represents the sensitivity of company’s return to a 1% change of the oil price, after controlling for changes of other independent variables. The letter βmi refers to the firm’s sensitivity to the market return and βdp andβts refer to the firm’s sensitivity to default premium and the changes in term structure, respectively. The αi is the expected return of company i and
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
eit is the random error of company i on time t.
The construction and the calculation of variables, including the market, oil price, default premium, term structure and the Oil & Gas stocks’ return, are presented in Table 2.
3.2.3 An integrated multi-factor APT model
The main difference between the macroeconomic model and the integrated multi-factor model is that the former one only captures the “systematic” factor risk that determines asset return, while the latter one captures both the “systematic” and the “unsystematic” or “firm specific” factor risk. According to Fama & Frenchs’ (1992) argument, the return on common stocks is related to firm fundamental factors like size, the earning price ratio ( or E/P) and the book-to-market ratio (or BE/ME). Their finding suggests that high-BE/ME firms tend to be persistently poor earners relative to low-BE/ME firms and small size firms tend to be persistently poor earners relative to large size firms. To take into consideration these additional fundamental factors that affect a firm’s return, we modify the model (1) by introducing twoadditional firm fundamental factors: Re/pt (return on the mimicking portfolio
for E/P factor) and Rb/m,t (return on the mimicking portfolio for BE/ME factor). This model
employs the E/P factor to construct the portfolio instead of the Size factor which was originally adopted by Fama and French in their three-factor model. The absence of the Size factor is due to the fact that most of the oil companies, especially the oil producers, are large companies, so that the unmodified Fama & French (1992) three-factor model will undermine the representativeness of the estimating samples. The integrated multi-factor APT model is shown as follows:
Rit = αi +βoil Roil,t+βm Rm,t+βdp DPt+βts TS+βep Re/p,t +βb/m Rb/m,t +eit (2) Where, the βep and βb/m refer to the firm’s sensitivity to the Fama & French control portfolio returns and other variables are defined in the same way as equation (1). The detailed explanation of the construction of the Fama & French control portfolio and the calculation of the factor return is presented in Appendix 2.
relative-distress risk or an involuntary leverage risk of the firm. Nevertheless, they point out that the relation between E/P and average return is due to the positive correlation between E/P and BE/ME. We also find the significant positive correlation between E/P and BE/ME in our study as illustrated in Table 5. However, we should not exaggerate the link between the two control portfolio factors as the correlation coefficient (0.26 in our case) is not extreme.
3.2.4 Two-step cross-sectional regression
To answer the research question whether the oil price risk is priced by the market we employ a two-pass cross-sectional regression model which is popularized by Fama and MacBeth
(1973). This model was originally used as a test of CAPM. First, they estimate betas of each
month from a first-pass regression by using the similar procedure as equation 1. Then the second-step regression for each month over testing period allows them to study how the parameters change over time. Likewise, the second-step regression for each Oil & Gas firm allows us to study whether oil risk sensitivity as well as other factors’ sensitivities are priced in a universe of the Oil & Gas industry. Under the absence of risk-less arbitrage opportunities, the second-step cross-sectional regression model arises from the first-pass time series regression model or return-generating process andcan be written as follows:
αi-Rf = λ0 +βm λm+βoil λoil+βts λts+βdp λdp +βe/p λe/p+βb/m λb/m (3)
This model depicts that the excess expected return of individual Oil & Gas firm (αi -Rf ) is explained by the sensitivities of its return to macroeconomic and firm’s fundamental factors. αi is the expected return of the ith company derived from equation (2). Rf is the one-month T-bill rate. The βoi , βmi , βts , βdp, βe/p and βb/m are the firm’s sensitivities to the oil price return, market return, changes in term structure, default premium, E/P portfolio return and Book/Market portfolio return, respectively. The λo , λm , λts , λdp , λe/p and λb/m refer to the risk premium for bearing each of the factor sensitivities, respectively.
3.2.5 Other issues in model selection
The use of weekly data in this research can bring in ARCH effects due to “volatility clustering”. Volatility clustering describes the tendency of large changes in asset prices to follow large changes and small changes to follow small changes (Brooks, 2002). Moreover,
Ahn & Gadarowski (1999) argue that one drawback of the studies based on the two-pass
cross-sectional regression model is that statistical inferences are often made ignoring the potential conditional heteroskedasticity or/and the autocorrelation in asset returns and factors. Therefore, it is necessary to test the heteroscedasticity and the normality of the residuals for the purpose of assessing the appropriateness of the estimate model. To approach this assessment, we use the Ordinary Least Squares regression to estimate the parameters of the
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
equations in the first-pass regression. Then we will apply a Bera-Jarque test and a White test to test for normality and heteroscedasticity of the error term for all the companies, respectively.
If the conditions of homoscedasticy and normally distributed error term are not met, it will indicate that the Ordinary Least Squares (OLS) regression is not appropriate in our case. In this case, one way to avoid the model misspecification is to use monthly data and a longer sample period instead of weekly data and a4 years sample period.
Another way is to use one of the two non-linear models GARCH (1, 1) (Brooks, 2002). GARCH is the model that does not assume constant variance and takes into account the variance of the error term evolves over time. Therefore, we will use GARCH (1, 1) in the presence of heteroscedasticity based on the test results. We adjust the two APT models (the example is in terms of integrated APT model) to fit the GARCH (1, 1) specification:
Rit = αi +βoil Roil,t+βm Rm,t+βdp DPt+βts TS+βep Re/p,t +βb/m Rb/m,t +eit (4)
σ
it2= α0 +α1 eit-12 +β1σ
it-12 (5)Where, equation 4 is written as a function of exogenous variables with an error term. Since
σ
it2 is the one-period ahead forecasted variance based on past information, it is called theconditional variance. The conditional variance specified in equation (5) is a function of three terms: a constant term,a GARCH term which specifies the equation for the standard error and an ARCH term which specifies the equation for conditional variance as dependent on the value of the previous period error term.
3.2.6 Robustness Tests
3.3 Data and Model Summary
In summary, the multi-factor APT model and the VECM dominate the models applied in empirical researches on the relationship between oil price risk and the stock market. Owing to the particular application and complex implementation of VECM, multi-factor APT model deliver a more suitable and understandable approach to carry out our purpose of research - i.e. the market premium of oil risk. More importantly, since none of the previous studies has used Fama & French factors in their APT models, our research will provide valuable information about firm’s oil risk sensitivities and market premium of oil risk controlling for both macroeconomic and firm’s fundamental factors. Finally, among the previous works, no study has been done based on data of all U.S. Oil & Gas firms. Our research hence delivers an insight into how the oil price changes affect the return of the U.S. oil stock and the U.S. Oil & Gas industry as a whole.
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
4. Results
In this section, the main test results from first-step time series regression as well as second-step cross-sectional regression are presented and a related interpretation is given. Then we will discuss the results of a variety of robustness tests such as a split half test, an oil producer versus oil service provider test, a future oil price versus gasoline price test and an inflation factor test.
4.1 Main Test Results
The two main regression results and the results from the robustness tests will be discussed in this chapter. Subsection 4.1.1 and 4.1.2 will present the results and a related interpretation of first-step time series regression and second-step cross-sectional regression, respectively. The results from a variety of robustness tests will be discussed in the last subsection.
4.1.1 Time Series Regression Results
As mentioned in thelast chapter, we test the heteroscedasticity and the normality of the residuals for the purpose of assessing the appropriateness of the estimating model by using White and Jarque-Bera tests respectively. As expected, the results indicate that 63 out of 96 firms have significant White test statistics and 69 out of 96 firms have significant Jarque-Bera test statistics, which suggests most of the firm have heteroskedasticity and non-normality in the error terms. Consequently, a GARCH (1, 1) model is employed in the first-path time series regression estimation. Because it does not assume constant variance and takes into account the variance of the error term evolves over time.
Table 6: Estimates of Macroeconomic Factor APT Model
Rit = αi +βoil Roil,t+βm Rm,t+βdp DPt+βts TSt+eitFirm a βm βdp βts βoil R2 Firm a βm βdp βts βoil R2
AHC 0.25 0.65* 10.10 4.40 0.27* 0.11 MRO 0.08 0.52* 1.50 2.58 0.21* 0.14 APA 0.11 0.40* 6.93 3.43 0.38* 0.26 MUR 0.19 0.51* -2.15 -0.98 0.21* 0.08 APC 0.01 0.35* -4.66 2.92 0.25* 0.13 MVK 0.40 1.06* -14.92 0.65 0.27* 0.06 APL -0.23 0.05 -6.72 2.08 0.19* 0.04 NBL 0.12 0.52* 3.77 4.67** 0.27* 0.18 BHI 0.04 0.41* -7.77 2.73 0.28* 0.08 NBP -0.24 0.13 -14.38*** 0.18 0.08** 0.03 BPL -0.20 0.10 -9.71 0.85 0.04 0.03 NBR 0.15 0.75* -1.93 2.07 0.32* 0.12 BRY 0.13 0.29** -19.37** -0.38 0.15* 0.08 NE 0.14 0.65* 2.18 1.13 0.32* 0.12 CAM 0.10 0.55* 7.10 1.11 0.21* 0.07 NFX 0.19 0.42* -4.79 2.98 0.28* 0.14 CHK 0.50*** 0.58* 0.67 4.85 0.45* 0.19 NOV 0.31 0.66* -10.87 1.48 0.35* 0.11 CKH -0.09 0.54* -8.31 -0.36 0.18* 0.15 NR -0.54 0.37 -18.15 -0.04 0.17 0.00 COG 0.31 0.57* -16.74*** 4.41 0.33* 0.20 NTG 0.24 0.06 1.12 4.68*** 0.17** 0.01 COP 0.02 0.48* -0.74 3.34 0.18* 0.17 OGE -0.10 0.44* -0.80 -1.27 -0.01 0.01 CPE 0.28 -0.19 -30.61** 8.42** 0.21** 0.01 OII 0.17 0.72* 7.18 2.39 0.20* 0.07 CRK 0.46 0.81* -2.66 1.73 0.30* 0.13 OIS 0.43 0.69* 24.83** 3.69 0.19* 0.08 CRR 0.12 0.43* 2.61 5.61 0.28* 0.11 OXY 0.31 0.60* 4.63 4.01 0.25* 0.24 CVX -0.11 0.33* 3.14 2.95 0.16* 0.15 PAA 0.00 0.24* -13.06* -0.19 0.06*** 0.05 DNR 0.52*** 0.24 -17.66 5.00 0.23* 0.07 PKD 0.80 0.71* -13.50 3.07 0.33* 0.00 DO 0.05 0.52* -6.99 1.20 0.38* 0.16 PPP 0.08 0.55* 10.04 4.23 0.25* 0.11 DRQ 0.11 0.84* -13.27 2.58 0.27* 0.08 PQ 0.18 0.54*** 32.26 10.85 0.27** 0.02 DVN 0.30 0.53* 5.06 3.81 0.37* 0.17 PVA 0.32 0.31** -19.28** 0.37 0.26* 0.09 DYN 1.11** -1.21* -26.69 -3.55 -0.15 -0.10 PXD 0.14 0.55* 2.98 -1.20 0.31* 0.14 EAC 0.32 0.58* -18.98 4.91 0.23* 0.15 RDC 0.01 0.74* 1.15 1.95 0.39* 0.12 EEP -0.14 0.00 -12.39** 0.34 0.08** 0.00 REM -0.02 0.25*** -15.94 5.36** 0.42* 0.20 EOG 0.29 0.42* -6.22 4.32 0.35* 0.19 RES 0.44 0.96* -20.83 1.59 0.10 0.04 EP 0.10 1.04* -11.33 5.08 0.22* 0.06 RIG 0.39 0.65* -11.58 1.36 0.39* 0.13 EPD -0.22 0.23*** -4.28 -0.41 0.05 0.03 RRC 0.68** 0.42* -7.90 1.86 0.30* 0.08 EPL 0.11 0.28*** -11.60 6.41 0.38* 0.12 SFY 0.03 0.43** -10.34 2.11 0.35* 0.08 ESV -0.08 0.64* -5.14 2.36 0.32* 0.10 SGY -0.18 0.39* -11.89 3.98 0.25* 0.10 FST -0.02 0.42* -9.23 3.90 0.36* 0.17 SII 0.12 0.58* -0.06 -0.97 0.29* 0.07 FTO 0.32 0.44* 2.72 5.69 0.16** 0.05 SLB -0.05 0.50* -12.06 -1.24 0.17* 0.07 GDP 0.42 0.39 -21.05 1.74 0.18 0.01 SM 0.45** 0.61* 4.73 4.27 0.26* 0.12 GI 0.79 0.24 -7.75 0.65 0.17 0.01 SPN 0.18 0.70* -8.74 1.65 0.25* 0.06 GRP 0.59*** 0.87* -0.50 5.23** 0.45* 0.12 SUN 0.52** 0.51* 3.84 3.21 0.16* 0.11 GSF 0.09 0.58* -6.71 2.85 0.29* 0.09 SWN 1.54* 0.58* -5.58 -0.45 0.27* 0.06 HAL 0.51 0.73* 4.36 2.26 0.19* 0.05 THX -0.09 0.47* -3.54 2.38 0.23* 0.13 HC -0.14 0.84* -16.91 1.75 0.19** 0.07 TMR -0.14 0.51*** -31.83 6.10 0.40* 0.05 HNR 0.14 0.39 2.60 4.75 0.31* 0.00 TPP -0.10 0.20** -5.35 -1.90 0.07** 0.02 IO 0.16 1.01* -9.76 4.15 0.09 0.02 TSO 1.09** 0.95 -1.56 6.83 0.38* 0.03 KCS 0.93** 1.18* -9.59 5.64 0.42* 0.06 TTI 0.33 0.61 -2.37 2.34 0.17* 0.04 KMG 0.01 0.39* -2.56 3.13 0.22* 0.12 UCO 0.13 0.75 -6.40 -0.03 0.14** 0.05 KMI 0.02 0.45* -6.64 1.45 0.15* 0.07 UNT 0.25 0.69 -13.26 2.23 0.34* 0.15 KMP -0.17 0.08 -9.47 1.51 0.10** 0.02 VLO 0.57*** 0.54 -9.47 2.25 0.27* 0.09 KMR 0.02 0.13 -12.01 2.34 0.12* 0.04 WFT 0.11 0.69 5.36 1.41 0.27* 0.09 KWK 0.74** 0.63* 0.07 6.21*** 0.26* 0.09 WGR 0.19 0.49 -26.63* 0.20 0.23* 0.15 LSS 0.21 0.86* -38.53 -1.09 0.31** 0.07 WHQ 0.07 0.48 8.32 3.04 0.28* 0.05 MDR 2.39 0.48 60.61* -3.73 0.25*** -0.05 WMB 0.57 1.03 -15.85 3.92 0.22* 0.01 MMP -0.05 0.20** -18.33** 0.23 0.13* 0.03 XOM -0.05 0.59 6.43 2.72 0.14* 0.19 MMR 0.38 1.00* -0.14 2.35 0.34* 0.01 XTO 0.55** 0.58 5.95 3.82 0.34* 0.17 ALL MEAN 0.23 0.51 -5.35 2.46 0.24 0.09
Note:*significant at 1%, **significant at 5%, *** significant a 10%. Estimated by GARCH (1, 1), Total Observation=209 weeks.
R2 are adjusted for degrees of freedom. MEAN= average number of estimate coefficients.
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
Table 7: Estimates of Integrated Multi-Factor APT Model
Rit = αi +βoil Roil,t+βm Rm,t+βdp DPt+βts TS+βep Re/p,t +βb/m Rb/m,t +eitFirm a βm βoil βts βdp βb/m βe/p R 2
Firm a βm βoil βts βdp βb/m βe/p R2
MDR 1.38 0.76 0.24 -8.82 47.57*** 0.30 1.23** 0.04 WMB 0.62** 0.67* 0.17* 1.44 -13.06 1.19* 0.97* 0.12 MMP -0.26 0.09 0.15* 1.61 -15.80** 0.25** 0.39* 0.04 XOM -0.04 0.57* 0.14* 2.84 6.97 0.01 0.08 0.20 MMR 0.47 0.90* 0.36* 1.05 7.02 0.38** 0.65* 0.02 XTO 0.50 0.51* 0.35* 4.38 9.09 0.03 0.45* 0.19
ALL
MEAN 0.20 0.39 0.24 2.56 -1.55 0.16 0.58 0.15
Note: *significant at 1%, **significant at 5%, *** significant a 10%. Estimated by GARCH (1, 1), Total Observation=209 weeks.
R2 are adjusted for degrees of freedom. MEAN= average number of estimate coefficients.
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
In order to further investigate how the firm’s oil price sensitivity might relate to firm’s other characteristics, such as its BE/ME, E/P and size, we plot the scatter-diagrams against the firm’s oil price sensitivities and other factors for two APT models. As can be seen from Figure 2 below, a roughly negative linear relationship exists between the firm’s oil price sensitivities and the firm’s BE/ME. On the individual firm level, we find an outlier in Figure 2 namely the Dynegy Inc. which has the highest BE/ME and lowest oil sensitivity in both models. The unusually high BE/ME may result from its own capital structuring policy while the negative oil sensitivity might be caused by the fact that the company’s business operations are focused primarily on the wholesale power generation sector of the energy industry10. One can therefore imagine that the high oil price raises the operating cost of power plants in which the crude oil is used as raw material and consequently reduce the return of the company’s stock. Moreover, there is no similar linear relationship being detected between the oil price sensitivity and the E/P and size factor. Fama & French (1996) claim that firms with low earnings tend to have high BE/ME and firms with high earnings tend to have low BE/ME. Based on their argument, it is conceivable that an oil firm with low BE/ME and high earnings might be more sensitive to the oil price changes, because the influence of an increase or decrease in oil price should increase or decrease its earnings on a larger scale than that of an oil firm with high BE/ME and low earnings. As these two appealing findings have not been examined by previous studies and a further investigation of possible explanation is beyond the scope of this paper, our results may contribute valuable information for future study on related issues.
As expected, market returns do have a statistically significant effect on most of the oil firms, which is supported by economic theory and previous studies (Al-Mudhaf & Goodwin, 1993; Sadorsky, 2001; Boyer & Filion, 2004). The average size of the market beta is similar in the two models (0.51 in macroeconomic APT and 0.39 in integrated APT) and is slightly larger than the oil betas. This result is consistent with Sadorsky (2001) who also argues the positive sign of the oil beta of these firms suggesting that the oil stocks are not a good hedging tool for market risk. Moreover, as the size of market betas in both models is less than 1, it is indicated that the Oil & Gas firms are less risky than the market as a whole over the testing period.
2.000000 1.000000 0.000000 -1.000000 B E /M E 0.500000 0.400000 0.300000 0.200000 0.100000 0.000000 -0.100000 -0.200000 Oi l
F it line for O IL2 B E M E F it line for O IL B E M E O IL2 B E M E O IL B E M E
oil price sensitivity V S B E /M E
F ig u re 2
R S q Linear = 0.048
R S q Linear = 0.04
Surprisingly, the regression coefficients of the default premium and the term premium are both statistically insignificant for nearly all oil firms. It seems that the oil firms are insensitive to these two macroeconomic factors. Furthermore, the signs of default premium sensitivities are opposite to the Chen et al’s (1986) study in which a positive sensitivity to default premium was found. However, they also find a negative correlation between the market return and the default premium over the period 1978-1983 which suggests that the positive sensitivity to the default premium may not stable over time. If we jointly consider the positive correlation between the oil stock return and the market return and the negative correlation between the market return and the default premium, the adverse result may not be surprising. Because it is possible that the increase in the default premium drives down the market return and therefore the oil stock return as the oil stock return positively relates to the market return. In addition to default premium, the sign of term premium is also contrary to the claim of
Sadorsky (2001) that a higher term premium increases borrowing costs thereby lowering Oil
& Gas stock returns. We would like to interpret this positive coefficient in the same sense of
Sadorsky (2001) but from a different point of view, that is when borrowing costs increase
investors may ask for higher stock returns as compensation for their loss on the fixed-income investment.
Furthermore, the estimated coefficients of E/P factor returns are highly significant for most of the oil firms while only one third of the firms have significant coefficients to the BE/ME factor returns. This implies that the oil firm’s earning is more likely to be affected by the changes of oil price. It makes sense because a large part of the oil firms’ earning comes from
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
the selling of oil or oil related products. Although Fama & French (1992) argue that firms with poor prospects signaled by low stock prices and high BE/ME have higher expected returns than firms with strong prospects, it seems that investors have a different perception to the default risk of some oil stocks. Due to the failure of rejecting the possibility that BE/ME just captures the investors’ irrational perception about the prospects of firms in Fama &
Frenchs’ (1992) study, our explanation based on investors’ diverse perception of firms’
perspective makes sense. The positive sign of the coefficients of E/P and BE/ME is in line with the evidence of Fama & French (1992), whereas the average size of the coefficients of our study (0.16 for BE/ME and 0.58 for E/P) is relatively smaller than theirs (0.33 for BE/ME and 0.87 for E/P)
Finally, the adjusted R squares in both models are low for most of the oil firms. The average adjusted R square in Table 1 of Appendices indicates that only 9% of the variation in oil stock returns can be explained by macroeconomic APT model though the figure is increased in integrated APT to 15% as the inclusion of firm specific factor returns. Compared to R square of previous studies, our result is smaller than Al-Mudhaf & Goodwin (1993) and Sadorsky
(2001). One reason resulting in comparatively low R square is the insensitivity of oil stocks to
the default premium and the term premium. Another reason might be that Sadorsky (2001) focuses on an energy industry index which has closer connection with the market portfolio and Al-Mudhaf & Goodwin (1993) focus on large integrated oil companies which are important components of the industry index and the market index and therefore share more co-movement with the market.
Thus the results presented in sub-section 4.1.1 indicate that the oil stock returns are sensitive to the oil price return, the market return, the E/P factor return and the BE/ME factor return but not sensitive to the default premium and the term premium.
4.1.2 Cross-Sectional Regression Results
Table 6: Regression Results of Macroeconomic Factor APT Model
αi -Rf = λ0 +βm λm+βoil λoil+βts λts+βdp λdpλ0 λm λoil λts λdp Adj. R DW
1 Full sample period 2002.01-2005.12 (1.5868) 0.1604 (0.0664)0.0096 (1.6313) 0.7388 (-1.3768) -0.0241 (3.0881) 0.0093* 0.0981 1.9885 Total Observation (number of the firms)=96
2 2002.01-2003.12 Sub period: (1.7555) 0.1535 (-1.6382)-0.1932 (0.4286) -0.3338 (-3.6241) -0.0143* (6.9321) 0.0076* 0.3278 2.1707 Total Observation (number of the firms)=96
3 2004.01-2005.12 Sub period: (-2.9587) -0.3539* (4.2162)0.5001* (2.6932) 1.1067* (0.6279) 0.0115 (2.3012) 0.0058* 0.3291 1.7763 Total Observation (number of the firms)=96
4 Full sample period Producer (2.7555) 0.4170* (0.6788)0.1574 (-0.6004)-0.4485 (-0.8569) -0.0201 (0.4767) -0.0024 -0.0420 2.2695 Total Observation (number of the firms)=49
5 Full sample period Equipment (-1.22837) -0.133676 0.60004*(0.0015) (1.155292)0.540438 (-1.878102)-0.046727 0.017985*(5.83257) 0.5342 1.7574 Total Observation (number of the firms)=47
6 Full sample period Future oil price (0.2209) 0.0214 (1.8552)0.2559 (3.4935) 0.0095* (1.5647) 0.0056 (-0.9689)-0.0145 0.2215 2.0967 Total Observation (number of the firms)=96
7 Full sample period Gasoline price (-0.8080) -0.0785 (1.7876)0.2541 (2.9452) 0.0163* (0.0503) 0.0007 (2.3919) 0.0066* 0.2721 2.2094 Total Observation (number of the firms)=96 8.Inflation 8.Adj. R 8 Full sample period Inflation included (4.1010) 0.8995* (-4.2293)-0.0176* (-1.3755)-0.0105 (-2.1053) -0.0571* (4.2854)0.0295* -0.0017* (4.2854) 0.4508
Total Observation (number of the firms)=96; Monthly Data; DW=2.2241
Note: * significant at 5%; Estimated by OLS; DW=Durbin-Watson statistic; Figures in parentheses are T-statistics.
In other words, the investors require high return for the oil firm with high oil price sensitivity under the integrated model. This is consistent with our expectation. The adjusted R square, which is adopted as a measure of performance fit of the two models, is considerably low for the macroeconomic model (0.0981) while it is improved by nearly 18% under the integrated model. The difference of adjusted R square between two models implies that the absence of market premium of oil risk in the macroeconomic model may result from model misspecification. However, without the support from robustness tests, we can not rush into the conclusion. It should be noted that the Durbin- Watson test statistics of all industries fall into the non-rejection range suggesting the autocorrelation in the residual series is not a problem within two models. In addition, the fact that oil risk is not priced under the macroeconomic model is consistent with the finding of Chen et al (1986). Using a similar APT model, they also find that the oil betas are insignificant for the asset pricing over the entire sample period. Nevertheless, the significant market premiums of the oil risk found in the sub-periods of two
Oil Risk and Oil Stocks L. Wang An Empirical Research on NYSE listed Oil & Gas Firms
previous studies (Chen et al 1986, Al-Mudhaf & Goodwin 1993) call for robustness tests of internal consistency (see Section 4.2).
Furthermore, we find in the macroeconomic model that the market return sensitivity fails to have a statistically significant effect on asset pricing while in the integrated model such an effect becomes significant. Recalling the fact that most of the oil firms are sensitive to the market return in the time series regression, it seems that the oil firm’s market beta does not explain cross-sectional differences in average returns. This is, to some extent, consistent with the evidence from Chen et al (1986). Applying the similar macroeconomic APT model, they find an insignificant market risk premium during the overall test period as well. However, the inclusion of firm specific factor returns in the integrated model improved the model fit and thus augmented the effect of the market risk on asset pricing. This suggests that the explaining power of market return may have less to do with the macroeconomic factor returns and more with the firm specific factor returns.
Finally, the constant terms under both models do not significantly differ from zero. It implies that one can not find an oil stock that offers extra return and has zero factor sensitivity –i.e. no sensitivity on market return, oil price change, default premium and other factor return. This finding, to some extent, contradicts to Burmeister, Roll, & Ross (1994) who find a zero-factor-sensitivity portfolio that offers a higher return and lower risk than the market portfolio.
Sub-section 4.1.2 shows us that the oil risk is positively priced by the market under the integrated APT model whereas it is not priced under the macroeconomic APT model. Moreover, the inclusion of the E/P and the BE/ME factor returns in the integrated model improve the model fit.
4.2 Robustness Tests Results
Knowing the evidence derived from the full sample period is not sufficient to conclude that oil risk is priced and the integrated APT model outperforms the macroeconomic APT model. Several robustness tests have been conducted.
