Determinants of Oil Price Exposure
An Analysis of the Petroleum Industry in the United States for 2003-2008
Jens Baarveld
Master Thesis University of Groningen Faculty of Economics and Business MSc Business Administration, Finance
Studentnumber:
s1516434
Supervisor:
Dr. Ing. N. Brunia
Determinants of Oil Price Exposure
An Analysis of the Petroleum Industry in the United States for 2003-2008
Abstract
In this study the determinants of oil price exposure of 63 U.S. oil producers are investigated over the period 2003 through 2008. Oil price exposure is found as a significant variable that affects the equity value of the firm. In this study I identify and prioritize the determinants that affect oil price exposure, which, in its turn affects the equity value of the firm. By the increased knowledge about how actual exposures are set and how they influence the equity value of the firm, corporate managers are able in altering the influences of exposures by adjusting firm-specific determinants. The determinants of oil price exposure are approached by the assumption that firms possess managerial flexibility. The sensitivity of the future value of the equity of the firm to the future value of oil prices (definition of oil price exposure) is a function of macroeconomic factors as well firm-specific factors. The results of this study show that oil price exposure is mainly influenced by macroeconomic factors but the firm-specific factor financial leverage does also influence the exposure to unexpected oil price changes. More specific, oil price exposure is a decreasing function of the level of oil prices. This determinant increases revenues and thus lowering the probability that a firm defaults, which would result in a zero value to equity holders. Oil price exposure is increased by volatility, because the likelihood to default is increased due to the increased probability to suspend production, which brings enormous costs by executing this managerial option. Oil price exposure is also increased by interest rates and the firm’s leverage ratio because of the increased debt payments. Hedging activities are not found as significant determinant of oil price exposure. A possible explanation is that the petroleum industry is homogeneous and oil producers disclose much value-relevant information and therefore investors hedges the exposures themselves.
JEL classification: G12, G15, G30
1. INTRODUCTION ...4
2. LITERATURE REVIEW ...8
2.1 Oil price exposure ...8
2.2 Determinants of oil price exposure ...9
3. MEASURING TOTAL OIL PRICE EXPOSURE ...15
3.1 Methodology used to estimate oil price exposure ...15
3.2 Sample ...17
3.3 Data used to estimate oil price exposure ...18
3.3 Estimated oil price exposure ...21
4. THE DETERMINANTS OF OIL PRICE EXPOSURE ...24
4.1 Methodology used to estimate the determinants ...24
4.2 Data used to estimate the determinants ...27
4.3 Empirical results ...34
5. CONCLUSION ...40
5.1 Determinants of oil price exposure ...40
5.2 Limitations to the study and further research ...41
REFERENCES ...42
APPENDICES...44
Appendix A: Results of Ordinary Least Squares (OLS) regression ...44
Appendix B: Results of the Hausman (1978) test...44
Appendix C: Correlation Between ‘quantity of production’ and ‘reserves’...44
Appendix D: Results of the random regression model with replacement of determinants by a dummy...45
Appendix E: Results of the Generalized Regression Procedure (GLS) ...46
1. Introduction
The primary goal of this study is to identify and prioritize the determinants of oil price exposure of the U.S. oil producers, wherein the determinants are categorized into macroeconomic factors (level of oil prices, return volatility of oil prices, and interest rates) and firm-specific factors (leverage ratio, production costs, quantity of production, reserves, percentage hedged, and hedge price). No previous research has investigated the firm-specific determinants for the petroleum industry. The consideration of firm-specific determinants is in interest to risk managers, investors, and managers who are compensated partly with equity (Hong and Sarkar, 2008). By increased knowledge on how exposures affect firm’s cash flows and how actual exposures are influenced by their determinants, corporate managers can adjust the sensitivity of the firm’s equity value to unexpected oil price changes. For example by making less use of debt financing or by engaging in hedging activities.
The interest is also underlined in previous studies (e.g. Faff and Brailsford, 1999; Sadorsky, 2001; Lanza et al., 2005; Boyer and Filion, 2007; Nandha and Faff, 2008; and Scholtens and Wang, 2008) that show the significant impact of unexpected oil prices on the equity value of the firm. Moreover, its interest is underlined by the prominent examples about falling stock prices due to unexpected oil price
changes that have ‘headlined’ the financial press in the 21st century (Faff and Hillier, 2002). That firms
actually do care about unexpected oil price changes is confirmed in the annual reports wherein they report that they care and try to manage exposure to commodity prices. I quote from the annual report 2005 of Brigham Exploration Co: “We use derivative instruments to manage exposure to commodity prices. Our objectives for holding derivatives are to achieve a consistent level of cash flow to support a portion of our planned capital spending. Our use of derivative instruments for hedging activities could materially affect our results of operations in particular quarterly or annual periods since such instruments can limit our ability to benefit from favorable price movements. We do not enter into derivative instruments for trading purposes”. So evidence exists that firms should care and also that firms actually do care about unexpected oil price changes.
Considering that oil price exposure has a significant impact on the equity value of the firm, one could argue that every factor affecting oil price exposure is also determinative for the future value of the equity of the firm. Keeping in mind this consideration and major importance of oil price exposure, the
research question in this thesis is as follows:
macroeconomic but also firm-specific determinants that affect the sensitivity of the firm’s equity value to unexpected changes in the oil price. As discussed at the beginning of this chapter, there has been some research looking into the relationship between oil prices and equity value. But only little research is published that is looking into its determinants. And these few researches looking into its determinants did all focus at the gold industry (e.g. Blose and Shieh, 1995; Tufano, 1998; and Hong and Sarkar, 2008). Except for Faff and Hillier (2002) who did focus at the petroleum industry and conclude that the underlying macroeconomic factors of the theoretical valuation model are also valid to the petroleum industry. They did, however, not consider firm-specific factors, as is done in this study. The determinants that I use in this study are derived from the theoretical valuation model of Brennan and Schwartz (1985). They developed a model that considers both macroeconomic and firm-specific factors as determinants of commodity price exposure. Furthermore they consider that firms have managerial flexibility. More specific, firms are able to temporarily or permanently suspend production. Basic valuation models, such as the discounted cash flow model, do not consider the flexibility in business decisions (Koller et al, 2005). Using the theoretical valuation model that considers managerial flexibility is more realistic, and therefore of increased interest to risk managers, investors, and corporate managers. Consequently it is of more interest to this study.
unexpected changes in oil prices is still possible, but simply equal to the market average. The advantages of using the ‘residual’ regression model are twofold. First, the precision of exposure sensitivity estimates is improved due to the reduction of residual variance. Second, oil prices could be correlated with macroeconomic factors and the insertion of a “market” portfolio controls for the impact on equity value (Bodnar and Wong, 2003). In this study I estimate ‘total’ exposure by using the regression of ‘residual’ exposure to be able to benefit from these advantages. And in order to assure the interpretation is still possible in absolute terms, I make sure the unexpected change in oil prices are negligible to the included “market” portfolio, which is described in chapter 3 in a greater detail. Moreover, I estimate the oil price exposure over a daily, weekly, and monthly return horizon in order to control for the sensitivity of estimating over longer return horizon. Bodnar and Wong (2003) argue that different return horizons could cause significant differences in empirical research, because of the presence of time varying (possibly spurious) correlations of the exposure with other (macroeconomic) variables. Subsequently, the determinants of oil price exposure are also estimated over these three return horizons.
In the second step the determinants are regressed against the estimated exposures using random regression modeling. This regression method gives the most efficient estimation, because degrees of freedom are saved and therefore more information can be used to find the best fit of the model (Baltagi, 1995). Finally, the determinants by using a Generalized Least Squares (GLS) procedure is estimated in altering the possibility that the variance of the disturbance terms is not constant (called heteroscedasticity) and that the disturbance terms are correlated with one another (called autocorrelation).
In the regressions I use panel data (time-series and cross-sectional data) of 63 U.S. listed oil producers at the NASDAQ, NYSE, and AMEX over the period 2003 through 2008. The main reasons for this sample is that focusing on oil producers offer a cleaner capture of the underlying determinants; it is compulsory for firms listed at the U.S. exchange markets to report firm-specific variables; and these three major exchange markets best reflect events in the firm’s equity value. In the data section of chapter 3 the choices for my sample are discussed in greater detail.
industry combined with the value-relevant information reported by oil producers. Subsequently, they can hedge the firm’s oil price exposures themselves.
2. Literature review
Many empirical studies about the subject of exposures are available. These studies can be clustered into three groups of contribution – commodity price exposure, exchange rate exposure, and interest rate exposure. The first, and most relevant to this study, evaluates determinants of exposure to exhaustible resource prices for oil producers and gold mining firms (e.g Tufano, 1998; and Faff and Hillier, 2002). The second evaluates the determinants of exposure to exchange rate fluctuations for multinational corporations (e.g. Chow and Chen, 1998; and Dominguez and Tesar, 2006). The third group considers the determinants of exposure to interest rate changes for financial institutions (e.g. Flannery and James, 1984; Elyasiani and Mansur, 1998; and Staikouras, 2003). In this section the focus is at the determinants of exposure to exhaustible resource prices. First, oil price exposure in its context is defined. Second, the theoretical models that analytically predict the determinants of commodity price exposure are discussed. Finally, the empirical results of previous research are given for each determinant, based on studied literature. The theory forms the base for the formulated hypotheses in the last section of this chapter.
2.1 Oil price exposure
To define oil price exposure a sharp distinction should be made between risk and exposure. Adler and Dumas (1985) argue that risk is identified with statistical quantities that summarize the probability that the actual price on a given future date will differ from its originally anticipated value. Risk, also called uncertainty, is a question of randomness, in explanation, unexpected price variations. Exposure, in contrast, is the effect of unexpected price variations on the firm’s equity value. In this study oil price exposure is defined as the sensitivity of the future value of the equity of the firm to the future value of the oil price.
terms of corporate profit.
The magnitude of oil price exposure is empirically tested by Faff and Hillier (2002) and found surprisingly small in comparison with the gold price exposure. In general, they found that an unexpected oil price change of 1 percent typically results in an increase in the equity value of the firm by 0.2 percent. These findings are in contrast with the gold mining industry (e.g. Blose and Shieh, 1995; Tufano, 1998; and Hong and Sarkar, 2008), where unexpected gold price changes are reflected in the equity value of the firm by greater than one. It should be noted that attempts to measure the magnitude of exposures often results in limited success regarding statistical significance. This phenomenon is given the term “exposure puzzle” (Bartram and Bodnar, 2007). Examples of attempts to solve the exposure puzzle are in the selectivity of firms (targeted subsamples versus entire populations), in the choice of the measurement of the explanatory variable (e.g. exchange rates), and by using lagged exposures instead of contemporaneous exposure effects. According to Bartram and Bodnar (2007) none of these considerations satisfactorily explain the low numbers of firms with significant exposures. As alternative to the attempts to solve the exposure puzzle they argue that exposures may be difficult to detect empirically, because firms take rational actions and therefore hedge their exposures. In explanation, firms that are highly exposed to unexpected oil price changes are most likely involved in hedging activities. The estimated exposures are therefore the exposure that remains after hedging activities. Nevertheless, estimated exposures remain economically meaningful and therefore can be related to macroeconomic and firm-specific factors in a way that is consistent with economic theory (Bartram and Bodnar, 2007). The objective of this study – estimating the determinants of oil price exposure – is not influenced by the possible insignificance of estimated exposures. In conclusion, despite possible insignificances in exposures as a consequence of hedging
activities, overall, exposures are expected to be positive and small in magnitude.
2.2 Determinants of oil price exposure
Theoretical valuation model with managerial flexibility
option to temporarily or permanently delay production and the possibility where firms are able selling forward its entire production. Blose and Shieh (1995) develop a similar theoretical valuation model as Brennan and Schwartz (1985). Their model estimates the theoretical sensitivity of the future equity value of the firm to unexpected changes in the gold price. A sample of 23 publicly traded gold mining firms over the period 1981 through 1990 is used for empirically testing the variables in their model.
Tufano (1998) develops a theoretical valuation model that is basically the model of Brennan and
Schwartz (1985). Tufano (1998) underlines the importance of real option (managerial flexibility) in theoretical valuation models, because discounted cash flow models systematically overestimate exposures, which is possibly the consequence of their failure to reflect managerial flexibility. A sample of 48 North American gold mining firms over the period 1990 through 1993 is used for empirically testing the variables in the model. Hong and Sarkar (2008) adjust the theoretical valuation model for a levered firm with a mean-reverting output price process such as the prices of commodities; they argue that no rigorous theoretical model in this context is yet developed. The model
is tested empirically using a sample of 30 U.S. gold mining firms.1
In these studies the corporate managers are provided with a guideline about how managerial decisions – in particular, hedging decisions – affect the commodity price exposure and to respond to investor’s concern that a firm’s hedging decisions may affect exposures (Tufano, 1998).
All these theoretical valuation models are applied to the gold mining industry. However, Faff and
Hillier (2002) find evidence that the macroeconomic determinants of the theoretical valuation model
of Brennan and Schwartz (1985) are similar for oil producers. A sample of 127 oil companies, across 21 countries, is used over the period 1990 through 2005.
Empirical findings of previous studies and hypotheses formulated
In Table 2.1 the empirical findings of previous studies about the determinants of commodity price exposure are shown. The macroeconomic determinants are commodity prices, return volatility of commodity prices, and interest rates. The firm-specific determinants are financial leverage, production costs, quantity of production, reserves, percentage hedged, and hedge price. Below Table 2.1 the empirical findings and theoretical explanations are given in more detail. At the end of each paragraph wherein a determinant is discussed, the hypotheses are formulated.
1 The mean-reverting rate is not included in my model, because the stationary oil prices in my sample suggest that oil prices
Commodity price exposure is found by most authors to be a decreasing function of the level of
commodity prices. Oil producers and gold mining firms benefit from an increase in the commodity
price due to the higher net cash flow per output. If the commodity price falls, considering the real option model that includes managerial flexibility, the decision to temporarily or permanently suspending production becomes more realistic, and therefore increases the uncertainty of cash flows. In the reverse case, if commodity prices increase, positive net cash flow become more certain and therefore reduces the sensitivity of the equity value of the firm to unexpected commodity price changes (Hong and Sarkar, 2008).
However, Faff and Hillier (2002) found a positive effect between commodity price exposure and the level of commodity prices. They argue that this unexpected sign possibly reflects a decrease in value of the real options to temporarily or permanently suspend production. Or, in case production is already
closed, it reflects an increase in value of the real entry options to reopen the production process.2 As a
result, the effect of commodity prices on commodity price exposure is influenced by effects outside the model and is no longer a correct capture of the effect. Based on this theory that commodity price increases make positive net cash flows more certain, I formulate the first hypothesis as follows:
H1: The level of crude oil prices has a negative significant effect on oil price exposure.
Commodity price exposure is also found as a decreasing function of return volatility of oil prices for oil producers. The explanation is found in the basic option theory (Hong and Sarkar, 2008; Tufano, 1998). The sensitivity to cash flows decrease when volatility is high due to the possibility of anticipation by firms, to produce when oil prices are high and shut down when oil prices are low. The
2 Faff and Hillier (2002) argue that this topic deserves a separate treatment, which they leave for further research.
Tabel 2.1: Empirical findings of previous studies about the determinants of commodity price exposure
Blose and Shieh Tufano Hong and Sarkar Faff and Hillier
(1995) (1998)* (2008) (2002)
Dependent Commodity price exposure Gold Gold Gold Oil
Independent Commodity price - *** - *** - *** (+) **
Return volatility of commodity price - *** - *** - **
Interest rates +/- ** (-) ** - ** Financial leverage + *** + Operating costs + *** + + *** Quantity of production -Percentage hedged - *** Hedge price - ** Control Reserves - *** 0
*** Indicates statistical significance at the 0.01 level ** Indicates statistical significance at the 0.05 level * Indicates statistical significance at the 0.10 level
(…) The results given in brackets are unexpected signs (i.e. the contrary sign that is predicted by the author)
Notes: * Significance results of Tufano (1998) are based on the joint estimation model (i.e. regressing the determinants directly on the return of stock prices).
degree of volatility contributes to this process by optimal exploitation of the real option. This result is similar to the elasticity of the value of a call option; at default the payoff is zero. The second hypothesis of this research is formulated as follows:
H2: The return volatility of oil prices has a negative significant effect on the oil price exposure.
The expected sign of the interest rate is found to be ambiguous. Brennan and Schwartz (1985) make a distinction between different interest rates. They argue that the 10-year Treasury bond is used to discount expenses and the gold lease rate (cost of physical ownership of crude oil) to discount revenues. Empirical results of Tufano (1998) and Faff and Hillier (2002) show that the 10-year Treasury bond has indeed the predicted inverse relationship with commodity price exposure. Tufano (1998) also empirically shows that the gold lease rate has the predicted positive effect on commodity price exposure. Hong and Sarkar (2008) use the yield on 3-month Treasury bills for discounting revenues and determinant of coupon payments in case a firm is levered. The empirical evidence though, shows a negative relationship between commodity price exposure and interest rates, suggesting that rising interest rates lower oil price exposure. This finding is surprising since Boyer and Filion (2007) argue that the capital needed for operating in the petroleum industry is enormous. Oil producers need large investments to renew and find reserves to meet their growth objectives and to be profitable. Considering this capital intensity makes external financing unavoidable and hence the use of debt, one would expect that rising interest rates increases the coupon payments and thus increases oil price exposure. Taking into consideration the 3-month Treasury bill as proxy for discounting revenues and determinant of coupon payments, I formulate the third hypothesis as follows:
H3: The interest rates have a positive significant effect on the oil price exposure.
Moreover, considering the discount rate of revenues, it is no surprise commodity price exposure is also a decreasing function of the leverage ratio. By increasing the debt level, the discount rate increases and the firm becomes more financially constrained, which, in turn increases the likelihood to become unable to payout their equity holders. If the firm defaults the bondholders take over and the equity value is zero. Hong and Sarkar (2008) find no significant relation between the leverage ratio and commodity price exposure and argue this result possibly occurred due to the endogenously determination of the leverage ratio by the firm itself. The fourth hypothesis is formulated as follows:
Commodity price exposure is found to be an increasing function of operating costs (both constant and proportional), because increasing operating costs reduce the certainty to positive net cash flows and equity value of the firm becomes more sensitive to unexpected commodity price changes (explanation is similar to decreasing commodity prices) (Hong and Sarkar, 2008). Contrary to the prediction of the theoretical valuation models, firm’s operating costs are found insignificant in Tufano’s (1998) study. A possible explanation could be incorrect data (Tufano, 1998). In this study I focus solely on the effect of proportional costs and define the fifth hypothesis as follows:
H5: The production costs have a positive significant effect on the oil price exposure.
Commodity price exposure is found as decreasing function of the quantity of production, holding reserves constant in the model. An increase in production should be beneficial for a firm since it increases its revenue and therefore reduces the probability that firms become financially constrained (Hong and Sarkar, 2008). Besides, an increase in production is often associated with an increase in commodity price, which is also predicted and empirically shown to be negatively related to commodity price exposure (Tufano, 1998). Keeping reserves constant is crucial to obtain a clean capture of the quantity of production, because fluctuating reserves suggest that the firm anticipates to increase profits from temporally price fluctuations and/or local shortages of the commodity (Blose and Shieh, 1995). Include reserves as control variable assures the clean capture of anticipating in the quantity of production to investment opportunities. The sixth hypothesis is formulated as follows:
H6: The quantity of production has a negative significant effect on the oil price exposure.
Commodity price exposure is also a decreasing function of both the percentage hedged and the hedge
price. The percentage hedged is measured in the previous studies (e.g Tufano, 1998; and Jin and
H7: The percentage hedged has a negative significant effect on the oil price exposure.
Additionally, in the case the exposure is not completely eliminated by the percentage hedged, the
forward price at which the commodity has been sold forward also plays a significant role (Tufano,
1998). The explanation is similar to that of the level of commodity prices. Although Tufano (1998) finds significant results for the forward price, the coefficient is quite small. An argument for the small coefficient is that the investors most likely do not understand how to correctly interpret this “second-order” hedging information. They do consider percentage hedged but the hedge price is less important by an investor’s judgment. Despite the small coefficient, considering the fact that a higher forward price reduces the sensitivity of the equity value of the firm to unexpected commodity price changes, because of the certainty of fixed income. Hypothesis eight is formulated as follows:
3. Measuring total oil price exposure
In this chapter the first step of the two-step regression model is estimated. First, the issues associated by measuring oil price exposure are discussed. Second, the sample for this study is described. The sample is used in both the first and second step of the two-step regression model. Also, the data used to estimate the exposures are described in the third section. And finally, the sign, magnitude, and statistical significance of the estimated exposures are presented, which is the input for the second step of the two-step regression model.
3.1 Methodology used to estimate oil price exposure
Now the oil price exposure is defined as the sensitivity of the future value of the equity of the firm to the future value of the oil price, the question at hand is “how to measure it”. Adler and Dumas (1984) argue that the impact of unexpected changes in the oil price on the firm’s cash flows is the most precise measure for a firm itself. This measurement requires intern data, which is hard to obtain by researchers if these cash flows are not publicly presented. As alternative, the unexpected change in oil prices is usually related to the firm’s stock price, since stock prices represent expected cash flows (Adler and Dumas, 1984). Keeping this in mind suggests that the oil price exposures can be estimated by the use of regression analysis wherein the changes in oil prices are regressed on the firm’s stock returns.
I use the following econometric specification for estimating oil price exposures (Equation 1):
€
R
i,t=
α
i+
β
1,iΔR
o,t+
β
2,iR
w,t+
ε
i,t (1)where Ri represents the total stock price return (including dividends), ∆Ro is the change in the price of
the 3-month future oil prices: West Texas Intermediate (WTI), εi,t is the disturbance term, Rw is the
return on the Morgan Stanley Capital International (MSCI) world portfolio, the i subscript refers to the
firms, and the t subscript refers to the time aspect. The coefficient β2 is the firm’s beta with respect to
the market portfolio; β1 is the sensitivity in the firm’s stock price (equity value) that can be explained
by unexpected movements in the oil price after conditioning on the return on a world portfolio. Thus
β1 reflects the oil price exposure.
Several important questions regarding this regression equation are addressed before turning to the data
description and estimation results. First, why is the term β2,iRw,t included in Equation 1 to measure the
why is the coefficient β1,i in Equation 1 still be interpretable as an estimate of total oil price exposure,
as the inclusion of the term β2,iRw,t seems to suggest that β1 is only a measurement of residual oil price
exposure? Finally, what regression technique should be used?
To address the first question why the world portfolio is included in Equation 1, note that excluding a
“market” portfolio (β2,iRw,t) is problematic, because particular variables (e.g. other macroeconomic
variables) are most likely correlated with the oil price change (∆Ro,t,) and uncorrelated with the
disturbance term (εi,t) (Chue and Cook, 2008). To make consistent and unbiased estimates a variable
must be uncorrelated with the disturbance term (εi,t) (Brooks, 2008). It should be noted that other
macroeconomic variables, besides the oil prices, could also be correlated to firm’s stock prices. By excluding a world portfolio, any affect of other macroeconomic variables on firm’s stock prices could be captured in the disturbance term. This possibility implies that the correlation between the oil price change and the disturbance term can be nonzero, violating the assumptions of a valid regression model. Therefore I include a world portfolio to absorb any remaining correlation between other macroeconomic variables and the disturbance term.
To answer the second question whether we can still interpret the coefficient β1,i in Equation 1 as an
estimate of total oil price exposure, note that the exogenous movement of oil price changes should be negligible on the world portfolio. To test whether oil price changes are negligible on the world portfolio, the following econometric specification is used (Equation 2):
€
R
w,t=
α + β
1ΔR
o,t+
ε
t (2)where Rw,t is the return on the MSCI world portfolio, ∆Ro,t is the change in the price of the 3-month
future oil prices: WTI, εt is the disturbance term, and the t subscript refers to daily, weekly, and
monthly data. Results are shown in Appendix A and show a significant result for each return horizon that an exogenous unexpected oil price change does affect the world portfolio and is therefore not negligible on the world portfolio. By using the uncorrelated disturbance terms of Equation 2 as proxy for the world portfolio in Equation 1, unexpected oil price changes become exogenous and therefore it
is still possible to interpret the coefficient β1,i as total oil price exposure (Grinblatt and Titman, 2002).
- The variance of the disturbance terms is constant; - The disturbance terms are uncorrelated with one another;
- The explanatory variables are not correlated with the disturbance terms, and; - The disturbance terms are normally distributed.
It is desirable to estimate with OLS if possible, since its behaviour is well known (Brooks, 2008). Moreover, the most likely violation in the regression model is endogeneity, which is countered by Chue and Cook (2008) by estimating exposures using the instrumental variables (IV) technique that can be validly used in the presence of endogeneity. Chue and Cook (2008) argue that the exposures estimated using OLS are similar in sign and magnitude to the exposures estimated using the method of IV. They do note the importance of controlling for endogeneity. Since I control for endogeneity by including a world portfolio, estimating Equation 1 using OLS seems appropriate.
3.2 Sample
In order to analyze the effect of the determinants on oil price exposure, a sample of 63 U.S. oil producers (SIC code 1311 “Oil and Gas Extraction: Crude Petroleum and Natural Gas” – upstream) listed at the NASDAQ, NYSE, or AMEX is used over the period 2003 through 2008. In total 378 firm-years are obtained. This study focuses at oil producers because these firms are operating in the upstream. Firms that are net buyers of crude oil do probably react differently to the underlying determinants of oil price exposure. As discussed before, oil producers should benefit from unexpected oil price rises due to the favorable effect on their output. In the reverse case, net buyers should experience a disadvantage from unexpected oil price rises due to the unfavorable effect on their output (Nandha and Faff, 2008). In explanation, an increasing oil price should increase the profit margin of oil producers and decrease the profit margin for net buyers of oil. Also firms that operate in all three segments, up- mid- and downstream, do react most likely different to the underlying determinants of
oil price exposure, because they have less assets related to the oil price.3 As result of less assets related
to the oil price the possibility exists the power of the tests are weakened (Faff and Hillier, 2002). Thus focusing only at the oil producers captures the underlying determinants the best, which is an important advantage for obtaining significant results. Focusing at firms listed at the U.S. exchange markets is because both macroeconomic and firm-specific variables are guaranteed collectable. The U.S. Securities and Exchange Commission obligates U.S. listed firms to report reserves; production quantity; production costs; and hedging activities. The three major American stock exchange markets – NASDAQ, NYSE, and AMEX – are most frequently traded, which reduces the possibility of
3 The upstream refers to exploration and production activities; midstream to the transportation sector – often the shipping of
estimation error due to the lag to respond to new information that became available about the stock prices. Consequently, the stock price is most likely consistent with price trends and therefore no adjustment is made to the estimated exposures to correct for infrequently traded stocks (Grinblatt and Titman, 2002; Dimson, 1979; Fowler and Rorke, 1983).
3.3 Data used to estimate oil price exposure
Data collection of oil price exposure
The sample is collected with DataStream. Data should be available from the beginning of year 2002 to be able to estimate the regressions over the period 2003 through 2008, because the use of natural logarithms in the variables forces the exclusion of 2002. First, an initial sample was extracted from DataStream of 110 oil and gas producers. In total 47 firms are excluded because of the highly restricted data necessary such as production costs, quantity of oil production, and reserves. The reasons for exclusion of firms are:
- No operations are related to oil exploration and production but are labeled as ‘oil and gas producer’, because they focus at commodities other than oil, such as gas or copper;
- No operations are related to oil exploration and production but are labeled as ‘oil and gas producer’, because a shift in operations to the mid- and/or downstream of oil did take place during the years and not yet corrected for;
- Firms are income trusts4 and therefore do not report their firm-specific variables or solely the
estimated quantities that are attributable to the firm;
- Firms are operating in multiple segments, up- mid- and downstream, which has the possibility to weaken the power of the tests;
- Firms have non-December fiscal year-ends, which is required in order to match the equity returns to unexpected price changes in crude oil;
- All firm-specific variables are unavailable. Appendix F lists all the firms included in this study.
4 Income trusts do not operate in the petroleum industry themselves, but receive interests, royalties or lease payments from oil
Each variable is collected by daily, weekly, and monthly data to determine the sensitivity of return horizon by estimating exposure, which is discussed as important feature of the ‘exposure puzzle’ by Bartram and Bodnar (2007). Although measuring exposures by different return horizons should be irrelevant due to the direct and accurate reflection in the firm’s stock prices assuming market efficiency and complete information, empirical results have pointed out the return horizon is relevant (Bodnar and Wong, 2003). They argue that estimating exposures by different return horizons could cause significant differences in empirical results due to time varying (possibly spurious) correlations between exposures and macroeconomic variables, which also affect the equity value of the firm. In their empirical results they find increasing exposures in their magnitude (in absolute terms) and a larger number of significant exposures. In this study no significant differences in the empirical results are expected as consequence of time varying (possibly spurious) correlations, because I add a world portfolio in the regression model to absorb any remaining correlations between the other macroeconomic factors and the disturbance term. The exposures and their determinants are estimated over different time horizons, because they are a solid robustness check to the estimation sensitivity. More specific, estimated yearly exposures are the result of the use of 262 data points by using daily data; 52 data points by using weekly data; and 12 data points by using monthly data.
The data for stock prices are total stock prices (including dividends) to prevent biases due to the consequences of firms that pay dividends and influence stock prices by this policy. The data used for oil prices is the 3-month NYMEX future prices on WTI crude oil ($ per barrel). Future prices are preferred over spot prices of crude oil, because exposure is a matter of the future and therefore should depend on the covariance of future equity value of the firm with future oil prices (Adler and Dumas, 1984). The idea of a covariance between a non-random current equity value of the firm and random future oil price changes is illogical. The reasons for using WTI futures are twofold. First, the WTI is most widely used in North America. Second, if oil producers use derivative instruments for hedging to affect their exposures, the majority uses futures, forwards, and other derivative contracts based on the WTI oil prices (Boyer and Filion, 2007). Thus if firms are engaged in hedging activities, the appropriate effect on exposures (as pleased by the firm itself) is captured due to the use of the same benchmark to estimate the exposure. And finally, the reason to use the 3-month future prices is
Tabel 3.1: Collection of variables to estimate the oil price exposure Variable Definition and Source
Dependent Stock Prices Total stock prices (including dividends) of firm i (Datastream)
Independent Oil Prices 3-month NYMEX future oil prices (Energy Information Administration)*
Control World Portfolio Price index of the MSCI World Index (Datastream)
because the majority of futures with maturities beyond the 3-months are such low trading that their use would be impractical (Haushalter, 2002). Moreover, it is reasonable to assume that the maturities are highly correlated. In Figure 3.1 the oil prices are shown for my sample. An increasing oil price is seen and an increased volatility. The average trading in 2008 is above $100 a barrel, which is more than triple the amount what it was traded at in 2003.
In Table 3.2 is the construction of the variables used in Equation 1 summarized.
The natural logarithm is taken of the stock prices, oil prices, and world portfolio index to estimate oil price exposures. Consequently, it is reasonable to assume that taking the natural logarithm results in a constant mean, variance, and autocovariance for each variable included. Hence, fulfilling the requirements of stationarity and making valid inferences possible.
3.3 Estimated oil price exposure
In this sample part of the variation in oil price exposure (the dependent variable) is attributable to time-series variation and the remainder to cross-sectional differences, as shown in Figure 3.2 and Figure 3.3. As predicted, the oil producers are positively exposed to unexpected oil price changes. The median exposures vary over the years with a minimum of 0.12 and a maximum of 1.06, as shown in Figure 3.2. Furthermore, the median and spread of oil price exposures increases as the time horizon for the return calculation lengthens. Bartram and Bodnar (2007) argue that exchange rate exposures increase in absolute terms if return horizon lengthens, because equity returns tend to theoretically grow linearly with time and exchange rate changes tend to mean revert towards zero. Although, no mean-reversion in oil prices is found in my sample, equity values do increase faster than oil prices with time since oil price exposures increases as return horizon lengthens.
Tabel 3.2: Construction of variables to estimate the oil price exposure
Variable Symbol Construction
Dependent Stock Prices Ri ln[(Stock price of individual firm at period t) / (Stock price of individual firm at period t-1)
Independent Oil Prices !Ro ln[(3-month NYMEX future oil price at period t) / (3-month NYMEX future oil price at period t-1)
significant. None of the found negative values are significant, which underlines the theory about the positive role of an unexpected oil price change on the equity value of the firm. Insignificant exposures are most likely the consequence of hedging activities. But as argued before, although exposures are insignificant they remain economically meaningful (Bartram and Bodnar, 2007).
Finally, Table 3.4 shows the number of firms that are constantly below or above the median exposure. If a firm’s exposure is constantly below or above the median exposure, it could be a first indication that the exposures are mainly influenced by macroeconomic variables.
These results can be interpreted as a first indication that exposures are not solely influenced by macroeconomic variables but also by firm-specific variables. In the next section the determinants are investigated in greater detail.
To summarize, in this chapter the first step of the two-step regression model is carried out. The signs are found to be positive and confirm the expectation that oil producers benefit from rising oil prices. The median oil price exposure over the daily return horizon is 0.38. The magnitude increases if return horizon lengthens to a median of 0.47 for weekly data and 0.76 for monthly data. The increasing magnitude underlines the theory that oil prices grow at a slower speed than stock price returns. Furthermore, movements in a firm’s exposure is not equal to movements in the averaged sample exposures, suggesting exposures are influenced both by macroeconomic and firm-specific variables.
Tabel 3.3: Significance of oil price exposures expressed in percentages of the total sample
1% significance level 5% significance level 10% significance level
Year daily data weekly data monthly data daily data weekly data monthly data daily data weekly data monthly data
2003 33% 11% 13% 44% 32% 33% 56% 41% 41% 2004 73% 48% 2% 81% 63% 6% 83% 71% 8% 2005 86% 65% 24% 89% 81% 48% 90% 81% 60% 2006 90% 79% 35% 90% 87% 73% 90% 87% 81% 2007 78% 48% 11% 86% 60% 48% 87% 63% 65% 2008 79% 84% 52% 84% 92% 79% 87% 92% 83% Average 73% 56% 23% 79% 69% 48% 82% 73% 56% Notes: daily data refers to the oil price exposures that are estimated over the daily return horizon; weekly refers to the weekly return horizon; and monthly to the monthly return horizon.
Tabel 3.4: Firm's exposure continu above or below exposure median daily data weekly data monthly data
Number of firms 8 4 5
Percentage of firms 13% 6% 8%
Observations 63 63 63
24
4. The determinants of oil price exposure
In the previous chapter the sign, magnitude, and statistical significance of oil price exposure is estimated, which is used as input for the second step of the two-step regression model. In this chapter the second step is estimated, which is the objective of this study. First, the appropriateness of the random regression model is discussed in the methodology. Second, the data collection and construction for the determinants are described. Also, the descriptive statistics and correlations are presented in the data section. Finally, the sign, magnitude, and statistical significance of the determinants are estimated and discussed in the empirical results section. To enhance the confidence in the results explanatory variables are replaced by dummies and the generalized least squares estimation procedure is used to control for estimation sensitivity.
4.1 Methodology used to estimate the determinants
By measuring the determinants that are mentioned in the hypotheses and comparing them to the estimated exposures, contributes to the knowledge about how actual exposures are set. In this section I describe how the determinants are compared to the estimated exposures.
The determinants of oil price exposure are investigated by using balanced panel data (cross-sectional and time-series), in which the regression has a double subscript on its variables (Brooks, 2008). The choice of using panel data is because by combining cross-sectional and time series data; the number of degrees of freedom is increased, and therefore increases the power of econometric estimates and also decreases problems with multicollinearity among determinants. The increased power of econometric estimates and decreased problems with multicollinearity is because the gap between the information requirements of the model and the information provided by the data decreases (Baltagi, 1995). The setup of a balanced panel data is used when the cross-sectional elements are observed over the entire sample period and no randomly missing observations are present, resulting in a dataset containing 378 firm-years.
Using a regression technique for panel data can develop an understanding of the relation of the macroeconomic and firm-specific characteristics with the estimated exposures. The simplest form to analyze panel data is in a pooled regression, which assumes that the intercept is equal for each firm and each year. The model can be specified as (Equation 3):
estimated on the explanatory variables, X is a 1 x k vector of observations on the explanatory
variables, and ui,t the disturbance term. The t subscript refers to the time-series dimension (t = 1, 2, 3,
..., T) and the i subscript refers to the firms (i = 1, 2, 3, ..., N).
While this is a simple way to analyze panel data because only one equation is required, the pooled regression has the same strict assumptions as an OLS regression, as described in section 3.1. The most important assumption is that the pooled regression method implicitly assumes homoscedacticity between the variables across time and firms in the sample (Brooks, 2008). Since such behavior of variables is uncommon in financial research (Brooks, 2008), two broad classes of panel estimator approaches can be used to overcome its limitations: fixed effects model and random effects model. Both are explained next.
The fixed effects model decomposes the disturbance term (ui,t), such that it allows an intercept in the
regression to differ among firms but not over time, with the individual specific effect (µi) (Brooks,
2008). Carrying out the fixed effects model results in a constant for every firm that is included in the regression model instead of having one intercept (α) for the regression. The fixed effects model can be mathematical expressed the same as the pooled regression, but with the disturbance decomposed into (Equation 4):
€
u
i,t=
µ
i+ v
i,t (4)where ui,t is the disturbance term, decomposed into µi, an individual specific effect that is not included
in the regression and is constant over time, and vi,t, the remainder disturbance, that varies over time
and cross-sectional units (capturing everything that is left unexplained about y) and can be thought of as the usual disturbance in the regression. A drawback of the fixed effects model is that it suffers from a large loss of degrees of freedom as the number of variables increase because the model creates a dummy for each explanatory variable in order to obtain the different intercepts among firms. By creating dummies the gap between the information provided by the data and the information required by the model increases (Baltagi, 1995). Therefore, in this study the random effects model is preferred over the fixed effects model since degrees of freedom can be saved and thus a more efficient estimation will be reached (Baltagi, 1995). Similar to the fixed effects, the random effects model also assumes varying intercepts among firms, which are constant over time. The difference between a fixed and random effects model is that under the random effects model the intercepts for each firm are
assumed to arise from a common intercept (α) and a random variable (εi) (Brooks, 2008). The random
term (α). The heterogeneity of the cross-sectional dimension is captured by the random variable (εi) where the dummy variables do this in the fixed effects model. The random effects model can be specified as (Equation 5):
€
y
i,t=
α + βX
i,t+ω
i,t,€
ω
i,t=
ε
i+ v
i,t (5)where α is the common intercept, β is a k x 1 vector, Xi,t is a 1 x k vector, ωi,t is the composite
disturbance term, εi is the random variable, and vi,t is the remaining disturbance.
It should be noted that the random effects model is only valid when the composite disturbance term
(ωi,t) is uncorrelated with all of the explanatory variables, because otherwise the estimates will be
biased and inconsistent (Brooks, 2008). The Hausman test (1978) is performed to investigate whether
the composite disturbance term (ωi,t) is uncorrelated with explanatory variables (Xi,t). The test results,
shown in Appendix B, indicate that the p-value for daily, weekly, and monthly return horizons are insignificant and conclude that the random effects model is the appropriate regression technique for my sample. All hypotheses are tested using the random effects models.
Furthermore, in the results robustness section I transform the regression technique to a generalized
least squares (GLS) procedure, in order to control for heteroscedasticity and autocorrelation. The
GLS subtracts a weighted mean (φ) of the dependent variable (yi,t) and independent variables (xi,t) over
time, and therefore ensures that there is no correlations between the firms in the disturbance terms.
The weighted mean (φ) is a function of the variance of the remaining disturbance term (vi,t) and of the
firm-specific random variable (εi). Since the coefficients of the determinants remain greatly the same
in sign, magnitude, and statistical significance (Appendix E), it is reasonable to assume that the estimates of the random effects model are unbiased, efficient, and consistent. The results of the GLS procedure are discussed in section 4.4 in greater detail.
To conclude, in order to obtain the determinants of the oil price exposure the following econometric specification is used (Equation 6):
€
β
i,t=
δ +
φ
jF
j,i,t+ω
i,tN
∑
,€
ω
i,t=
ε
i+ v
i,t (6)Data collection of the determinants
The data covers the period January 2003 till December 2008. In Table 4.1 is the collection of the raw data for the determinants of oil price exposure given.
Data for oil prices and return volatility of oil prices is the 3-month NYMEX future oil prices, as discussed in section 3.2. The data for interest rates is the yield on the 3-month U.S. Treasury Bills, which is also used by Hong and Sarkar (2008) to discount revenues and to determine the coupon rates. The leverage ratio is composed by the end-of-year book value of long-term debt and the end-of-year market value of the equity of the firm. The market value of the equity of the firm represents the share price multiplied by the number of ordinary shares in issue. Production costs in dollars ($) per Bbl (barrel of oil) are obtained manually from annual reports. I searched the body of the annual reports by keywords such as “production costs”, “operating costs”, “production expense”, and “operating expense”. In case firms report their production costs of oil and gas in a merged value, it is often reported in dollars ($) per Mcf (thousand cubic feet of natural gas). To convert the production costs in dollars ($) per Bbl (barrel of oil) I multiply the value by six (six thousand cubic feet of natural gas is equal to one barrel of oil). Although the theoretical valuation model of Brennan and Schwartz (1985) considers both the variable production costs and the fixed production costs as determinant of commodity price exposure, fixed costs are excluded in this study due to unavailability of the data. The quantity of production is also obtained manually from annual reports. The quantity of production is expressed in Bbl (barrel of oil) at the end of each year. The body is searched by keywords such as “Bbl”, “production”, and “operating results”. The amount of reserves can be distinguished in proved and probable. Proven reserves have a certainty of at least 90% and are the next reserves to be developed. Probable reserves have a certainty of 50% to be produced (Boyer and Filion, 2007). Since the proven reserves are more reliable than probable reserves, I use therefore the amount of proven reserves as proxy for this variable. The amount of reserves is also expressed in Bbl (barrel of oil) and manually obtained from annual reports. Finally, data for the total amount of barrels hedged in Bbl (barrels of oil) and the hedge price in dollars ($) per Bbl (barrel of oil) are also manually obtained from annual reports. The derivative program is often reported in ‘Item 7A. Quantitative and
Tabel 4.1: Data collection of the determinants of oil price exposure Determinant Definition and Source Dependent Oil price exposure Total oil price exposures (Excel)
Independent Oil Prices 3-month NYMEX future oil prices (Energy Information Administration) Volatility 3-month NYMEX future oil prices (Energy Information Administration) Interest rate Yield on the 3-month U.S. Treasury Bills (DataStream)
Leverage ratio End-of-year book value of long-term debt and the end-of-year market value of the equity of the firm (DataStream) Production cost Cost to extract oil in dollars ($) per barrel of oil (Bbl) (Annual reports)
Qualitative Disclosures About Market Risk’. Furthermore, I searched the entire body by keywords such as “hedge”, “hedging”, “swap”, “collar”, “derivative”, and “commodity risk”. If firms did not purchase any futures contracts or any derivative financial instruments they often report in their annual report “The Company did not enter into any hedging agreements to limit exposure to oil and gas price fluctuations”.
Data construction of the determinants
Before estimating the sign, magnitude, and significance of determinants, one should test for
non-stationarity, because applied econometric analysis requires constancy of means, variances, and
autocovariances to validly undertake hypothesis tests (Seddighi et al., 2000). Generally speaking,
stationarity is that previous values of the disturbance term (εi,t) do not have a lasting effect on the
dependent value (yi,t) as time progresses (Brooks, 2008). Non-stationarity expressed in mathematical
terms is (Equation 7):
€
y
i,t=
ρ
1,iy
i,t −1+
ρ
2,iy
i,t −2+
ρ
3,1y
i,t −3+ ...+
ρ
n,iy
i,t −n+
ε
i,t (7)Table 4.2 shows evidence of nonstationary data. By repeating the analysis on the first differences of the determinants, results are found significant (stationary) and therefore the data in first differences are used in the regressions. The data for volatility remains unchanged, though, because the first and second order does not increase the t-statistics. In my opinion taking the first differences of the volatility seems illogical at first sight, because the variable is estimated from the stationary returns of oil prices. Hence, volatility is used unadjusted in the regression analysis.
In Table 4.3 the construction of the determinants of oil price exposure is given, which fulfill to the stationarity requirements.
The oil price is the year average from the total data points. More specific, the estimated yearly oil prices are the result of the average of 262 data points by using daily data; 52 data points by using weekly data; and 12 data points by using monthly data. Consequently, the first difference of the natural logarithm of oil prices is taken to fulfill to the stationarity requirements. The return volatility of oil prices is constructed by running a GARCH(1,1) model on the returns of the oil prices. Volatility can be measured by other methods such as implied volatility series imputed from oil options data, which is future-looking data and therefore argued as more precise (Faff and Hillier, 2002).
Tabel 4.2: Stationarity of the determinants of oil price exposure
Determinant Daily return horizon Weekly return horizon Monthly return horizon Observations
Dependent Oil price exposure 189.00*** 246.08*** 210.04*** 315
Independent Oil Prices 6.32 (154.55**) 6.13 (151.96*) 8.38 (154.17**) 315
Volatility 34.71 42.56 54.58 315 Interest rate 57.01 (189.44***) 58.65 (193.97***) 52.35 (183.70***) 315 Leverage ratio 163.35*** 315 Production cost 46.29 (283.83***) 315 Quantity of production 96.93 (199.91***) 315 Percentage hedged 105.85* 315 Hedge price 34.62 (80.81***) 315 Control Reserves 92.66 (274.68***) 315
Notes: The test statistics are obtained from a sample period of 2003-2008. The first differences are taken from a sample period of 2002-2008. The sample period is increased to assure the year 2003 is not excluded in the test due to taking the first difference of the data. The values within the brackets are the results of taking the first differences.
*** significant at 1% level ** significant at 5% level * significant at 10% level
Tabel 4.3: Data construction of the determinants of oil price exposure
Determinant Construction
Dependent Oil price exposure Estimated using the Ordinary Least Squares (OLS) regression technique
Independent Oil Prices ln[(averaged oil price in year t) / (averaged oil price in year t-1)]
Volatility Running a GARCH(1,1) model on the returns of oil prices using Eviews
Interest rate ln[(averaged interest rate in year t) / (averaged interest rate in year t-1)]
Leverage ratio (book value of long-term debt of firm i in year t) / (market value of the equity of firm i in year t)
Production cost ln[(production costs of firm i in year t) / (production costs of firm i in year t-1)
Quantity of production ln[(quantity of production of firm i in year t) / (quantity of production of firm i in year t-1)]
Percentage hedged (amount hedged of firm i in year t) / (reserves of firm i in year t)
Hedge price ln[(averaged forward price of firm i in year t) / (averaged forward price of firm i in year t-1)
Control Reserves ln[(oil reserves of firm i in year t) / (oil reserves of firm i in year t-1)
Unfortunately, the data for oil options are unavailable. Although, it seems reasonable to assume that the GARCH(1,1) model gives convenient results, since Hansen and Lunde (2004) have done research to 330 different volatility models and none of the theoretical “best” volatility models do significantly outperform the GARCH(1,1) model. The interest rates are constructed in the same way as oil prices are. They are averaged and consequently the first difference of the natural logarithm of interest rates is taken in order to fulfill to the stationarity requirements. The leverage ratio is a fraction between the book value of long-term debt and the same year market value of the equity. The production costs,
quantity of production, and reserves are all constructed by taking the first difference of the natural
logarithm. Percentage hedged is a fraction between the quantities of oil barrels hedged divided by the barrels reported as oil reserves. Tufano (1998) also measures percentage hedged with the alternative ‘percentage of next-year oil production’. This alternative is not utilized in this study because of data unavailability of production levels in year 2009. Including the alternative measure would decline the sample period to 2003 through 2007. This reduction in sample size is undesirable due to the consequently reduction in the number of degrees of freedom, and therefore decreasing the power of econometric estimates and increasing the problem of multicollinearity, as discussed before. Moreover, production and reserves are highly correlated (0.94), as shown in Appendix C, which makes it reasonable to assume no significantly differences would be found. Hedge price is the weighted average of linear financial instruments (e.g. forwards, futures, and swaps), where the weights are the barrels of oil under contract. This measure excludes non-linear financial instruments (e.g. options and collars), which are also widely used by firms. These instruments require using Black’s option pricing model for calculating the hedge ratio (Jin and Jorion, 2006). Tufano (1998) uses both measures and find similar results. Hence, I can reasonably assume that my constructed variable of only including linear financial instruments is robust and is therefore labeled as proxy for hedge price.
Descriptive statistics
In Table 4.5 are the firm-specific determinants shown, wherein strong differences over time and cross-sectional are seen as suggested by the standard deviations. The Jarque-Bera indicates non-normality in the disturbance terms, with the exception of oil price exposure over the daily return horizon. It should be noted that no data points are deleted as outlier, since preliminary checks have shown that adjusting the data points had no significant effect. And as argued in Section 3.3, non-normality is practically inconsequential for my sample size.
Remarkable is that 209 firm-years are engaged in hedging activities and no hedging activities are
found in 169 firm-years.5 Jin and Jorion (2006) give a possible explanation for the firm-years that are
not hedged and argue that it is the consequence of the transparency in the oil and gas industry and therefore investors hedges the oil price exposures themselves.
5 Data is available upon request by the author.
Tabel 4.5: Firm-specific determinants estimated by different data/return frequencies
Leverage Production Quantity of Percentage Hedge price Reserves
costs production hedged
Mean 0.41 0.12 0.10 0.03 0.04 0.09 Median 0.22 0.13 0.03 0.01 0.00 0.05 Maximum 7.21 0.85 5.96 0.93 0.80 2.85 Minimum 0.00 -0.89 -1.53 0.00 -0.17 -2.40 Std. Dev. 0.73 0.21 0.50 0.09 0.12 0.48 Skewness 5.38 -0.43 5.06 7.35 3.53 1.22 Kurtosis 41.61 6.86 54.53 66.18 17.13 13.75 Jarque-Bera 25309.92 246.54 43426.25 66273.05 3925.35 1914.87 Probability 0.00 0.00 0.00 0.00 0.00 0.00 Observations 378 378 378 378 378 378
Oil price exposure Oil price Volatility Interest rates
Daily Weekly Monthly Daily Weekly Monthly Daily Weekly Monthly Daily Weekly Monthly
Correlations
As final topic in the data section the correlations between the determinants are discussed, which gives a comprehensive indication of which determinants are expected to have a significant role on oil price exposure. More specific, it gives a comprehensive indication about the determinants that fit the plotted exposure lines in Figure 3.1 the best. The correlation matrix is shown in Table 4.6 and has some noteworthy findings.
First, strong correlations are found among the macroeconomic factors and indicate possible signs of near multicollinearity. Dealing with the existence of near multicollinearity could be to ignore it, or drop one of the collinear factors, or to transform the highly correlated factors into a ratio and include only the ratio and not the individual factor in the regression (Brooks, 2008). No adjustments are made in the variables and thus ignored, since strong a priori theoretical reasons for including all determinants are present. Namely, the explanatory variables are analytically derived from a theoretical valuation model that requires the inclusion of all parameters. In the empirical results section I transform explanatory variables into dummies to control for possible influences of collinear determinants.
Second, oil price exposure is significantly positively correlated with the macroeconomic factors, except for oil prices over the monthly return horizon. Positive correlations for oil prices and volatility with oil price exposure are contrary to the predictions that are made in my hypotheses. These positive correlations suggest that firms operating in an environment with higher oil prices and more volatile oil prices do decrease oil price exposure. Of course, from these correlations no direction of causality can be estimated. This is further investigated by regressing the data in section 4.3.
4.3 Empirical results
In this section important results obtained from the performed random regression model are discussed, which is also the groundwork for answering the research question “Which macroeconomic and firm-specific determinants influence oil price exposure of oil producers?”.
Sensitivity to return horizon
