The green paradox of climate policy
Investigating shareholder value in the oil and gas industry
Ruben Nooijen*
Supervisor: prof. dr. M. (Machiel) Mulder Thesis: MSc Economics and MSc Finance
January 2016
Abstract
Current climate policies might fail in addressing the climate change problem by inducing resource owners to bring forward their extraction plans. This paper is the first to empirically investigate this so-called green paradox theory in the oil and gas industry. I use panel data on the stock prices, exploration and production activities of 25 of the world’s largest companies over the 1996-2014 period. The results indicate a clear negative impact of the announcement of climate policy targets on shareholder value, so that incentives are provided for green paradox outcomes. Subsequently the effects for the exploration and production activities are mixed, showing slight evidence of an increase in oil production levels. These results have implications for policy makers regarding the effectiveness, timing and scope of climate policies.
Keywords: green paradox, climate policy, oil and natural gas, shareholder value, exploration and production
JEL classification: D22, G12, L71, L78, Q58 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1. Introduction
At the recent 2015 Paris climate conference, the first-ever global, legally binding climate deal was adopted, aiming to keep the increase in global average temperature to well below 2°C. At the same time, the Organization of Petroleum Exporting Countries (OPEC) is overproducing and oil prices are reaching historically low levels of below $35 per barrel. This is hampering the transition towards sustainable energy, as renewable alternatives to oil are becoming relatively more expensive. This is a perfect example of the already long-standing battle between the interests of business and environmental policy makers. There is a general consensus that the oil and gas industry has a significant impact on the environment. The U.S Environmental Protection Agency (EPA) reports petroleum and natural gas systems and refineries to be the second- and third-highest contributors to greenhouse gas emissions in 2014, following power plants1. Inevitably linked with the energy transition is the discussion on climate policy design, and there is a great interest in how these policies will impact the competitiveness, productivity and profitability of the industries to which they are applied. Already in 2002, the World Resources Institute (WRI) indicated the prospect of policies to combat climate change to be one of the major environmental issues the oil and gas industry would have to deal with in the next decade, besides the constrained access to oil and gas reserves. These climate deals can have serious financial consequences for oil and gas companies by affecting sales, operating costs, asset values and shareholder value. Back then the impact on corporate performance was determined to be substantial, but not yet reflected in stock prices (Austin and Sauer, 2002). Nowadays, however, there is an increasing awareness among investors that fossil fuel reserves may even become ‘stranded assets’, as current reserves are already exceeding the emissions budget that is allowed for avoiding global warming of 2°C. This induces serious risks of write-offs or downward revaluations of the reserves of oil and gas companies.
A recent theory states that the aforementioned increase in oil production could be the optimal response of the industry to these potential negative financial effects of climate policy. It was termed the ‘green paradox’ by Sinn in 2008, and in Sinn (2015) the concept is explained clearly: “policies !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
aimed at reducing future demand for fossil fuels could backfire by inducing resource owners to bring forward their extraction plans, thus accelerating global warming”. The increased level of awareness and commitment to climate policies worldwide makes it critical to provide a better understanding of how companies are affected financially, and thus whether these policies could unintentionally lead to green paradox outcomes. This is of great importance to the policymaking process, specifically for the levels and timings of environmental regulations (Bushnell et al., 2013). Obviously, it is crucial to know whether the risk of green paradox effects exists, as it would mean that most of the current climate policies not only fail to address the climate change problem, but actually aggravate it by increasing emissions today.
In this paper, I study the impact of climate policy on the stock prices and exploration and production activities of 25 of the world’s largest publicly traded international oil and gas companies. Before starting the empirical investigation, I develop a theoretical model to illustrate the impact of climate policy on the value and extraction paths of the oil and gas companies. The model presents two testable hypotheses. The first hypothesis is that oil and gas companies experience a decrease in shareholder value following the announcement of climate policy, despite the possible increase in extraction levels. The second hypothesis is that the companies indeed increase their production levels following these same announcements, constituting a green paradox. To test the first hypothesis, event study methodology is used to estimate the companies’ cumulative abnormal stock returns following the announcement of climate policy targets. Subsequently I use dynamic panel data models to determine green paradox effects, estimating the impact of climate policy on the companies’ exploration and production activities.
The empirical results provide confirmation of the first hypothesis. Statistically significant negative abnormal returns are measured across the sample of oil and gas companies as a result from the announcement of climate policy targets. These policies are thus expected to have financial consequences for these companies and could provide incentives for green paradox outcomes. Evidence for the second hypothesis of an increase in production levels is however mixed. The climate policy variables show negative coefficients in the models for exploration and development activities, which are the preceding stages to production in the oil and gas supply process. Looking at production, the climate policy variables do show the expected positive signs consistent with green paradox effects, but not all variables are statistically significant. The impact seems to be more profound for the production of oil than for natural gas.
The paper will proceed as follows. Section two reviews the existing literature. Section three discusses the climate policies to be considered in the study. Section four develops the theoretical model and the hypotheses. Sections five and six present the empirical models and the data. Section seven follows with the results, and section eight concludes.
2. Literature review
This section starts by discussing the existing theory and empirical papers regarding the impact of climate policy on shareholder value in general, as well as for oil and gas companies specifically. Afterwards, an overview is presented of the little empirical research there has been to date regarding the green paradox.
2.1 Climate policy and shareholder value
compliance costs leading to a comparative disadvantage in international competition. However, amongst others Dowell et al. (2000) present an opposing view stating that companies may find it more costly not to adhere to the higher environmental standards, which also holds for making new investments. Moreover, firms can reduce pollution by making changes in the production process instead of incurring direct costs and some fringe benefits may be associated with adhering to higher environmental standards, such as heightened employee morale and thus productivity. Also nowadays, being environmentally sustainable as a company provides the opportunity to enhance corporate reputation, and there is evidence of a positive relationship between reputation and value creation (Roberts and Dowling, 2002).
An empirical study by Linn (2010) uses stock prices to estimate the change in expected profits of electric power plants under the Nitrogen Oxides Budget Trading Program (NBP) in the U.S. Using an event study, the author concludes that the market capitalization of NBP firms declined by as much as $25 billion. The results suggest that most of the decline in profits could be attributed to coal generators in restructured states. Another article by Bushnell et al. (2013) studies impacts on firms of the EU Emissions Trading Scheme (EU-ETS). The authors examine the impact of a sharp devaluation in CO2 prices in late April 2006 as an event study on the share prices of affected
firms. Using daily returns for 552 stocks from the Dow Jones STOXX 600 index, the authors conclude that several industrial sectors benefited from the ETS rather than being hurt by the imposition of CO2 regulation. Indeed,
when CO2 prices fell (a relaxing of regulation), the sharpest declines in equity
prices occurred within industries that are the most carbon-intensive, including oil and gas. The article by Ramiah et al. (2013) investigates the impact of 19 announcements of environmental regulation on the stock prices of companies listed on the Australian Stock Exchange over the period 2005-2011. The study shows that shareholder wealth of the electricity industry, being the biggest polluter, was not affected by the introduction of climate policy. Oil and gas stocks did experience negative abnormal returns, but only for two of the announcements. The authors assume that the polluters are passing costs on to consumers, and that therefore climate policy in its current form may not be effective.
several different climate policy scenarios, the financial impacts are found to range from a five percent loss in shareholder value to a slight gain, but were not yet reflected in stock prices. Within the industry, companies have different exposure to these policies depending on (i) the oil and gas mix of its production, proven reserves and acreage; (ii) the relative focus on upstream or downstream activities; and (iii) the regional spread of its operations. The oil and gas mix is of importance as oil produces more carbon per unit of energy than natural gas. And although climate policy aims to reduce consumption of fossil fuels including natural gas, it might also lead to substitution of carbon-intensive coal with natural gas in electricity markets. Regarding the position along the value chain, pure upstream companies and pure refining companies could be affected differently depending on the set-up of the climate policy. Lastly, regional differences in policies could lead to significant differences in profitability in these regions. These differences are said to be less important for oil than for natural gas, as oil effectively trades on a global market.
To summarize, these studies show that there is no general agreement on whether climate policy creates or destroys value for shareholders of affected companies. The results indicate that the impact on a firm’s expected profits depends on the type of policy and market structure, which differ among industries and regions. Therefore, empirical research is required to determine how climate policy affects value in the oil and gas industry. Based on the results of the studies presented above, one would rather expect these effects to be negative than positive.
2.2 Green paradox
In the theoretical literature regarding the green paradox, authors use different models of fossil fuel extraction to examine the impacts of a variety of announced or implemented climate policies. The authors are focusing on the roles of stock-dependent extraction costs, spatially differentiated policies, and backstop technologies, to identify when well-intentioned policies are likely to backfire and result in a green paradox outcome. Overall, this theoretical literature seems to confirm the risk of a green paradox and strengthens the case for taking a supply-side view of the climate problem (Sinn, 2015).
price, quantity and quality of coal delivered to U.S. power plants in the period between the announcement of the Acid Rain program (ARP) and its implementation. As the program put a nation-wide limit on sulphur-dioxide (SO2) emissions, it acted as a signal to owners of coal reserves that it would
be harder to sell their product in the near future. Using data on coal deliveries to U.S. coal-fired power plants, they find strong evidence of a price decrease and of an increase in the sulphur premium (i.e. the fall in price for high-sulphur coal was larger than for cleaner low-high-sulphur coal). An increase in coal use by utilities is only identified in a subset of plants and no evidence was found for an increase in the sulphur intensity of the coal purchased by the coal-fired utilities. Overall, the authors conclude that their results regarding the existence of a green paradox are somewhat mixed, although the results do suggest that fossil fuel prices react to policy announcements as predicted by the theory. The other econometric study by Lemoine (2013) uses the unexpected collapse of the U.S. Senate’s 2010 climate effort to establish the existence of a green paradox in U.S. energy markets. He uses this exogenous shift in carbon price expectations to identify the effect of the anticipated climate policy on current commodity prices. He finds that coal futures showed an increase of $1.30 per short ton upon the weekend collapse of the Senate’s 2010 climate effort, and the price of natural gas futures was found to rise also. This response suggests that the proposed policy was increasing greenhouse gas emissions in the years leading up to the program’s implementation in 2013.
3. Climate policy
Previous work mainly relies on the use of event studies to try and capture the impact of a single policy on a set of companies or industries in a certain country. This study focuses specifically on the announcements of climate policy targets, which I elaborate on in this section. Also, the impact of the targets is briefly discussed by looking at their objectives.
3.1 Climate policy targets
The oil and gas industry is a very internationally oriented market with companies operating in multiple countries and regions. Worldwide, these companies face different policymakers, using a mix of regulations and policy measures in an attempt to combat global climate change. This makes it inherently difficult to determine which climate policies have the biggest impact on the industry’s profitability and are most likely to result in green paradox outcomes. Looking at the European Union for example, multiple policy instruments are currently in place as the Emissions Trading Scheme (EU-ETS) as well as renewable energy subsidies. Determining the individual impact of all of these past and existing policies can prove to be a tough exercise. Successful environmental legislation typically unfolds over months or even years, which makes it impossible to separately identify the impact of legislation from the countless other factors affecting the companies during these months (Kahn and Knittel, 2003).
3.2 Impact of climate policy targets
Targets usually relate to a reduction in greenhouse gas emissions, an increase in the share of renewable energy in the energy mix, an increase in energy efficiency or a combination of these three aspects. In 2014, for the first time in at least 40 years, global energy-related CO2 emissions remained stable while
the global economy grew, achieving a so-called relative decoupling of CO2 and
economic growth (IEA, 2015). Non-hydroelectric2 renewable energy capacity has experienced exponential growth over the last few years so that it comprised roughly 10.6% of total world electricity generating capacity at the end of 2014. This is an estimated 27.7% even if one were to include hydroelectric power. Energy intensity3, as an aggregated indicator for energy efficiency, decreased globally over the period 1990-2013 at a steady rate of about 1.25% (REN21, 2015).
There is no doubt that climate policy has contributed to the discussed changes in greenhouse gas emissions, renewable energy and energy efficiency, and that these work to reduce the demand for fossil fuels. However, on a global level, the consumption of fossil fuels is still increasing and will continue to do so, not least because demand is still growing in developing countries. In preparation of the climate change conference in Paris in December 2015, countries submitted their Intended Nationally Determined Contributions (INDCs), outlining the post-2020 climate change actions they intend to take under the new international agreement. Even taking into account these INDCs, the share of fossil fuels in the world energy mix is still expected to be around 75% in 2030. Growth of the demand for oil and coal slows down but the consumption volumes do not decline. The demand for natural gas increases significantly as a substitute for carbon-intensive coal in electricity markets. Yet, renewable energy is expected to become the main source of electricity by 2030 (IEA, 2015). These are important facts to bear in mind when assessing the impact of the climate policy target announcements. The projections again indicate that green paradox effects might more likely be present in oil as opposed to natural gas markets, and it is crucial to take these differences into account in the empirical investigation.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2 Includes geothermal, solar, tide and wave, wind and biomass and waste.
3!A country’s energy intensity is defined as the ratio of energy consumption to national gross
4. Methodology
To be able to illustrate the impact of climate policy on the value of oil and gas companies and their extraction paths, I set up a simple theoretical model of non-renewable resource extraction based on the book by Jon M. Conrad (1999). This theoretical model allows me to develop intuition on the mechanisms behind the theory previously discussed and to set up hypotheses before conducting the empirical research.
4.1 Developing the theoretical model
In the model, oil and gas companies have a maximization problem
π = 1 (1+r)t(pt−ct)qt s.t. qt =R0 t=0 T
∑
t=0 T∑
R0 given, RT =qT = 0linear inverse demand curve pt =a−bqt
(4.1)
Here π denotes profits. pt, qt and Rt are the unit price, level of extraction and
stock of reserves in period t respectively. r is the real interest rate and the marginal cost of extraction ct is assumed to be constant. The single constraint
requires that cumulative extraction exhausts initial reserves R0. These initial
reserves are known and there is no exploration and discovery. The linear inverse demand curve implies a maximum ‘choke-off’ price at the intercept pt
= a where qt = 0, which is the price of a ‘backstop’ substitute. When the unit
price reaches this level, demand is zero, as alternative resources will have become cheaper. There is perfect competition in the industry and exhaustion occurs at time t = T, at which time the remaining reserves and extraction will be zero. Optimizing the equation specified above subject to the single constraint leads to the following intermediate result
(pt−ct)= (1 +r)t(p
0−c0) ⇔
[pt−ct]−[p0−c0]
first demonstrated in his 1931 paper ‘The economics of exhaustible resources’. As discussed in the paper by Jensen et al. (2015), fossil fuels are non-renewable resources and therefore their prices not only reflect production costs but also their scarcity. Because of these scarcity rents, the owners of fossil fuels maximize their profits by deciding when to extract their coal, gas or oil reserves. The Hotelling rule then states that for resource owners with tenure security, this profit maximizing extraction path is such that the resource rent increases at a rate equal to the interest rate. Present value maximizing oil and gas companies plan their extraction according to this rule, because if they would not a reallocation of extraction to periods with a higher discounted price increases present value. Knowing that pT = a and in the end solving for
qt4 allows me to specify the fundamental equation
qt = (a−c b )(1−(1 +r) t−T)=R 0 t=0 T
∑
t=0 T∑
(4.3) !Here, I have used the specified constraint that cumulative extraction exhausts initial reserves. Using this equation, one can then calculate the date of exhaustion T and numerically plot the extraction and price paths.
With regard to implementing climate policy, on the one hand policy makers can use environmental taxes to directly address the market failure of environmental externalities of producing fossil fuels by incorporating these externalities into prices. This leaves economic agents the flexibility to decide for themselves how to change their behaviour and reduce these externalities. On the other hand, policy makers can make use of environmental subsidies to stimulate the use of environmentally beneficial alternatives as renewable energy. This involves the government steering the economy towards certain environmental solutions (OECD, 2010). In this theoretical model, the effect of both types of environmental policy tools is exactly the same, which can be seen when looking at the specified inverse demand curve
incorporates the environmental externalities into prices in a similar way. Both policies have the same impact on relative prices, the subsidy making renewables cheaper as compared to fossil fuels, and the tax making fossil fuels relatively more expensive. However, as Jensen et al. (2015) showed, it does matter whether the implemented tax or subsidy is constant over time, increases at the rate of interest or is increasing at a rate even faster than the interest rate. A subsidy or tax increasing faster than the interest rate aggravates green paradox effects as it continuously provides incentives to extract now rather than later. On the other hand, a subsidy or tax that is constant over time increases the value of future extraction thereby extending the extraction path over an additional few years.
4.2 Numerical illustration
For the numerical illustration, the values of the parameters in the model are roughly based on the current crude oil market for six oil and gas supermajors included in the study. These are BP, Chevron, ConocoPhillips, ExxonMobil, Royal Dutch Shell and Total. The data has been supplied by Evaluate Energy Ltd. At the end of 2014, the combined crude oil reserves of these companies amounted to 46,701 million barrels5, and therefore initial reserves are set at 50 billion barrels. The initial level of extraction is set at 3 billion barrels (per year) resulting from a combined production of 9,154 thousand barrels per day in 2014. Based on the cost of supply per barrel of oil equivalent in 20146, the marginal cost of extraction is set at $30. I use the relatively stable oil price of about $100 per barrel of the few years prior to the 2014 drop in prices and an accompanying real interest rate of 3 percent. Lastly, for the demand curve I use a = 200 (price of the backstop substitute) and b = 33 1/3 so that demand matches the determined initial level of extraction and the oil price.
Looking at Figures 1 and 2, one can see the effects of climate policy on the extraction and price paths and the present value of profits of the oil and gas companies. Following the style of Jensen et al. (2015), I use a tax that is levied on the carbon content of fuels. I assume that this carbon tax of $30 per barrel is to be introduced in 2020 and that the tax will rise at the rate of
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
5 Includes natural gas liquids, oil sands/synthetic crude oil and associates.
6 Data source: financial reports. Comprises exploration expenditure, production costs and
Figure 1: impact of climate policy on extraction and price paths
Impact of the announcement of a carbon tax of $30 to be introduced in 2020, rising at the rate of interest. Solid and dashed lines represent the extraction (qt) and price (pt) paths for a group of six oil and gas supermajors before and after announcement of the climate policy (cp). For the demand curve, I use a = 200 and b = 33 1/3.
Figure 2: impact of climate policy on present value of profits
interest. In the two figures, the impact of climate policy is evident. Under the Hotelling rule it is assumed that economic agents base their behaviour on expectations of future prices. Therefore, a policy aimed at curbing resource-derived revenues in the future induces resource owners to bring forward their extraction plans in anticipation of the implementation date, which in turn depresses market prices. As can be seen in Figure 1, the level of extraction clearly increases ahead of implementation of the policy, constituting a green paradox. The results will now be examined more in detail to set up my hypotheses for conducting the empirical research.
4.3 Hypotheses
To investigate the companies’ profitability, it is important to make a distinction between short-term profits and long-term value, as the implied results for both are different. The effect on short-term profits, ahead of implementation of the policy, depends quite strongly on the value of the tax or subsidy and its implementation date. This effect, which is illustrated by the first bar in Figure 2, can already be considered as being relatively small, and it becomes even smaller the further away implementation is. In the years leading up to the implementation of more distant policies, short-term profits can even be temporarily higher than without the tax or subsidy. In these years, the increase in extraction by the oil and gas companies allows to more than offset the negative price effect of the climate policy, resulting in higher profits. Theoretically, these effects on short-term profits could be reflected in the company’s accounting earnings, which assess short-term performance from a shareholder viewpoint. However, because there is no single answer regarding the impact of climate policy on the profitability of these companies in the short term, it is not informative to test for such effects empirically.
future profits should then be reflected in the companies’ equity prices. As Koller et al. (2010) argue, the stock market indeed tracks the long-term fundamental performance of companies. Using discounted cash flow (DCF) models, the company’s stock price is determined by future earnings and cash flows. These earnings can then be paid out in dividends or be reinvested to generate more earnings in the future, creating long-term value for shareholders. Determining whether such pronounced effects of climate policies on the stock prices of oil and gas companies exist is important as it shows whether these policies are indeed expected to reduce long-term value, and thus whether they could have provided incentives for green paradox outcomes. Using the results from the model and taking into account these guiding principles, the following first hypothesis is set up:
Hypothesis 1: despite the possible increase in extraction levels, oil and gas companies experience a decrease in shareholder value following the announce-ment of climate policy.
Now I will discuss the oil and gas companies’ supply behaviour more in detail. The extraction paths with and without climate policy are shown in Figure 1. As was previously discussed, the negative price effect causes the oil and gas companies to extract faster because the reserves must be exhausted before the backstop price is reached. Put differently, an announcement of climate policy results in an increase in supply thereby decreasing prices, as companies perceive the program as a threat to their future profitability bringing their supply forward. As was mentioned previously, resource prices are difficult to explain because they are affected by a sheer number of factors, making it all but impossible to identify the Hotelling rule. Therefore, it seems more suitable to focus directly on the extraction paths of the oil and gas companies instead of determining a decrease in resource prices. These extraction paths are reflected by the companies’ production decisions. The level of production is also a variable over which the oil and gas company as the decision-maker has direct control (Walls, 1992). This has resulted in the following second and final hypothesis:
5. Empirical model
The two key hypotheses both require a separate approach and thus two different empirical methodologies. For looking at the shareholder value effects of climate policy announcements, I use event study methodology. Next, the econometric model for investigating the impact on oil and gas exploration and production is specified, which involves dynamic panel data models.
5.1 Event study methodology (hypothesis 1) 5.1.1 Model selection
To examine the impact of climate policy announcements on the profitability of international oil and gas companies, their abnormal stock returns have to be estimated. These abnormal returns can be measured by subtracting the expected return from the announcement period return. The expected return is the return that would have accrued to shareholders in the absence of climate policy. The finance literature considers multiple models of expected returns, which can be classified as statistical or economic models. The first category of models are derived from statistical assumptions about the behaviour of asset returns, in contrast to the second category which relies on economic theories of asset price formation (Polinsky and Shavell, 2007). The literature favours the use of statistical models as several studies have found evidence inconsistent with the economic models. Also, the gains of using an economic APT model are said to be small, as the most important factor is similar to a market factor and additional factors add relatively little explanatory power. The model does eliminate the biases introduced by using the CAPM, but the statistical models succeed in doing this also (MacKinlay, 1997). However, as I am focusing on a specific industry, it might be that the international oil and gas companies perform differently in response to the known risk factors. Therefore, next to a statistical market model, an economic multifactor APT model is used for means of sensitivity analysis.
5.1.2 Statistical market model The market model can be specified as
Here Rit is the return of stock i for time t. Rm,t is the return of the market
portfolio for time t. α and β are firm-specific factors and εit is the error term.
5.1.3 Economic APT model
The specification of the APT model, which allows for the presence of multiple factors in determining stock returns, is less straightforward. These models are based on the law of one price and thus assume no arbitrage because all the return generating factors are included in the model. According to Elton et al. (2014), one approach to define this set of factors is to first hypothesize a set of macroeconomic variables that might affect the returns of these oil and gas companies on the basis of economic theory. Consequently, regression analysis can be used to estimate the factor loadings.
For hypothesizing the set of variables, I turn to the rich literature on assessing the financial determinants of oil and gas company stock returns, with a specific focus on the sensitivity to crude oil and natural gas prices7. The mentioned studies use both common macroeconomic and company-specific fundamental factors to estimate the sensitivity of oil and gas stocks to these variables. Although a large number of factors would be necessary to explain the entire return of these equities, including four or five is stated to be enough to obtain notable explanatory power (Boyer and Filion, 2007). I choose to include five common macroeconomic factors that are used in the literature as important explanatory variables determining oil and gas equity prices. These are the return of the market, the exchange rate, the interest rate and the oil and gas price8. First of all, following the theory of Sharpe (1964), the return of the market portfolio should influence the return of the oil and gas stocks. Second, because all companies are multinationals the exchange rate is included to isolate the impact of currency risk. Additionally, the capital intensity in the industry is high due to the scale of the investments !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
7 See e.g. Al-Mudhaf and Goodwin (1993); Rajgopal and Venkatachalam (1998); Boyer and
Filion (2007); Scholtens and Wang (2008).
8 According to Chen et al. (1986), the rate of inflation should also be included as it impacts
necessary to operate. Because of external financing, variations in the interest rate could then be a significant risk factor (Boyer and Filion, 2007). Lastly, oil and gas prices are said to be a common risk factor in the industry, as fluctuations in these prices directly affect revenues, profits, investments and cash flows.
Having defined the factors to be included in the multifactor APT model, the following equation can be set up
Rit = αi+ βmRm,t+ βerRer,t+ βirRir,t+ βoilRoil,t + βgasRgas,t+ εit (5.2)
Here Rit is the excess return of stock i for time t. Rm,t represents the return of
the market index, Rer,t is the exchange rate return, Rir,t is the interest rate
factor and Roil,t and Rgas,t are the oil and gas price return, all for time t. αi is
the expected return for company i and εit is an error term.
5.1.4 Estimating (cumulative) abnormal returns
Using the two models specified above, the abnormal returns for the oil and gas companies during the event window can be estimated. These are calculated using the following formulas:
Market model
ARiτ =Riτ−(ˆαi+ ˆβiRm,τ) (5.3)
APT model
ARiτ =Riτ−(ˆαi+ ˆβmRm,τ+ ˆβerRer,τ+ ˆβirRir,τ + ˆβoilRoil,τ+ ˆβgasRgas,τ) (5.4)
Here ARiτ is the abnormal return in event time that is indexed using τ.
announced over the weekend) and there is a five-day event window [τ1 = 0, τ2
= 4]. The cumulative abnormal return for each company over the five trading days can be defined as
CARiτ = τ=0ARiτ
4
∑
(5.5)As in Brown and Warner (1985), a maximum of 250 daily observations is used for the period surrounding the announcement date, starting at day –245 and ending at day +4. The first 245 days in this period (–245 through –1) is then defined as the estimation window used to calculate the expected return for the stock during the event window. The significance of the (cumulative) abnormal returns is determined using the standard t test, which assumes that the returns are normally distributed. The tests can be specified as
(5.6)
Here SD(ARiτ) is an estimate of the standard deviation of the abnormal
returns, which is estimated from the time-series of abnormal returns during the estimation window (Brown and Warner, 1985). The standard deviation of the cumulative abnormal returns SD(CARiτ) is calculated as usual using the
estimated abnormal returns in this five-day period. To draw overall inferences of the impact of climate policy announcements on the entire oil and gas industry, the abnormal returns have to be aggregated over the individual companies to obtain the industry ARIτ and CARIτ. This is done by estimating
intercept only models for the (cumulative) abnormal returns, which essentially average the abnormal returns of each company, but allow for the use of robust standard errors.
5.2 Exploration and production methodology (hypothesis 2) 5.2.1 Model specification
Regarding the model for investigating the second hypothesis, I take a look at the literature concerning the econometric modelling of oil and gas supply, which estimate the historical relationship between variables such as the oil
tAR
iτ =
ARiτ
SD(ARiτ) tCARiτ =
price, drilling and production9. Some mainly attempt to explain exploration activities, while others develop integrated models of oil and gas production. According to Kemp and Kasim (2003), a key component of the oil and gas supply process is that it takes place in stages, so that decisions at any stage are based on information obtained at the preceding stage. Therefore, I will focus here specifically on the well-defined stages of exploration, development and production (i.e. extraction). Modelling the multiple stages allows me to capture the integrated nature of the decision-making process in the oil and gas industry, and to see where climate policy might have made an impact. A clear distinction is made in the literature between the models of physicists and economists, both focusing on different explanatory variables. The former are dominated by geological factors such as cumulative discoveries and technological conditions. The latter stress the importance of expected profits, including prices, regulations and other economic variables. However, Moroney and Berg (1999) concluded that combining physical and economic factors yields superior results in modelling oil supply. A partial adjustment model of oil production based on reserves, lagged production and the oil price, was regarded as being the most reliable. Kemp and Kasim (2003) also stress the partial adjustment assumptions underlying their model specification.
Hence, the models used here are specified in a similar way, also including lagged dependent variables in the equations. The panel data then allows me to estimate a dynamic model on an individual level. However, in a dynamic model fixed effects estimation is substantially different. The within transformed lagged dependent variable is correlated with the within transformed error, which leads to biased and inconsistent estimates. To solve this problem, I make use of the Anderson-Hsiao instrumental variables (IV) and the Arellano-Bond generalized method of moments (GMM) estimators. These estimators start by taking the first-difference of the equation and consequently instrumenting for the lagged dependent variable yi,t-1, the former
using yi,t-2 and the latter using further lags of the dependent variable as
instruments (Verbeek, 2008). Additionally, a substantial part of the literature uses a specification of the regression equations in logarithms, which allows for non-linear relationships and facilitates interpretation of the coefficients as !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
9 See e.g. Pindyck (1974); Pesaran (1990); Walls (1992); Favero and Pesaran (1994); Moroney
elasticities. Therefore, the equations are estimated using logarithms besides the regular specification, and the results of both will be presented.
5.2.2 Economic and physical variables
As was mentioned above, major forces driving oil and gas exploration and production are the expected oil and gas prices as they determine revenues. In the methodology, I discussed the profit maximizing extraction path for resource owners, which involved reallocating extraction to periods with higher prices. In an attempt to capture this mechanism, a price expectations variable is included in the models. This variable is defined as the present future price divided by the spot price for both oil and natural gas. Here, I assume the present future price to be the expected future spot price. This view, also used by Fama and French (1987), defines the present future price as a forecast of the future spot price plus an expected risk premium. It states that the present future price is determined by the market participants’ beliefs about the present and future trends of supply and demand, and that the cost of carry seems to be much less relevant (De Almeida and Silva, 2009). A negative sign for this variable would then be consistent with the discussed reallocation mechanism, as it predicts oil and gas companies to increase current extraction if future prices are expected to decline relative to current prices.
5.2.3 Climate policy variables
To be able to determine green paradox effects, I make use of dummy and trend variables. The theoretical model illustrated the impact of climate policy on extraction in the ideal case of a single isolated announcement, but of course the reality is quite different. This was discussed previously and resulted in the specification of climate policy target announcements of the world’s largest emitters of greenhouse gases (see Table 1 in the next section). These announcements allow for multiple specifications of variables representing the impact of climate policy. First of all, similar to the theoretical model, one can choose a single year representing the most significant increase in the intensity of climate policy. It becomes clear from the table that this year is 2009, in which the climate policy targets spread much more worldwide with all major countries participating. This results in a dummy variable that is zero in all years before 2009 and equal to one afterwards. Second, one can include a trend to proxy for the increase in the intensity of climate policy over time. A quadratic trend is specified to reflect the fact that climate policy has been evolving at increasingly higher rates with more countries participating and more stringent targets10. Lastly, one could specify a climate policy dummy time series encompassing the impact of all announcements together. Unfortunately, the different types of targets are difficult to compare across countries and even more difficult to translate into a reduction in demand for fossil fuels. One option could be to determine the relative contributions of the targets according to the countries’ share in total emissions; the calculations are presented in Appendix B. However, this is only a very rough estimation as best so one must be careful in drawing conclusions. The three discussed variables are included separately in the models and the results should provide an exhaustive overview of the effects of climate policy. World GDP is included as an additional economic control variable to capture shifts in the global demand curve and to isolate these from climate policy impacts11. Obviously, positive signs on the coefficients for the climate policy variables !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
10 Also, including a linear trend is not possible due to the specification in fist differences. 11 Next to world GDP, several other control variables affecting the global oil and gas market
would then be an indication of evidence supporting the green paradox hypothesis.
5.2.4 Model equations
The set-up of the models is then as follows: Exploration model
XPLit = αi + β1XPLi,t−1+ β2OPRICEtEXP+ β
3GPRICEt EXP
+ β4GDPt+ γCPt+ εit (5.7) Development model
DEVit = αi + β1DEVi,t−1+ β2SXPLi,t+ β3OPRICEtEXP+ β
4GPRICEt EXP
+ β5GDPt + γCPt+ εit (5.8)
Production model (estimated separately for oil and natural gas)
PRODitOIL/GAS = α i + β1PRODi,t−1 OIL/GAS + β 2RESi,t−1 OIL/GAS + β 3SDEVt + β4OPRICEtEXP+ β 5GPRICEt EXP + β 6GDPt+ γCPt + εit (5.9)
Here the dependent variables XPLit and DEVit are the number of exploration
and development wells drilled respectively, and PRODit is production of oil or
natural gas by company i for time t. SXPLi,t and SDEVi,t denote the number
of successful exploration and development drillings. OPRICEtEXP and
GPRICEtEXP are the price expectations variables for both oil and natural gas.
GDPt is world GDP for time t and CPt encompasses the three climate policy
variables specified above, which are defined as CPdum2009, CPtrend and CPdum. Lastly, αi is the constant and εit is an error term.
5.3 Overview variables
6. Data
Here, I first describe the climate policy targets and announcement dates to be included in the study. Next, I discuss the characteristics of the oil and gas companies in the panel data set. Afterwards, I will present an overview the variables included in both empirical methodologies, including their exact definition and source of the data, and show descriptive statistics.
6.1 Announcements of climate policy targets
Table 1 presents an overview of the key announcements of climate policy targets over the 1996-2014 estimation period. To be consistent and complete, the list has been compiled with help of the book by Joyeeta Gupta (2014), who discusses the key events and outputs in the different phases of climate governance history. After having defined the key targets, I used LexisNexis Academic to determine the appropriate announcement dates. However, as Polinsky and Shavell (2007) note some events may have multiple event dates. In the case of the EU, formal approval of the climate policy targets after announcement involved meetings by several entities such as the Council of Ministers, the European Commission and the European Parliament. These different types of events might all provide new information to the investor about the likelihood of passage. Therefore, in this case, besides the earliest announcement date to be found, related events that followed are also included in the estimations.
climate policy as the drop in shareholder wealth and the increase in production are set to occur directly after announcement. Therefore, only the announcement date is needed for empirical investigation.
6.2 Characteristics oil and gas companies
The panel data used in the study consists of 25 of the world’s largest publicly traded international oil and gas companies12. The sample is nicely diversified as one can see from their characteristics in Table 2. Together, these companies !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
12 Some of the current largest public companies could not be included as their public offering
Table 1: announcements of climate policy targets
Climate policy targets and announcement dates since 1996 until 2014 of the EU, the US, Russia, Brazil, India and China. Dates were determined using LexisNexis Academic. /1990 stands for compared to 1990 levels.
Number Date Region Climate policy target announcement 1 03-03-97 EU 15% reduction emissions by 2010/1990 2 22-10-97 US Stabilizing emissions by 2008-2012/1990
3 11-12-97 World Kyoto Protocol – 5.2% reduction emissions developing countries by 2008-2012/1990
4 07-11-05 China 15% renewable energy by 2020
5 05-03-06 China 20% reduction energy intensity by 11th Five-Year Plan 2006-2012
6 10-01-07 EU 20% reduction emissions, 20% improvement energy efficiency and 20% renewable energy by 2020/1990 7 09-03-07 EU Council of Ministers agrees on 2020 targets 8 23-01-08 EU European Commission adopts 2020 targets 9 04-06-08 Russia 40% reduction energy intensity by 2020/2007 10 18-12-08 EU European Parliament approves 2020 targets 11 19-06-09 Russia 10-15% reduction emissions by 2020/1990 12 21-09-09 India 20% renewable energy by 2020
13 13-11-09 Brazil 36% to 39% reduction emissions by 2020/business-as-usual
14 18-11-09 Russia 25% reduction emissions by 2020/1990 15 25-11-09 US 17% reduction emissions by 2020/2005
16 26-11-09 China 40-45% reduction carbon intensity by 2020/2005 17 03-12-09 India 20-25% reduction emissions intensity by 2020/2005 18 05-03-11 China 16% reduction energy intensity and 11.4% renewable
energy by 12th Five-Year Plan 2011-2015
19 22-01-14 EU 40% reduction emissions, 27% improvement energy efficiency and 27% renewable energy by 2030/1990 20 24-10-14 EU Council of Ministers agrees on 2030 targets 21/22 12-11-14 US 26% to 28% reduction emissions by 2025/2005
Table 2: characteristics of the oil and gas companies
Characteristics of the sample of companies. E&P stands for exploration and production, an * notes discontinued or spun-off refining operations in the last few years. ME and CA stand for Middle East and Central Asia respectively. A ✓ marks a non-zero level of production in 2014. Data source: Evaluate Energy Ltd. Company Original listing Type
Oil-heavy
US Canada Latin America
Europe Africa & ME
Asia Pacific
Russia & CA
Anadarko Petroleum US E&P ✓ ✓ ✓
Apache US E&P ✓ ✓ ✓ ✓ ✓ ✓
BG Group UK (EU) E&P ✓ ✓ ✓ ✓ ✓ ✓
BP UK (EU) Integrated ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cabot Oil & Gas US E&P ✓
Canadian Natural Resources Canada E&P 70-80% ✓ ✓ ✓ Chesapeake Energy US E&P ✓
Chevron US Integrated 60-70% ✓ ✓ ✓ ✓ ✓ ✓ ✓ ConocoPhillips US E&P* 60-70% ✓ ✓ ✓ ✓ ✓
Devon Energy US E&P ✓ ✓
Eni Italy (EU) Integrated ✓ ✓ ✓ ✓ ✓ ✓
EOG Resources US E&P ✓ ✓ ✓
ExxonMobil US Integrated ✓ ✓ ✓ ✓ ✓ ✓
Hess US E&P* 60-70% ✓ ✓ ✓ ✓
Imperial Oil Canada Integrated >80% ✓
Marathon Oil US E&P* 60-70% ✓ ✓ ✓ ✓
Murphy Oil US E&P* 70-80% ✓ ✓ ✓
Noble Energy US E&P ✓ ✓ ✓
Occidental Petroleum US E&P 70-80% ✓ ✓ ✓
OMV Austria (EU) Integrated ✓ ✓
Repsol Spain (EU) Integrated ✓ ✓ ✓ ✓ ✓
Royal Dutch Shell Netherlands (EU) Integrated ✓ ✓ ✓ ✓ ✓ ✓ ✓
SM Energy US E&P ✓
Suncor Energy Canada Integrated >80% ✓ ✓ ✓
provide an ideal overview of the international oil and gas industry, as they are from several countries and have different business volumes and targeted markets. Austin and Sauer (2002) discussed that oil and gas companies have different exposure to climate policy depending on their oil and gas mix, their position along the value chain and the location of their operations and sales. The table provides an overview of these characteristics for the companies in the sample. Original listings include EU countries, Canada and the US. The vast majority of the companies are multinationals, but there are also some which operate solely in the US or Canada. There are independent exploration and production companies as well as integrated companies, the latter also being involved in refining activities. The fourth column shows whether the company is oil-heavy, indicating the company’s average share of oil in the resource reserves portfolio over the years. The average share of oil in the oil and gas mix for this set of companies was just over 50%, which is why I have indicated those with a share of over 60%.
6.3 Definition of included variables and descriptive statistics 6.3.1 Event study variables (hypothesis 1)
time period. The Brent oil price and Henry Hub natural gas price are chosen because they serve as the main benchmarks in the world markets.
Table 3: definition and source of variables
Dependent and explanatory variables included in the event study and exploration and production methodologies. Stock and market returns for the statistical market model are calculated without subtracting the risk-free rate. Original stock listings are used and all monetary variables are expressed in US dollars. Data frequency for the event study is daily and for exploration and production it is yearly. Exploration and development wells drilled, reserves and production are worldwide. Productive wells are those with proved reserves allocated and thus exclude dry wells. NGL stands for natural gas liquids. bcf: billions of cubic feet; 000 b/d: thousands of barrels per day; mmcf/d: millions of cubic feet per day. Variable Definition Source
Event study methodology Stock return
per firm
Ri = ln((stock price)t/(stock price)t-1) – 3 month
US Treasury bill rate
Datastream
Market return Rm = ln((price MSCI world index)t/(price MSCI
world index)t-1) – 3 month US Treasury bill rate
Datastream
Exchange rate Rer = ln((exchange rate USD/EUR)t/(exchange
rate USD/EUR)t-1)
Datastream
Interest rate Rir = ln((10 year US Treasury note rate)t/(10
year US Treasury note rate)t-1) – 3 month US
Treasury bill rate
Datastream
Oil price Roil = ln((Brent spot price)t/(Brent spot price)t-1) Datastream
Gas price Rgas = ln((Henry Hub spot price)t/(Henry Hub
spot price)t-1)
Datastream
Exploration and production methodology Exploration
per firm
XPL = total number of exploratory wells drilled
SXPL = number of productive exploratory wells drilled
Evaluate Energy Ltd.
Development per firm
DEV = total number of development wells drilled
SDEV = number of productive development wells drilled
Evaluate Energy Ltd.
Proved reserves per firm
RESOIL = world oil/NGL reserves (mln barrels)
RESGAS = world natural gas reserves (bcf)
Evaluate Energy Ltd. Production
per firm
PRODOIL = world oil/NGL production (000 b/d)
PRODGAS = world natural gas production
(mmcf/d)
Evaluate Energy Ltd.
Price expectations OPRICEEXP = (Brent 12-month future
price)t/(Brent spot price)t
GPRICEEXP = (Henry Hub 12-month future
price)t/(Henry Hub spot price)t
Datastream
GDP GDP = world gross domestic product Datastream Climate policy
variables
CPdum2009 = 0 before 2009, 1 afterwards
CPtrend = quadratic trend
CPdum = see Appendix B
I also chose to separately estimate the models using regional variables for means of sensitivity analysis. This allows for segmented markets instead of integrated world capital markets To this end, I include the MSCI North America index and the West Texas Intermediate (WTI) oil price for those companies originating from the US and Canada. For the European companies, I use the MSCI Europe index and German government short- and long-term bond rates. I remain using the Henry Hub gas price because, unfortunately, data on prices on the European gas hubs did not go back as far as 1996. The US Treasury bill and German government bond used as risk-free rates both have a one-year maturity.
The descriptive statistics for the stock returns of the oil and gas companies included in the study are presented in Table 4. I also performed the Jarque-Bera (JB) normality test and the augmented Dickey-Fuller (ADF) test for a unit root of which the results can also be found in the table. The former test assumes a null hypothesis of a normal distribution, the latter test of non-stationarity. These tests were also performed for the other variables included in the event studies, being the returns of the market, the exchange and interest rates and the oil and gas prices, yielding the same results as for the stock returns. The null hypothesis of the unit root test is strongly rejected in favour of stationarity, but the null of a normal distribution is also soundly rejected so that non-normality seems to be an issue. However, Brown and Warner (1985) find that non-normality of daily returns has no obvious impact on event studies and that standard tests for significance of mean excess returns are well specified, even in samples of only five securities. The authors do note that recognition of autocorrelation in the time-series of daily mean excess returns can be advantageous for hypothesis tests over multi-day intervals. Therefore, I chose to use heteroskedasticity and autocorrelation robust standard errors in estimating the abnormal returns.
6.3.2 Exploration and production variables (hypothesis 2)
Table 4: descriptive statistics for the stock returns of the oil and gas companies
Descriptive statistics and normality- and unit root tests for the returns from 1996 until 2014 (4957 observations). *** p<0.01, ** p<0.05, * p<0.1.
Brent and Henry Hub spot and future prices are used being the main international benchmarks. The 12-month future is the most distant contract for which data is available over the entire estimation period. The specification of the climate policy variables has been discussed previously.
In the Figures 3, 4 and 5, the development over time of oil and gas production and reserves as well as exploration and development drillings is depicted. As showing graphs for all 25 companies would be too cumbersome, the data has been aggregated to get a rough feeling for the general trends in the market. Also, I included rescaled data on worldwide production of oil and natural gas as reported by the EIA to get an idea of how the sample of companies compares to the global oil and gas industry. As can be seen in the figures, growth in production levels of the sample is fairly consistent with world levels. However, there is a striking deviation in oil production after 2010, at which time oil production declines sharply with world levels steadily increasing. The growth of production levels also seems to be broadly consistent with the growth of the physical inventory of reserves, except for a similar deviation near the end of period under investigation. Looking at the 2009 financial crisis, the production of oil was much more heavily affected than natural gas, providing signs of different market conditions for both resources. Additionally, in 2009 there is a sharp decline in the number of development wells drilled, which is not present for exploration drillings. This might be explained by the fact that that development is the preceding stage to production and that investment decisions here are more flexible to adjust to changing demand conditions.
Left: the solid line represents the oil production of the total sample of companies together. A few missing values were filled in using extrapolation. The dashed line represents world production as reported by the EIA, rescaled for ease of comparison. Right: oil reserves of the total sample of companies together. Data source: Evaluate Energy Ltd.
Figure 4: natural gas production and reserves Same set-up as Figure 3 but for natural gas. Data source: Evaluate Energy Ltd.
Figure 5: exploration and development wells drilled
Left: exploration wells drilled. Right: development wells drilled. For both variables only the companies that have the full range of data available were included to show consistent figures (15 out of the 25 companies in the sample). Data source: Evaluate Energy Ltd.
stationary in first-differences, and the regression results will therefore not be spurious. Arellano-Bond tests for second-order autocorrelation in first differences (first-order correlation in levels) showed signs of autocorrelated errors for some models. Hence, again heteroskedascticity and autocorrelation robust standard errors are used.
7. Results
In this section, the results of the event studies and the exploration and production models are discussed. First, I take a look at the abnormal returns resulting from the climate policy target announcements. Second, I determine the impact of the climate policy variables in the estimated models for exploration, development and oil and natural gas production.
7.1 Event study results (hypothesis 1)
As can be seen in Table 5, statistically significant negative abnormal returns were measured across the sample of oil and gas companies as a result from the announcement of climate policy targets. I focus mainly on the cumulative abnormal returns as they show deviations from the mean process over a longer period of time, and are thus more likely to reflect long-term effects. The insignificant results for several of the announcements could be due to multiple reasons, ranging from misspecification of the event date because of information leakage, to targets that are not credible or are lacking ambition. Therefore, I mainly look at the bigger picture instead of drawing conclusions from the (in)significance of individual announcements. Of the statistically significant CARI’s, the vast majority proves to be negative and losses range
from –1.61% up to –6.47% over the five trading days. Notable is that all of the announcements by China as well as most of those by the EU made quite a significant negative impact. Comparing the results of the market model with those of the APT model, the latter results generally seem to be smaller in magnitude. Also the CARI’s for announcements 9 and 10 by Russia and the
Table 5: abnormal returns for the oil and gas industry of climate policy announcements
Results of the estimations of the industry abnormal returns (ARI) and cumulative abnormal returns (CARI) following climate policy target announcements
using world data. Abnormal returns are computed using (1) the market model and (2) an Arbitrage Pricing Theory (APT) model: (1)
Rit= αi + βiRm,t+ εit (2) Rit= αi+ βmRm,t+ βerRer,t+ βirRir,t+ βoilRoil,t+ βgasRgas,t+ εit
Here Rit is the (excess) return of stock i for time t. Rm,t represents the return of the market index, Rer,t is the exchange rate return, Rir,t is the interest rate
factor and Roil,t and Rgas,t are the oil and gas price return, all for time t. αi is the expected return for company i and εitis an error term. Abnormal returns of
each company are essentially averaged and t-tested for significance. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. Market model APT model
Announcement Date Region ARI (%) Std. err. CARI (%) Std. err. ARI (%) Std. err. CARI (%) Std. err.
Table 6: sensitivity analysis of abnormal returns
Same methodology as Table 6 but using regional instead of world data as a sensitivity analysis. The MSCI world index is replaced by the MSCI North America and Europe indices. For companies originating from the US and Canada, I use US Treasury rates and the WTI oil price. For the European companies, German government bond rates and the Brent oil price are used. The exchange rate (USD/EUR) and the gas price (Henry Hub) remain the same. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
Market model APT model
Announcement Date Region ARI (%) Std. err. CARI (%) Std. err. ARI (%) Std. err. CARI (%) Std. err.
the sensitivity analysis, using regional instead of world data. For the market model, the differences in the estimated coefficients for the cumulative abnormal returns are relatively minor. However, there are two additional announcements that have resulted in significantly negative CARI’s, namely
number 2 by the US and number 14 by Russia. For the APT model, the differences are larger, with announcements 1, 6 and 17 by the EU and India losing their significance, and 2, 7 and 10 by the US and the EU becoming significant. In the case of the APT model, the results of the announcements by the EU thus seem to be relatively sensitive to the data specification. Table 7 presents an overview of the results, showing the percentage of CARI’s being
positive, negative or insignificant in the four possible combinations of the different models and data. The high percentage of statistically significant negative CARI’s in each of the specifications provides confirmation of the first
hypothesis. Oil and gas companies indeed experience a decrease in shareholder value following the announcement of climate policy, despite the possible increase in production levels.
7.2 Exploration and production results (hypothesis 2) 7.2.1 Exploration and development wells
In Tables 8 and 9, I present the results of the models explaining the number of exploration and development wells drilled. Comparing the R-squared revealed that the specification in logarithms greatly enhances the overall fit of the models, which is why these results are presented here. The coefficients can thus be interpreted as elasticities. The results of the regular specification can be found in Appendix D.
As can be seen in the two tables, both processes of drilling exploration and development wells seem to exhibit a substantial degree of persistence. The
Table 7: overview of event study results
Summary of the industry cumulative abnormal returns (CARI) resulting from climate
policy announcements. The percentage of announcements resulting in significantly negative and positive, or no CARI’s for the market and APT models using world and regional data.
Market model APT model
World data Regional data World data Regional data Negative reaction 52 62 52 47 Positive reaction 10 10 5 10
No reaction 38 28 43 43
Table 8: exploration model
Results of the model explaining the number of exploration wells drilled including the three different climate policy variables. The Anderson-Hsiao IV and Arellano-Bond GMM estimators involve first differencing and instrumenting for yi,t-1, the former using yi,t-2 and
the latter using yi,t-2, yi,t-3, … (for each t). Only 24 companies are included in the
estimations as there is no exploration and development data available on BG Group, the panel estimation removes missing values from the analysis. The variables are specified in logarithms, regular specification can be found in Appendix C. Data source: Evaluate Energy Ltd. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1.
Anderson-Hsiao IV Arellano-Bond GMM Variables (1) (2) (3) (4) (5) (6) L.XPL –0.550* –0.532* –0.556* –0.491*** –0.512*** –0.498*** (0.313) (0.305) (0.314) (0.101) (0.104) (0.103) OPRICEEXP –0.482* –0.497** –0.442* –0.247 –0.141 –0.185 (0.255) (0.253) (0.251) (0.240) (0.246) (0.234) GPRICEEXP –0.283* –0.268* –0.280* –0.139 –0.156 –0.132 (0.166) (0.162) (0.166) (0.122) (0.124) (0.121) GDP –1.303 –0.798 –0.565 –0.109 –0.456* –0.0346 (1.075) (0.870) (1.117) (0.152) (0.256) (0.130) CPdum2009 –0.120 –0.00751 (0.214) (0.0997) CPtrend –0.471 –0.165* (0.484) (0.0950) CPdum –0.199 –0.162 (0.365) (0.169) Constant –0.119 –0.0279 –0.0549 (0.0849) (0.138) (0.0922) Observations 334 334 334 359 359 359 R-squared 0.413 0.529 0.630 Number of I 24 24 24 24 24 24
Table 9: development model
Results of the model explaining the number of development wells drilled including the three different climate policy variables. Same estimators and data issues as Table 9, again specification in logarithms. Data source: Evaluate Energy Ltd. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1. Anderson-Hsiao IV Arellano-Bond GMM Variables (1) (2) (3) (4) (5) (6) L.DEV –1.095 –0.880 –1.024 –0.594*** –0.624*** –0.598*** (1.692) (1.632) (1.617) (0.0795) (0.0686) (0.0779) SXPL –0.0931 –0.0928 –0.0878 –0.133*** –0.124*** –0.125*** (0.0725) (0.0629) (0.0652) (0.0408) (0.0419) (0.0416) OPRICEEXP –0.643*** –0.773*** –0.697*** –0.330 –0.406** –0.373* (0.240) (0.258) (0.231) (0.220) (0.167) (0.194) GPRICEEXP –0.541 –0.457 –0.517 –0.438*** –0.499*** –0.460*** (0.711) (0.685) (0.678) (0.128) (0.118) (0.120) GDP –1.477 –1.899 –1.320 –0.506** –0.760** –0.537** (2.594) (2.134) (2.851) (0.222) (0.292) (0.212) CPdum2009 –0.274 –0.224 (0.266) (0.135) CPtrend –0.468 –0.167** (0.776) (0.0782) CPdum –0.468 –0.376* (0.570) (0.193) Constant –0.0808 –0.000459 –0.0575 (0.281) (0.403) (0.308) Observations 298 298 298 324 324 324 R-squared 0.809 0.838 0.819 Number of I 24 24 24 24 24 24
Bond GMM estimator. This is also the case for the climate policy dummy time series in the development model. These results suggest that the number of exploration and development wells drilled decreases as the intensity of climate policy increases. My theoretical model does not include new discoveries of oil and natural gas fields, and therefore makes no specific predictions about exploration and development drillings as a result of climate policy. However, as exploration and development leads to the physical inventory feeding production, the negative coefficients for the climate policy variables seem to be inconsistent with the green paradox theory.
7.2.2 Oil and natural gas production
