The Relationship Between Oil Price and Costs in the Oil Industry

Abstract We propose a simple structural model of the upstream sector in the oil industry to study the determinants of costs with a focus on its relationship with the price of oil. We use the real oil price, data on global drilling activity and real cost of drilling to estimate a three-dimensional VAR model. We use short run restrictions to decompose the variation in the data into three structural shocks. We estimate the dynamic effects of these shocks on drilling activity, costs of drilling and the real price of oil. Our main results suggest that (i) a 10% increase (decrease) in the oil price increases (decreases) global drilling activity by 4% and costs of drilling by 3% with a lag of 4 and 6 quarters respectively; (ii) positive shocks to drilling activity affect the oil price negatively within a year; (iii) shocks to cost of drilling have a relatively small and statistically insignificant effect on the price of oil.


INTRODUCTION
The economic profession is still struggling to fully understand the crude oil price and the shocks driving it (Hamilton, 2008;Kilian, 2009;Anderson et al., 2014).The goal of this paper is to contribute to understanding of the determinants of drilling costs and the relationship between drilling costs and the real price of oil.To do that we use data on individual wells from Wood MacKenzie on drilling activity and costs of drilling (Wood Mackenzie, 2015).The information provided allows us to construct two quarterly time series capturing (i) the total number of exploration wells drilled in the oil industry and (ii) the average cost of drilling these wells.We use the constructed time series in combination with the real price of oil to estimate a three-dimensional structural VAR model.
The proposed structural model of the upstream sector allows us to decompose the variation in the reduced form errors of the estimated VAR into three structural shocks.To identify the shocks we assume a recursive structure.The first structural shock is an oil price shock which is defined as an unpredictable innovation to the oil price.Oil price shocks capture demand shocks in the upstream sector as an increase in oil prices boosts cash flows of the oil companies and raises the profitability of marginal projects.To identify the shock we assume that the oil price is predetermined to drilling activity and costs of drilling.This assumption is plausible because it takes more than three months, and typically more than a year, following the drilling of an exploration well before the global supply of oil can be affected.Moreover, expectations of the future oil price are unlikely to be formed contemporaneously because it takes often more than a year before reasonable estimates of the newly discovered reserves are available.
The second structural shock is an activity shock, which is defined as an unpredictable innovation to the number of wells drilled.Technological advances in the upstream sector (e.g.deep sea offshore drilling) and an expansion of area available for leases allowed oil companies to explore regions of the world which have not been accessible before.On the other hand, the nationalisation of an oil industry or tougher safety regulations may force oil companies to revert to sources of lower quality and higher costs of extraction.We assume that shocks to activity are predetermined to changes in costs.This assumption is plausible because drilling costs depend on geological conditions and the duration of the drilling which may both heavily vary even within a field.As a result drilling costs are rarely known ex ante even by the operating company.
The remaining variation in the errors-after accounting for the variation in the oil price and drilling activity-is referred to as a cost shock.By construction these shocks are orthogonal to the contemporaneous oil price shocks and shocks to drilling activity and, thus, may be used to study the effect of exogenous shocks to costs on the price of oil.
We have three main results.First, following a 10% increase (decrease) in the oil price, drilling activity increases (decreases) by 4% and costs of drilling increases (decreases) by 3% with a lag of 4 and 6 quarters respectively.Second, activity shocks affect the oil price negatively within a year but have only a small and insignificant effect on drilling costs.Third, we do not find that cost shocks have a significant effect on the price of oil.
Our paper is most closely related in methodology to Kilian (2009) who uses a threedimensional structural VAR and short-run restrictions to understand the different nature of shocks driving the oil price.As opposed to our approach, he models the world market for crude oil rather than the upstream sector of the oil industry.To do this he uses global oil production to capture supply shocks.To model demand shocks for industrial commodities, he uses dry cargo single voyage ocean freight rates.After accounting for shifts in demand and supply, he argues that the residual variation in the oil price is driven by precautionary demand shocks rather than by supply shocks as has been previously believed.Our work may be thought of as an extension of his work as we take the variation in the oil price (consisting of demand and supply shocks in the crude oil market) as given and use it to decompose the observable variation in the number of wells drilled and costs of drilling.
Our paper is also related to Anderson et al. (2014).Using data from Texas on drilling activity and rig rents, they present evidence that drilling activity and drilling costs significantly respond to a change in the oil price.On the other hand, they are not able to find any significant relationship between oil price changes and the contemporaneous extraction of oil.They use these results to motivate a theoretical model in which drilling activity is at the centre of deriving an optimal extraction path.In doing so, they are able to rationalize their empirical findings.Our work differs from their contribution in two main aspects.First, and most importantly, their main contribution is theoretical whereas our focus is on the causal identification of shocks and the estimation of the dynamic responses of the variables in the system to shocks.Second, we use data on drilling and costs of drilling covering most areas of the world (excluding onshore drilling in the US), whereas their data is constraint to activity and rig rents in Texas.
Finally, our paper is also related to a more applied literature which is focusing on the identification and the estimation of the price elasticity of drilling in the oil and gas sector.See Dahl and Duggan (1998) for a comprehensive review of the literature covering also other interesting elasticities such as the size of the discoveries or the success rate of discovering a productive well.
The price elasticity of drilling typically reported in these studies ranges between 0.5 and 2.5.We estimate a drilling elasticity of around 0.5 which is at the lower end of the distribution of the estimated coefficients.
The remainder of this paper is structured as follows.In the next section we provide a discussion on the institutional framework of the upstream sector in the oil industry and drilling costs.In the third section we describe the data.In the fourth section we discuss our identification and estimation strategy.In the subsequent section our results are presented before we conclude.

COSTS IN THE UPSTREAM SECTOR OF THE OIL INDUSTRY
Activity in the oil industry can be broadly divided into three sectors: an upstream sector, which involves locating and extracting the product located under the Earth's surface; midstream sector, which mainly involves the transportation and storage of the product; and downstream sector, which involves the processing, distribution and selling of the final product.In what follows we focus on the upstream sector as we are interested in the determinants of drilling costs and their interaction with the oil price.
Reservoirs containing the product are typically buried under many layers of rock and may be located onshore and offshore.Thus, drilling represents the core of activities in the upstream sector as it is essential to access the product.Types of wells drilled differ in their purpose.We broadly differentiate between two types of wells: exploratory wells 1 and production wells.The former have the purpose of identifying new reserves.The latter are mostly drilled in known reservoirs to maintain or increase oil production.
Once the well has been drilled the drilling rig is no longer required and can be moved.Drilling activity of any producer fluctuates with outcomes from recently drilled wells and the firm's success in locating new reserves.A successful exploration well drilled attracts subsequent drilling from the own or a competing company, and a dry hole does not.Thus, outsourcing drilling rigs reduces the overall capacity requirements of rigs and greatly reduces transportation and mobilization costs.This explains why companies in the oil industry are not vertically integrated and why drilling is typically outsourced (Kellog, 2011).
Total cost of developing and operating a successful well consist of capital expenditures and operational expenditures.Costs accrued due to the drilling of a new well are subsumed in capital expenditures and represent an important share of fixed costs of developing a new well (app.40-50% in recent years according to IHS (2014)).Operational expenditures are required to ensure a day-to day functioning of the well and represent the marginal costs of a productive well.Usually, marginal costs represent only a small part of overall costs which justifies our focus on capital expenditures (Adelman, 1962).
Standard economic theory predicts a strong bi-directional relationship between the price of crude oil and marginal costs of production.It is easy to see that a productive well should stop producing if the costs of producing an additional unit exceeds the price of oil.On the other hand, wells which are expected to exhibit higher marginal costs may be activated if the oil price is sufficiently high and the profitability of the well appears more likely.In the intermediate run such a relationship should theoretically also exist between drilling costs and the oil price.We think of the intermediate run as the time period in which we can treat exploration and development of individual wells as variable in the production process.Ceteris paribus, an increase in drilling costs shifts the average cost curve up such that the break even point (marginal costs equal average costs) of the well can only be reached with a higher price of crude oil.With this simple theory in mind we hypothesise that a positive shock to the price of oil will allow companies to explore less accessible regions of the world which were not considered to be profitable before.Thus, an increase in the price of oil changes the composition of exploration wells drilled by adding more costly wells and thus pushing up the average cost of wells drilled, and vice versa.To re-optimise the drilling activity a company will typically require some time.Our approach allows us to quantify how many quarters energy companies need on average to respond to a change in the price of oil and how the change in drilling activity is associated with the average cost of drilling.
On the other hand, we would also expect that negative and positive shocks to drilling costs eventually translate into changes in the price of oil.Negative shocks to drilling costs might be the result of technological advances (e.g. the recent improvements in horizontal drilling).This theoretically expected downward pressure on drilling costs is counter balanced by an increased scarcity of easily accessible resources.For two reasons we expect the identification of such shocks to costs and the quantification of their effect on the price of oil to be quite challenging.First, the existence of large economic rents in the oil industry.Economic rents can adjust to capture shocks to drilling costs which arguably dilutes the propagation mechanism.Second, the depletion of available resources and technological advances are gradual processes and difficult to statistically identify in the short run.This is particularly true in our context in which we focus on the variation in the growth rates rather than levels.Nevertheless, our approach allows us to identify unanticipated changes to cost of drilling a well and to estimate the cumulative effect of these shocks on the price of oil.

DATA
The raw dataset was obtained from Wood Mackenzie, and compiled using a range of methods: (i) meetings with energy companies, annual reports and industry specific publications, (ii) state publications and information from public institutions, (iii) historical data, investor presentations and different types of media sources.The data is subject to several limitations which we discuss below.
Unfortunately, the sample does not contain any information on production wells.However, due to our focus on growth rates in drilling activity and cost of drillings, the missing information does not appear to be a serious drawback.Global data suggests that the share of exploration drillings in the total number of wells drilled amounts to 20-25% offshore and 7.5-12.5% onshore over the last 10 years (IHS, 2014).The share of total drilling expenditure spent on exploration drillings amounts to 30-35% onshore and offshore.Overall, the numbers suggest that the development of new reserves requires a constant ratio of exploration and production wells to be drilled.If we are willing to assume a constant ratio over time, growth rates in exploration drillings and cost of drillings represent reasonable proxies for growth rates in production drillings and costs. 2 3.The correlation coefficient between the depth of the well drilled (logged) and the cost of drilling (logged) is above 0.5.
4. The share of cost variables missing in the subsample since 1999 is less than half a percent.Only in 195 cases the information is not available for both variables.This represents less than 1% of our sample.
5. We have information on the starting day of drilling and on the completion day of drilling.We use the former because we are interested in how quickly companies in the oil sector respond to an increase in the crude oil price.
6. Nominal cost of drilling is transformed into real values using US CPI.Quarterly CPI for the US is taken from OECD statistics with the base period in 2010.
7. The calculation of costs per well can vary depending on the region and the company involved.However, there is a great deal of effort from company's employees to construct comparable costs of drilling and exclude "back office" costs.
8. Additionally or alternatively, we could have normalised the variable by the depth of the well drilled.However, that would not capture an important part of the mechanism we have in mind.Ceteris paribus, the necessity to drill deeper to locate new reserves implies that a unit of oil extracted is more expensive.
We are also completely missing information on US onshore drillings of exploration wells.Unfortunately, there is little we can do about it.However, as before, we argue that by focusing on growth rates we are able to limit the consequences of this drawback.IHS (2014) provides yearly information on the total number of exploration wells drilled onshore and the total amount spent on those drillings for the US and the rest of the world.We use that information to compute correlation coefficients between the growth rates in the US and the rest of the world.The computed correlation coefficient is above 0.95 suggesting that the missing information may be ignored for our purposes.
We limit our data in the time dimension by starting our analysis in 1995 because the quality of data appears to decrease significantly over time from 2000 backwards.We also exclude successful gas wells from our sample because gas markets are not sufficiently well integrated globally and gas is not a good substitute for crude oil.The exclusion of gas wells does not alter our result significantly.
Approximately 4% of the reported cost of drilling is missing.To account for the missing observations we proceed as follows.We start with the premiss that the most important predictor of drilling costs is the depth of the well drilled 3 and whether the well is drilled onshore, on the shelf or deep offshore (henceforth location of drilling).In Figure 7 (see Appendix) we plot the estimated means of the depth of the wells for individual years and locations by differntiating between observations for which the information on costs is missing and for which it is not.Because most of the data missing is form the period before 1999 we only plot the estimates for the years 1995-1998. 4 Eyeballing the estimated means in Figure 7 suggests that for our purposes (calculation of averages) we may treat the missing observations as missing at random.We replace the missing values with the group specific mean of the available information when calculating the quarter specific means.
The raw data for drilling activity and cost of drilling is presented in Figure 9 and in Figure 10.Note that Figure 9 suggest that the total number of exploration wells drilled was subject to short term fluctuations but remained reasonably stable in the last 20 years.However, the composition of drilling activity changed over time.In particular, oil companies reallocated their effort from onshore to deep offshore drilling in the end of the 1990s.Figure 10 differs in this respect and suggests that costs of drilling experienced a continuous increase since approximately 2000, whereas the increase in costs was particularly pronounced in deep offshore drilling.
We use the data to construct two variables.First, quarter specific growth rates in drilling activity.Drilling activity is captured by the total number of wells drilled in a quarter. 5Second, quarter specific growth rates in cost of drilling.Cost of drilling is the average cost of drilling one exploration well in real 6 $US in a particular quarter. 7Costs of drilling a well varies considerably across locations of drilling (onshore, shelf and deep offshore) as presented in Figure 10.Thus, before calculating the quarter specific averages, cost of drilling is normalized by the average location specific real cost of drilling in our sample period. 8More formally, the normalized real cost of drilling  a well is given by , whereas denotes real cost of drilling a well in location with c wlt c ˆ= c w l wt wlt c l {onshore, shelf, deep offshore} in quarter . 9The constructed time series are then logged and l ∈ t first differenced.Both time series result in 75 observations and are presented in the second and third panel of Figure 1.
In the first panel of Figure 1 the percentage change of the real oil price is presented.The nominal daily price of crude oil 10 is taken from EIA and the arithmetic mean is used to calculate quarter specific values.As before the oil price is transformed into real values using US quarterly CPI.Descriptive statistics of our variables are presented in Table 1.

IDENTIFICATION AND ESTIMATION
We use the growth in the real price of oil, , the growth in the number of wells is a vector of constants capturing the average growth rate.
is a matrix of the respective α A i coefficients in period .is a three-dimensional vector with serially uncorrelated and mutually ti e t uncorrelated errors.We assume that matrix has a recursive structure such that the reduced form A 0 errors can be decomposed according to : - We suggest to decompose the variation in the time series into three structural shocks.The first structural shock is an oil price shock which is defined by unpredictable innovations to the oil price.Oil price shocks represent demand shocks in the upstream sector.An increase in oil prices boosts cash flows and raises the profitability of marginal projects.Both channels trigger a rise in capital expenditure and drilling activity.To identify the shock we assume that the oil price does not respond contemporaneously (within a quarter) to shocks originating in the upstream sector, shocks to drilling activity or cost of drilling.This assumption is justified by the fact that increased activity in drilling does not immediately translate into a change in oil supply.It takes typically longer than a quarter before the drilling of an exploration well translates into an increased production of oil (Jahn et al., 2008).Alternatively, one could argue that expectations of market participants about new discoveries effect the oil price.But it usually takes more than a quarter before a reasonable estimate of the size of an oil field is formed, and it takes even longer before total drilling costs can be determined (Adelman, 1990;Jahn et al., 2008).Formally, we assume and to be zero.a a The second structural shock is referred to as an activity shock and is defined as an innovation to drilling activity which cannot be explained by oil price shocks.These shocks may be driven by technological progress and a change in regulatory constraints.Technological advances in deep sea offshore drilling and an expansion of area available for leases allowed oil companies to explore regions of the world which have not been accessible before.On the other hand, the nationalisation of an oil industry or tougher safety regulations may decrease drilling activity in certain regions of the world.We assume that shocks to activity are predetermined to changes in costs.This assumption is plausible because it is very difficult to estimate drilling costs accurately and often drilling costs are only known after the drilling has been completed (Jahn et al., 2008).Formally, we assume to be zero.a 23 We refer to the remaining variation in cost of drilling-after accounting for the variation in the oil price and drilling activity-as cost shocks.Shocks to drilling costs may represent random variation in drilling costs over time.But they also capture the continuous trade-off between technological progress and increasing project complexity.By construction, cost shocks are orthogonal to oil price shocks and drilling shocks and, thus, may be used to evaluate the impacts of these shocks on drilling activity and the price of oil.
We use the LR sequential test to determine the optimal number of lags (the results are presented in Table 2).In our preferred specification we use 8 lags.The reduced-form VAR model is consistently estimated by the least-squares method (Lu ¨tkepohl, 2011).We rely on asymptotic theory to construct the 90% confidence intervals which are used for inference.
We conduct several robustness tests.First, we reestimate our VAR by changing the ordering of the first two variables.Thus, we assume that drilling activity does not respond to oil price shocks within a quarter.This assumption appears reasonable because companies typically do not change investment decisions based on short term variations in the oil price.Second, instead of using the number of wells as a measure for drilling activity we use the Baker Hughes rig count to proxy activity in the upstream sector.Third, we reestimate our model with 12 lags and 4 lags instead of 8 lags.The estimated cumulative IRFs are presented in Figure 13 and Figure 12.Fourth, instead of relying on asymptotic theory we bootstrap the standard errors.Our results are robust to all these changes and in most of the cases remain nearly identical.

RESULTS
Dynamic responses of the variables in the system to structural shocks-In Figure 2 we present the responses of the real oil price, drilling activity and costs to the structural shocks.We prefer to present the cumulative responses of the variables because we are interested in the new equilibrium reached rather than short run fluctuations.For the interested reader the non-cumulative impulse responses are presented in Figure 11 (see Appendix).
Following an oil price shock, , the oil price increases immediately but then adjusts within p e t the next three years to approximately half of the magnitude of the initial increase.The initial 13% The Relationship Between Oil Price and Costs in the Oil Industry / 245 Copyright ᭧ 2015 by the IAEE.All rights reserved.
11. See last paragraph of section 2 for a discussion of these reasons.increase in the price of oil increases the number of wells drilled by approximately 6% within a year and average cost of drilling by nearly 4% within 6 quarters.In the long run (15 quarters) the initial increase in the number of wells drilled nearly returns to the initial level.On the other hand, drilling costs increase permanently by less than 4%.
Following a structural activity shock, , the number of wells drilled increases by approx-d e t imately 8% instantaneously but then drops to 2% in the long run.Costs of drilling is only affected contemporaneously and does not differ significantly from the initial level in the subsequent periods.On the other hand, the oil price seems to respond within a year to an increased activity in the upstream sector.The negative response in the oil price is consistent with an increase in crude oil supply following an increase in drilling activity.An initial increase in drilling activity by 8% decreases the oil price by approximately 3% permanently.
Drilling costs increase immediately and persistently by approximately 8% following a structural cost shock, .These shocks are associated with a temporary increase in drilling activity c e t which dies out within a year.Our results further suggest that cost shocks do not have a significant effect on the price of oil.However, the estimated coefficient of the cumulative response remains always positive while being surrounded by a reasonably large confidence interval.Thus, a potential interpretation of our results is that we do no find any evidence for the theoretically expected response of the oil price to cost shocks because for several reasons 11 it is difficult to estimate the dynamic response of the oil price precisely.
Evolution of structural shocks and their cumulative effect-In Figure 3 we plot the structural shocks implied by the suggested model.The presented residuals are smoothed by a local polynomial estimator to ease interpretation.Figure 4-Figure 6 display the cumulative contributions of individual structural shocks to the development of the variables in the system.We will discuss the development of the individual structural shocks and their contribution to the variables in the system in turn below.
The structural oil price shock presented in the first panel of Figure 3 exhibits comparably large swings and, not surprisingly, is the main source of the oil price variation as can be seen in the first panel of Figure 4. Structural oil price shocks also explain a large chunk of the variation in drilling activity where they generate prolonged swings (see Figure 5).In comparison to the cumulative effect on drilling activity the contributions of the structural oil price shock to cost of drilling appear smaller, relatively smooth and more persistent (see Figure 6).In particular, the increase in the oil price since approximately 2003 (see Figure 3 and Figure 8) persistently increased cost of drilling but had a less persistent positive effect on drilling activity which is consistent with our analysis of the cumulative impulse responses.
The historical development of the activity shock is presented in the second panel of Figure 3.The development of the activity shock appears smoother in comparison to the oil price shock with a large negative shock in the end of the 1990s and a period of smaller but successive positive shocks between 2007 and 2010.Note, that the negative activity shock coincides with several big mergers in the oil industry, e.g.Exon and Mobile, BP and Amoco, Chevron and Texaco.Thus, besides the necessity to decrease capital expenditure due to an extreme low of the oil price in 1999 the decrease in drilling activity might reflect the purpose of the mergers and acquisitions to reduce costs and increase profitability.Both of these shocks had an effect on the price of oil by creating an upward pressure on the oil price in the end of 1999 and a downward pressure since 2007 (see   4).On the other hand, activity shocks did not seem to contribute much to the variation in drilling cost in the last 20 years.
The smallest variation among the structural shocks is displayed by the structural cost shock.Given the potential source of cost shocks, that does not come as a surprise.Above we have argued, that besides random variation, shocks to drilling cost may reflect the continues trade-off between technological advances and the increased scarcity of easily accessible oil fields.Both are gradual trends, which disseminate only slowly and which are unlikely to be reflected in large structural shocks but are more likely to result in small successive shocks.However, the cumulative effect of a series of small shocks might still exhibit an observable and persistent effect on the variables in the system.Indeed, the time series presented in panel 3 of Figure 6 suggests that a series of small and negative structural shocks in the end of the 1990s had a negative and persistent effect on the cost of drilling.We suspect that the series of negative cost shocks in the end of the 1990s might reflect technological advances in the oil industry such as improvements in seismic technology (Dahl and Duggan, 1998).More importantly, panel 3 in Figure 4 suggests that this cumulative effect has also translated into a downward pressure on the price of oil.Thus, even though we were not able to pick up a statically significant impulse response of the oil price to a structural cost shock the historical decomposition of the oil price suggests that an accumulation of structural cost shocks had a small but still measurable effect on the price of oil.

CONCLUSION
We use micro data from Wood MacKenzie to construct two quarterly time series capturing the total number of exploration wells drilled and (ii) the average cost of drilling.In combination with the real price of oil we estimate a three dimensional structural VAR model.By assuming a recursive structure we are able to decompose the variation in the reduced form errors into three structural shocks: an oil price shock, an activity shock and a cost shock.We estimate the dynamic effects and present the historical cumulative effects of these shocks on drilling activity, costs of drilling and the real price of oil.
We have three main results.First, our results suggest an upwards sloping supply curve in the upstream sector of the oil industry.In particular, following an oil price shock of 10% costs of drilling increases permanently by approximately 3%.The increase in drilling activity following an oil price shock is larger, ca.4%, but much less persistent.Second, activity shocks affect the oil price negatively and permanently within a year.Third, shocks to cost of drilling have only a small and statistically insignificant effect.However, the historical decomposition of the structural shocks reveals that a series of successive negative cost shocks in the end of the 1990s had a small but persistent effect on the price of oil.

Figure 1 :
Figure 1: Time series of the main variables (growth in percent)

Figure 2 :
Figure 2: Responses to One-Standard-Deviation Structural Shocks

Figure 3 :
Figure 3: Historical Evolution of the Structural Shocks

Figure
Figure 4: Historical Decomposition of the Oil Price

Figure
Figure 8: Real Oil Price (logged)

Table 1 : Descriptive Statistics
9. Alternatively, we construct the time series without the normalization.Our main remain are robust to this change.10.Brent Spot Price FOB (Dollars per Barrel)