Applied Drilling Circulation
Applied Drilling Circulation
Applied Drilling Circulation
Rates
Abstract
We present a simple real options model that illustrates how changes in uncertainty can result
empirically test the model, we examine a panel of oil rig rental rates. Our empirical analysis
conrms that after we control for several relevant economic variables, price uncertainty
Real options are generally dened as assets that provide their owners with economic
opportunities, but not obligations. For example, a vacant lot provides its owner the op-
portunity to construct a building (Titman (1985)), and mineral rights provide the owner
with the opportunity to extract natural resources (Brennan and Schwartz (1985)). The key
insight from this analysis is that, ceteris paribus, increased uncertainty increases the value
of the real option and decreases the tendency that it will get exercised (see also McDonald
other aspects of the real option exercise decision. For instance, consider the option to drill
I
Khokher and Venkatesan are at the A.B. Freeman School of Business, Tulane University, 7 McAlister
Drive, New Orleans, LA 70118 USA. Morovati is at Stanford University. Titman is with the McCombs School
of Business, The University of Texas at Austin, 2110 Speedway, Stop B6000 Austin, TX 78712 USA. We
thank workshop participants at the 2015 Financial Research Association meeting for comments. Remaining
errors are ours alone.
an oil well in the Gulf of Mexico, which is the focus of this paper. Because the supply of
drilling rigs is at least temporarily xed, a shock to uncertainty about oil prices, or even
a shock to the level of oil prices, will not necessarily aect the amount of drilling. As our
simple model illustrates, such a shock may, instead, cause the rental rates on the oil rigs
to adjust. In the language of real options, shocks to option payos (the oil revenues) will
aect the option exercise prices (the cost of drilling), but may not aect the exercise choices
(whether or not to drill). If this situation is indeed the case, a more appropriate test of the
real options model will examine the cost of drilling rather than the level of drilling.
In our simple model, the supply of drilling rigs is xed and producers can extract their
oil reserves immediately, or they can delay extraction untill the next period. As uncertainty
increases, this timing exibility becomes more valuable and producers are less likely to extract
reserves immediately. In this case, the producers' demand curve for oil rigs shifts downward
To test the model, we analyze the rental rates of drilling rigs in the Gulf of Mexico. Our
dataset, which covers virtually all oshore drilling projects from 2000 to 2014 in the Gulf of
Mexico, contains over 2,000 detailed rental contracts for drilling equipment. Our database
contains 256 active oil companies, a large majority of which are not publicly listed. We have
data on the rental rates of 250 separate drilling rigs: we control for rig-type specic xed
eects and identify the eect of macro variables on the rental rates.
As expected, rig rental rates are low when oil prices are low. In addition, as the water
depth increases, the forward looking eect (reected in the rental rates' responses to futures
prices) becomes dominant, while the eect of nancial constraints decreases. Furthermore,
our empirical results show that rig rental rates are low during periods of increased uncertainty.
We perform a host of econometric tests to establish that rig operators respond to changing
market conditions by adjusting rig rental rates, and also to show that volatility has a causal
impact on these investment costs. In our tests, we control for a variety of market variables
2
and the technical features of the rigs.
This paper adds to the literature on exhaustible resource investments. Litzenberger and
Rabinowitz (LR 1995) use Tourinho's (1979) characterization of reserves as call options to
determine oil prices. Both Tourinho and LR take extraction costs as given. In contrast, our
focus is on the determinants of these investment costs in equilibrium. This paper is also
related to studies that examine granular capital expenditure data. For example, Paddock et
al. (1988) apply option valuation techniques to data on oshore lease auctions in the Gulf
of Mexico. Kellogg (2015) reports that in, onshore Texas oil elds, the response of drilling
activity to price uncertainty is quantitatively consistent with his reduced-form model. Gilje
and Taillard (2016) use natural gas drilling activity to compare the investment decisions of
public and private rms. However, these studies do not focus on the drivers of investment
costs per se; they rely on nancial statement or intensity-of-drilling based measures to
proxy for investment. Compared to these approaches, our rig rate data gives us a more
Section 1 presents a simple real options model of the market for oil rigs. Section 2
reviews the institutional setting of the oshore oil and gas industry in the Gulf of Mexico
and describes the rental rate data. Section 3 presents the empirical results. Section 4
concludes.
1. The Model
In this section, we present a simple model to show that in some cases, a shock to the
volatility of oil prices has no eect on the number of oil rigs in service or on the quantity of
oil produced. Instead, a volatility shock causes the market to react by adjusting rig rental
rates.
3
1.1. The Economy
We consider a two period economy (t = 0, 1) with a reserve of crude oil and N drilling
rigs.
1 The oil is located underground at depths that lie within (0, ∆]. At each depth δ (0, ∆],
an unlimited quantity of oil is available. All of the N rigs can be used in both periods. Even
if a rig is used in period 0, it becomes available to be reused by period 1. The rigs are of
dierent types; each type of rig is appropriate to extract reserves located at a particular
depth. For example, rig type N can only be used to extract oil located at depth ∆.
The economy contains two types or categories of agents: oil producers (or drillers), and
rig operators. Agents in each of these categories exist in a continuum. We identify a producer
in the continuum by the depth (δ) at which her reserve is located. Each producer in the
continuum owns equal quantities of reserves. Each rig operator in the continuum owns equal
Because the supply of rigs in the economy is limited (N ), producers can only extract Q of
the total reserve in any period. Producers rent rigs to extract reserves; to extract a quantity
small part of the total supply, so we take oil prices as exogenous in our model. All market
participants observe the spot price of oil(P0 ) at the initial period (t = 0) and the distribution
of the oil price at the terminal period (t = 1). In particular, the time 1 price of oil P ˜1 is
lognormally distributed with E lnP ˜1 = µ, and V ar lnP˜1 = σ 2 . We denote the rate of
because each producer's reserve is located at a dierent depth and the costs of extracting
reserves increase with the depths at which the reserves are located. The time t extraction
4
cost xδt for producer δ lies within (0, Xt ]. Producers extract low cost reserves rst, and so
The producers' extraction costs consist of the costs of renting rigs and additional costs,
for example labor. A producer δ pays xδt = (Rt + C) δ to extract her reserve at time t. The
labor cost parameter C is a given scalar. We determine the specic values (Ret ) of the rig
rigs depends on the time 0 value of the rig rate variable (R0 ). The relation between the
number of oil rigs that producers demand and R0 constitutes the demand curve for oil rigs
at period 0. The optimal number of rigs for a particular level of R0 is given by,
ˆm ˆ∆
δ
0 ∆−δm
Here, δm is the marginal producer for the specied level of R0 , and E0Q (.) denotes expected
δm N
producers demand for the specied level of R0 is then immediate as, ND (R0 ; P0 , σ) = ∆
.
At this initial period, each producer can extract her reserve for sale in the spot market
or leave the reserve underground. Producers do not extract oil for above ground storage. If
the producer leaves her reserve underground, she retains the option to extract her reserve
in period 1. The producer's decision involves comparing the value of the extracted oil to
the value of the underground oil reserve. If the payo from immediate extraction is greater
2 Agents at time 0 observe the distribution of oil prices and rig rates at time 1. They observe the
distribution of rig rates at time 1, because we solve the model recursively starting at time 1. In this case,
agents can make their time 0 decisions conditional upon the distribution of oil prices and rig rates at time 1.
5
than the value of the underground reserve, the producer extracts the oil. In this case, we
will have a marginal producer, δm , such that all producers with lower extraction costs will
extract their reserves at t = 0, and those with higher costs will not. This value of δm enters
into the calculation of the optimal number of rigs demanded in period 0, N D (R0 ; P0 , σ).
This marginal producer, δm , will vary with the particular choice of R0 . The demand for rigs
to extract their oil. Rig operators decide upon the number of rigs, NS (R0 ; P0 , σ), they will
supply. This number will, in general, depend upon the level of R0 . The relation between
NS (R0 ; P0 , σ) and R0 constitutes the supply curve for oil rigs at period 0.
At this initial period, each rig operator compares the revenue she can earn if she rents her
rig immediately against her revenue if she chooses not to rent her rig. If she rents her rig, she
earns δR0 ; she earns nothing otherwise. The aggregate of each rig operator's optimal decision
to rent or not will yield the optimal number of rigs supplied in period 0, NS (R0 ; P0 , σ).
1.4. Equilibrium
At period 0, producers and rig operators bid on and supply rigs to determine the specic
value R0e of the rig rental rate variable R0 .3 The quantity of rigs demanded [ND (R0 ; P0 , σ)]
and supplied [ND (R0 ; P0 , σ)] is a function of the rig rental rate variable, R0 . In equilibrium,
the quantity of rigs demanded will equal the quantity supplied, (ND (R; P0 , σ) = NS (P ; P0 , σ)).
This equilbrium is characterized by the value of the rental rate variable, R0e (P0 , σ), that
equates demand and supply. Proposition 1, which follows later, documents key characteris-
3 We consider the current price of oil (P0 ) is suciently high, so that it does not constrain this optimal
choice.
6
At period 1, agents observe the realized price of oil (P1 ). Producers and rig operators
again bid on and supply rigs. This process determines the specic value R1e of the rig rental
rate variable R1 that equates demand and supply at this terminal period. The specic value
R1e of the rig rental rate variable R1 will depend on the realized price of oil, P1 . At this time,
producers will rent all rigs that are economically viable. Specically, producers will rent rigs
to extract reserves as long as extraction costs are below the realized price of oil, P1 . Figure
2 explicitly relates producer δ 's payo at time 1 to the price of oil at that time (P1 ).
Proposition 1. In this equilibrium, the economy fully utilizes the available supply of rigs
(N ) and aggregate production (q) is at capacity (Q) . In particular, ND (R0e ; P0 , σ) =
NS (R0e ; P0 , σ) = N , reserve ∆ is extracted and q(∆) = Q. Furthermore,
i. The equilibrium rig rental rate decreases in the riskiness of the spot price and increases
in the spot price level. Specically, given σ0 > σ and P00 > P0 , then R0e (P0 , σ0 ) < R0e (P0 , σ),
and R0e P00 , σ > R0e (P0 , σ) .
ii. The equilibrium rental rate adjusts to a shock to the riskiness of the spot price so that
the economy fully utilizes the available supply of rigs, NDe (R0e ; P0 , σ0 ) = NSe (R0e ; P0 , σ0 ) = N
, and continues to produce at capacity (Q).
The intuition for these results is as follows. First, the available supply of rigs will be fully
utilized at the initial period because each rig operator will optimally choose to rent her rig.
The full utilization of the rigs follows because, each rig operator can and will rent her rig in
both periods. Because all rigs are used, the economy produces at capacity.
Furthermore, producers can extract their oil reserves immediately, or they can delay ex-
traction untill the following period. These reserves grant producers the option to delay their
investment decisions. An increase in the riskiness of oil prices makes this timing exibil-
ity more valuable and, thus, producers are less likely to extract reserves immediately. In
7
this case, the producers' demand curve for oil rigs shifts downward, and the equilibrium rig
rental rate will decline. This result implies that, ceteris paribus, equilibrium rig rental rates
should decrease with oil price volatility. Figure 3 illustrates this intuition. Analogously, an
increase in the spot price of oil makes extracting producers' oil immediately attractive, thus
raising the demand for rigs. Hence, equilibrium rental rates will rise. Whether the level
or the volatility of the spot price changes, rig rental rates will adjust endogenously, but oil
able beneath the surface of the earth. Extracting resources out of these reservoirs requires
sophisticated techniques that depend on several factors, including the geology of the area
and the type of resource. In conventional reservoirs, which are still by far the dominant
source of energy in the world economy, extraction requires drilling one or several wells into
the reservoirs.
4 The wells often pass through several thousand feet of rocks to reach the
reservoirs, and once the wells are completed, crude oil usually starts owing out.
5 Oil and
gas professionals refer to a tract of sea or land that contains oil reservoirs as an oil eld,
4 These stocks are often trapped within porous rocks under immense pressure. A majority of conventional
oil and gas production projects involve drilling oil wells into these high-pressure reservoirs and the resources
simply start owing, at least during the early stages of production, due to their high pressure.
5 Recent advances in drilling techniques has been a major enabling component of shale gas revolution.
Horizontal drilling, once beyond imagination, requires directing the drilling bit thousands of feet under the
surface to make a vertical turn and penetrate the hydrocarbon layers, often parallel to the earth's surface.
6 Most of the hydrocarbon reservoirs contain both oil and gas. If a hydrocarbon reservoir mainly contains
natural gas, the eld is called a gas eld. In the rest of this paper, we will simply refer to crude oil for much
of our study, but our results are easily extended to natural gas, as well.
8
Oshore drilling targets the hydrocarbon resources that exist under the seabed. The
Bureau of Ocean Energy Management (BOEM) is responsible for managing and administer-
ing petroleum production in the Federal regions of the Gulf of Mexico. BOEM has divided
the Gulf of Mexico into block grids; each grid contains several square miles, usually in the
compete in the bidding process to win the leases. Energy companies can perform seismic
analyses in advance to evaluate the potential for hydrocarbon discovery in each tract. The
winning bidder pays the sale price and obtains the right, but not the obligation, to start
developing the petroleum eld for a certain period of time, usually ve or eight years. If a
company starts developing a eld, it keeps the right to production from that eld for as long
as it wishes to pay the associated fees and royalties. If a lease owner decides to abandon a
tract and not to develop it, the tract will return to the stock of BOEM tracts after the lease
expires and becomes available for potential future auctions. The tract lease is analogous to
a call option. To own the lease, the winning bidder has to pay certain fees, similar to the
price of a call option. Owning the lease gives the owner the right, but not the obligation, to
start drilling to make the tract productive and create future streams of income. The owner
will only start developing if she nds the development worthwhile or, in other words, nds
the expected future cash ows more valuable than the development costs. The development
costs are similar to the strike price of a call option, and the discounted expected future cash
ows are analogous to the value of the underlying asset in a call option.
Oshore projects require large upfront investments, but yield uncertain production rates
and cash ows to the investors. The major cost of developing an oshore oil eld is the
rental cost of drilling rigs; this may account for 60-70 percent of the total development cost.
Oshore drilling rigs are huge mobile structures that rms use to drill wells in the seabed to
7 For a precise map of the grid blocks, please download the following pdf le:
http://www.boem.gov/uploadedFiles/BOEM/Oil_and_Gas_Energy_Program/Mapping_and_Data/visual1.pdf
9
reach the petroleum reservoirs. Petroleum service companies (e.g. Transocean) usually own
this equipment and rent it out to oil companies (e.g. BP). Often, oil companies contract
rigs for short periods of time, ranging from several weeks to a few months. Included in the
rig rental fees are costs of highly skilled labor provided by the service company, specialized
equipment and material needed for the specic rig, and other overhead items like helicopters,
which are necessary for employees to commute to and from the rigs. Once drilling is complete,
the oil company installs the production facility and the petroleum service company transfers
the rig to its next drilling location. Although companies can transfer drilling rigs virtually
anywhere, the high opportunity cost of the time spent on long routes eectively creates a
and risks associated with severe weather are some of the other aspects that are relevant to
oshore drilling.
8
unique dataset gives us precise estimates of investment costs in the Gulf of Mexico oshore
drilling market. An important point to note is that we are mainly focusing on private sector
in this paper, in eect, contribute to capacity building within a competitive fringe that is a
Analysts often use the capital expenditures of publicly listed oil and gas companies as
a measure of investment costs. Government regulations require public rms to make their
balance sheets available. However, using balance sheet data alone imposes serious limitations
on the empirical analysis. First, the frequency of publicly available data on capital expen-
8 The DeepWater Horizon disaster in April 2010 provides an example of the immense potential risks
involved.
10
ditures is very low (annual, for most cases). In addition, the number of public exploration
and production rms (E&P) is small. This small number limits the available data and, as
a result, decreases the power of empirical tests. Second, the published data on capital ex-
penditures is aggregated and does not relate to a specic investment opportunity. The data
aggregates assorted types of investments. These investments may include building reneries,
installing pipelines, expanding current production plants, or developing new elds. Because
of this aggregation, identifying the factors that will impact new investment opportunities is
dicult. Third, many active private companies do not publish their balance sheets.
Our detailed dataset allows us to evaluate investments at the smallest unit of investment
in the oil industry: an oil well. The rental cost of drilling equipment constitutes the majority
of the cost of developing an oil eld. The cost of drilling alone accounts for 60 to 70 percent
of the oshore drilling contracts between service companies and oil companies from 2000 to
2014. The contracts include the names of the service companies that owned the rigs, the
names of the oil companies that rented the equipment, the contracts' start and end dates,
and the rig rental rates. The dataset also includes the xture date (contract signing date)
of each contract, cost and age of the equipment, and the rated water depth. The technical
specications and rig type allow us to follow each group of rigs with great precision.
10
The dataset includes almost 2033 contracts for drilling in the Gulf of Mexico region.
Our database contains 256 active oil companies, a majority of which are not publicly listed
companies. The publicly listed companies operate mainly in deep water, primarily, because
of the great technological sophistication and investment costs necessary to operate in deep
waters. Both of these factors are prohibitive for small companies. Furthermore, large public
9 https://www.eia.gov/analysis/studies/drilling/pdf/upstream.pdf
10 A detailed description of the data elds present in our data is available at the data vendor's website:
http://www.riglogix.com/RigLogix_Data.aspx
11
rms can access the capital markets more easily than small concerns, and they face fewer
nancial frictions. To evaluate the eect of nancial constraints on the rms in our sample,
we divide our sample into three dierent water depths. We run our tests on each bin
The drilling data consists of more than 250 drilling rigs that have been repeatedly con-
tracted. Hence, they are unavailable at times and become available again after their rental
periods expire. The unit of observation is at the individual contract level. Accordingly, each
rig appears several times in our data. We match each contract to the OV X , V IX , spot,
f utures, treasury , and AAApremium levels that existed on the rst date of the contract
signing month. We use the CBOE Crude Oil ETF Volatility Index (OV X) as a proxy for
the volatility in the oil prices. OV X is a market estimate of expected 30-day volatility of
crude oil futures prices and is calculated using the CBOE Volatility Index (V IX ) methodol-
ogy applied to options on the United States Oil Fund, LP (USO). Unfortunately, the OV X
series goes back only untill June 2007.
To increase the statistical power of our tests, we also use V IX as a proxy for price
reected by the S&P 500 stock index option prices and, importantly, it is highly correlated
with the OV X .11 Signicantly, unlike the OV X series, the V IX series spans the entire
Spot and f utures are the spot price of Brent Crude and the slope of the 12-month
futures price of Brent Crude, respectively. Although we have the exact contract signing
dates, businesses often make investment decisions using information from several months
prior to and following considerable negotiations with the rig operating companies. To allow
for this dierence in timing, we take the past six-month moving average of volatility, spot
11 The correlation between OVX and VIX is greater than 0.83. More information about OVX, VIX, and
their computational methodology is available at http://www.cboe.com/
12
price, and futures price. We use the yield on the treasury bill (treasury) and the excess yield
of three month-to-maturity investment grade bonds over treasury bills of the same maturity
Lynch provide the US Corporate AAA Index (AAAyield). This index is the yield on dollar
denominated investment grade corporate debt publicly issued in the US domestic market.
Table 1 provides a summary of the overall matched data. In Panel A, we provide the
distributional details about the contract-related variables, and also of the other relevant
nancial variables. The median age of the rig is 27 years, and the median days for which it is
rented is 39. Panel B presents summary information by the dierent rig types. Jackup and
Semisub are the two most popular rig types that drillers use in the Gulf of Mexico region.
Because the complexity of drilling increases with the water depth, the rigs are distinguished
primarily by the maximum level of water depth at which they can operate. Commensurate
with the challenges of deep water drilling, companies need superior and expensive technology
to successfully drill at these depths. Therefore, rigs that can drill in the deep waters are
expensive to construct and, hence, also have high daily rental rates. For example, the average
rental rate for a Submersible rig, which operates in relatively shallow water, is $61,646 per
day. Compare this amount to $406,343 for renting a Drillship, a rig that operates at depths
of close to 10,000 feet. Also note the high correlation in operable water depths, construction
We recognize that there exist great dierences in the rental rates across the dierent
types of rigs, and we deal with these dierences in our multivariate regression specications.
In Panel C, we highlight the variation in the daily rental rates, one of the main points of our
paper. The standard deviation in the rental rates, across each of the rig types, is at least
35 percent of the average rental rate. Standard real options models do not account for such
variation in investment costs. In the following section, we identify the dierent factors that
13
3. Empirical Analysis
In this section, we empirically investigate several factors that aect the investment deci-
sions of oil and gas companies, and we document the results of our tests.
sion specication:
For every contract, i, and rig type, j, the rental rates are denominated in dollars per day.
OV X and V IX are our primary measures of volatility in oil prices. Yield on the Treasury
of Brent Crude oil (spot) and the slope of the futures contract price for Brent Crude to be
Equation (2) also uses some rig-specic characteristics as control variables. We include
age of the rig (Age), and construction cost of the rig (Construct) as independent variables
to explain the variation in daily rental rates. We expect the rental rates to decrease with
the age of the rig and to increase with the construction cost of the rig. In addition, we
also add contract-specic variable(s). A potentially important predictor of the daily rental
rate should be the water depth at the location of the well (LocDepth). Based on our earlier
14
discussion, we expect the rental rates to increase with the water depth and hence, predict
To capture the dierences across rig types, all regression specications have a rig-type
specic xed eect. We cluster the standard errors by every calender month, rig manager,
and by rig owner to correct for any correlation in error terms. Further, we take the natural
log of all of the explanatory variables. Taking the natural log helps us to interpret the
marginal eects of the coecients as a percent change in the dependent variable to a percent
denominated variables like rental rate and spot by removing the eect of general ination.
We use the monthly consumer price index (CPI) to adjust prices for ination.
Table 2 presents the results from a pooled OLS regression. Consistent with standard
economic intuition, age of the rig is negatively related to rental rates and construction cost
of the rig is positively related to rental rates. Further, we also nd that the the rental costs
increase with the water depth at the drilling location. We observe this eect because the
technical sophistication and risks of drilling increase as the water depth increases. Note, this
eect captures the variation in rental rates after controlling for the rig-type.
The cost of capital and price volatility are the two key variables in Table 2. High cost of
capital, as reected by the eective yield on a treasury bill, is positively correlated with rig
rental rates. As the cost of capital in the aggregate economy increases, the rig owners' costs
of running the business will go up. In a perfectly competitive economy, we would expect the
rig owners to pass these additional costs on to their eventual customers, the lessees. Our
results are consistent with this expectation. Importantly, in column (I) of Table 2, which
uses the whole sample, we also observe a negative relationship between future price volatility
15
The nancial risks associated with drilling, and the nature of companies that engage in
drilling, vary substantially with water depths and well locations. To further illustrate the
dierent reasons for the variations in rig rental rates, we split the data into sub-samples,
based on water depths. In columns (III), (IV), and (V) of Table 2, we report our results
for rigs that operate in depths lower than 100 feet, between 100 and 1000 feet, and depths
greater than 1000 feet, respectively. The results for low and medium water depths are not
very dierent from our main results. However, for deep water, the results are in sharp
contrast.
Firms that operate in deep water are large public rms that face little or no nancial
frictions. As a result, such rms are mostly indierent to marginal changes in price volatil-
ities. Further, rigs that operate in deep water, Semisubs and Drillships, are signicantly
more expensive equipment to rent than shallow water rigs. Therefore, the lessor rms that
lease deep water rigs are also, on average, large public rms.
15 These large lessor rms do
not exhibit much sensitivity to changes in cost of capital because the daily rig rental rates
do not change in response to changes in treasury bill rates. However, given the large amount
of initial investment necessary to operate in deep water areas, future expectations about oil
prices matter a great deal for rental rates. Among the three depth categories, the slope of
futures prices is positive and most signicant for the rigs that operate in deep water. post1
is an indicator variable for contracts signed between January 1, 2004 and April 20, 2010,
and post2 is an indicator variable for contracts signed after April 20, 2010. Having these
two variables in the regression controls for time trends in rig rental rates. The negative
coecients for both of these variables indicate that rig rental rates declined in these periods
compared to the base group, which consists of contracts signed prior to 2004.
To further substantiate our results, we recreate the results presented in Table 2 by using
15 Well over 63% of the rig owners in the deep water category were publicly listed rms. Compare this
percent with only 30% for leased rigs in the below 100 feet category.
16
ovx, an alternate measure of the market's expected volatility. By using ovx, we lose nearly
half of our sample and therefore, also lose considerable statistical power. Nevertheless,
despite using ovx, we nd that the results remain qualitatively similar to those we presented
earlier.
16
We also perform a falsication test to provide robustness to our results. To each rental
contract, we randomly assign a volatility number. We pick this volatility number from the
empirical distribution of ovx and vix. After randomizing the associated volatility, we should
not expect to nd any relationship between such volatility and rig rental rates. We run the
same specication as in equation (2). Consistent with our expectations, we do not nd any
are an endogenous variable. To further illustrate this point, we use the exogenous event of
the British Petroleum (BP Deepwater Horizon) oil spill that occurred on April 20, 2010 to
On April 20, 2010, Deepwater Horizon, a Semisub rig, exploded while drilling a well in
the Gulf of Mexico region. Although the company nally capped the well, by the time it
completed the process, the explosion had resulted in the largest oshore oil spill in U.S.
history. This catastrophic accident led to substantial loss of human life and threatened the
ecological balance in all of the southern states that surround the Gulf of Mexico. In response
to the incident, the U.S. Department of the Interior issued a moratorium on new deepwater
oil drilling permits in 500 feet of water or more. Soon after, Hornbeck Oshore Services
and several other companies that were engaged in oshore drilling challenged this order. A
16 Because no OVX data was available prior to 2007, our regression specication cannot have both post1
and post2. Therefore, we included only post2 in the specication. The coecient on post2 now shows the
marginal change in the rental rates after April 20, 2010, compared to earlier dates.
17
Federal judge, Judge Martin Feldman, overturned the moratorium, and an appelate court
subsequently upheld his ruling. The Interior Secretary then issued a second ban in June,
The Deepwater Horizon explosion and the subsequent litigation activity created a great
deal of uncertainty about future drilling activities and their associated cash ows. We predict
that as a response to these events, companies would lower rig rental rates for deep water rigs
to induce manufacturers to drill. Because the moratorium did not apply to depths lower than
500 feet, we should not expect to see any eect on the rental rates for such rigs. However,
in an eort to use the available capacity, we might nd that some rms move to drilling in
shallower water (less that 500 feet) and therefore, demand shallow water rigs. In these cases,
this substitution could raise the rental rates of the shallow water rigs.
where post is an indicator variable that takes the value one if the contract signing date is
in the post-event period, and zero otherwise. deepwater is also an indicator variable that
takes the value one if the water depth at the well location is greater than 500 feet, and
zero otherwise. P ost ∗ deepwater is the interaction of the two indicator variables mentioned
above. The above specication includes all of the covariates we used earlier in Table 2.
The dierence-in-dierence estimator of the treatment eect, (β3 ), is more ecient in the
17 Additional details about Hornbeck Oshore Services LLC v. Salazar are available at
https://en.wikipedia.org/wiki/Hornbeck_Oshore_Services_LLC_v._Salazar
18 See Roberts and Whited (2013) for a discussion of dierence-in-dierence estimators and their eciency
with additional exogenous controls.
18
captures the marginal eect of the event on the rig rental rates after we control for any
inherent dierences between the regions (shallow or deep), and for any systematic changes
in the economic conditions over time. A negative β3 would be consistent with our previous
results.
Table 3 presents the values for the dierence-in-dierence estimator from the estimating
equation (3). For column (I), April 20, 2010 to April 19, 2011 represents the post-event
period. The coecient on deepwater is positive, conrming that rigs that work in deeper
water have higher rental rates than shallower water rigs. Most importantly, the coecient
on the interaction term, as predicted, is negative and highly statistically signicant. This
nding suggests that the rig rental rates endogenously adjusted to accommodate for the
Amidst all of the legal controversy, on October 13, 2010, the Interior Secretary nally
lifted the moratorium on deep water oil drilling. Strict new rules, including one-on-one
worker training, accompanied this relaxation. On account of the new compliance require-
ments and because of increased scrutiny in the permit review process, companies encountered
substantial delays in obtaining permits. The delays were such that even two months after
the Interior Secretary lifted the ban, no new permits were awarded.
19 Overall, in the after-
math of the oil spill, a great deal of uncertainty surrounded deep water drilling and experts
expected this uncertainty to continue well into early 2012. In column (II) of Table 3, we
redene our post-period to be two years, ranging from April 20, 2010 to April 19, 2012. We
dene the pre-period, in both columns (I) and (II), as the two years from April 20, 2008 to
April 19, 2010. The results of column (II) are consistent with our earlier results from the
one-year horizon.
In column III of Table 3, we provide the results from a falsication/placebo test. Instead
19
of the actual date of the BP oil spill, we assign April 20, 2009 as the exogenous event date.
The result in column III does not show the same negative coecient for interact as in
columns I and II. Further, we also provide bootstrap results we obtain from applying the
in each iteration, we pick a random date between January 1, 2004 and April 19, 2009. We
avoid the period before January 2004 because the number of contracts was pretty sparse.
Then, for each selected date, we dene the post-event period as the one year period after
such a date. We run the same specication as in Table 3 and collect the coecient on
interact. Because we used these randomly picked dates, our prediction is that we should
not nd any signicance on the interact variable. The mean of the point estimates over
the 500 iterations is 0.220. The 5th and the 95th percentiles of the distribution are -0.009
and 0.509, respectively, and the inverse quantile function of zero is 0.072. Collectively, these
gures imply that close to 93 percent of the estimates were positive, which validates that
pre-existing dierences in the rig rental rates did not drive our results. Overall, these results
rms should consider the value of continuing to hold options and, perhaps, delay investments.
In this context, we should naturally test the response of drilling activity to changes in price
volatility. However, an important assumption underlies the expected relation between these
two variables. In particular, this relation assumes investment costs are constant and inelastic.
Above, we have demonstrated that considerable variation exists in rig rental rates and that
they dynamically adjust to the macro-economic environment. Therefore, our thesis is that
in such an environment, drilling activity and price volatility might not be negatively related.
20
In addition to having contract level details, our dataset also provides month-level informa-
tion on the utilization rates of the dierent rig types. In other words, it gives us information
about the percent of available rigs for GOM exploration that producers are actually using.
The utilization rates tell us a great deal about the drilling activity in each month. Figure 5
plots the time series of the utilization rates for the two most-used rig types in the GOM.
20
The average utilization rates are extremely high, in excess of 80 percent, and show limited
variation.
21 Table 4 provides results from an OLS regression, where the (log) rig utilization
rate is the dependent variable. Utilization rates are very sticky; therefore, we also include a
lagged utilization rate in the specication. Additionally, we include the average rental rate
of the specic rig type in the previous month as one of the explanatory variables. We lag
other regressors by three months to account for the delay between the decision to drill and
the actual commencement of drilling. Columns (I) and (II) of Table 4 use dierent measures
of price volatility. However, in neither of these cases can we reject the null hypothesis that
These results are in sharp contrast to the ndings of Kellogg (2014). The dataset in
Kellogg (2014) is for onshore drilling activity and, importantly, it lacks information on the
amount of investment needed for the projects he studies. In other words, it does not have
data on the cost of drilling. In fact, the inherent dierences between the onshore and oshore
drilling industries could also be an important reason for the dierences in the results. First,
20 We have made no adjustments for seasonality or any kind of smoothening to the raw utilization rates.
21 We chose to keep our model simple and focus on full utilization at date 0. Because most of the literature
stresses the waiting to drill eect, we wanted to provide a simple model where all of the volatility eect goes
through the rig rate channel and drilling activity is unaected. Of course, in reality, increases in volatility
are likely to result in a lower utilization rate as well as lower rig rates in some markets. To match the
utilization data, we will need rigs that have dierent qualities the low quality rigs require high labor costs
and are thus taken out of service when oil prices are low. Including this feature in the model would add
further complications that would make it less tractable. In reality, rigs may not be utilized for a short period
because they are being transported from one location to another. Another plausible consideration is that
80% utilization may be close to full utilization, in the same sense as 5% unemployment is considered full
employment.
21
because the oshore rig rental costs are extremely high, both in absolute dollar terms and
also in terms of their proportions of total costs, negotiating these rates is a binding constraint
on companies' investment decisions. Second, in the short run, the supply of oshore rigs is
xed. Commissioning a new rig happens rarely; besides, building new rigs is a long process.
Additionally, oshore rigs are massive, making moving them from one place to another very
expensive. Therefore, their geographical mobility is restricted to a very small area. Given
these constraints on the supply of rigs, companies sign contracts to rent them several months
before the actual start day of drilling. Typically, these contracts also have severe penalties
for cancellation. As a result, the rate of drilling cannot easily adjust to changes in volatility.
Third, the barriers to entering the onshore rig rental market are substantially lower than
those of the oshore market, resulting in an excess supply of onshore rigs. Because of
this situation, the rental market for onshore rigs is extremely competitive, leaving very low
margins for any further price drops. Finally, the opportunity costs of keeping expensive
oshore rigs idle are prohibitive and encourage the rig owners to enter the bargaining game.
Overall, the above results are consistent with our view that supply side dynamics matter a
causal eect on the rental rates of rigs. Ideally, we would like to compare the contracts that
companies make in times of high volatility to those they make during other times. However, a
potential selection bias confounds our eorts to identify the causal eect by simply comparing
sample means. Furthermore, the fundamental characteristics of the contracts that companies
make during times of high volatility and normal volatility might be dierent, making the
We overcome this problem by comparing the rental rates of contracts that companies
22
make during a high volatility regime (the treatment group) with those of matched samples
of contracts made in a normal volatility regime (the control group). For the purposes of
these tests, we label the contracts as treated if companies entered into them while volatility
was in the highest quintile (above 80%) of the distribution. All other contracts are the control
group.
22 We use the contract characteristics and other exogenous macro variables to match
the treated and control groups. We use the nearest-neighbor, as well as the optimal match
algorithm.
23
We match the contracts on age of the rig (age), construction cost of the rig (construct),
depth at the location of the well (locdepth), six month moving average of Brent Crude price
(brentavg ), spot price of Brent (spot), yield of the treasury bill (treasury ), and rig type.
We include an indicator variable for contracts signed between January 1, 2004 and April 20,
2010 (post1), and also an indicator variable for contracts signed after April 20, 2010 (post2).
to treatment and control groups are random (unconfoundedness). Further, in the matching
process, we enforce an exact match regarding the rig type and for the period in which the
company signs the contract (post1 and post2). Figure 4 is a love plot that provides the
extent of balance between the two groups. Readers can see that after we complete the
matching process, the two groups are very similar in the observed dimensions. Following the
matching process, we proceed to compute the overall average treatment eect (AT E ). This
estimate points to the causal eect of price volatility on rig rental rates. We compute AT E
1 1
P P
as
N i (Yi (1) − Ŷi (0))+ M j (Ŷj (1) − Yj (0)), where N and M are the number of treated and
22 A simple t-test for the dierences in mean of the two sample shows that the mean for the treated group
is 0.164 lower than for the control group. Because we are using log rental rates, this ndign implies that the
rental rates in a high volatility regime are, on an average, 16.4% lower than than in regular times.
23 The nearest-neighbor matching method is a greedy matching algorithm, this algorithm chooses the
closest control match for each treated unit individually, without trying to minimize a global distance measure.
In contrast, optimal matching nds the matched samples with the smallest average absolute distance across
all of the matched pairs.
23
controls in the matched sample, and Ŷj (0) and Ŷj (1) are the imputed potential outcomes for
each observation j under the counterfactual condition.
Table 5 reports the AT E for the dierent matching methods that we used. The rst row
of Table 5 refers to the results we obtained by using the nearest-neighbor matching method.
We use the inverse of the variance matrix as the weighting scheme for the covariates.
24
Companies made 374 contracts during the times of high volatility. In our sample, this
number refers to contracts companies made in the months where the VIX was 23.6 percent
or higher. Before undergoing any matching procedure, the control sample had 1431 contracts.
In the second specication, we impose stricter matching requirements than those imposed
previously. We stipulate that all matches are equal to or within 0.25 standard deviations
of each covariate. This restriction obviously reduces the matched number of contracts. The
rst two rows report the AT E after adjusting for bias (see Abadie and Imbens (2012))
and report the correct Abadie-Imbens standard error. Both of the specications lead to a
negative coecient that is also highly statistically signicant. The point estimates for AT E
are -0.047 and -0.042, respectively. This negative sign on the coecient is consistent with
our prediction regarding volatility and conrms that it has a negative causal eect on the
4. Final remarks
The exhaustibility of natural resources, like crude oil, brings the inter-temporal decision
to extract the resource to the core of a rm's investment decision. McDonald and Siegel
(1986) show that because the developer of an unexploited oil eld has discretion over the
timing of her opportunity to invest, this waiting option creates value that is central to the
cost-benet decision. In addition, the capital expenditure for developing an oshore oil eld
24 In unreported results, we also use Mahalanobis distance as the covariate weighting scheme. The results
are not qualitatively dierent than those reported here.
24
is almost entirely irreversible because the scrap value of these projects is close to zero. This
irreversibility feature and the timing option inherent in exhaustible resource investments has
made the real options approach standard in the literature (for example Dixit and Pindyck,
Our evidence suggests an important change in the way that oil and gas rms should ap-
proach their investment decisions. The rental costs of oshore drilling equipment constitute
a major share of the costs of developing new wells. Consequently, rental rates may aect the
decisions of oil and gas companies to either act on particular investment opportunities or to
wait for better market conditions. We nd signicant variation in rig rental rates. An impor-
tant conclusion to draw from this evidence is that standard real options techniques, which
do not consider variations in investment costs, are likely to produce sub-optimal investment
decisions. To address this issue, researchers should build and test real option models that
25
Table 1: Summary of the Data
This table provides the summary of our data. Panel A provides details about the contract parameters and about the
other economic variables we used in this article. Geographical region represents the dierent parts of the world where the
drilling activity occurred. Contract length is the amount of time for which the rig is leased. Location water depth is the
actual depth of water where the oshore drilling activity is going to take place. Age is the time, in years, since the rig was
contructed. AAAyield is the yield on AAA rated US corporate bonds provided by Bank of America Merrill Lynch. Panel
B reports the dierent types of rigs that are leased out, the number of contracts for each of these rig types, and a mean for
dierent variables relating to the rigs. Daily rental rate is the rent that is charged by the rig owner for every day that the
rig is rented out. Max WD is the techincally specied maximum water depth in which each of the rigs can operate. The
dollar denominated construction cost of the rig is provided in the nal column. Panel C provides the summary statistics
of the daily rental rate distribution for each of the dierent rig types.
26
Table 2: Explaining the Variation in the Daily Rental Rates.
This table shows the results from the pooled OLS regression. The dependent variable, in all of the columns, is the natural
log of the daily rental rate. The variable volatility is the log of the six month moving average of CBOE Volatility Index
computed from the S&P 500 stock index option prices (VIX); spot is the log of the six-month moving average of the spot
price of Brent Crude; f utures is the log of the six-month moving average of the slope of the futures contract on Brent
Crude expiring in 12 months; treasury is log(1+yield), where the yield is that of the treasury bill; age is the log of the
time, in years, since the rig was contructed; construct is the log of the dollar denominated contruction cost of the rig;
aaapremium is the log of the dierence between yield on AAA rated US corporate bonds and yield on a treasury bill;
locdepth is the log of water depth at the location of the well; post1 is an indicator variable for contracts signed between
January 1, 2004 and April 20, 2010; post2 is an indicator variable for contracts signed after April 20, 2010. All of the
specications include rig-type xed eect. The standard errors are clustered by rig managers, by rig owners, and by each
calendar month. The signicance levels are denoted by *, **, and *** and indicate whether the results are statistically
dierent from zero at the 10%, 5%, and 1% signicance levels, respectively.
27
Table 3: Endogenous Rental Rates.
This table provides the results from a dierence-in-dierence analysis. The dependent variable is the natural log of the
daily rental rate. The variable volatility is the log of the six-month moving average of the CBOE Volatility Index computed
from the S&P 500 stock index option prices (VIX); spot is the log of the six-month moving average of the spot price of
Brent Crude; f utures is the log of the six-month moving average of the slope of the futures contract on Brent Crude
expiring in 12 months; treasury is log(1+yield), where the yield is that of the treasury bill; age is the log of the time, in
years, since the rig was contructed; construct is the log of the dollar denominated contruction cost of the rig; aaapremium
is the log of the dierence between yield on AAA rated US corporate bonds and yield on a Treasury Bill. post is a dummy
variable that takes the value one when the contract signing date is in the post-event period, and zero if the contract signing
date is in the pre-event period. For column (I), the post-event period is one year from April 20, 2010 and for column (II)
post-event period is two years from April 20, 2010. The pre-event period is the two years prior to April 20, 2010. For the
falsication test in column (III), the post-event period is one year after April 20, 2009. deepwater is a dummy variable
that takes the value one when the water depth at the well is greater than 500 feet, and takes the value zero otherwise.
post ∗ deepwater is the product of two dummy variables, post and deepwater. All of the specications include rig-type
specic xed eects. The standard errors are clustered in two dimensions, by each calendar month, and by rig owner. The
signicance levels are denoted by *, **, and ***, and indicate whether the results are statistically dierent from zero at
the 10%, 5%, and 1% signicance levels, respectively.
28
Table 4: Rate of Drilling
This table shows the results from the pooled OLS regression where the log monthly utilization rates is the dependent
variable. In column 1, volatility is ovx, which is the log of the six-month moving average of the CBOE Crude Oil ETF
Volatility Index. In column II, volatility is vix, which is the log of the six-month moving average of CBOE Volatility Index
computed from the S&P 500 stock index option prices. The variable laggedU til is the log utilization rate of the previous
month; spot is the log of the six-month moving average of the spot price of Brent Crude; f utures is the log of the six-month
moving average of the slope of the futures contract on Brent Crude expiring in 12 months; treasury is log(1+yield), where
the yield is that of the treasury bill; aaapremium is the log of the dierence between yield on AAA rated US corporate
bonds and yield on a treasury bill; avgrate is the log of the average rig-specic rental rate in the previous month; post1
is an indicator variable for the period between January 1, 2004 and April 30, 2010; post2 is an indicator variable for
months after May 1, 2010. ovx, vix, spot, f utures, treasury, and aaapremium are all lagged by three months. All of the
specications include rig-type xed eects. The standard errors are clustered by rig type and by each calendar quarter.
The signicance levels are denoted by *, **, and ***, and indicate whether the results are statistically dierent from zero
at the 10%, 5%, and 1% signicance levels, respectively.
(I) (II)
volatility 0.060 0.001
(0.048) (0.029)
post1 -0.018
(0.024)
29
Table 5: Causal Eect of Volatility on Rental Rates
This table provides estimates of the mean dierence between the rental rates on contracts made during "high" volatility
times and rental rates on contracts made at all other times. Contracts made during "high" volatility (treated group) are
the contracts that were made in the months when the volatility was in the top most quintile (above 80%) of the volatility
distribution. We match the two groups, treatment and control, in many dimensions. The observable covariates used are
age, which is the age of the rig in years; construct, which is the dollar value of constructing the rig; locDepth, which is the
water depth at the site of the drill. aaapremium is the log of the dierence between yield on AAA rated US corporate
bonds and yield on a treasury bill; spot is the log of the spot price of Brent Crude; f utures is the log slope of the futures
contract on Brent Crude expiring in 12 months; treasury is log(1+yield), where the yield is that of the treasury bill; post1
is an indicator variable for contracts signed between January 1, 2004 and April 20, 2010; post2 is an indicator variable for
contracts signed after April 20, 2010; and rigtype is the type of rig. We use two dierent matching algorithms, Nearest
Neighbor and Optimal match. Results are provided for both matching schemes. We report the estimate of the overall
average treatment eect (AT E ) along with the estimates of Abadie-Imbens standard error in paranthesis below. Row
2 reports the AT E after the treatment and control group covariates have been constrained to be within 0.25 standard
deviations of one another. The signicance levels are denoted by *, **, and ***, and indicate whether the results are
statistically dierent from zero at the 10%, 5%, and 1% signicance levels, respectively.
30
Figure 1: Model Assumptions and Timeline
31
Figure 2: The Oil Producers' Payos with Endogenous Rig Rates
This gure illustrates the producers' payos at time 1. The payos at time 1 for producer δ is dened as P ayof f1δ =
P1 − (R1 + C) δ . Here, R1 is the time 1 value of the rig rental rate variable. We use the equilibrium condition equating
the quantity of rigs demanded and supplied to determine R1 ; we can then express R1 as a function of the oil price at time
1, P1 .
32
Figure 3: Equilibrium in the Market for Oil Rigs and Oil Price Volatility
This gure illustrates equilibrium at t=0 in the market for oil rigs. The gure also shows the eect of changing the
exogenous oil price volatility variable σ on the endogenous rig rate variable R0 .
33
Figure 4: Covariate Balance
The graph below plots the covariate balance between the control group and the treated group. The treated group contains
the contracts that companies made in the months when the volatility was in the top most quintile of the volatility
distribution (above 80%). The black points are the pre-matching dierences between the covariates. The red points are
the dierences in the covariates in the matched sample.
34
Figure 5: Utilization Rates
The graph below plots the utilization rates of two main rig types in our data. The utilization rate represents the number
of rigs in use during a given month as a percentage of total available rigs.
35
AppendixA. Proofs
At t = 1, agents observe the realized price of oil (P1 ). At this time, each one of the
producers δ(0, ∆] can spend xδ1 = (R1 + C)δ and extract her reserve. Or, she can leave her
δ
if her reserve is economically viable P1 − x1 > 0 . She will not rent a rig to extract her
δ
reserve if the current price of oil is below her extraction cost, P1 − x1 < 0 . There is then
a marginal producer with extraction cost xδ1m who is indierent between extracting the oil
and leaving it underground. For this marginal producer, P1 − xδ1m = 0. Hence, this marginal
P1
producer can be uniquely identied as, δm = R1 +C
.
The marginal cost producer at t=1 depends on the value of the rig rental rate variable
R1 . If the rig rental cost variable (R1 ) is suciently high, δm < ∆ and not all producers will
extract their reserve. With a decrease in R1 it will be optimal for more producers to extract
Given the choice of the marginal producer, δm , we derive the optimal quantity of rigs
δm N P1
demanded at t = 1 , ND1 (R1 ; P1 , σ). In particular, ND1 (R1 ; P1 , σ) = ∆
= N
∆ R1 +C
. If
δm < ∆, the number of rigs used is N δm /∆ , and this number of rigs is less than the number
of rigs available (N ). In this case, the economy-wide production (δm Q/∆) is below capacity
δm N
(Q). A decrease in R1 raises the value of δm and hence this also raises ND1 (R1 ; P1 , σ) = ∆
.
Therefore, other things equal, the quantity of rigs demanded, ND1 (R1 ; P1 , σ), is decreasing
they do not rent their rig they earn nothing. Therefore, the optimal quantity of rigs supplied
so NS1 =ND1 . 1
BecauseNS = N, this equilibrium condition requires that all rigs are rented,
36
will extract her reserve and the economy produces at capacity. Furthermore, denoting the
P1
equilibrium rig rental rate variable at t=1 as R1e , we obtain that R1e = ∆
−C . However,
P1
+
because rig rates are positive, we note that R1e = ∆
−C .
As we show next, producer δ can use the time 1 value of the rig rate variable, R1e , to
initial period, each producer δε(0, ∆] computes her time 1 extraction costs to be, xδ1 =
P1
+
(R1e + C)δ = δ ∆
−C + Cδ or:
δ P1
∆
P1 > C∆
xδ1 =
Cδ
P1 < C∆
Thus, at t = 0, each producer δ understands that her extraction costs at t=1 will increase
with the price of oil realized at that time (P1 ). Morover, extraction costs (xδ1 ) increase with
0
the depth of the reserve(δ); xδ1 > xδ1 for δ 0 > δ.
At t = 0, producer δ also understands that because she has the option but not the
+
obligation to extract her reserve at period 1, her time 1 payo is P1 − xδ1 or:
δ
(1 − ∆
)P1 P1 > C∆
+
P1 − xδ1 = P1 − Cδ Cδ < P1 < C∆
0
P1 < Cδ
Producer δ 's reserve at t = 0 is equivalent to a long position in a call with an exercise price
δ
of Cδ and a short position in
∆
calls with an exercise priceC∆. Producer δ can compute the
value of her underground reserve (Vδ ) at t = 0 by comparison with a schedule of option prices;
these option prices follow from the model parameters. For example, call options for the pur-
37
chase of oil at t=1 with exercise price E are traded at a price CE . Therefore,Vδ = CCδ −
δ
C . In addition, with
∆ C∆
P1 ∼ N (µ, σ 2 ) the reserve value Vδ can be computed directly as fol-
lows:
(P0 −e−r Cδ) (P0 −e−r Cδ)
r
r
P0 (∆−δ) C∆−P
Vδ = + erf √ 0e − erf Cδ−P
√ 0e + k.
2 2∆ 2σ 2 2σ
´x
!
P 2 e2r −2Cδer P0 +Cδ 2 P 2 e2r −2C∆er P0 +C∆2
e−t
2 √ − r+ 0 2 − r+ 0 2
Here, erf (x) =
0
√
π
dt and k= 2π e 2σ
−e 2σ
+
P0 e2r −2C∆P0 er +C∆2
σ(∆−δ) − r+ P0 (∆−δ)
2σ 2
√
∆ 2π
e − 2∆
.
At t = 0, each of the producers δ(0, ∆] can spend xδ0 = (R0 + C)δ and extract their
reserve or leave it underground. A particular producer δ(0, ∆] makes this decision at this
time as follows:
There is a marginal producer with extraction cost xδ0m who is indierent between extracting
the oil and leaving it underground. For this marginal producer, P0 − xδ0m = Vδm and hence
P0 −Vδ
this marginal producer can be uniquely identied as, δm = R0 +C
.
The marginal cost producer at t=0 depends on the value of the rig rental cost variable
R0 . If the rig rental cost variable (R0 ) is suciently high, δm < ∆ and not all producers will
extract their reserve. With a decrease in R0 it will be optimal for more producers to extract
Given the choice of the marginal producer, δm , we derive the optimal quantity of rigs
δm N P0 −Vδ
demanded at t = 0 , ND (R0 ; P0 , σ). In particular, ND (R0 ; P0 , σ) = ∆
= N
∆ R0 +C
. A
portion of the available supply of rigs (N ) can remain unused at time 1. If δm < ∆, the
38
number of rigs used is N δm /∆ , and this number of rigs is less than the number of rigs
available (N ). In this case, the economy-wide production (δm Q/∆) is below capacity (Q).
δm N
A decrease in R0 raises the value of δm and hence this also raises ND (R0 ; P0 , σ) = ∆
.
Therefore, other things equal, the quantity of rigs demanded, ND (R0 ; P0 , σ), is decreasing
they do not rent their rig they earn nothing. Therefore, the optimal quantity of rigs supplied
so NS =ND . BecauseNS = N, this equilibrium condition requires that all rigs are rented,
N P0 −Vδ
NS =ND = N . In particular, .
∆ R0 +C
= N. In this case, the highest cost producer,∆, will
extract her reserve and the economy produces at capacity (Q). Denoting the equilibrium rig
P0− V∆
rental rate parameter at t=0 as R0e , we obtain that R0e = ∆
−C . We abstract from
corner solutions where R0e can be negative, by assuming that the current price of oil (P0 ) is
P0 −V∆
acterized by the time 0 rig rental rate of the marginal producer, R0e (P0 , σ) = ∆
− C.
In this equilibrium, the highest cost rig operator will clear the market for rigs, δm = ∆.
The economy in this case fully utilizes the available supply of rigs (N ) and produces at full
capacity (Q).
An increase in the oil price volatility to σ0 raises Vδ . Each reserve increases in value
because the likelihood that it will be economically feasible at the terminal date increases
P0 −Vδ
with volatility. Recall that the marginal producer is determined as, δm = R0 +C
. When
a rise in volatility increases the option value of the reserves, Vδ , then the value of δm will
decline. In this case, it will no longer be optimal for high cost producers to extract their
39
δm N
reserves. As a result, the optimal quantity of rigs demanded at t = 0 , ND (R0 ; P0 , σ) = ∆
,
will also decline. This willl be reected in a downward shift in the demand for oil rigs,
ND (R0 ; P0 , σ), and a lower equlibrium rig rental rate, R0e (P0 , σ 0 ). The decline in rental rate
will be sucient so that economy again fully utilizes the available supply of rigs (N ) and
produces at full capacity (Q). Analogously, an increase in the spot price of oil (P0 ) makes it
attractive for producers to extract their oil, this raises the demand for rigs and their rental
rates.
References
Abadie, A., and G. W. Imbens. 2012. Bias-corrected matching estimators for average treat-
Brennan, M. J., and E. S. Schwartz. 1985. Evaluating natural resource investments. Journal
of Business pp. 135157.
Kellogg, R. 2014. The eect of uncertainty on investment: evidence from Texas Oil Drilling.
McDonald, R., and D. Siegel. 1986. The Value of Waiting to Invest. The Quarterly Journal
of Economics 101:70727.
Titman, S. 1985. Urban land prices under uncertainty. The American Economic Review
75:505514.
40