How Flexible is US Shale Oil? Evidence From North Dakota∗ Hilde C. Bjørnland†

Frode Martin Nordvik‡

Maximilian Rohrer§

January 20, 2016

PRELIMINARY VERSION PLEASE DO NOT DISTRIBUTE

Abstract The steep decline in the price of oil since 2014 has spurred renewed interest in the dynamics of oil production, in particular, to what extent the producers in unconventional oil pools will respond to the negative price incentives by reducing oil production and drilling activity. Yet, little is known of how unconventional oil producers respond to price signals. Using a novel well-level monthly production data set covering more than 16,000 crude oil wells in North Dakota, we find crude oil production from unconventional oil wells to respond four times as strongly to oil price changes as conventional wells, with the strongest output response at oil prices between 39-50 USD/bbls. Also, contrary to conventional wells, the monthly number of fracked wells reacts strongly to changes in the slope of the future curve, but not to spot prices. For a 10 percent increase in the 12 month calendar spread, fracking producers immediately hold back 75,000 barrels of crude oil, effectively storing it underground, awaiting the higher expected price. JEL-codes: Q35, Q33, Q30, Q40, Keywords: Crude oil prices, oil extraction, US oil shale boom

∗

The authors would like to thank seminar and conference participants at BI Norwegian Business School and the 9th International Conference on Computational and Financial Econometrics (CFE 2015) in London for valuable comments. This paper is part of the research activities at the Centre for Applied Macro and Petroleum economics (CAMP) at the BI Norwegian Business School. The usual disclaimers apply. The views expressed in this paper are those of the authors and do not necessarily reflect the views of Norges Bank. † BI Norwegian Business School and Norges Bank. Email: [email protected] ‡ Corresponding author: BI Norwegian Business School. Email: [email protected] § BI Norwegian Business School. Email: [email protected]

1

1

Introduction

In the second half of 2014, the price of oil plunged, partly as a result of the robust supply growth from US shale oil. These events have spurred renewed interest in the dynamics of oil production, in particular, to what extent producers in unconventional oil pools will respond to the negative price incentives by reducing oil production and drilling activity. Previous studies addressing this issue typically find oil production to be price inelastic in the short run, see e.g. Anderson et al. (2014), Griffin and Teece (1986), Jones (1990), Ramcharran (2002) Hogan (1989) and Dahl and Y¨ ucel (1991).1 Common to these studies, however, is the fact that they analyse aggregate output responses from conventional oil producers. Very little (if anything) is known about the dynamics of unconventional crude oil production. Although production is still one a small scale (centered on the US), the use of hydraulic fracturing technology is likely to spread to other oil producing countries, potentially making unconventional oil a much larger share of total production than it is today. Knowledge of price elasticity in both conventional and unconventional oil production is therefore important. We aim to fill this gap by examining the short-term response of unconventional2 and conventional crude oil wells to short-term price changes. Using a novel and rich monthly panel data set from 1974 to 2015, covering more than 16,000 oil wells, both conventional and unconventional in the North Dakota oil patch, we are able to study the response of crude oil wells along two margins. First, we ask if entry 3 of new wells into production is sensitive to spot price and/or to the slope of the futures curve, distinguishing between conventional and unconventional oil wells. Second, we ask if production from existing wells respond significantly to innovations in the spot price and/or to the slope of the futures curve, and compare the two types of well technology. Figure 1 motivates such a study. The figure displays aggregate supply changes in North Dakota oil production together with the changes in oil prices, focusing on two specific events; the oil price decline in 1986 (left panel) and the oil price decline in 2014. The price decline are of roughly similar magnitudes, but the production environment was very different. In 1986, North Dakota oil production was entirely conventional, while in 2014 it was almost entirely unconventional. The figure shows that while aggregate production showed no visible response to the price collapse in 1986, production in 2014 had the biggest drop in a single month since 2001 following the most recent oil price collapse. The use of microeconomic data to infer about the price elasticity of aggregate output has several advantages. First, by constructing a rich panel data set, we can eliminate any 1

One exception is Rao (2011), which find production in oil wells in California to respond significantly to tax changes. 2 For the sake of variation, we will refer to both fracking wells and unconventional wells interchangeably, but we always mean hydraulically fractured horizontal wells. 3 For unconventional wells, this is also the time of fracking.

2

Figure 1. Supply responses, 1986 and 2014 oil price crash

Note: The panels show the monthly percentage change in aggregate crude oil production in North Dakota plotted together with oil price change, at two different time periods. The oil price changes are lagged one month.

potential negative bias of aggregation over well’s production rates when estimating the empirical model. Second, by using panel data we are able to explore the cross-sectional variation in for instance well type, age, location or other characteristics of interest, and we can investigate the potential heterogeneity in production behaviour across technologies (conventional or unconventional). Lastly, having a large cross-section in a panel is beneficial for statistical inference when analyzing a relatively short time-period as we do here. Our results supports the stylized facts pertained in Figure 1. That is, using the full panel of monthly well-level production, we document that the production response to an oil price change from unconventional oil wells is more than four times as strong as the response from conventional oil wells. The average monthly supply elasticity of unconventional wells to changes in the spot price of oil is 0.19, while it is only 0.04 for conventional wells. We find that the supply elasticity from unconventional wells is the strongest for oil prices between 39-50 USD/bbls4 . This indicates an average marginal cost in this range for the majority of shale oil wells in North Dakota. Furthermore, using the monthly number of wells entering into production from 2007 until 2015, we show that unconventional oil producers react strongly to changes in the slope of the futures curve when deciding on the timing of fracking, whilst they do not react to spot prices as such. Shale oil producers reduce the number of wells they frack when the futures curve is upward sloping, and increase the nr. of wells when it is downward sloping. A natural interpretation is that fracking producers effectively use the ground as crude oil 4

This refers to the WTI-price, but in the analysis we use the first-purchase oil price in North Dakota which sells at a 5 USD discount, on average.

3

storage when market conditions favor storing the oil and selling it at higher prices in the futures market. For a 10 percent increase in the WTO 12-month calendar spread, producers immediately reduce fracking equivalent to a withdrawal of 75,000 barrels of first-month crude oil production from the market, and effectively store it in the ground awaiting a higher expected price. However, we find no such relationships for conventional oil producers. What makes unconventional oil producers different? We document a key difference between conventional and unconventional wells, namely that the first-month production rate from the average fracked well is almost four times as high compared to a conventional well. Moreover, the decline rate of fracked wells is on average about five times higher than for conventional wells. This production front-loading increases the incentive for unconventional producers to optimize the timing of fracking. If a well can be fracked and the quantity sold at a higher expected price on the futures market one year from now, producers can profit from holding back on fracking if the expected price gain is larger than the cost of storing the oil in the ground plus interest payments. Conversely, if the futures curve is downward sloping, there is a substantial gain from fracking the well immediately rather than later. In contrast, an average conventional well has a production rate much more thinly spread over its lifetime, reducing the incentive for producers to optimize the timing of putting the well into production. This is the first study to document that shale oil producers can effectively use short-term storage below ground to profit when futures curves slope upwards. Another characteristic that potentially sets unconventional oil producing wells apart, is the cost structure. From a microeconomic perspective, only an upward sloping shape of the marginal cost curve will lead to any response of supply to changes in the oil price. As Anderson et al. (2014) note, the marginal cost of conventional oil production is very low and rarely enters the range of oil prices observed in the data. Therefore, they note, most conventional wells are produced at maximum until they encounter a capacity constraint determined by geology. Fracking oil producers, however, arguably have a higher marginal cost of extraction due to the process of hydraulic fracturing that is necessary in order to produce oil from the shale reservoirs. The finding that supply from unconventional oil wells are highly elastic at prices between 39-50 USD/bbl suggest that these prices are in the range of marginal costs for unconventional oil producers. The finding that there are large and substantial differences in the price elasticity of crude oil production between conventional and unconventional oil wells has important policy implications. For instance, when designing tax policies, policymakers should take into account that producers adjust differently to price sensitive news (such as a tax on revenue). Hence, any tax policy that targets operators at the well level will have to take into account that fracking producers are more sensitive to price changes than conventional well operators. In a similar way, a decision of whether or not to lift the existing export ban for crude oil in the Unites States could influence production decisions differently for

4

conventional and conventional oil producers, at least in the short run. Our study informs on the magnitude of such potential short run supply responses. Our findings have also implications for how one can identify and analyse supply and demand disturbances in oil market models. An assumption typically made to identify shocks in the oil market models is that production is price inelastic in the short run, see e.g. Kilian (2009). While this may be correct for conventional oil, the difference in price elasticity between conventional and unconventional oil producers suggests that the reported supply responses in existing oil market models may be on the lower end, in particular as unconventional oil takes up an increasing share of total oil production.5 . The paper proceeds as follows: Section 2 describes the data environment, while Section 3 explains the empirical model. In Section 4 we present the results, while concluding remarks follow in Section 5

2

Oil and gas in North Dakota

Our focus is to use microeconomic data on the production side to infer about macroeconomic relationships; i.e., price elasticity of aggregate output. The use of such data offers several advantages. First, it allows us to eliminate the potential negative bias of aggregation over well’s production rates when estimating the empirical model. Second, by using microeconomic data we are able to explore the cross-sectional variation in for instance well type, age, location or other characteristics of interest, and we can investigate the potential heterogeneity in production behaviour across technologies. Lastly, having a large cross-section in a panel is beneficial for statistical inference when analyzing a relatively short time-period. In the following subsections, we describe in detail the data environment.

2.1

Conventional and unconventional geology and drilling

A conventional oil and gas bearing rock is typically porous, such as sandstone or washed out limestone. When crude oil forms in a permeable6 rock, gas will naturally gather at the top of the reservoir and the crude oil will be trapped in the porous rock underneath it. At the very bottom, there is water. The crude oil is pressurized in the pores of the formation rock, so that when a well is drilled into the reservoir, the hydrostatic formation pressure and the permeability of the formation rock naturally drives the hydrocarbons out of the rock and up into the well. Conventional oil well technology can produce effectively 5

In 2013, the Energy Information Administration (EIA) released a world shale oil and gas reserve assessment that showed 32 countries outside the US have substantial reserves, with Russia and China among the countries with the largest reserves, see https://www.eia.gov/analysis/studies/worldshalegas/ A range of countries such as UK, Colombia, Australia, Scotland and Botswana have now initiated licence rounds for unconventional oil reservoirs. 6 Permeability is a geological term to describe how easily oil flows naturally through rock

5

from these types of reservoirs, aided by the natural pressure and the permeability in the reservoir. In practice, this involves drilling a conventional vertical well straight into the reservoir and producing the oil that flows by itself into the well7 . The earliest oil and gas exploration in North Dakota began before 1910, but was hampered by primitive technology. The first commercial oil discovery was made in 1951 in the Williston basin, and it was followed by several other discoveries in the 1950s. Discoveries continued to be made in the 1960s and 1970s, and production gradually increased until it peaked in the 1980s. In terms of wells drilled, this first conventional ”oil boom” peaked in 1981, with 834 wells drilled in one year. In 1999, only 34 wells were drilled, the lowest number of wells since oil drilling began. In terms of production rates, this first conventional oil boom petered out towards the end of the 1990s and beginning of the 2000s. In April 2004, the production reached a minimum at around 75,000 barrels produced pr. day. The conventional oil boom was a vertical drilling boom. Widespread horizontal drilling did not emerge until the recent unconventional oil boom, as it was too costly and technically difficult to apply it on a large scale until then. Contrastingly to a conventional oil reservoir, when crude oil is trapped in a rock formation that has zero permeability, the natural pressure in the reservoir formation is not enough to make the oil flow into the well once a well is drilled - because the oil is trapped in small pockets inside the shale rock formation. Such reservoirs require unconventional wells in order to be depleted, and this is where fracking and horizontal drilling technology is applied. These types of reservoirs, often called tight oil reservoirs, require stimulation once the well has been drilled, most commonly in the US is hydraulic fracturing, or fracking, combined with horizontal drilling. While the conventional oil boom in North Dakota was a vertical drilling boom, the shale oil boom in North Dakota is a horizontal drilling boom (see e.g. Miller et al. (2008) for details). The main reason for this is the thickness of the middle Bakken layer. It is at most around 150 feet thick, which is fairly thin for an oil producing zone (see e.g. Meissner (1984)), making it inefficient to produce by vertical wells. Entering a thin geological layer is more efficient when the well can enter into the layer horizontally, and more of the well-bore can be exposed to the producing zone. Since horizontal drilling has become less costly the past 10-15 years, the combination of this technology together with hydraulic fracturing, made operators able to extract resources from the middle Bakken layer at a cost below the prevailing oil prices. The unconventional oil boom in North Dakota relates mainly to oil fields discovered and produced in the Bakken formation.8 The Bakken is a rock formation occupying an area of about 520,000 km2 , about the size of Spain. According to the US Geological Survey 7

See Oil and Gas Production Handbook. An Introduction to Oil and Gas Production, H. Devold (ABB), 2009 for further details 8 The oil discovery that marked the beginning of the unconventional oil boom in North Dakota was the Parshall oil field in the middle Bakken layer, discovered in 2006. The field was developed by horizontal wells combined with hydraulic fracturing, and by 2013 the field had more than 200 producing wells.

6

(USGS), it is the largest continuous oil resource in the lower 48 states. In april 2013, the USGS estimated that the amount of oil that could be economically recovered from the Bakken would be at 7,4 billion barrels. The formation stretches out over regions such as North Dakota, Montana, Saskatchewan and Manitoba. Even though the conventional oil boom in North Dakota took place within the Bakken area, the geological rock layer known as Middle Bakken was too costly and technically difficult to produce from before the 2000s.

2.2

Well technology and front-loading

For both conventional vertical wells and horizontal fracking wells, the process of drilling a well starts with a a rig drilling a vertical well into the ground, to depths up to 10,000 feet. Conventional vertical wells will stop drilling at this point and produce from the vertical well. However, the unconventional oil wells will be drilled further, but with an angle, or a so-called ”bend” to it. This initiates the horizontal part of the well, which can extend up to 10,000 feet in the horizontal direction. After the well is encased with a metal pipe, a so-called casing, throughout the entire well, to keep the formation wall from caving into the well-bore and to be able to control the flow of oil to the well-head, it can be completed and made ready for production. At this point, activity stops - the drilling rig and its crew leaves the site, and the well must be fracked by a fracking crew in order to actually start producing. The fracking crew arrives with a so-called ”missile”, a manifold that is placed on the well-head to pump the fracking water and chemicals down into the well at high pressure. Several large trucks containing the horsepower to maintain the high pressure are needed in this process, and in addition a blender truck that blends the chemicals, sand and water together. The blender truck then sends the water mix to the missile, which sends the mix down the well at very high pressure in order to crack the tight oil reservoir formation open. When the formation is cracked open, the oil starts to flow up to the well head at high pressure. The fracking process can take as little as 2-3 days to be completed, after which the well will produce for months or years9 . As can be seen in Figure 2 from the data set we construct (details about the data environment are given in Section 2.3), fracked well production rates are very high initially, and the production profile heavily front-loaded relative to conventional well production. A fracked well in North Dakota has an average production rate of 10,500 barrels pr. day in the first full month of production, while for conventional wells the average first month production rate is about 3,000 barrels pr. day. Furthermore, the average monthly decline rate is 1,6 percent for unconventional wells and just 0,03 percent for conventional wells. This front-loading of production from fracked wells means a higher fraction of the total 9

For a short introduction, see http://www.greeleytribune.com/news/9558384-113/drilling-oil-equipmentwellbore

7

Figure 2. Production profiles of horizontal and vertically drilled wells

Note: Cross sectional average of the production profile of wells in the sample plotted against well age measured in months, divided into average production for horizontal and vertical wells.

output is produced in the first months of the well’s lifetime, compared to a conventional wells, which have output more thinly spread over the lifetime of the well.

2.3

Data

The data set is provided by the North Dakota Industrial Commission (NDIC), Oil and Gas Division and gives production figures on a monthly frequency for 16,639 crude oil wells in North Dakota. The time period is from January 1974 to April 2015. Table 3, reported in the appendix for brevity, displays the main summary statistics. As can be seen from the table, about two thirds of all wells are unconventional. While the average age of a horizontal well is 43 months, the average age of a conventional well in the sample is 217 months. This of course underlines the need to control for well age in the regressions. As a consequence of this, average production is higher for unconventional wells than conventional wells. The variability in production is not very different when examining the coefficients of variation. If anything, the variability of conventional well production seem slightly larger on average. Figure 3 details all of North Dakota’s crude oil wells generated from the well data set of each well’s coordinates within the state. As is visible from the map, it is the western and north-western part of North Dakota where the Bakken field is located that is the most resource rich part of the state. The black dots represent unconventional wells, and the blue represent conventional wells. While the conventional well are spread around in small clusters, the unconventional wells are almost all located within the Bakken pool. We distinguish between conventional and unconventional (fracking) wells according

8

Figure 3. North Dakota: Individual Wells

Note: Map of North Dakota plotting each individual oil well’s location since 1952 according to latitude and longitude

to whether the wells are vertical or horizontal wells. Hence, the number of horizontal wells drilled in North Dakota is an approximation of the number of fracked wells there. As described above, this is plausible because the shale oil boom in North Dakota is a horizontal drilling boom, while the conventional oil boom that peaked in the 1980s was a vertical drilling boom. In the Bakken field, 98.4 percent of all wells are horizontal wells, and have been drilled in the past 10 years. Figure 4 shows the complete history of monthly well entries in North Dakota dating back to the very first well entering into production in 1952. As can be seen, the recent boom in the number of wells entering into production surpasses anything previously seen in North Dakota. Furthermore, the boom is entirely caused by horizontal drilling and fracturing. Of the 16,639 wells in our sample, roughly two thirds are horizontally drilled wells. Essentially all of the horizontal wells have been drilled the past 5-10 years, and are hydraulically fracked wells. We limit our sample to crude oil wells that have recorded production each month. This implies that shut-in production, when operators shut down oil production entirely is not part of our sample. In the elasticity estimates, we wish to limit our inference to wells that are in production, and the dynamics of oil producers when wells are actually in operation. Since it is well known that some well operators shut in their wells when oil prices are low, we believe that excluding these wells from our sample will, if anything, bring the elasticity measure below its real value. Finally, the total number of operator firms in the sample is 533 companies, reflecting 9

Figure 4. New Wells over time.

Note: The number of new wells entered into production on a monthly frequency, separated into well technology type.

a fragmented industry structure with many firms operating only a few wells.

2.4

Oil price

The oil price series that we use is the monthly spot price of crude oil sold in North Dakota, provided by the Energy Information Administration (EIA). This so-called first-purchase price is the price of oil at that the oil producer receives for the crude oil transaction at the well-head, when oil is first removed from the lease where it has been produced. The series is the monthly average of all contract prices for crude oil sold from leases in North Dakota. The contract price must be reported and submitted to the EIA by firms that assume ownership of crude oil as it leaves the lease on which it was produced. The reporting period is within 30 days. Hence, firms observe their own lease-level price that they obtain for a given volume delivered to the buyer in real time. However, the average for all producers within the month is observed with a time lag of about one month. The North Dakota first-purchase price sells at a discount relative to the WTI benchmark price, mainly due to higher transportation cost for North Dakota crudes. The average dollar discount relative to the WTI in our time period is about 5 USD. We use the difference between the futures price for a contract with delivery at time

10

t+12 and the front-month contract10 to measure the future curve slope, or the expected annual price change. If this meaure is negative, the price is expected to be lower than today one year from now. If it is positive, the price is expected to increase during the next 12 months. The price series is the West Texas Intermediate (WTI) crude oil delivered in Cushing, Oklahoma. The use of futures prices to measure firm’s price expectations is done for several reasons. First, NYMEX futures are traded liquidly at the time horizons considered here, and with many risk-neutral traders, the futures price should equal the expected future spot price. Also, oil well operators are believed to use the futures market to make price projections. The futures prices included in this paper are real prices, so that the annual expected price change reflect real rather than nominal changes. Throughout the analysis, we use the real prices, deflated by the US CPI. The reason for using real prices is to allow producers to react to changes in the real value of oil, and for the price elasticity of supply partly to depend on the price level of oil at any time, relative to other goods.

3

Empirical model

Below we describe in detail the two main empirical models that we estimate. Model 1 is a panel data model with prices and output in first-differences. The main motivation for using such a model is to estimate the average percentage response of monthly welllevel production to a percentage change in the crude oil price. Model 2 is a time series regression model of the aggregate monthly number of unconventional wells entering into production as a response to two main price indices, the expected annual price change and the spot price.

3.1

Elasticity of producing wells

In model 1, we regress the monthly percentage change in barrels produced pr. well in North Dakota on the current and lagged percentage change in the first-purchase price of crude oil in North Dakota.11 Since a well production rate, and to some extent its output variance, varies over the well’s lifetime, the age of the well, measured in months since production start, is a key characteristic. We therefore control for the age of the well. Since well production variation as a function of age could exhibit non-linearities, we also add a quadratic age term. We also control for seasonal variation in production by adding monthly dummies. North Dakota experiences extremely cold temperatures during 10 11

The first unexpired future contract price We make the assumption that the oil price is exogenous to the monthly number wells put into production at time t in North Dakota. This is plausible, since there is about 533 operators in our sample, and no single operator is able to exert any market power in the global market for crude oil. North Dakota produces just over one million barrels of crude oil per day, which is about the production level of Colombia, or roughly one per cent of the global annual output of oil.

11

winter time and it is natural to assume that activity might be hampered slightly by such climate conditions. We also include a linear time trend. We further extend the analysis by controlling for all well fixed effects. Such fixed effects can be unobserved geological characteristics of the reservoir, production costs, well quality, depth and operator firm characteristics. We also want to check for any level-dependence in the price elasticity of supply by adding an interaction between the oil price change and the price level. Furthermore, we add the expected change in the price of oil, based on the difference between the 12 month futures contract and the front month contract. We estimate the price elasticity of supply using the following regression: ∆ Qit = ∆ Qit−1 + β1 ∆Pt−n + β2 ∆Pt × Pt + β3 Et [∆ P(t+12)−t ] + Xit γ + vit

(1)

where ∆Qit is the percentage change in quantity of oil produced on a monthly frequency for well i at time t. The lagged dependent variable is added to allow for well production changes to be autocorrelated. The dynamic panel bias which may be present for panels with small t is less of a concern in our setting since t is equal to 378 in our baseline estimation. This should be sufficient for the right-hands side variables to be asymptotically uncorrelated. ∆Pt is the percentage change in the real price of crude oil at the well-head in period t, when n is equal to zero, or in period t − 1, when n is equal to one. ∆Pt × Pt is the interaction between the percentage change in the price of oil at time t and the level of the oil price. Et [∆P(t+12)−t ] is the expected change in the oil price the next 12 months, as measured by difference between the 12 month futures contract price and the front month contract. X is a set of time-varying controls, and vit is the error term clustered at the well level. Estimates of β1 captures the average well output response at time t to a percentage change in the real price of oil in the current and preceding month. β3 captures any oil price level-dependency in the elasticity of output at the well level, β4 measures the response of output to the expected change in the oil price 12 month ahead, as contained by the WTI 12 month calendar spread. When well operators decide to produce more (or less), this may be reflected in the quantity of output actually observed. However, the short-term production rate of oil is subject to geological constraints as well as other operational constraints that may distort the mapping from the producers’ output decision to observed output. We therefore repeat the baseline regressions, using only number of days a well is producing pr. month as a dependent variable. The production decision of how many days to produce each month is more of a choice variable for the well operator than the volume that may be produced at any instant, which is subject to the mentioned constraints. Hence, if unconventional well operators are more price-elastic than conventional produces, it should be reflected also in their production choices measured by the number of days the wells are producing each month. 12

3.1.1

Level dependence

Producers’ willingness or ability to respond to price changes may be non-linear in price levels. This could for instance be related to capacity constraints. In times when the price level is high, it may be that more of the production capacity is utilized, rig rental rates and fracking crews are expensive, and labour markets are tight, relative to when the price level is low. It could also be that if the price level enters the range of marginal extraction costs, it may be in such price intervals that one would expect to see the highest price responsiveness among producers. Hence, we ask if it is reasonable to assume oil producers will react similarly to an increase in the oil price when the price is, say, 100 USD/bbl as when the price is, say, 50 USD/bbl. To address this, we investigate if there are any non-linearities in the elasticity of supply on the well-level. The first exercise is to add an interaction term as stated in model 1, where the effect of a price change on output depend explicitly on the oil price level. The second exercise is to divide the historical monthly oil price level into 10 quantiles according to oil price level, and estimate the regression model 1 for each quantile, and compare the coefficients for each quantile. If the supply elasticity is linear with respect to oil price level, the coefficients should be roughly the same across quantiles. If there are non-linearities in the response of output depending on the level of the oil price, we should se different coefficients for different quantiles.

3.2

Well entry

In order to examine the entry of wells for the two types of oil producers, we investigate two linear regression plots of the expected annual change in the price of oil and the monthly nr. of wells entering into production in North Dakota (see Figure 4) for the period from 2008 until 2015, for conventional and unconventional producers. If there is a negative relationship, the producers react to a positive expected annual real price appreciation by reducing the nr. of wells they put online today. If there is no slope, producers do not react to changes in the future curve slope. In addition, we estimate a second model, model 2, which is a time-series regression where we control for seasonality and time trends. We report the results from the estimation of model 2 in Table 2. We estimate the variation in the monthly number of fracked wells via the following baseline regression in logs: Ft = β1 Ft−1 + β2 Et [∆ P(t+j)−t ] + us + vt + t

(2)

where Ft is the log of total monthly nr. of fracked wells in North Dakota, Et [∆ P(t+j)−t ] is, as before, the expected rate of price change j periods ahead, us are monthly dummies to control for any seasonal variation in fracking intensity, vt is a linear time trend to take into account the upward trend in the number of fracked wells since 2007. t is the error 13

term, which is heteroscedasticity robust. The time period is 91 months from 2007:M10 until 2015:M4.

4

Results [To be completed]

In Section 4.1, we report the results from the estimation of model 1, where we estimated the response of monthly well production to a percentage change in the price of oil, lagged one month. In addition, we show graphically the results from estimating model 1 on 10 quantiles of the oil price. In Section 4.2, we report the results from the estimation of model 2.

4.1

Supply elasticity

Table 1 displays the results from analysing the response of producing wells to an oil price change. Estimates are from the estimation of model 1 on a sample of 15,272 producing wells from 1974 to 2015. Column 1 displays the average response of all wells, independent of well technology, to a percentage change in the spot price of oil at time t. The overall combined elasticity of output for all 15,272 wells in the panel is 0.23. The elasticity point estimate is level-dependent, and a 10 percent increase in the oil price reduces supply elasticity by 0.3 percentage points. The expected annual oil price change is also significant and negative, indicating that producers reduce production when the future curve has a positive slope. The economic significance of the estimate is however small, and stays small throughout. Column 2 shows the response when limiting the sample to the 4,084 conventional wells producing in this period. While the coefficient on the lagged percentage change in the spot price is 0.04, the coefficient on the contemporaneous oil price change is not significant. The other estimators stay roughly the same. The elasticity estimate of 0.04 is small in economic terms, and in line with previous studies that find a low supply elasticity for conventional oil. Note that the elasticity point estimate does not depend on the level of the oil price for conventional oil wells. Column 3 shows the same regression, only limiting the sample to the 10,718 unconventional wells in the sample. The elasticity for this type of well is 0.18, which is much higher and significantly different from the elasticity of conventional wells. For the unconventional wells, there is considerable level-dependence. For a 10 percent increase the oil price, the short-term elasticity drops by 0.5 percent. Unconventional wells also respond more strongly to changes in the future curve, with a coefficient of -0.05, which means they reduce production when the expected price one year from now is higher than the current price. Note also that both type of wells do not respond significantly to contemporaneous spot price changes, indicating that it is operationally unfeasible to react to price innovations in period t.

14

Table 1. Price Elasticity of Oil Production (1) All wells

(2) Conv. wells

(3) Unconv. wells

(4) Unconv. wells

-0.33∗∗∗ (-156.50)

-0.35∗∗∗ (-108.25)

-0.32∗∗∗ (-107.77)

-0.33∗∗∗ (-107.25)

∆ Log spot pricet

0.14∗∗∗ (4.35)

0.02 (0.47)

0.10 (1.21)

-.13 (1.20)

∆ Log spot pricet−1

0.09∗∗∗ (16.72)

0.04∗∗∗ (5.70)

0.18∗∗∗ (17.30)

0.19∗∗∗ (15.49)

∆ Log spot price X spot pricet

-0.03∗∗∗ (-10.73)

0.00 (0.15)

-0.05∗ (-2.30)

0.00 (0.11)

Age of well

0.00∗∗∗ (43.20)

0.00∗∗∗ (16.58)

0.00∗∗∗ (32.43)

0.00∗∗∗ (15.27)

Age of well sqd

-0.00∗∗∗ (-30.06)

-0.00∗∗∗ (-12.56)

-0.00∗∗∗ (-22.09)

-0.00 (-14.71)

Annual exp. price change

-0.03∗∗∗ (-10.73)

-0.03∗∗∗ (-6.68)

-0.05∗∗∗ (-7.96)

-0.01 (-1.53)

-0.02∗∗∗ (-13.15) yes no yes 1,269,577 15,272 1.1974-4.2015

-0.04∗∗∗ (-14.84) yes no yes 759,634 4,084 1.1974-4.2015

-0.06∗∗∗ (-9.79) yes no yes 429,793 10,718 1.1974-4.2015

-0.07∗∗∗ (-23.45) yes yes yes 364,987 10,476 1.2005-4.2015

Dep. variable: ∆ Log Productiont

∆ Log Productiont−1

Constant Time trend Well Fixed effects Seasonal dummies N Nr. of wells Time period

Note: The dependent variable is the percentage change in the monthly production for well i. ∆ Log Fr. month price is the percentage change in the front month contract WTI oil price. Age of well is the age of each well, measured in months since first production month. Standard errors are heteroscedasticity robust and clustered on well level. t-statistics in parentheses. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

15

Column 4 shows the result when we control for well fixed effects as well, and the elasticity point estimate declines slightly to 0.6, but is still four times higher than the conventional well elasticity estimate. Figure 5 graphs estimates for short-term supply elasticity for 10 quantiles of the North Dakota first-purchase (wellhead) price, for unconventional wells. The upper left panel shows the elasticity estimate for unconventional wells in terms the percentage change in barrels of oil produced pr. month. There is a peak elasticity of supply of around 0.4 in the price interval between 34-45 USD/bbl. Since there is an average price discount of 5 USD in North Dakota relative to the WTI benchmark, this price level roughly corresponds to a WTI price range between 39-50 USD/bbl. For unconventional wells, therefore, it is natural to interpret this interval as a ”break-even” price range, or alternatively, the range of prices which cross the marginal cost curve for the average unconventional oil well. The interpretation of this result is that for a 10 percent increase in the price of oil at these levels, the production will increase by 4 percent in the following month. There is a second peak elasticity in the price interval between 79-85 USD/bbl, which corresponds to 84-90 USD/bbl in WTI prices. This suggests that a large share of wells in North Dakota are marginally profitable at such price levels. In total, the upper left panel indicates that there are two main break-even price levels in North Dakota and that the average well has a marginal cost in the range between 39-50 USD/bbl. The right panel in Figure 5 shows the elasticity of barrels produced pr. month to a positive change in the expected annual oil price change, as measure by the WTI 21 month ahead future price spread. The negative spikes in the price interval between 34-45 USD/bbl indicates that in this price range, producers react strongly to changes in the slope of the future curve by reducing output, possibly storing it until the higher expected price materialize. The magnitude of the elasticity is about the same as for the spot price, and in the same price intervals. This supports the notion of this interval as a marginal cost level for the average well in North Dakota. The lower left panel in Figure 5 shows the response of the nr. of days a well produces oil each month to changes in the spot price of oil. As can be seen, the operators decision to produce on more days pr. month in response to an oil price increase corresponds almost exactly to the output elasticity estimate of about 0.4, and at the same price interval between 34-45 USD/bbl. This estimate means that for a 10 percent increase in the price, the operator on average decides to increase the nr. of producing days pr.month by 4 percent, which means that if your well is currently producing 20 days pr. month you will keep the oil flowing one more day following such an oil price increase. The lower right panel shows the response of days producing to a change in the expected annual oil price change, as measure by the WTI 21 month ahead future price spread. A positive increase in the expected annual oil price change, means that producers on average reduce the number of days that they produce. The negative peak point estimate is 0.2 in the same price interval as before, between 34-45 USD/bbl.

16

Figure 5. Elasticity at different oil price levels

Note: The upper two panels show price elasticity of supply at different levels of the spot price (left), and to changes in the future price curve (right). The bottom two panels show the same relationships, only for the nr. of days a well is producing pr. month. Quantiles, 1: 11-21 USD, 2: 21-25 USD, 3: 25-28 USD, 4: 28-34 USD, 5: 34-45 USD, 6: 45-56 USD, 7: 57-68 USD, 8: 69-79 USD, 9: 79-85 USD, 10:

4.2

Well entry

We now turn to examine the response in the monthly number of wells entering into production in North Dakota to expected annual change in the price of oil for the period from 2008 until 2015, for conventional and unconventional producers. If there is a negative relationship, the producers react to a positive expected annual real price appreciation by reducing the nr. of wells they put online today. If there is no slope, producers do not react to changes in the future curve slope. Figure 6 shows linear regression plots of the WTI 12-month calendar spread, or the expected annual rate of price change, on a monthly frequency on the y-axis and the total number of wells that are completed each month on the x-axis. Being above zero on the yaxis means the futures curve slope is positive, and below zero means the slope is negative, and the price is expected to fall. The negative slope of the regression line in the right panel supports the notion that unconventional well operators on average reduce the number of wells they frack in months when the slope of the futures curve is positive. Conversely, they increase fracking in months when the slope is negative. The slope coefficient of the simple regression is -73, which means that if the monthly expected price one year ahead is 10 percent higher than today, fracking producers immediately reduce fracking by 7 wells, on average. In terms of average first-month production,

17

Figure 6. Fracked wells and expected annual oil price change

Note: The figure to the left shows the correlation between expected change in the price of crude oil (spread between 12 month future contract and the first month future contract) and the number of conventional oil wells entering production each month (corr.coef = 0.0. The figure to the right shows the same relationship only for monthly nr. of fracked wells. (corr.coef.=-0.6). The time period is from 2008-2015. this means that about 75,000 barrels are immediately withheld from the market, and remains in underground storage until the expected higher prices materialize. The left panel shows the same relationship for conventional well completions, and as can be seen, there appears to be no such dynamics in the production from conventional oil wells. Table 2 shows the results from the estimation of model 2 from Section 3.2, which confirms that the negative relationship between the futures curve slope and the monthly number of fracked wells is significant also when we control for time trend, seasonality and autocorrelation in the monthly series of fracked wells. In column 1, a regression of log nr. of fracked wells on the expected 12 month price change yields a negative and significant coefficient, supporting the main result. Column 2 shows the same regression only using the expected 6 month price change. The negative effect is even larger in magnitude at this future price horizon. The coefficient of -0.56 implies that if the 6 month future price lies 20 percent above the front month price, fracking producers immediately reduce the number of wells they frack by nearly 12 percent. In column 3, the spot price of oil is added as a control, but the coefficient is not significant at conventional levels, and the effect of the 6 month expected oil price change becomes even larger. Hence, fracking producers do not appear to react to spot prices, but rather to the slope of the futures curve. Column 4 runs the regression, but with the log nr. of wells fracked in first-differences as a robustness 18

Table 2. Fracked wells regression: (1) log(wells)

(2) log(wells)

(3) log(wells)

(4) ∆log(wells)

Lag nr. of wells

0.77∗∗∗ (8.97)

0.77∗∗∗ (8.93)

0.76∗∗∗ (8.71)

Expected 12 month price change

-0.35∗∗ (-2.21) -0.56∗∗ (-2.06)

-0.63∗∗ (-2.17)

-0.64∗∗ (-2.23)

∆ Log spot pricet

0.40 (0.28)

0.89 (0.65)

∆ Log spot price X spot pricet

-0.15 (-0.41)

-0.25 (-0.70)

Dep. variable: Nr. of fracked wells

Expected 6 month price change

Time trend

0.00 (1.62)

0.00 (1.59)

0.00 (1.62)

-0.00∗ (-1.82)

Constant

-1.255 (-1.13) Yes 0.93 91

-1.240 (-1.11) Yes 0.93 91

-1.25 (-1.10) Yes 0.93 91

0.95∗ (1.80) Yes 0.24 91

Seasonal dummies R squared N

Note: This table shows the results of a regression of the log of monthly nr. of fracked wells on expected price changes at different time horizons. The future spreads are based on the percentage difference between the 6 and 12 month contract and the front month contract. The price series are deflated and reflect real quantities. Time period is from 2007:M10 to 2015:M4. t statistics are in parentheses. ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

check. The results do not change. The overall interpretation of these results is that producers react to intertemporal relative prices, not spot prices. Furthermore, producers reduce fracking, effectively storing oil, when the future price is above the current spot price. This type of flexibility of when to start producing is not seen for conventional wells. It supports the notion that frontloading increases the economic incentive to optimize the timing of when to frack wells.

19

5

Conclusion

This paper has quantified large and substantial differences in the elasticity of crude oil output between conventional and unconventional oil wells in the North Dakota oil patch. First, we examined the response of operators along the extensive margin, i.e. how new wells entering the production phase react to changes in the expected annual change in the price of crude oil. We found a strong negative relationship between the entry of new fracking wells and positive changes in price expectations, but a zero correlation for conventional oil wells, supporting the view that fracking well operators are more priceresponsive than conventional well operators. When the price one year ahead is expected to be higher than todays price, operators of fracking wells are systematically reducing the number of wells they put on line. Conversely, when the price one year ahead is expected to be lower than todays price, operators increase the number of wells they put into production. Along the intensive margin, we estimated the response of producing wells to a change in the price of crude oil in the previous month. We found that the unconventional oil wells have an output elasticity with respect to price of 0.19, while the response from conventional oil wells is zero. The estimate for unconventional oil wells is well above existing elasticities in the literature.

20

6

Appendix

Table 3. Summary Statistics: This table provides summary statistics for the time series of monthly data, the cross-section of wells, and individual observations on wellmonth level. Mean well production is in barrels pr. month Time Series Start End Mean Mean Mean Mean Mean

well production change well production front month price price change 12-month spread

1974-01-01 2015-05-01 1560.6 -0.018 62.87 -0.0044 -0.038

Nr month

Min

Max

0 9.16

77838 9.59

497

St.dev St.dev St.dev St.dev St.dev

2993.9 0.55 23.09 0.08 0.1

Nr. of unconv. wells Nr. of conv. wells

11,112 5,013

-0.40 -0.33

0.36 0.59

Min

Max

4085.2 0 0.68 -9.17 49 0

62421.66 9.59 206

Total cross Section Nr. of wells Nr. of pools Observations

16,639 37 1,566,794

Unconventional wells Nr. of wells Mean well production (bls) Mean change well production Mean well age (months)

11,112 3294.3 St.dev -0.029 St.dev 43 St.dev

Conventional wells Nr. of wells Mean well production (bls) Mean change well production Mean well age (months)

5,013 774.65 St.dev -0.014 St.dev 217 St.dev

21

1838.12 0 77052.58 0.49 -7.64 7.61 152 0 668

References Anderson, S. T., R. Kellogg, and S. W. Salant (2014). Hotelling under pressure. Technical report, National Bureau of Economic Research. Dahl, C. and M. Y¨ ucel (1991). Testing alternative hypotheses of oil producer behavior. The Energy Journal , 117–138. Griffin, J. M. and D. J. Teece (1986). Opec behavior and world oil prices. Hogan, W. W. (1989). World oil price projections: a sensitivity analysis. Harvard University, Energy and Environmental Policy Center, John F. Kennedy School of Government. Jones, C. T. (1990). Opec behaviour under falling prices: implications for cartel stability. The Energy Journal , 117–129. Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply shocks in the crude oil market. The American economic review 99 (3), 1053–1069. Meissner, F. F. (1984). Petroleum geology of the bakken formation williston basin, north dakota and montana. Miller, B. A., J. M. Paneitz, M. J. Mullen, R. Meijs, K. M. Tunstall, M. Garcia, et al. (2008). The successful application of a compartmental completion technique used to isolate multiple hydraulic-fracture treatments in horizontal bakken shale wells in north dakota. In SPE Annual Technical Conference and Exhibition. Society of Petroleum Engineers. Ramcharran, H. (2002). Oil production responses to price changes: an empirical application of the competitive model to opec and non-opec countries. Energy economics 24 (2), 97–106. Rao, N. (2011). Taxation and the extraction of exhaustible resources: Evidence from california oil production.

22