This jupyter notebook examines the relationship between the number of drilling rigs actively developing oil and the crude oil production level in the United States. While controlling for a deterministic time trend, I empirically verify the impact the rig count has on production. I find that the relationship between the rig count and the production level is linear when operators are expanding and growing their rig count: New rigs are needed to be put to work in order for production to grow. On the contrary, I find that the relationship between the rig count and the production level is non-linear when the rig count is plummeting: With a falling rig count operators can focus their drilling activity in the sweet spots, resulting in increased rig productivity.
In the past, the number of oil-oriented drilling rigs was conclusively indicative of oil production growth. However, technological advances have resulted in operators finding ways of increasing production without adding more rigs. This jupyter notebook examines if the rig count is still relevant: if you can’t trust the rig count report, what should you trust? By using the rotary rig count, I conduct two different forecasting models for the United States crude oil production level. The first model uses the rotary rig count to predict the U.S. production level 26 weeks into the future. The second model uses the rotary rig count and a time variable to predict the U.S. production level 26 weeks into the future.
Oil drilling rigs are used to identify geologic reservoirs and to create wells that allow the extraction of oil from those reservoirs. A drilling rig is a necessary evil for production to take place, making the U.S. rotary rig count a fundamental determinant of current and future production capabilities.
Figure 1 displays crude oil production and the rotary rig count over time for the United States. A casual review of the data suggests a predictable relationship between the number of drilling rigs and the production output; however, the magnitude and extent of this relationship is unclear. Domestic production generally lags the rig count and leads oil and gas employment.
Figure 2 is identical to figure 1, except that the U.S. rotary rig count is now shifted 26 weeks forward. There are several reasons that explain the presence and relevance of leads in drilling rigs. Companies bring new rigs online a few months before they expect to need the output in order to give themselves time to get them into full production. To bring a rig online is a process, and even when a rig is in production mode it isn’t necessarily running at full capacity. Hence, analysts and investors should expect an estimated lag time from drilling activity to new-well production. Examining the data, I observe a high correlation between the rig count and production with a 26 week lag. In some cases, where either data or industry reports suggest otherwise, a different lag may be used.
Figure 3 displays a scatterplot of the U.S. rotary rig count and the U.S. crude oil production level. The vertical axis is inverted for comparision with figure 4. The scatterplot suggest that there is no clear linear relationship between the number of drilling rigs and production. With that said, there appears to be a linear relationship between 2010-2015 and 2017-today.
Figure 4 shows the evolution of U.S. rig productivity and the number of drilling rigs over time. Rig productivity is defined as U.S. crude oil production devided by the rotary rig count. Examining figure 3 and figure 4, we can see that rig productivity is the endogenous factor that is contributing to the nonlinearity between the number of drilling rigs and production. Today, improvements in pad drilling (the ability to drill multiple wells from one rig) and increased efficiency or speed of drilling are examples of innovations that continue to help companies to increase rig productivity. Furthermore, better acerage quality contributes to a higher level of initial production which gives the appearance of increased rig productivity.
Between 2010-2014 the rig count was increasing with Shale operators expanding aggresivley. Operators were expanding into worse acerage quality and rig productivity was falling. Following the oil crash in late 2014, the operators decided to focus their drilling activity to the sweetspots, which caused rig productivity to surge. Over the last years, rig productivity has been stabilizing around similar levels to 2014, with most drilling activity being focused in the Permian.
Figures 5 shows the relationship between the number of drilling rigs and the crude oil production level between 2010 to mid 2015. Over the chosen time period, the rig count predicts U.S. production quite well.
Figures 6 shows the relationship between the number of drilling rigs and the crude oil production level between late 2016 until today. Over the chosen time period, the rig count predicts U.S. production quite well. Figures 6 is equivalent to forecasting model 2.
Figures 7 shows the realtionship between the number of drilling rigs and the crude oil production level while controlling for a time trend. The time trend is extracted from the U.S. crude oil production level by regressing the production level on a time variable. The idea is that the time trend will capture rig productivity which is changing over time. Note that when normalizing for a time trend, the horizontal axis loses its economical interpretation. Figures 7 is equivalent to forecasting model 1.
In this section, I present the two different regression models that are used to predict the U.S. crude oil production level. The models rest on the notion that the number of active drilling rigs determines the total output level. Model 1 will be using data between 2010 until today while Model 2 will be using data between late 2016 until today.
Model 1 is expressed as:
where PROD is the weekly U.S. crude oil production level at time t, RIGS are the number of rotary rigs that are activley drilling for oil and T is a time trend variable tracking rig productivity and technological advances.
Model 2 is expressed as:
where PROD is the weekly U.S. crude oil production level at time t and RIGS are the number of rotary rigs that are activley drilling for oil.
A database covering the rotary rig count is a prerequisite for this work. The Baker Hughes North American Rotary Rig Count is a weekly census of the number of drilling rigs actively exploring for or developing oil in the United States. A rotary rig rotates the drill pipe from surface to drill a new well (or sidetracking an existing one) to explore for, develop and produce oil. The Baker Hughes Rotary Rig count includes only those rigs that are significant consumers of oilfield services and supplies and does not include cable tool rigs, very small truck mounted rigs or rigs that can operate without a permit. Non-rotary rigs may be included in the count based on how they are employed. For example, coiled tubing and workover rigs employed in drilling new wells are included in the count. To be counted as active a rig must be on location and be drilling or 'turning to the right'. A rig is considered active from the moment the well is "spudded" until it reaches target depth. Rigs that are in transit from one location to another, rigging up or being used in non-drilling activities such as workovers, completions or production testing, are not counted as active. Since 1940 the highest weekly US rig count was 4,530 recorded on December 28, 1981. The lowest U.S. rig count was recorded in 2016.
For data covering the U.S. crude oil production level, I use the EIA Weekly Petroleum Status report. The weekly production data is published on Wednesday while the rotary rig count data is published on Friday. Figure 8 is displays the weekly U.S. crude oil production level between 2010 to late 2018. Figure 8 illustrates that the weekly production data contains outliers created by exogenous events like hurricanes etc.
In figure 9 the production level is smoothend to remove any short-run noise. The blue line represent the long-run trend, or structural component, that is used when implementing the regression.
Model 1 performance well both in-sample and out-of sample. The disadvantage with the model is that the time trend captures most of the price response.
An alternative to using the rig count to predict production is to use well completion data to predict production. This choice would eliminate the endogenity problem that arise from changes in rig productivity over time. Yet well data is not readily accessible and doesn't lead production with a couple of months. By tracking the rig count, investors can observe fundamental changes that will have an effect in production months later. That is, investors can use reliable data to develop production forecasts months into the future.
Figure 15 illustrate how the oil price affect the oilrig activity. Changes in oil prices leads changes in the rig count by up to one quarter. There are several reasons that explain the presence and relevance of lags in rig drilling. During periods of lower oil prices, oil companies initially revisit their resources that they reckon un-economic. There are also rig contracts and rigs rented for a number of years, which stand in the way of suddenly terminating drilling activity. The lags are also present during higher oil periods as it takes more time to acquire new leases/concessions, carry out seismic surveys, recruit workers, etc.