The immediate outlook for US oil production has rarely been more important for the financial world, with traders scrutinising every scrap of data for signs of a sustained pull-back in output. It has also rarely been harder to predict.
Until late last year, a handful of energy analytics firms had honed the art of real-time oil production forecasts to a near science, running reams of information through complex models that account for everything from a well's production curve to weather patterns. With the price slump, however, these firms are struggling to keep up with the rapid pace of change.
"Things were much more straightforward at $100," says Bentek Energy analysis manager Anthony Starkey, whose Colorado-based firm publishes end-of-month production data, one of several closely monitored private reports.
Last week's US government data showing a marginal drop in weekly output was a case in point. Some thought it marked a turning point and crude prices rallied more than $1 right after the data, despite an otherwise bearish rise in crude stocks.
The problem is the figure is not what it might seem to be. Rather than a survey-based data like most other weekly government statistics, the number is an estimate based on weekly production in Alaska and US Energy Information Agency's month-old forecasts, according to the agency's methodology.
A comparison of EIA's weekly production figures for December and the monthly figure based on state data published two months later, underscores the risk of taking the weekly numbers at face value.
While the weekly numbers never exceeded 9.13 million barrels per day, the monthly number came in at 9.3 million.
The latest weekly estimate is due on April 8.
State well production statistics are the industry standard, but they vary widely in terms of timeliness and detail. Some states such as North Dakota and Texas release the information within about two months, others, such as Ohio do so once a year.
TIMELY SNAPSHOT
Energy analytics firms have adopted an array of techniques to overcome those limitations and offer markets a more timely snapshot of US oil production.
The firms are deploying cameras on pipelines, using natural gas flows to help predict crude output and relying on new algorithms that take into account production potential.
The techniques can work remarkably well when producers follow established patterns. Things get tricky though, when they scramble to keep up with tumbling crude prices that have fallen about 60 per cent since June 2014.
Drillinginfo Inc., of Austin, Texas, for example, has recently began producing its own monthly index for "new production capacity," representing the likely future output and based on a model that uses everything from daily rig information to well types and production curve data.
Given that most shale wells begin pumping only about five months after they were drilled, the index has effectively served as a pretty accurate five-month leading indicator, according to co-founder and chief executive Allen Gilmer.
"If you did a backward comparison, the index is within a per cent," he said.
But that time lag is getting longer and much less predictable, as more oil companies opt to leave new wells idle to await a recovery in prices, making calculating output harder. The share of wells that start pumping oil within five months has fallen to 72 per cent from 78 per cent a year ago, Gilmer says.
Gilmer says their data suggests US production will peak in May and then enter a five-month decline.
Forecasts more risky
The rise in the numbers of drilled but uncompleted wells, however, makes forecasts more risky. Many bigger energy companies have continued to drill new wells, but delayed the hydraulic fracturing process waiting for prices to rebound. In North Dakota there were 825 wells waiting for completion, up from 585 at the end of June, according to state data.
As the backlog grows, it is becoming much harder to tell how many of the wells will be left idle, or for how long, Bentek's Starkey says.
Bentek's view is that US production is near a turning point, but not quite there yet. "We'd be surprised to see it happening already," says Starkey.
PointLogic Energy, an analytics firm bought last year by energy industry publisher OPIS, offers national daily oil production estimates based on models that combine historical well production data and real-time natural gas pipeline flow information. Oil wells also produce natural gas, and increases or decreases in gas flows can be predictive of crude outputs.
"The old historical method that applied up to two years ago is slowly deviating," vice president John Uxer said. "We started out as an engineering model looking at the capability of what a given field can produce. All of a sudden there's a difference between that and what's happening in the market."
The company's model - which generates updates on a daily basis that are then compiled into weekly and monthly estimates - is showing a decline over recent weeks, but Uxer does not believe oil production has tapered off just yet.
Energy industry intelligence service Genscape uses cameras placed along pipelines to gauge energy use by the big pumps needed to force crude down the pipe, another source of real-time data that helps calculate production figures, said Jodi Quinnell, manager of crude analytics at Genscape.
The company expects production to peak at 9.5 million bpd in April - slightly above EIA's 9.44 million estimate - before heading down.