View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

VOL. 12, NO. 13 • NOVEMBER 2017


Real-time Data Inaccuracies Pose
Challenges to Gauging the Oil Market
by Justin J. Lee and Jesse Thompson


ABSTRACT: Initial estimates of
global oil market balance, or
the implied change in global
inventories, are frequently
used to identify supply
shortages or surpluses. These
have important implications
for future oil prices. Initial
inventory data undergo
various revisions, which may
contribute to inefficiencies in
oil pricing.


here is no direct measurement
of global crude oil inventory
outside of domestic stocks
estimates prepared by the
Organization for Economic Cooperation
and Development (OECD).1
The implied change in global petroleum inventories (or stocks) is often
derived from the difference between
estimated global daily production and
This is an indicator of the extent to
which the market anticipates production surplus or shortage. Often, initially
released production and consumption
data are revised. These changes frequently send conflicting signals of surplus or shortage, potentially contributing
to inefficient pricing.
Market participants use this and other data to assess profitability of futures
or equity. Moreover, they look closely at
consumption figures, which tend to be
initially less reliable than those depicting

Tracking World Inventory
Inventory changes signal whether
the market is in surplus or shortage. The
United States is one of the only countries
with large and frequently measured
petroleum stocks. Other members of
the OECD also provide measurements
of inventory changes that account for a

large portion of total global stocks.
By comparison, non-OECD countries,
particularly China, members of OPEC
and many emerging economies—where
global energy demand growth is centered—don’t provide such information.
Thus, market participants use periodto-period global implied changes in
world petroleum inventory as a proxy for
changes in actual global inventory.
The U.S. Energy Information
Administration (EIA) publishes the
Short-Term Energy Outlook, containing
monthly estimates of world petroleum
production and consumption.2 The
implied change in inventory can be calculated by taking the difference between
the two EIA series. For each month,
the U.S. agency releases its most recent
initial estimates, as well as revisions to
earlier data.
Initial estimates are a guide that market participants use to form or update
beliefs about current and future prices.
Understanding data revisions provides
context with which to interpret initial
estimates of market balance.
Such revisions mainly correct survey reporting errors by the EIA and the
International Energy Agency (IEA).
For example, initial estimates of
monthly U.S. production from the EIA
are drawn from a sample of more volatile
weekly data. A second round of rev-

Economic Letter
Tracking Revisions

sions comes from aggregated production
data compiled into monthly state-level
data. Finally, there are annual revisions
to historical data that can go back several years, as well as revisions based on
methodological changes.3



To track the revisions, the monthly
releases of world production and consumption and U.S. production and consumption from the EIA are collected.4
Because these datasets provide timely

Current Prices Consistent with Revised Changes in Inventory

Index, 2000:Q2,
implied change in inventory since 2000:Q2

Spot price of crude oil (in inverted scale),
nominal dollars per barrel




in inventory




in inventory



dated pricing











NOTES: Chart shows data ended June 2016, which are revised after at least 12 months. Brent crude oil price is computed
by averaging daily nominal price of dated Brent into quarterly frequency.
SOURCES: Federal Reserve Bank of Dallas; Federal Reserve Economic Data–Federal Reserve Bank of St. Louis;
International Energy Agency, Oil Market Report (July 2017); Energy Information Administration, Short-Term Energy
Outlook (July 2017); authors’ calculations.



Initial Estimates Unreliably Identify Imbalance

Implied monthly change in inventory,
million barrels per day*


Confidence band
Estimates–revised (12 months)


Initial Data, Wrong Signal











*Three-month centered moving average.
NOTE: The confidence band is calculated by measuring one standard deviation of revisions. (Revisions are defined as the
difference between initial estimates (blue line) and revised estimates after 12 months (red line).)
SOURCES: Federal Reserve Bank of Dallas; Energy Information Administration; authors’ calculations.


figures, it becomes possible in real time
to track revisions as more complete data
become available.5 Similarly, the IEA’s
monthly Oil Market Report data are
tracked through various revisions.6
Market participants, partly relying on
these sources of information, make decisions whether to buy or sell crude oil or
other petroleum-related equities. If the
initial estimates are subject to substantial
revisions or if the IEA and EIA data are
at odds, trades may be completed that
might not otherwise occur. This is especially true when participants are unaware
or uncertain of how to account for potential future revisions.
Inaccuracies in petroleum data,
therefore, may have real-time impacts on
the price of crude oil, motor fuels such
as gasoline and distillates. Market participants also compile information from
myriad other sources that sometimes
influence their analysis of unrevised
All else constant, the implied crude
oil inventory is indirectly related to the
price of crude oil. When the world produces more oil than it consumes, the
excess is stored—in inland oil tanks,
ships and pipelines. This is generally
supportive of lower petroleum prices.
The spot price of Brent crude oil, an
international benchmark, closely followed the evolution of implied global
stocks in real time, despite market participants seemingly being unaware of the
likely magnitude or direction of future
revisions (Chart 1).
After 12 months of revisions, both
data-releasing agencies’ implied stock
changes followed the same trend: The
EIA and IEA indicated that the implied
inventory in mid-June 2016 had risen
to around levels last seen in 2010 in the
aftermath of the global recession. The
nominal, or stated price, of crude oil
declined accordingly.

A data point can be revised years
after the initial estimate, but the magnitude of revisions beyond the initial
12 months is relatively small, and the
qualitative interpretation of the implied
inventory change is little altered.7
A confidence band of one standard
deviation—a measure of the dispersion

Economic Letter • Federal Reserve Bank of Dallas • November 2017

Economic Letter
of values around the mean—of revisions
helps assess whether the initial estimates
of market imbalance are substantive
(Chart 2).8 Historically, 17 percent of
revised implied change in global inventory fell outside of the confidence band,
sending false signals about the trajectory
of global inventory.
Typical revisions to production and
consumption estimates are small relative to the size of the global oil market
(roughly 97 million barrels consumed
daily) but large relative to the size of
implied inventory changes.
Initial estimates of implied changes
in inventory were revised by an average
840,000 barrels per day from late 2007
through early 2016, based on calculations using EIA data. This data produced
a largely inaccurate result—a quarter of
the time, the market indicated surplus
(or shortage) when the opposite was
true. Furthermore, over three-quarters of
the time, initial estimates of imbalance
were indistinguishable from zero using
the confidence band.

Three Large Episodes
Three episodes of inaccurate signals
were particularly large relative to historical norms.
Beginning in early 2009, roughly
coinciding with the end of the global
recession, initial data suggested a significant decline in global petroleum stocks.
However, revised estimates later
showed that the implied stock level was
essentially flat.
Initial estimates from mid-2011
implied relatively flat inventories, but
the revised data indicated that significant draws were occurring. In late 2014,
implied stock builds were also shown to
be much larger than initially estimated.
On the whole, revisions to estimated
consumption levels tended to be slightly
larger than revisions to production.
Consumption was the main contributor
to revisions following the global recession; the OECD consumption was
regularly overestimated from 2008 to
2015. Subsequently, consumption in
non-OECD countries exceeded what initial estimates suggested, driven by solid
economic growth and rising gasoline and
diesel demand, particularly from China
from 2011 to 2015.

Implied inventories have risen sharply since 2015. Steady increases in petroleum demand failed to match substantial
U.S. and OPEC production growth. Most
notably, the EIA’s report of monthly U.S.
crude oil production has been systematically revised upward since 2015, as
reflected in the overall implied level of
U.S. crude oil inventory (Chart 3).



The same exercise can be repeated
using the quarterly data produced by
the IEA. The initial IEA quarterly report
incorporates some revised monthly data
before its initial estimates are issued.
The IEA revisions in this sample had
an absolute average revision of about
423,000 barrels per day, with a standard
deviation of 302,000 barrels per day.

Underestimated Production Drove Revisions Upward

Implied change in inventory since August 2007,
million barrels per day

Confidence band

Estimates–revised (12 months)













NOTE: The confidence band is calculated by measuring one standard deviation of revisions. (Revisions are defined as the
difference between initial estimates (blue line) and revised estimates after 12 months (red line).)
SOURCES: Federal Reserve Bank of Dallas; Energy Information Administration; authors’ calculations.



Revisions to Oil Market Imbalance Vary Substantially

Difference between revised, initial estimates of implied change in inventory,
million barrels per day












NOTE: Shaded regions show episodes where initial estimates significantly mischaracterized imbalance given the size of
the EIA revisions.
SOURCES: Federal Reserve Bank of Dallas; Energy Information Administration; International Energy Agency; authors’

Economic Letter • Federal Reserve Bank of Dallas • November 2017


Economic Letter

The magnitude of these revisions is substantially lower than those for the EIA’s
data series—whose initial estimates are
timelier—but this is largely a function of
a timing of the releases.
When interpreting initial data from
the different agencies, analysts should be
aware of the differences in methodology
and the behavior of revisions, not just the
magnitude of the revisions (Chart 4).

Prices Reflect Multiple Revisions
The scale of revisions to implied
global inventory changes suggests that
initial estimates of imbalances between
production and consumption need to be
substantial to conclude market surplus
or shortage.
Market participants should account
for the uncertainty of revisions when
interpreting these initial estimates.
However, even with the uncertainty in
the real-time data, oil prices—on average—tend to move as expected based on
revised global implied stocks.
Lee is a research analyst in the Research
Department and Thompson is a business
economist in the Research Department at
the Houston Branch of the Federal Reserve
Bank of Dallas.

Analyzing fluctuation of crude oil inventory is crucial to
understand the oil market dynamics. See, “The Role of
Inventories and Speculative Trading in the Global Market
for Crude Oil,” by Lutz Kilian and Daniel P. Murphy,


Journal of Applied Econometrics, vol. 29, April/May
2014, pp. 454–78.
The EIA’s Petroleum Supply Monthly is another source
of the data. The data from the Short-Term Energy Outlook
and the Petroleum Supply Monthly may differ due to
timing of the releases.
For example, OPEC production numbers would be revised because of membership changes over time; reclassification of products might occur when blending natural
gas liquids into crude streams causes mismeasurement
of crude inputs into refineries. In 2015, the EIA changed
its methodology and started collecting production information directly from universal oil and gas firms in the
U.S.; significant changes to the magnitude of revisions
are not identified yet as a result of this change.
The U.S. consumption measure is a measure of refinery
intake volume. The U.S. production measure includes
domestic production and imports (foreign production)
of crude oil. Exports of crude oil from the U.S. were
negligible over the sample period.
For more extensive application, consult “Frontiers of
Real-Time Data Analysis,” by Dean Croushore, Journal
of Economic Literature, vol. 49, no. 1, March 2011,
pp. 72–100.
Different agencies have different definitions of
petroleum products. For example, the EIA’s measure of
petroleum includes crude oil (as well as lease condensates), natural gas liquids, biofuels, other liquids and
refinery processing gains. The IEA defines petroleum as
components of crude oil, natural gas liquids and nonconventional oils. This discrepancy is one of the sources
of difference in measure.
Previous literature documented that petroleum production revisions largely occur within 24 months of initial
release. The literature documented substantial revisions
after the first year and no substantive revisions after
two years. For full documentation, consult “Real-Time

Economic Letter

is published by the Federal Reserve Bank of Dallas.
The views expressed are those of the authors and
should not be attributed to the Federal Reserve Bank
of Dallas or the Federal Reserve System.
Articles may be reprinted on the condition that
the source is credited to the Federal Reserve Bank
of Dallas.
Economic Letter is available on the Dallas Fed

Forecasts of the Real Price of Oil,” by Christiane Baumeister and Lutz Kilian, Journal of Business & Economic
Statistics, vol. 30, December 2011, pp. 326–36.
The magnitude and direction of revisions must neither
be applied to statistical nor probabilistic interpretation.
There is no evidence that the revisions conform to any
well-defined distributional properties. The revisions are
not independent and identically distributed.

Marc P. Giannoni, Senior Vice President and Director of Research
Jim Dolmas, Executive Editor
Michael Weiss, Editor
Dianne Tunnell, Associate Editor
Ellah Piña, Graphic Designer

Federal Reserve Bank of Dallas
2200 N. Pearl St., Dallas, TX 75201