This is the accessible text file for GAO report number GAO-09-700 
entitled 'Energy Markets: Refinery Outages Can Have Varying Gasoline 
Price Impacts, but Gaps in Federal Data Limit Understanding of Impacts' 
which was released on July 30, 2009. 

This text file was formatted by the U.S. Government Accountability 
Office (GAO) to be accessible to users with visual impairments, as part 
of a longer term project to improve GAO products' accessibility. Every 
attempt has been made to maintain the structural and data integrity of 
the original printed product. Accessibility features, such as text 
descriptions of tables, consecutively numbered footnotes placed at the 
end of the file, and the text of agency comment letters, are provided 
but may not exactly duplicate the presentation or format of the printed 
version. The portable document format (PDF) file is an exact electronic 
replica of the printed version. We welcome your feedback. Please E-mail 
your comments regarding the contents or accessibility features of this 
document to Webmaster@gao.gov. 

This is a work of the U.S. government and is not subject to copyright 
protection in the United States. It may be reproduced and distributed 
in its entirety without further permission from GAO. Because this work 
may contain copyrighted images or other material, permission from the 
copyright holder may be necessary if you wish to reproduce this 
material separately. 

Report to Congressional Requesters: 

United States Government Accountability Office: 
GAO: 

July 2009: 

Energy Markets: 

Refinery Outages Can Have Varying Gasoline Price Impacts, but Gaps in 
Federal Data Limit Understanding of Impacts: 

GAO-09-700: 

GAO Highlights: 

Highlights of GAO-09-700, a report to congressional requesters. 

Why GAO Did This Study: 

In 2008, GAO reported that, with the exception of the period following 
Hurricanes Katrina and Rita, refinery outages in the United States did 
not show discernible trends in reduced production capacity, frequency, 
and location from 2002 through 2007. Some outages are planned to 
perform routine maintenance or upgrades, while unplanned outages occur 
as a result of equipment failure or other unforeseen problems. GAO was 
asked to (1) evaluate the effect of refinery outages on wholesale 
gasoline prices and (2) identify gaps in federal data needed for this 
and similar analyses. 

GAO selected refinery outages from 2002 through September 2008 that 
were at least among the largest 60 percent in terms of lost production 
capacity in their market region and lasted at least 3 days. GAO 
developed an econometric model and tested a variety of assumptions 
using public and private data. 

What GAO Found: 

While some unplanned refinery outages, such as those caused by 
accidents or weather, have had large price effects, GAO found that in 
general, refinery outages were associated with small increases in 
gasoline prices. Large price increases occurred when there were large 
outages; for example, in the aftermath of hurricanes Katrina and Rita. 
However, we found that such large price increases were rare, and on 
average, outages were associated with small price increases. For 
example, GAO found that planned outages generally did not influence 
prices significantly—likely reflecting refiners’ build-up in 
inventories to meet demand needs prior to shutting down—while for 
unplanned outages, average price effects ranged from less than one cent 
to several cents-per-gallon. Key factors influenced the size of price 
increases associated with unplanned outages. One such factor was 
whether the gasoline was branded—gasoline sold at retail under a 
specific refiner’s trademark—or unbranded—gasoline sold at retail by 
independent sellers. Our analysis showed that during an unplanned 
outage, branded wholesale gasoline prices had smaller price increases 
than unbranded, suggesting that refiners give preference to their own 
branded customers during outages, while unbranded dealers must seek out 
supplies in a more constrained market. Another factor that affected the 
size of price increases associated with outages was the type of 
gasoline being sold. Some special blends of gasoline developed to 
reduce emissions of air pollutants exhibited larger average price 
increases than more widely used and available conventional gasoline, 
suggesting that these special gasoline blends may have more constrained 
supply options in the event of an outage. 

Existing federal data contain gaps that have limited GAO’s and 
Department of Transportation’s (DOT) analyses of petroleum markets and 
related issues. For example: 

* Data linking refiners to the markets they serve were inadequate for 
GAO to fully evaluate the price effects of unplanned outages on 
individual cities, limiting the analysis to broader average effects. 

* Pipeline flow and petroleum product storage data were inadequate for 
DOT to fully address a January 2009 Congressionally mandated study to 
identify potential pipeline infrastructure constraints, and limited GAO’
s ability to identify re-supply options for cities experiencing outage 
disruptions. 

Federal agencies generally have continued to update their data 
collection surveys to meet their respective needs and emerging changes 
in the energy sector. However, in some cases the individual agency 
efforts have resulted in the collection of information that does not 
necessarily meet the data needs of other agencies or analysts who 
monitor petroleum product markets. 

What GAO Recommends: 

We recommend that the Administrator of the Energy Information 
Administration (EIA) convene a panel of agency officials, industry 
representatives, and experts to determine if existing data meet the 
current and future needs of the Congress and analysts who use such 
data. We provided a draft of this report to EIA, the Environmental 
Protection Agency (EPA), and the Department of Transportation (DOT). 
EIA agreed with our recommendations, and EPA and DOT made technical 
comments only. 

View [hyperlink, http://www.gao.gov/products/GAO-09-700] or key 
components. For more information, contact Frank Rusco at (202) 512-3814 
or ruscof@gao.gov. 

[End of section] 

Contents: 

Letter: 

Background: 

While Refinery Outages Can Have Large Price Effects on Rare Occasions, 
in Most Instances and on Average, Price Effects of Outages Are 
Relatively Small: 

Gaps in Federal Data Constrain Analyses of Outage Effects and Other 
Related Issues: 

Conclusions: 

Recommendations for Executive Action: 

Agency Comments and Our Evaluation: 

Appendix I: Scope and Methodology: 

Appendix II: GAO's Quantitative Methodology for Determining Impacts of 
Refinery Outages on Wholesale Prices: 

Appendix III: GAO Contact and Staff Acknowledgments: 

Tables: 

Table 1: Special Fuel Blends that Experienced Price Increases Greater 
than Conventional Gasoline Due to Unplanned Refinery Outages: 

Table 2: Special Fuel Blends that Experienced Price Increases About the 
Same as Conventional Clear Gasoline in the Event of Unplanned Refinery 
Outages: 

Table 3: Data Used In Our Econometric Model: 

Table 4: Regression Results for Effect of Unplanned Outages on 
Unbranded Gasoline Prices--Dependent Variable is the Logarithm of 
Unbranded Gasoline Price: 

Table 5: Regression Results for Effect of Unplanned Outages on Branded 
Gasoline Prices--Dependent Variable is the Logarithm of Branded 
Gasoline Price: 

Abbreviations: 

CARB: California Air Resources Board: 

CBG: Cleaner burning gasoline: 

DOE: Department of Energy: 

DOT: Department of Transportation: 

EIA: Energy Information Administration: 

EPA: Environmental Protection Agency: 

FERC: Federal Energy Regulatory Commission: 

HHI: Hirschman Herfindahl Index: 

IIR: Industrial Information Resources, Inc. 

MTBE: Methyl Tertiary Butyl Ether: 

OPIS: Oil Price Information Services: 

PADD: Petroleum Administration for Defense Districts: 

PSRS: Petroleum Supply Reporting System: 

RFG: Reformulated Gasoline: 

RFS: Renewable Fuel Standard: 

RVP: Reid Vapor Pressure: 

[End of section] 

United States Government Accountability Office: 
Washington, DC 20548: 

July 30, 2009: 

Congressional Requesters: 

The 150 refineries in the United States play an important role in the 
nation's economy and energy security by supplying consumers and 
industry with needed petroleum products. Unplanned refinery outages-- 
such as those caused by hurricanes, fires, or refinery equipment 
failures--have raised questions about the stability and cost of U.S. 
gasoline and other petroleum product supplies. In addition to unplanned 
outages, refineries must periodically undergo planned outages, during 
which they shut down major pieces of equipment to perform maintenance, 
overhaul, and repair operations. In October 2008,[Footnote 1] we 
reported that, with the exception of impacts in 2005 related to 
Hurricanes Katrina and Rita, refinery outages across the United States 
generally did not show discernible trends in reduced production 
capacity or in the frequency and location of outages from 2002 through 
2007. In addition, in March 2007, the Department of Energy's (DOE) 
Energy Information Administration (EIA) reported that unplanned 
refinery outages can result in local supply disruptions that result in 
temporary price increases; however, refinery outages do not always 
affect prices.[Footnote 2] Moreover, analyses by EIA and the California 
Energy Commission have described how an unplanned refinery outage under 
certain conditions--for example, a tight market supply and demand 
balance for refined products coupled with low inventories or other 
sources of re-supply to meet demand in the event of an unplanned 
outage--can trigger price increases. Still, the impacts on gasoline 
prices due to refinery outages and other disruptions are not fully 
understood. In particular, while it is well understood that extreme 
events that disrupt crude oil or petroleum product supplies can have 
significant effects on the prices of these commodities, the price 
effects of less dramatic disruptions, such as routine refinery outages, 
are not well understood. 

Further compounding the potential impact of refinery outages, in recent 
years prior to the current economic recession, global demand for crude 
oil and petroleum products such as gasoline, diesel fuel, and jet fuel 
had grown more quickly than available capacity to produce them. 
Furthermore, some refiners had been running near capacity, particularly 
during the peak-demand summer months. During tight market conditions, 
unexpected refinery outages could stress the petroleum product supply 
system, affecting operations at refineries, pipelines, and storage 
terminals. In addition, the proliferation of special fuel blends-- 
gasoline that has special characteristics designed to meet federal, 
state, and local air quality requirements--as well as the increasing 
use of biofuels such as ethanol as a component of gasoline, have 
complicated the manufacturing and distribution processes for petroleum 
products. Once produced, the various blends of petroleum products must 
be kept separate throughout shipping and delivery. Other disruptions, 
such as a pipeline break, can hamper the ability of the supply 
infrastructure to deliver the steady supply of gasoline and other 
petroleum products that U.S. consumers and businesses depend on. In the 
past, local supply disruptions could be addressed more quickly because 
additional fuel of the same formulation could be purchased from 
numerous sources, but with the proliferation of special fuel blends, 
replacement supplies of a special blend might not be as readily 
available, and refineries with the capability to produce them could be 
hundreds of miles away. 

A number of federal agencies--including EIA, the Environmental 
Protection Agency (EPA), the Department of Transportation (DOT), and 
the Federal Energy Regulatory Commission (FERC)--have a role in 
monitoring the effects of outages and ensuring the safe, efficient, and 
adequate supply of petroleum products during and after those outages. 
In this context, you asked us to study and evaluate (1) how refinery 
outages have affected U.S. wholesale gasoline prices since 2002 and (2) 
to what extent available federal data allow for the evaluation of the 
impacts of refinery outages on petroleum product prices and reflect 
emerging trends in petroleum product markets that may be important to 
future analytical needs. 

To evaluate how refinery outages have affected U.S. wholesale gasoline 
prices since 2002, we purchased data that included detailed information 
on refinery outages from Industrial Information Resources, Inc. (IIR); 
data estimating the quantity flows of gasoline and other petroleum 
products produced at most U.S. refineries and then transported to those 
U.S. cities that make up the main markets for those products from Baker 
& O'Brien; and weekly wholesale price data for 75 U.S. cities from the 
Oil Price Information Service (OPIS). We also obtained and analyzed 
data from EIA's monthly refinery production survey, form EIA-810, and 
other EIA data collection surveys. We determined that these data were 
sufficiently reliable for the purposes of this report. Specifically, we 
determined these data were sufficient to complete our analyses of the 
immediate average wholesale price impacts associated with refinery 
outages on various gasoline types, but they were neither sufficient to 
determine the effects experienced by individual cities nor the longer 
term or dynamic effects of outages on prices.[Footnote 3] We developed, 
and extensively tested, an econometric model that examined the 
statistical relationship between refinery outages and gasoline prices. 
We analyzed commercial data for 20 gasoline types and distinguished 
between branded and unbranded gasoline to determine if those factors 
influenced the price effect of an outage. We limited our analysis to 
outages that were determined to be 1) at least among the largest 60 
percent of outages in terms of lost production capacity within their 
market region and 2) that lasted at least 3 days.[Footnote 4] In our 
model, we limited the effect of an outage on prices to one week, after 
which time we assumed that petroleum products were supplied from an 
alternate source. As a result, our analysis evaluated the short-term 
effects of outages but did not evaluate the length of time those 
effects occurred.[Footnote 5] Limitations on available public data on 
the production and supply of petroleum products restricted our analysis 
to those cities for which Baker & O'Brien had collected and maintained 
data; these cities generally represented the United States 
geographically, but might not have been representative of all cities. 
To control for the effects on gasoline prices, our model incorporated 
data on numerous factors--such as gasoline inventory levels, refinery 
capacity utilization, and gasoline specifications. Our modeling results 
reflect a particular city's reliance on the refinery experiencing the 
outage.[Footnote 6] 

To assess the extent to which available federal data allowed for the 
evaluation of refinery outage impacts, we reviewed federal government 
data collection surveys from federal agencies including EIA, EPA, and 
FERC, as well as private companies. We reviewed the surveys for 
comprehensiveness, utility, potential gaps or limitations, and to 
understand the extent to which the surveys collect useful data to 
analyze the impacts of disruptions. We reviewed past GAO reports and 
other federal agency or intergovernmental agency studies on petroleum 
product markets to identify data gaps, limitations, or inconsistencies. 
Furthermore, we interviewed key industry and expert institution 
representatives regarding data utility and limitations in their own 
work. Our work was not a comprehensive evaluation of all federal energy 
data, but rather an assessment of key data GAO used in this and past 
reports, and other select data that we determined during the course of 
our review to have posed limitations for GAO's or other agencies' 
evaluations of important policy questions related to petroleum markets. 
See appendix I for a more detailed description of our objectives, 
scope, and methodology. 

We conducted our work from October 2008 through July 2009 in accordance 
with generally accepted government auditing standards. Those standards 
require that we plan and perform the audit to obtain sufficient, 
appropriate evidence to provide a reasonable basis for our findings and 
conclusions based on our audit objectives. We believe that the evidence 
obtained provides a reasonable basis for our findings and conclusions 
based on our audit objectives. 

Background: 

Refineries process crude oil into petroleum products through a 
combination of distillation and other processes. A single barrel of 
crude oil produces a varying amount of gasoline, diesel, jet fuel, and 
other petroleum products depending on the configuration--or complexity--
of the refinery and the type of crude oil being refined. 

This report focuses on the production of finished gasoline.[Footnote 7] 
Finished gasoline is primarily defined by three characteristics: 
blendstock, vapor pressure, and oxygenate content. Blendstock is the 
designation for the base gasoline produced so that other materials can 
be blended in to meet various air quality or other local 
specifications. Vapor pressure, also known as Reid Vapor Pressure 
(RVP), measures the gasoline's evaporation characteristics or 
volatility. Oxygenates are fuel additives, particularly alcohols and 
ethers, which increase gasoline octane levels and reduce carbon 
monoxide pollution associated with automobile emissions. The most 
widely used oxygenate in the United States is ethanol, which may be 
added to gasoline in varying percentages. Federal regulations specify 
that no more than 10 percent ethanol can be blended into gasoline. 
Ethanol is generally blended with gasoline at the terminal or wholesale 
"rack"--the distribution center between refineries and retail fueling 
stations. For the purposes of this report, conventional gasoline does 
not contain special federal, state, or local blendstock, RVP, or 
oxygenate requirements unless otherwise noted, while "special fuel 
blends" refer to blends of gasoline that are designed to be cleaner 
burning and generally contain either a certain blendstock, RVP, or 
oxygenate requirement to meet federal, state, or local fuel 
specifications. An example of a gasoline used to meet a state fuel 
specification is California Air Resources Board (CARB) gasoline, which 
is designed to reduce harmful exhaust emissions that cause smog and is 
used exclusively in California. 

Petroleum product markets are evolving in part as a result of the 
increasing use of biofuels--fuels derived from plant or animal 
products--throughout the country. The Energy Policy Act of 2005 
generally required that at least 7.5 billion gallons of biofuels be 
blended into motor vehicle fuels in the United States by 2012. These 
targets were later amended under the Energy Independence and Security 
Act of 2007, which increased the volume of biofuels to be blended with 
gasoline from 9 billion gallons in 2008 to 36 billion gallons in 2022. 
EPA was charged with implementing the Renewable Fuel Standard (RFS) 
program and issuing regulations to ensure that the annual volumes of 
biofuels specified by the legislation are being blended into motor 
vehicle fuels. In addition, some states require the use of biofuels. 
For example, in Minnesota all fuel must contain 10 percent ethanol, 
while a number of other states offer consumers incentives--such as tax 
credits and rebates--for purchasing ethanol or other biofuels. The 
steadily increasing use of biofuels in the United States has 
complicated the production and distribution of gasoline. Biofuels such 
as ethanol are produced at dedicated biofuel production facilities--not 
at refineries--and currently cannot be transported by most petroleum 
product pipelines.[Footnote 8] Therefore in order for ethanol to be 
blended with gasoline, it must be shipped to the terminal by truck or 
rail, where it is then mechanically mixed with gasoline as it is 
delivered into trucks for shipping to retail. 

Gasoline with or without biofuels is typically sold as either branded 
or unbranded. Branded gasoline is that supplied from major refiners and 
sold at retail stations under these refiner's trademarks, and often 
contains special additives. Contracts for branded gasoline tend to be 
less flexible than contracts for unbranded gasoline but guarantee a 
more secure supply. Conversely, unbranded gasoline may be supplied by 
major or independent refiners, but is not sold under a refining 
company's trademark. Buyers of unbranded gasoline may or may not have a 
binding contractual arrangement with a refiner. 

The supply infrastructure--which includes pipelines and terminals that 
hold supply inventories--is a critical component of the nation's 
petroleum product market in that it facilitates the flow of crude oil 
and petroleum products from one geographic region to another. Crude oil 
pipelines connect several large refining centers to crude oil sources, 
and petroleum product pipelines connect these refineries to population 
centers all over the country. Thus, a disruption in one geographic 
region can affect the supply and prices in another geographic region. 
To help mitigate the effects of potential supply disruptions caused by 
refinery outages or sudden increases in demand and to facilitate smooth 
supply operations, refiners, distributors, and marketers of petroleum 
products maintain inventories of crude oil and petroleum products. 
Inventories represent the most accessible and readily available source 
of supply in the event of a production shortfall, such as one caused by 
a refinery outage, or increase in demand. 

In October 2008,[Footnote 9] we reported that unplanned and planned 
refinery outages across the United States did not show discernible 
trends in the frequency or location of outages from 2002 through 2007, 
with the exception of impacts beginning in 2005 related to Hurricanes 
Katrina and Rita. During that study, however, we found that EIA does 
not collect information on refinery outages directly and thus the 
information it collects on its monthly refinery survey and uses to 
indirectly estimate outages has a number of limitations. Specifically, 
EIA's method of using EIA-810 data to estimate outages cannot 
distinguish between planned and unplanned outages, which could have 
different impacts on petroleum product prices for consumers.[Footnote 
10] Also, as we reported, because the monthly refinery survey data are 
monthly aggregate data, major outages that straddle the end of one 
month and the beginning of the next may be difficult to identify and 
the observable effects of those outages could be diluted. We further 
reported that the exact date and length of an outage may be difficult 
to determine from EIA's monthly refinery survey data, making it 
difficult to use the data to determine whether a specific outage had a 
significant effect on the production capacity for some petroleum 
products as well as market prices.[Footnote 11] 

Several U.S. agencies have jurisdiction over and monitor the U.S. 
refining and supply infrastructure industries and petroleum product 
markets. 

* Within the Department of Energy (DOE), the Energy Information 
Administration (EIA) collects and analyzes data, including supply, 
consumption, and prices of crude oil and petroleum products; inventory 
levels; refining capacity and utilization rates; and some petroleum 
product movements into and within the United States. Much of the data 
that the agency collects is obtained by surveys under EIA's Petroleum 
Supply Reporting System (PSRS). The PSRS is comprised of 16 data 
collection surveys and includes, among others, weekly and monthly 
surveys of refiners, terminals, and pipelines.[Footnote 12] The purpose 
of the PSRS is to collect and disseminate basic and detailed data to 
meet EIA's responsibilities and energy data users' needs for credible, 
reliable, and timely information on U.S. petroleum product supply. EIA 
generally updates its PSRS surveys every 3 years and has issued such 
updates in 2003, 2006, and 2009. EIA also conducts analyses in support 
of DOE's mission and in response to Congressional inquiries. For 
example, EIA recently conducted its semiannual forecast of planned 
refinery outage effects. EIA evaluates a wide range of trends and 
issues that could have implications for U.S. petroleum product trends 
and markets, and each year issues a publication known as the Annual 
Energy Outlook. 

* The Environmental Protection Agency (EPA), among other things, 
develops and enforces regulations that implement environmental laws 
that aim to control the discharge of pollutants into the environment by 
refiners and other industries. The EPA, with the concurrence of DOE, 
can grant waivers on fuel requirements that allow petroleum product 
markets to be more easily re-supplied should an "extreme and unusual" 
situation--such as a problem with distribution of supply to a 
particular region, a natural disaster, or refinery equipment failure-- 
occur.[Footnote 13] In addition, EPA oversees the Reformulated Gasoline 
(RFG) program. This program was developed in response to a requirement 
in the Clean Air Act that cities with the most severe smog pollution 
use reformulated gasoline--gasoline blended to burn cleaner and reduce 
smog-forming and toxic pollutants in the air--to reduce 
emissions.[Footnote 14] EPA is also responsible for implementing and 
issuing regulations to ensure that gasoline sold in the United States 
contains a minimum volume of biofuels, such as ethanol or biodiesel, 
and its reports, according to EPA officials, are geared toward 
collecting data on fuel quality which is enforced at the refinery. 
Under EPA's Renewable Fuel Standard (RFS) program, refiners, importers, 
and blenders are required to use a minimum volume of biofuels each 
year, determined as a percentage of the total volume of fuel the 
company produces, blends, or imports. Entities that are unwilling or 
unable to meet this percentage standard may purchase biofuel credits 
from other obligated parties in order to satisfy the requirement. EPA 
monitors RFS program compliance and has the authority to waive the 
standard if it determines that specified biofuel volumes would cause 
severe harm to the economy or the environment in a particular region, 
state, or the country or that there is an inadequate domestic supply. 

* The Department of Transportation's (DOT) Pipeline and Hazardous 
Materials Safety Administration focuses on pipeline safety and 
establishes standards for transmission and distribution systems for 
crude oil and petroleum product pipeline. Among other things, it 
oversees pipelines' design, maintenance, and operating procedures to 
maintain the safe, efficient, and reliable delivery of petroleum 
products. 

* The Federal Energy Regulatory Commission (FERC) monitors energy 
markets and regulates rates and practices of oil pipeline companies 
engaged in interstate transportation of natural gas, oil and 
electricity. It establishes and enforces the rates, known as "tariffs," 
for transporting petroleum and petroleum products by pipeline. 

While Refinery Outages Can Have Large Price Effects on Rare Occasions, 
in Most Instances and on Average, Price Effects of Outages Are 
Relatively Small: 

While it can be expected that some refinery outages have quite large 
price effects, the results of our analysis found that on average 
refinery outages were associated with small increases in gasoline 
prices. Based on our analysis of wholesale prices across 75 U.S. cities 
from 2002 through September 2008, planned outages generally did not 
influence prices, while unplanned refinery outages had generally small 
wholesale gasoline price effects in the cities they serve. Price 
increases varied depending on whether the gasoline was branded or 
unbranded and according to the gasoline type affected by the outage. 

Extreme Outage Events Can Lead to Large Temporary Price Increases: 

On rare occasions, refinery outages can have large temporary effects on 
gasoline prices. For example, as we recently testified, petroleum 
product prices increased dramatically following Hurricanes Katrina and 
Rita.[Footnote 15] This occurred in part because many refineries are 
located in the Gulf Coast region and power outages shut down pipelines 
that refineries depend on for crude oil supplies and to transport 
refined petroleum products, including gasoline to wholesale markets. 
DOE reported that 21 refineries in affected states were either shut 
down or operating at reduced capacity in the aftermath of the 
hurricanes. In total, nearly 30 percent of the refining capacity in the 
United States was shut down, disrupting supplies of gasoline and other 
products. Two pipelines that send petroleum products from the Gulf 
Coast to the East Coast and the Midwest were also shut down as a result 
of Hurricane Katrina. For example, Colonial Pipeline, which transports 
petroleum products to the Southeast and much of the East Coast, was not 
fully operational for a week after Hurricane Katrina due to large-scale 
power outages and flooding. Consequently, according to the Federal 
Trade Commission, average gasoline prices for the nation increased 45 
cents-per-gallon between August 29 and September 5, 2005; short-term 
gasoline shortages occurred in some places; and the media reported 
gasoline prices greater than $5 per gallon in Georgia. The hurricane 
came on the heels of a period of high crude oil prices and a tight 
balance worldwide between petroleum demand and supply, and illustrated 
the volatility of gasoline prices given the vulnerability of the 
gasoline infrastructure to natural or other disruptions. 

While extreme outages can cause large temporary price increases, such 
events were relatively uncommon during the period of our analysis. For 
example, for unbranded prices, of the approximately 1100 unplanned 
outages we evaluated, 99 percent of the time they were associated with 
wholesale price increases of no more than about 32 cents-per-gallon, 
and 75 percent of the time they were associated with price increases of 
less than 6 cents-per-gallon in the cities affected. 

On Average, Price Effects Associated with Outages Are Relatively Small, 
and Depend on Key Factors: 

Overall, our analysis indicated that planned outages--where refineries 
temporarily shut down to perform routine maintenance or equipment 
upgrades--generally did not have a significant effect on wholesale 
gasoline prices. As we reported in October 2008,[Footnote 16] planned 
outages are typically scheduled during periods of less demand and 
interspersed among refiners and refineries. In addition, the equipment 
and labor are generally booked months--or even years--in advance, and 
can be arranged with those customers with whom the refiners have long- 
term contracts at a cost less than would be required in an emergency or 
unplanned situation. Industry representatives told us that because a 
refinery must draw on a limited number of equipment makers and skilled 
laborers, the refinery's plans for maintenance eventually become public 
knowledge. In this case, the market "expects" the outage to occur, 
therefore planned outages do not generally trigger significant price 
responses, unless something unexpected occurs or the market is 
disrupted elsewhere. Furthermore, refineries stockpile petroleum 
products in preparation for planned outages and therefore do not 
experience the same shortage of production materials experienced during 
unplanned outages. 

Unplanned outages, on the other hand, were associated with gasoline 
price increases but these increases were generally small and depended 
on key factors, including whether or not the gasoline was branded or 
unbranded and the type of gasoline being sold. With respect to the 
distinction between branded and unbranded gasoline, our analysis showed 
that in the event of an unplanned refinery outage, unbranded gasoline 
was generally associated with greater wholesale price increases than 
branded gasoline. Specifically, we found that for conventional 
gasoline--the most common and widely available gasoline blend-- 
unbranded gasoline had an average 0.5-cents-per-gallon increase in 
price associated with unplanned refinery outages, while branded 
gasoline had a smaller--about 0.2-cents-per-gallon--increase. The price 
effects observed in these cases reflect an average increase in prices 
at the wholesale terminals in the 75 cities over the study period. 
These results suggest that--as some traders and other market 
participants have told us--during disruptions, refiners generally 
choose to give priority in supplying those customers with whom they 
have long-term supply contracts, which typically are for branded 
gasoline. Therefore, in such conditions independent marketers--which 
typically sell unbranded gasoline--may be forced to pay higher prices 
to obtain product to sell. On the other hand, industry experts told us 
that unbranded sellers may be able to buy wholesale gasoline at lower 
prices than branded sellers during normal market conditions.[Footnote 
17] 

With regard to the type of gasoline fuel blend being sold, our analysis 
shows that the price increases associated with an unplanned refinery 
outage were significantly greater for 8 of the 19 "non-base-case" 
gasoline types we identified than our "base case" conventional clear 
gasoline, while the price increases for other gasoline types were 
generally about the same as those of conventional gasoline. In our 
analysis, we selected conventional gasoline as our base case and used 
our model to determine whether there were significant differences 
between this base case and other fuel types with respect to the 
relationship between unplanned refinery outages and price changes. We 
looked at 19 other non-base case fuel types that were in use in the 75 
cities we reviewed. We compared the results of these 19 other fuel 
types to our conventional gasoline base case and measured the price 
differences. The price increases associated with unplanned refinery 
outages for various branded and unbranded gasoline types that were 
higher than our conventional gasoline base case are shown in table 1. 

Table 1: Special Fuel Blends that Experienced Price Increases Greater 
than Conventional Gasoline Due to Unplanned Refinery Outages[A]: 

Gasoline type: Conventional[B] base case; 
Cents-per-gallon increases for unbranded gasoline types: 0.5; 
Cents-per-gallon increases for branded gasoline types: 0.2; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Throughout the time period. 

Gasoline type: California Air Resources Board (CARB) gasoline with 2% 
Methyl Tertiary Butyl Ether (MTBE) as oxygenate[C]; 
Cents-per-gallon increases for unbranded gasoline types: 3.2; 
Cents-per-gallon increases for branded gasoline types: 
[Empty]; Locations that require this gasoline type: California; 
Time period sold: Beginning of study period (January 2002) to November 
2003. 

Gasoline type: CARB with no oxygenate; 
Cents-per-gallon increases for unbranded gasoline types: 10.1; 
Cents-per-gallon increases for branded gasoline types: 
[Empty]; Locations that require this gasoline type: None, although 
found in California[C]; 
Time period sold: Beginning of study period to May 2006. 

Gasoline type: Conventional with 5.7% ethanol as oxygenate; 
Cents-per-gallon increases for unbranded gasoline types: 4.1; 
Cents-per-gallon increases for branded gasoline types: 1.3; 
Locations that require this gasoline type: Pima County, (Tucson) AZ; 
Time period sold: Winters, from beginning of study period to present. 

Gasoline type: CARB with 5.7% ethanol as oxygenate; 
Cents-per-gallon increases for unbranded gasoline types: [Empty]; 
Cents-per-gallon increases for branded gasoline types: 1.4; 
Locations that require this gasoline type: California; 
Time period sold: Beginning of study period to present. 

Gasoline type: Conventional with 10% ethanol as oxygenate, 7.0 RVP; 
Cents-per-gallon increases for unbranded gasoline types: 8.0; 
Cents-per-gallon increases for branded gasoline types: [Empty]; 
Locations that require this gasoline type: Clay, Jackson, and Platte 
counties, MO; 
Time period sold: Summers, from June 2004 to present. 

Gasoline type: Conventional with 10% ethanol as oxygenate, 9.0 RVP; 
Cents-per-gallon increases for unbranded gasoline types: [Empty]; 
Cents-per-gallon increases for branded gasoline types: 1.5; 
Locations that require this gasoline type: Iowa, Minnesota, many parts 
of Oregon; 
Time period sold: Sold year round in Iowa, and in the summer in all 
other locations; used since the beginning of study period to present. 

Gasoline type: Low sulfur; 
Cents-per-gallon increases for unbranded gasoline types: 3.9; 
Cents-per-gallon increases for branded gasoline types: [Empty]; 
Locations that require this gasoline type: Georgia; 
Time period sold: April 2003 to present. 

Gasoline type: Reformulated gasoline (RFG); 
Cents-per-gallon increases for unbranded gasoline types: 1.3; 
Cents-per-gallon increases for branded gasoline types: 0.9; 
Locations that require this gasoline type: Numerous cities, counties, 
states; 
Time period sold: Beginning of study period to present. 

Source: GAO analysis of data from various sources, as described in 
appendixes I and II. 

Note: All reported figures are statistically significant at the 10 
percent level or less. 

[A] Price increases for special blends include the base case increases 
of 0.5 and .02 for unbranded and branded fuel types respectively. We 
calculated price effects using our model estimates of the impact of 
outages on wholesale gasoline prices. 

[B] Conventional gasoline used as the base case is gasoline that does 
not have a special RVP or oxygenate content specified to meet local air 
quality needs or preferences. 

[C] Although Methyl Tertiary Butyl Ether (MTBE) was not specifically 
required to be the oxygenate used in California, an oxygenate was 
required under federal RFG provisions. The use of MTBE was banned in 
California on December 31, 2003. Following the phase out of MTBE and 
the transition to ethanol, according to California Energy Commission 
(CEC), California refiners and gasoline marketers began using ethanol 
at a minimum concentration of 5.7 percent by volume. Although nearly 20 
percent of the gasoline sold could have been non-oxygenated, according 
to the CEC, due to segregation limitations in the distribution 
infrastructure system and concerns about maintaining fungible gasoline 
production for purposes of exchange agreements and periodic unplanned 
refinery outages, the gasoline market gravitated towards a near- 
unanimous mix of ethanol at roughly 6 percent volume by January 2004. 

[End of table] 

The results suggest that some special fuel blends that include such 
characteristics as unusual oxygenate requirements, lower RVP 
requirements, or unusual oxygenate/RVP combinations may be more 
sensitive to unplanned outages than other special fuel blends. For 
example, for unbranded gasoline, the prices of some special fuel 
blends--such as CARB, conventional gasoline with oxygenate formulations 
such as 5.7 percent ethanol, or uncommon oxygenate/RVP formulations 
such as conventional gasoline with 10 percent ethanol and a 7.0 RVP-- 
were more sensitive to unplanned refinery outages than conventional 
gasoline without such specifications. Specifically, the largest price 
differences between our conventional gasoline base case and special 
gasoline blends, were for CARB without oxygenate and conventional 
gasoline blended with 10 percent ethanol and a 7.0 RVP. In these 
instances, prices were about 10-cents and 8-cents-per-gallon higher 
than our base case. The results show that the prices of unusual 
oxygenate/RVP combinations that are not commonly produced at most 
refineries may be more sensitive to unplanned outages than conventional 
gasoline, which can be more readily re-supplied to a city experiencing 
an outage. 

Our analysis also shows that a number of other special fuel blends did 
not experience significant price increases associated with unplanned 
refinery outages above that of conventional gasoline, although the fuel 
types affected depended partly on whether the gasoline was branded or 
unbranded. These fuel types and the locations that require them are 
shown in table 2. 

Table 2: Special Fuel Blends that Experienced Price Increases About the 
Same as Conventional Clear Gasoline in the Event of Unplanned Refinery 
Outages: 

Fuel type: Conventional; base case; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Throughout the time period. 

Fuel type: Cleaner burning gasoline (CBG); 
Locations that require this gasoline type: Maricopa County (Phoenix), 
Arizona; 
Time period sold: March 2005 to present. 

Fuel type: CBG with 10% ethanol as oxygenate; 
Locations that require this gasoline type: Maricopa County (Phoenix), 
Arizona--winters; 
Time period sold: February 2005 to present. 

Fuel type: Conventional RVP 7.0; 
Locations that require this gasoline type: Jefferson and Shelby 
counties, Alabama; Johnson and Wyandotte Counties, Kansas; Livingston, 
Macomb, Monroe, Oakland, St. Clair, Washtenaw, Wayne, and Lenawee 
counties Michigan; El Paso, Texas; 
Time period sold: Summers only, from the beginning of study period to 
present. 

Fuel type: Conventional RVP 7.8; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Beginning of study period to present. 

Fuel type: Conventional RVP 9.0; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Beginning of study period to present. 

Fuel type: Conventional 7.7% ethanol as oxygenate; 
Locations that require this gasoline type: Although not required in any 
location, this gasoline occurs frequently in Iowa and Minnesota; and 
the cities of El Paso, Texas; Missoula, Montana; Fargo, North Dakota; 
and Sparks/Reno, Nevada; 
Time period sold: Beginning of study period to present. 

Fuel type: Conventional 7.7% ethanol as oxygenate, RVP 9.0; 
Locations that require this gasoline type: Although not required in any 
location, this gasoline occurs frequently in numerous cities, counties, 
and states; 
Time period sold: Beginning of study period to present. 

Fuel type: Conventional 10% ethanol as oxygenate; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Beginning of study period-present. 

Fuel type: Conventional 10% ethanol as oxygenate, RVP 7.8; 
Locations that require this gasoline type: Denver and Boulder, 
Colorado; Clackamas, Marion, Multnomah, Polk, and Washington counties, 
Oregon; 
Time period sold: Summers, beginning May 2004 in CO; May 2005 in OR 
counties. 

Fuel type: Low Sulfur RVP 7.0; 
Locations that require this gasoline type: Atlanta and 45 other 
counties in Georgia; 
Time period sold: Summer, since April 2003. 

Fuel type: Reformulated Gasoline with Methyl Tertiary Butyl Ether 
(MTBE) as oxygenate; 
Locations that require this gasoline type: Numerous cities, counties, 
and states; 
Time period sold: Beginning of study period to May 2006. 

Source: GAO analysis of data from various sources, as described in 
appendixes I and II. 

[End of table] 

Finally, it should be noted that individual outages may have different 
effects on prices depending on a variety of factors beyond those 
discussed above. As discussed previously in this report, and in work by 
EIA and the California Energy Commission, under certain conditions-- 
such as low inventories, high seasonal demand, certain special fuel 
requirements, and geographic conditions that may hinder easy re-supply 
to the market--an unplanned refinery outage could be expected to result 
in a price surge in some cases. However, in some cases, unobserved 
factors can mitigate the effects of outages, or even cause prices to 
fall, making it appear as if the outage caused prices to fall. For 
example, a large shipment of a particular special fuel blend located 
just offshore or beyond the Canadian border could be a significant 
source of re-supply in the event of a disruption. In addition, while 
our analysis examined the effect of about 1,100 unplanned outages and 
1,000 planned outages, our model did not differentiate between the 
types of refinery equipment that went out of service, which could have 
varying effects on wholesale gasoline prices. For example, an unplanned 
outage of a fluid catalytic cracker--a type of processing equipment 
used to maximize the production of gasoline--could be expected to have 
a more significant effect on wholesale gasoline prices than an 
unplanned outage on a piece of equipment--such as a certain type of 
hydrotreater--that is designed to maximize production of distillates 
such as diesel fuel or heating oil. Because our model does not 
distinguish between the type of unit experiencing an outage, our 
results show average impacts across different types of refining units, 
which means we tend to underestimate the effect of an outage at a unit 
such as a fluid catalytic cracker, and overestimate that of a non- 
gasoline producing unit.[Footnote 18] 

Gaps in Federal Data Constrain Analyses of Outage Effects and Other 
Related Issues: 

Existing federal data contain gaps that limit analyses of refinery 
outages on petroleum product prices and in some cases do not reflect 
emerging trends--although agencies continue to take steps to improve 
their data collection. These data gaps created challenges to our, and 
another federal agency's, analyses and ability to respond to 
Congressional inquiries. Specifically, we were limited in this report 
in our ability to fully evaluate 1) the price effects of unplanned 
outages at individual cities and 2) a city's gasoline re-supply options 
in the event of an outage. 

Our ability to fully evaluate the price effects of unplanned outages at 
individual cities--for example, price effects in Atlanta, Georgia 
associated with outages related to Hurricanes Ike and Gustav--was 
limited because federal data do not link refiners to the cities they 
serve. Although federal data exist regarding most refinery activities, 
the refiner-to-market link contains key gaps.[Footnote 19] 

* While EPA's annual reformulated gasoline area report requires each 
refinery to identify the cities the refinery believes it supplies with 
reformulated gasoline, this reporting is limited to reformulated 
gasoline. As such, the reports do not capture the estimated refiner-to- 
city link for a majority of gasoline types--including conventional 
gasoline and special fuel blends--sold in the United States. Further, 
the reports are not intended to identify the quantities of gasoline 
distributed. 

* EIA's monthly refinery survey, the EIA-810, collects data regarding 
the volume of certain petroleum products being produced at refineries, 
including gasoline and unfinished gasoline blending components, but 
does not distinguish among all types of gasoline, such as premium 
versus regular or summer versus winter RVP, or identify which cities 
refineries serve. 

* Our ability to identify a city's gasoline re-supply options in the 
event of an outage was also limited because of gaps in federal pipeline 
flow data.[Footnote 20] Although we identified flow data collected at 
three agencies, the data were of limited use because they did not show 
the volumetric entry, flow, and exit of specific petroleum products 
through the pipeline. These specific data are important to 
understanding which refiners can and cannot supply various cities in 
the event of an outage and thus can be used to help determine potential 
price impacts. 

* FERC's quarterly reports by pipeline operators specify the number of 
barrels of petroleum products pipeline companies transport, but these 
data do not identify the entry and exit points of petroleum products 
along the pipeline infrastructure system, or the specific type of fuels 
transported. 

* EIA's monthly pipeline survey collects data on pipeline shipments 
between Petroleum Administration for Defense Districts (PADD)--a 
geographic aggregation of the 50 states and the District of Columbia 
split into five districts--as well as pipeline inventories by PADD. 
However, data at the PADD level do not correspond to particular cities 
and therefore the data cannot be used to identify the states and/or 
cities in which petroleum product flows originate and terminate. 

* DOT's annual report on hazardous liquids collects pipeline flow data, 
but DOT officials told us, and we also found, that these data are 
highly aggregated and the annual collection of information is too 
infrequent to be informative in many cases. Further, these data are not 
designed to show the discrete movement of petroleum products through 
the pipeline infrastructure. 

To help address these gaps in federal data, we purchased commercial 
data for our analysis from the energy consulting company Baker & 
O'Brien (see appendix I). These data estimate the quantity flows of 
gasoline and other petroleum products produced at most U.S. refineries 
and transported to those U.S. cities that make up the main markets for 
these products. While we found the Baker & O'Brien data to be 
sufficiently reliable for the purposes of our analysis, these data are 
estimates only. Although we determined the commercial data that we 
purchased to perform our analyses were sufficient to describe the 
wholesale price impacts associated with refinery outages on various 
gasoline types, the data were not sufficient to accurately estimate the 
effects experienced by individual cities. Further, the 
comprehensiveness of the data we purchased was limited in part because 
private companies do not have the same ability as the federal 
government to require refiners to provide comprehensive and accurate 
information.[Footnote 21] 

Similar gaps in federal data also limited a recent effort by another 
federal agency to fully address Congressional concerns regarding 
potential pipeline constraints and agency concerns regarding refinery 
outages. 

* In a January 2009 Congressionally mandated study to identify 
potential pipeline infrastructure constraints, DOT was unable to fully 
address the study's objectives due to the lack of appropriate federal 
pipeline flow and petroleum product storage data.[Footnote 22] In its 
report, DOT noted that "a need exists to develop more robust metrics 
for such (pipeline flow) measurements." The report also stated that 
"reliable data on storage facilities is sparse" and emphasized the need 
for additional data on oil and petroleum product storage terminals, 
including the location, size, and volumetric capacity of existing 
facilities to assess whether stored petroleum products are sufficient 
to mitigate supply disruptions.[Footnote 23] In addition, the study 
noted that additional data regarding the changing location and 
arrangement of petroleum product pipelines would be necessary to 
evaluate volumes of petroleum products transported. DOT concluded that 
an analysis sufficient to address Congress's directives in the 2006 law 
would require further quantitative and analytical modeling. In 
particular, DOT officials told us the federal interagency effort to 
collect data would need to result in more comprehensive data--including 
volumetric pipeline entry, flow, and exit information[Footnote 24]--as 
well as more reliable storage terminal and inventory data in order to 
more fully assess the current and future reliability of the nation's 
pipeline infrastructure and ability to respond to market disruptions. 

The absence of key data also limits the ability of federal agencies to 
monitor the effect of emerging trends such as the use of biofuels-- for 
example, ethanol--in petroleum product markets. Specifically, we found 
that gaps in federal data do not allow agencies to track where gasoline 
blended with ethanol ultimately winds up in the fuel stream. Not having 
this information may be at odds with consumer's interests. Since, 
according to EPA, a gallon of ethanol contains two-thirds the energy of 
a gallon of gasoline, when gasoline blended with ethanol is sold in 
areas with no ethanol or oxygenate requirement, consumers may be 
purchasing fuel that provides fewer miles-per-gallon without being 
aware of it.[Footnote 25] Our analysis of gasoline sales data shows 
that from 2002 through 2008, conventional gasoline blended with ethanol 
had been sold in areas with no ethanol or other oxygenate mandates in 
at least 32 states.[Footnote 26] Agency and industry officials told us 
that as the volume of biofuels to be blended with gasoline continues to 
grow to 36 billion gallons in 2022, ethanol will increasingly be 
distributed in locations that do not have requirements for oxygenate 
content. 

Despite these gaps in federal data, individual agencies have generally 
continued to take steps to update their data collection surveys to meet 
their respective agency objectives or needs, and have often coordinated 
to more efficiently obtain petroleum product data needed for a variety 
of purposes at multiple agencies. 

* In 2009, EIA began collecting data regarding the production, stocks 
at production facilities, sales for resale, and end-use sales of 
biodiesel fuel.[Footnote 27] Also, three existing EIA forms were 
expanded to collect biodiesel imports and biodiesel blending and stocks 
at terminals and refineries. Our work indicates this new survey will 
help analysts identify how and where biodiesel is being used, a key 
emerging trend in the petroleum industry. In addition, these data will 
be used by EPA to help monitor the volumes of biofuel use specified in 
the RFS. 

* Effective January 2009, EIA consolidated reporting of inventory 
information at refineries, pipelines and terminals from two surveys to 
one. This action will permit a more detailed and reliable analysis of 
petroleum product terminal operations and provide a baseline for the 
volume of petroleum products at various terminal locations that can 
potentially re-supply a city in the event of a major disruption. While 
this partially addresses our need to have federal data that shows the 
re-supply options in the event of a disruption, it neither shows the 
refiner-to-market link nor does it provide detailed batch information 
on petroleum product flows that would facilitate future analyses. 
Comprehensive inventory information may be particularly useful to DOT 
should it be tasked with completing another study to identify potential 
petroleum product infrastructure constraints. 

* EPA officials told us they have worked with the Department of 
Agriculture and DOE in recent years regarding the recently issued 2007 
Renewable Fuels Standard program guidance. The aim of such guidance is 
to monitor biofuel use--a key emerging market trend--and monitor 
compliance with biofuels specified in the RFS. 

Nonetheless, in some cases the individual agency efforts have resulted 
in the collection of information that does not necessarily meet the 
data needs of other agencies or analysts who monitor petroleum product 
markets. For example, federal reporting efforts have evolved such that 
EIA maintains primary responsibility for collecting information on 
total gasoline supply, including gasoline blendstocks, while EPA 
maintains primary responsibility for capturing another key 
characteristic--RVP--of certain gasoline blendstocks. Specifically, 
EIA's surveys are structured to collect data on total gasoline supply, 
including blendstocks, on a monthly basis, whereas EPA collects RVP 
information on each batch of reformulated gasoline on a quarterly 
basis, and for all conventional gasoline supplied by a particular 
refiner on an annual basis. This means that companies report key 
information regarding gasoline components to two different federal 
entities, and analysts who need information regarding the blendstock 
and RVP of gasoline must go to two federal entities to obtain what is 
available; in addition, the data are not comparable in terms of 
periodicity. Finally, as described earlier, three different agencies 
collect a limited amount of pipeline flow data to meet their specific 
agency's objectives, but collectively these data do not allow analysts 
to fully monitor the flow of petroleum product markets. This limited 
not only our ability to identify a city's gasoline re-supply options in 
the event of an outage in this analysis, but also DOT's efforts to 
fully address a Congressional mandate. In sum, these separate pieces of 
data do not come together to form a complete picture of current 
petroleum product markets. 

Conclusions: 

To the extent reasonable, the collection of petroleum product data by 
federal agencies should allow these and other agencies and analysts to 
form a clear picture of U.S. petroleum product markets while minimizing 
the government's costs of collecting and maintaining, as well as the 
costs to industry of providing, these data. In our work we identified 
gaps in public data, some of which we could address by purchasing 
privately collected data, and some of which led to limitations to what 
our analysis could address. Specifically, we were unable, with publicly 
available data, to identify which refiners serve various cities across 
the country, and by extension, which refineries produce special fuel 
blends designed to meet federal, state, and local requirements. While 
the available public data, along with the commercial data we purchased, 
allowed us to analyze the broad impacts of refinery outages on various 
gasoline types on average; during the initial week of the outage, the 
data were not sufficient to determine the effects at individual cities. 
We also found an absence of some data on emerging market trends in 
biofuels that is troubling, given the rapid expansion of biofuel 
production and use in recent years. Some data gaps we identified may 
exist because data collection efforts generally reflect individual 
agency needs and, thus, may not necessarily consider the broader needs 
of other federal agencies and analysts. We recognize that agencies have 
a primary responsibility to perform their individual missions and that 
these agencies face their own specific budgetary constraints. However, 
we note the importance of each agency acknowledging that the collection 
of individual pieces of federal data contributes to a larger data 
universe and taking reasonable steps to ensure that the totality of 
these data allow for meaningful understanding and oversight of 
petroleum markets. In addition, agencies must be conscious of 
efficiency by considering the costs associated with gathering and 
maintaining data. 

Improving the usefulness and completeness of publicly held data--as 
well as reducing the associated costs--will require that each agency be 
aware of the part of the overall data picture it is responsible for, as 
well as the usefulness of these data beyond the immediate agency 
mission. Continued and improved coordination between such agencies, 
including EIA, EPA, DOT, and FERC, could improve the collective 
understanding and oversight of the refining industry and petroleum 
product markets. 

Recommendations for Executive Action: 

To evaluate existing, publicly held petroleum products market data and 
to determine if they are sufficient to meet the current and expected 
future missions and needs of the Congress, federal agencies, and other 
public and private stakeholders, we recommend that the Administrator of 
the EIA convene a panel comprised of agency officials from EIA, EPA, 
DOT, FERC, and other relevant agencies, industry representatives, 
public stakeholders, and other analysts and data users, to collect 
these data and develop a coordinated interagency data collection 
strategy. The panel should: 

* assess the costs and benefits of collecting: 

- more systematic information about which refiners serve which cities 
and: 

- more discrete reporting of the volumetric entry, flow, and exit of 
petroleum products through the pipeline infrastructure system; 

* identify additional data that would be useful to track and evaluate 
emerging market trends--such as the proliferation of biofuels and 
special blends--and assess the costs and benefits of collecting such 
data; 

* identify opportunities to coordinate federal data collection efforts 
so that agencies can respond fully to Congressional requests and meet 
governmentwide data needs to monitor the impact of petroleum product 
market disruptions; and: 

* identify areas where data collection is fragmented--such as multiple 
survey instruments collecting similar information--to determine if 
those efforts can be consolidated and modified to enhance the overall 
usefulness and improve the efficiency of collecting and reporting these 
data. 

Agency Comments and Our Evaluation: 

We provided a draft of this report to the Department of Energy (DOE) 
and its Energy Information Administration (EIA), the Environmental 
Protection Agency (EPA), and the Department of Transportation (DOT) for 
review and comment. DOE's EIA agreed with our recommendations and 
provided additional comments regarding the recommendations and the 
report's discussion of data gaps, which are summarized below. EIA also 
provided technical and clarifying comments, which we incorporated as 
appropriate into the report. EPA and DOT provided only technical 
comments, which we also incorporated as appropriate. 

Regarding our recommendations, EIA stated that it supports the 
recommendations, including our specific suggestions to review data 
adequacy, strengthen interagency coordination of data collection and 
use, and fully engage government, industry and public stakeholders. EIA 
stated that it believes it has a strong program to address all of these 
suggested actions, and is working closely with other federal entities 
through established joint programs, as well as informally to coordinate 
data collection. For example, the agency noted it has been working with 
an interagency group comprised of 40 federal agencies to facilitate the 
development of a trade processing system for U.S. Customs and Border 
Patrol. 

In commenting on the report's discussion of data gaps, EIA stated it 
agrees that a review of possible data gaps is necessary and noted that 
it is currently--as of July 2009--reviewing the adequacy and quality of 
currently collected and commercially available refinery outage 
information. The agency believes, and we agree, that the adequacy of 
refinery outage data for analysis is one that EIA has taken seriously. 
To this end, EIA noted it published Federal Register notices on 
December 9, 2008, and February 28, 2009, informing the public of the 
agency's intended review of refinery outage data. EIA plans to complete 
its review and provide its recommendation regarding additional 
government data collection this fall in its mandated semiannual 
refinery outage study. EIA stated it then plans to publish its 
analysts' assessment and recommendations to solicit the broadest 
possible comment. At that time EIA will consider the use of a panel of 
government, industry, and public stakeholders--as we suggested--to 
determine its future steps. We support EIA's efforts to address data 
issues and believe that its current plans are a step in the right 
direction toward ensuring that the best data are available to help 
achieve its mission of producing independent and unbiased research to 
help the Congress, public, and international community better 
understand energy markets and promote sound policy-making. 

We are sending copies of this report to interested Congressional 
committees; the Administrator of the Energy Information Administration, 
the Administrator of the Environmental Protection Agency; the Secretary 
of the Department of Transportation; and other interested parties. This 
report also will be available at no charge on the GAO Web site at 
[hyperlink, http://www.gao.gov]. 

If you or your staffs have any questions concerning this report, please 
contact me at (202) 512-3841 or ruscof@gao.gov. Contact points for our 
Offices of Congressional Relations and Public Affairs may be found on 
the last page of this report. Major contributors to this report are 
acknowledged in appendix III. 

Signed by: 

Frank Rusco: 
Director, Natural Resources and Environment: 

List of Requesters: 

The Honorable Charles E. Schumer: 
Vice Chairman: 
Joint Economic Committee: 
United States Senate: 

The Honorable Christopher J. Dodd: 
United States Senate: 

The Honorable Byron L. Dorgan: 
United States Senate: 

The Honorable Joseph I. Lieberman: 
United States Senate: 

The Honorable Joseph Courtney: 
House of Representatives: 

The Honorable Rosa DeLauro: 
House of Representatives: 

The Honorable John B. Larson: 
House of Representatives: 

The Honorable Christopher S. Murphy: 
House of Representatives: 

[End of section] 

Appendix I: Scope and Methodology: 

We addressed the following questions during our review: (1) How have 
refinery outages affected U.S. wholesale gasoline prices since 2002? 
(2) To what extent do available federal data allow for the evaluation 
of the impacts of refinery outages on petroleum product prices, and do 
these data reflect emerging trends in petroleum product markets that 
may be important to future analytical needs? For the purposes of this 
report, we define the various types of outages as follows: 

* Planned outages are periodic shutdowns of one or more refinery 
processing units or possibly the entire refinery to perform 
maintenance, inspection, and repair of equipment or to replace process 
materials and equipment that have worn out or broken, in order to 
ensure safe and efficient operations. 

* Unplanned outages are events where an entire unit or refinery must be 
brought down immediately and without advance notice and are caused by 
unplanned circumstances such as a fire or a power outage. 

To determine trends in refinery outages over the time period from 2002 
through September 2008, we purchased data from Industrial Information 
Resources, Inc. (IIR) that contained detailed information on refinery 
outages, including the estimated dates of the outages, whether the 
outages were planned or unplanned, and the amount of reduced production 
capacity due to each outage. We evaluated the data and found they 
provide reliable estimates of outages from 2002 onward. In our 
analysis, we counted an outage event as the halting of production 
capacity on any piece of equipment at the refinery; where multiple 
units such as a crude distillation and one or more secondary processing 
units were simultaneously down, we counted this as a single outage 
event in our model.[Footnote 28] 

To evaluate how refinery outages have affected U.S. wholesale gasoline 
prices we obtained and analyzed data from Energy Information 
Administration (EIA)'s monthly refinery production survey form, EIA- 
810, from 2002 through 2006, and other EIA surveys. We also purchased 
(1) data that included detailed information on refinery outages between 
2002 and 2008 from Industrial Information Resources, Inc. (IIR), a 
private company that provides research and forecasts for various large 
industries; (2) data estimating the quantity flows of gasoline and 
other petroleum products produced at most U.S. refineries' and 
transported to those U.S. cities that make up the main markets for 
these products from Baker & O'Brien, an energy consultancy company 
whose software is licensed to 9 of the 10 top U.S. refining companies; 
and (3) weekly wholesale price data for 75 U.S. cities gasoline markets 
from Oil Price Information Service, a private company that provides 
pricing and other data at the wholesale or "rack" level. We determined 
that these data were sufficiently reliable for the purposes of this 
report. We used the Baker & O'Brien quantity flow estimates to measure 
the proportion of each city's product that is generally supplied by a 
particular refinery. We developed, and extensively tested, an 
econometric model that examined the statistical relationship between 
refinery outages and gasoline prices. We limited our analysis to 
outages that (1) were determined to be of the largest 60 percent within 
their market region and that lasted at least 3 days, (2) had a 
corresponding market city in the Baker & O'Brien data, and (3) for 
which we had useful and complete gasoline price data at the wholesale 
terminal level. In our model, we limited the effect of an outage on 
prices to one week, after which time we assumed that petroleum products 
were supplied from an alternate source. As a result, our analysis 
evaluated the short-term effects of outages but did not evaluate the 
length of time those effects occurred. In our model, we incorporated 
data on numerous factors that could affect gasoline prices--such as 
gasoline inventory levels and gasoline specifications--in order to rule 
out, or "control" for their effects on prices. Because we were able to 
control for these other factors, we believe we were able to isolate the 
impacts of outages on prices given the inherent issues with the various 
data sets. There were some factors that potentially affected gasoline 
prices over time and city-specific information that we could not 
include, although we were able to use econometric techniques to control 
for some of these factors.[Footnote 29] After controlling for the 
additional factors that affected gasoline prices, we were able to 
estimate the average impact of outages on wholesale gasoline prices. 
The statistical significance of our findings are noted throughout the 
report. Although we focused our study on wholesale prices, we cannot be 
certain that the price effects at the retail level would be the same, 
although some research has shown that wholesale price changes are 
generally passed on to the retail level.[Footnote 30] In developing our 
model, we consulted with a number of economists and incorporated their 
suggestions wherever possible. Finally, we performed an analysis to 
test the robustness of our model, including changing various 
assumptions regarding the model in order to ensure that our results 
were not highly dependent on any single specification of the model. 

To assess the extent to which available federal data allow for the 
evaluation of the impacts of refinery outages and determine whether the 
data reflect emerging trends in petroleum product markets, we reviewed 
data collection instruments from federal agencies--including EIA, the 
Environmental Protection Agency (EPA), the Federal Energy Regulatory 
Commission (FERC), and the Department of Transportation (DOT)--and 
reviewed them for comprehensiveness, utility, accessibility, and 
potential gaps or limitations. In addition, we reviewed past GAO and 
other federal agency or intergovernmental agency studies on refined 
product markets to identify data gaps, limitations, or inconsistencies. 
Finally, we interviewed key industry, expert institution, and academic 
representatives regarding data limitations and utility in their own 
work and what other data concerns or needs they might have for future 
analyses. Our work was not a comprehensive evaluation of all federal 
energy data, but rather, an assessment of key data GAO used in this and 
past reports, and select other data that were determined during the 
course of our review to have posed limitations for GAO's or other 
agencies' evaluations of important policy questions. 

We conducted this performance audit from October 2008 through July 2009 
in accordance with generally accepted government auditing standards. 
Those standards require that we plan and perform the audit to obtain 
sufficient, appropriate evidence to provide a reasonable basis for our 
findings and conclusions based on our audit objectives. We believe that 
the evidence obtained provides a reasonable basis for our findings and 
conclusions based on our audit objectives. 

[End of section] 

Appendix II: GAO's Quantitative Methodology for Determining Impacts of 
Refinery Outages on Wholesale Prices: 

Introduction: 

We developed an econometric model to explain the impact of refinery 
outages on gasoline prices. Our model controlled for as many 
contributing factors as possible, however, there were not always 
sufficient data available to control for all possible factors affecting 
wholesale gasoline prices. Our model examined how wholesale gasoline 
city rack prices were affected in the week during which a large 
unplanned refinery outage occurred. 

Econometric Model Specification and Methodology: 

We examined weekly average data on wholesale city rack gasoline prices. 
We used data from 75 wholesale city racks from January 2002 through 
September 2008. We believe that the increased information from higher 
frequency data--for example, by using daily data--would be outweighed 
by the extra noise generated by such relatively high frequency data. 
Further, using lower frequency data, such as monthly data, runs the 
risk of obscuring some of the less extended but important effects of 
unplanned outages on gasoline prices. Another limitation of our 
analysis is that, in some cases, our data series for the control 
variables, described below, are generally available only on a monthly 
basis,[Footnote 31] in which case these values are assigned to the 
corresponding weekly observations. We consulted with government and 
academic experts to help develop our econometric model. 

The Dependent Variable-Wholesale Gasoline Price: 

Our variable of interest was the price of gasoline, specifically the 
wholesale rack price of gasoline. 

* Our dependent variable was the logarithm of the wholesale city rack 
price of gasoline. Note that we include a time dummy variable for every 
time period so we do not have to deflate the wholesale price by a price 
index such as the producer price index or the price of crude.[Footnote 
32] We used an Augmented-Dickey-Fuller test designed for panel data to 
test for stationarity in levels of our dependent variables, in the case 
of both unbranded and branded prices.[Footnote 33] Our tests showed 
that our unbranded and branded dependent variable was stationary in 
levels.[Footnote 34] 

* We examined separate models for unbranded and branded products to 
test for the consistency of our results. 

* There may be multiple gasoline prices reported for a given city rack 
on a given date. In general, we used the wholesale rack price of 
gasoline that is required in that specific city because we were 
interested in determining whether areas with non-standard gasoline 
specifications[Footnote 35] experienced larger gasoline price increases 
when a refinery that supplied their particular specification had an 
outage. 

* By including a complete set of time dummy variables-one for each 
week's observation in the data-our model controlled for factors that 
vary only over time (and are invariant across cities), such as the 
national average price level, the price of crude oil, and seasonal 
effects. 

Explanatory Variables--Measuring the Impact of an Outage on Gasoline 
Prices: 

Our primary interest was to examine the impact of refinery outages on 
gasoline prices. There are two key issues: 

1. Identifying an outage. We acquired data on outage occurrences from 
IIR. These data provide information about the outage, including whether 
the outage was planned or unplanned, the date of the outage, the 
duration of the outage, and the capacity of the unit that was offline 
due to the outage. 

2. Measuring the impact of a given outage on a particular city. For 
each city, we estimated the proportion of its product that it generally 
received from each refinery; a city may be served by one or more 
refineries. 

Our measure of an outage's impact was the proportion of a city's 
product that was generally supplied by the refinery (or refineries) 
experiencing an outage. If a city was generally estimated to receive no 
product from the refinery experiencing the outage, then the effect was 
zero, the explanatory variable was zero, and the refinery outage had no 
impact on that city's gasoline price. Alternatively, if, for example, a 
city received 20 percent of its product from said refinery, the 
explanatory variable had a value of 0.20 for that time period. It is 
also possible that a single city may have been impacted by more than 
one refinery outage at the same time, so in that case we would sum 
these effects. For instance, if in addition to the 20 percent impact 
example above, there was an outage at a refinery supplying 10 percent 
of the city's product, the explanatory variable would take a combined 
value of 0.30.[Footnote 36] 

Other Explanatory Variables: 

In addition to the impact of outages, our model includes other 
important variables that may influence the price of gasoline. 

* Volume of inventory of gasoline relative to the volume of sales of 
gasoline. This could affect the availability of gasoline at the 
wholesale level and hence affect prices. Prices should decrease when 
inventories are high relative to sales and should rise when inventories 
are low relative to sales. However, inventories and sales may 
themselves respond to changes in wholesale gasoline prices, so this 
variable may be endogenous. 

* Capacity utilization rate. This could affect the wholesale price of 
gasoline through changes in the availability of gasoline product. One 
possibility is that, when utilization rates are high, there would be 
more gasoline available, which would tend to lower prices; conversely 
if utilization rates are low, less gasoline would be available, which 
would tend to raise prices. However, it is possible that as utilization 
rates approach very high levels, there are significant increases in 
cost of production, which could then result in higher prices. Further, 
capacity utilization may react to changes in gasoline prices, so it is 
possible that this variable is endogenous. 

* Market concentration. Markets with fewer sellers of product or that 
are more highly concentrated, may be associated with higher gasoline 
prices. However, the direction of effect may run the other way too, 
such that markets with higher prices may attract entrants, which may 
reduce the level of market concentration. We treat market concentration 
as an endogenous variable. 

* Lagged dependent variable. Our model includes lagged values of the 
left hand side variable; namely, the logarithm of the wholesale price 
of gasoline. Gasoline price data may be serially correlated and it is 
reasonable to include the effect of past gasoline prices on current 
gasoline prices. 

* Time fixed effects. We included a dummy variable for each time period 
in the analysis. 

* City fixed effects. We included a dummy variable for each city in the 
analysis. These city fixed effects may assist in controlling for 
unobserved heterogeneity. 

* Product specification. We included a dummy variable for each of the 
different types of gasoline used in our model. 

* Interaction between the product specification dummy variables and the 
outage impact variable. We included a set of interaction terms to test 
whether cities that with special fuel requirements experience higher 
price increases due to outages. 

Econometric Model Specification: 

Our fixed effects model can be written as follows: 

yit = (xit, wit)B + ci + fi + uit, i = 1,2,...,N; t = 1,2,...,T (1), 

where: 

yit is the logarithm of wholesale rack gasoline price at city i in week 
t. 

xit is a vector of predetermined variables for city i in week t that 
are assumed to be independent of our error term, uit, including a 
lagged value of our dependent variable. 

wit is a vector of possibly endogenous variables at city i in week t. 

ci is the fixed effect or dummy variable for city i. 

ft is the fixed effect of dummy variable for week t. 

B is a vector of parameters to be estimated. 

Our key outage effect variable measures the percent of a city's product 
supply affected by an outage; that is: 

[Refer to PDF for formula] 

where Outageir't is equal to 1 when an outage occurs at time t in the 
r'-th refinery that serves the i-th city, and the remaining term is the 
proportion of product provided by that refinery to that city. When 
there is no outage, Outageir't is equal to zero. Thus, this variable 
measures a city's reduction in product due to an outage (or outages). 
In the extreme case, there may be a single refinery that supplies 100 
percent of a city's product, in which case the impact on product of an 
outage at that refinery on that city would be large, with a concomitant 
effect on that city's gasoline prices. 

The outage impact may also have varied according to the type of fuel. 
The variable, sirt measures the percentage of supply of product that 
was interrupted; it may not account completely for the difficulty in 
finding a replacement for that product. If a city used a fuel that is 
commonly produced, such as conventional clear gasoline, it would likely 
be more straightforward to find an alternative source of supply. 
However, if the city uses a special fuel, it may be more difficult to 
find an alternative refinery to supply that product. Therefore, in 
addition to a set of dummy variables for each fuel specification, we 
included a set of interaction terms of our outage supply affect 
variable with each of the fuel specification dummy variables. 

* We used xtiverg2 in STATA.[Footnote 37] The xtivreg2 estimation 
procedure allowed us to estimate standard errors that are robust to 
heteroskedasticity and autocorrelation. 

* We estimated the model using the logarithm of price as the dependent 
variable. Note that because we have time dummies, we do not need to 
control for crude oil prices or price inflation because these variables 
are invariant across cities for a given time period and so are 
collinear with the time dummies. Our specification necessarily subsumes 
the impact those variables that only vary over time and not vary across 
cities. 

* We used a C-statistic test to ascertain whether the inventory-sales 
ratio and the capacity utilization rate should be treated as endogenous 
or exogenous.[Footnote 38] In the case of both the unbranded gasoline 
prices and the branded gasoline price models, our test could not reject 
the null hypothesis that these variables were exogenous. 

* Measures of market concentration, such as the Hirschman Herfindahl 
Index (HHI), have been shown to be endogenous,[Footnote 39] so we 
tested for whether it was exogenous and use two-stage least squares 
when appropriate, using merger events as instruments. We used a C- 
statistic to test for the exogeneity of the spot market HHI. In the 
case of the unbranded gasoline price model, the test rejected the null 
hypothesis of exogeneity. In the case of the branded price model, the 
test could not reject the null hypothesis of exogeneity. We estimated 
both models treating the spot market HHI as endogenous, which we 
recognize might be a less efficient estimator but is nevertheless a 
consistent estimator. 

* We used Hansen's J-statistic to test for over-identification of our 
instruments; namely, that they should be correlated with the 
regressors, but uncorrelated with the regression errors.[Footnote 40] 
In every case, the J-statistic accepted the null hypothesis that our 
instruments were valid. 

* We estimated separate models for unplanned and planned outages. While 
unplanned outages can be reasonably viewed as exogenous--random-- 
events, planned outages need to be scheduled more than a year in 
advance and may be scheduled to coincide with time periods of typically 
lower seasonal demand. Therefore, we believe it was appropriate to 
model these two types of outages separately. 

* We estimated separate models for unbranded prices and branded prices. 

* We estimated the model (1) except that we dropped those observations 
where waivers were in effect. 

Table 3: Data Used In Our Econometric Model: 

Data Sources: 

Variable: Prices; 
Description: Wholesale gasoline price in cents-per-gallon. Branded and 
unbranded. Weekly averages; 
Source: OPIS. 

Variable: Spot market HHI; 
Description: Market concentration, measured by refinery capacity of 
corporations in each spot market. Monthly data; 
Source: EIA, GAO analysis. 

Variable: Merger dummy variables; 
Description: Dummy variable equal to 1 from the effective date of the 
merger. Equal to 0 before the effective date of the merger. Similarly 
for announced dates of each of the mergers; 
Source: OPIS, IHS Herold. 

Variable: Inventory-sales ratio; 
Description: Ratio of total motor gasoline inventories to finished 
gasoline product supplied. Monthly data at the PADD level; 
Source: EIA. 

Variable: Capacity utilization rate; 
Description: Capacity utilization rate. Monthly data at the PADD level; 
Source: EIA. 

Variable: Fuel type dummy variables; 
Description: Set of dummy variables for the gasoline fuel type. Details 
the main fuel type, presence of additives and RVP. Weekly data; 
Source: OPIS. 

Variable: Employment growth; 
Description: Percentage growth in employment at the state level. 
Monthly data; 
Source: Department of Labor. 

Variable: Unemployment rate; 
Description: Percentage unemployment rate at the state level. Monthly 
data; 
Source: Department of Labor. 

Variable: Real personal income growth; 
Description: Percentage growth in personal income at the state level 
deflated by the consumer price index. Quarterly data; 
Source: BEA. 

Variable: Consumer price index; 
Description: Consumer price index. Monthly data; 
Source: Department of Labor. 

Variable: Percent of general product supply affected by the outage; 
Description: Proportion of the usual amount of supply affected by an 
outage or outages. Quarterly data; 
Source: IIR for outage events. Baker & O'Brien for determining the 
amount of product generally supplied by each refinery to each city. 

Source: Baker & O'Brien, BEA, EIA, IHS Herold, IIR, Department of 
Labor, OPIS and GAO analysis. 

[End of table] 

Table 4: Regression Results for Effect of Unplanned Outages on 
Unbranded Gasoline Prices--Dependent Variable is the Logarithm of 
Unbranded Gasoline Price: 

Spot market HHI; 
Coefficient: 0.26751; 
Standard error: 0.08108[A]. 

Inventory-to-sales ratio; 
Coefficient: -0.00177; 
Standard error: 0.00168. 

Utilization as a % of operating capacity; 
Coefficient: -0.00033; 
Standard error: 0.00006[A]. 

Log of unbranded gasoline price lagged 1 period; Coefficient: : 
0.85919; Standard error: : 0.00573[A]. 

Fuel types: % supply reliance (% reliance on refinery experiencing 
outage); 
Coefficient: 0.00014; 
Standard error: 0.00004[A]. 

Fuel types: 

CBG; 
Coefficient: 0.00632; 
Standard error: 0.00391. 

CBG with 10% ethanol; 
Coefficient: 0.00956; 
Standard error: 0.00277[A]. 

CARB with 5.7% ethanol; 
Coefficient: 0.00871; 
Standard error: 0.00322[A]. 

CARB with 2% MTBE; 
Coefficient: 0.00681; 
Standard error: 0.00342[B]. 

Conventional RVP 7.0; 
Coefficient: 0.01048; 
Standard error: 0.00195[A]. 

Conventional RVP 7.8; 
Coefficient: 0.00314; 
Standard error: 0.00106[A]. 

Conventional RVP 9.0; 
Coefficient: 0.00343; 
Standard error: 0.00094[A]. 

Conventional 5.7% ethanol; 
Coefficient: 0.00903; 
Standard error: 0.00410[B]. 

Conventional 7.7% ethanol; 
Coefficient: 0.00465; 
Standard error: 0.00189[B]. 

Conventional 7.7% ethanol RVP 9.0; 
Coefficient: 0.01727; 
Standard error: 0.00205[A]. 

Conventional 10% ethanol; 
Coefficient: 0.00532; 
Standard error: 0.00116[A]. 

Conventional 10% ethanol RVP 7.0; 
Coefficient: 0.00894; 
Standard error: 0.00263[A]. 

Conventional 10% ethanol RVP 7.8; 
Coefficient: 0.00818; 
Standard error: 0.00176[A]. 

Conventional 10% ethanol RVP 9.0; 
Coefficient: 0.00680; 
Standard error: 0.00154[A]. 

Low sulfur; 
Coefficient: 0.00374; 
Standard error: 0.00191[B]. 

Low sulfur RVP 7.0; 
Coefficient: 0.00822; 
Standard error: 0.00239[A]. 

RFG 10% ethanol; 
Coefficient: 0.01572; 
Standard error: 0.00416[A]. 

RFG MTBE; 
Coefficient: 0.01433; 
Standard error: 0.00400[A]. 

Fuel types interacted with % supply reliance: 

CBG interaction with % supply reliance; 
Coefficient: 0.00007; 
Standard error: 0.00025. 

CBG with 10% ethanol interaction with % supply reliance; 
Coefficient: -0.00010; 
Standard error: 0.00007. 

CARB with 5.7% ethanol interaction with % supply reliance; 
Coefficient: 0.00009; 
Standard error: 0.00013. 

CARB with 2% MTBE interaction with % supply reliance; 
Coefficient: 0.00123; 
Standard error: 0.00049[B]. 

CARB without oxygenate interaction with % supply reliance; 
Coefficient: 0.00430; 
Standard error: 0.00033[A]. 

Conventional RVP 7.0 interaction with % supply reliance; 
Coefficient: -0.00026; 
Standard error: 0.00017. 

Conventional RVP 7.8 interaction with % supply reliance; 
Coefficient: 0.00000; 
Standard error: 0.00007. 

Conventional RVP 9.0 interaction with % supply reliance; 
Coefficient: 0.00001; 
Standard error: 0.00006. 

Conventional 5.7% ethanol interaction with % supply reliance; 
Coefficient: 0.00135; 
Standard error: 0.00033[A]. 

Conventional 7.7% ethanol interaction with % supply reliance; 
Coefficient: 0.00009; 
Standard error: 0.00013. 

Conventional 7.7% ethanol RVP 9.0 interaction with % supply reliance; 
Coefficient: -0.00033; 
Standard error: 0.00006[A]. 

Conventional 10% ethanol interaction with % supply reliance; 
Coefficient: -0.00001; 
Standard error: 0.00013. 

Conventional 10% ethanol RVP 7.0 interaction with % supply reliance; 
Coefficient: 0.00108; 
Standard error: 0.00041[A]. 

Conventional 10% ethanol RVP 7.8 interaction with % supply reliance; 
Coefficient: -0.00009; 
Standard error: 0.00012. 

Conventional 10% ethanol RVP 9.0 interaction with % supply reliance; 
Coefficient: 0.00004; 
Standard error: 0.00011. 

Low sulfur interaction with % supply reliance; 
Coefficient: 0.00084; 
Standard error: 0.00046[C]. 

Low sulfur RVP 7.0 interaction with % supply reliance; 
Coefficient: 0.00012; 
Standard error: 0.00024. 

RFG 10% ethanol interaction with % supply reliance; 
Coefficient: 0.00015; 
Standard error: 0.00007[B]. 

RFG MTBE interaction with % supply reliance; 
Coefficient: -0.00003; 
Standard error: 0.00012. 

R-squared; 0.996. 

J-test for over-identification (signif. level); 18.8%. 

Observations; 26325. 

Number of cities; 75. 

Source: GAO analysis of various data sources. 

Notes: The standard error estimates are robust to heteroskedasticity 
and autocorrelation. The regression model included fixed effects for 
cities and time dummies for each week of data. The model is estimated 
using two-stage least squares, treating the spot market HHI as 
endogenous. The fuel dummy variable for CARB without any MTBE or 
ethanol was not estimated due to collinearities. 

See table 3 for a list of data sources used in this table. 

[A] significant at the 1 percent level. 

[B] significant at the 5 percent level. 

[C] significant at the 10 percent level. 

[End of table] 

Table 5: Regression Results for Effect of Unplanned Outages on Branded 
Gasoline Prices--Dependent Variable is the Logarithm of Branded 
Gasoline Price: 

Spot market HHI; 
coefficient: 0.17803; 
standard error: 0.05804[A]. 

Inventory-to-sales ratio; 
coefficient: -0.00057; 
standard error: 0.00452. 

Utilization as a % of operating capacity; 
coefficient: -0.00020; 
standard error: 0.00004[A]. 

Log of branded gasoline price lagged 1 period; 
coefficient: 0.90106; 
standard error: 0.00452[A]. 

% supply reliance (% reliance on refinery experiencing outage); 
coefficient: 0.00006; 
standard error:0.00002[A]. 

Fuel types: 

CBG; 
coefficient: 0.00784; 
standard error: 0.00784. 

CBG with 10% ethanol; 
coefficient: 0.00429; 
standard error: 0.00460. 

CARB with 5.7% ethanol; 
coefficient: -0.00681; 
standard error: 0.00224[A]. 

CARB with 2% MTBE; 
coefficient: 0.00556; 
standard error: 0.00137[A]. 

Conventional RVP 7.8; 
coefficient: 0.00142; 
standard error: 0.00073[B]. 

Conventional RVP 9.0; 
coefficient: 0.00256; 
standard error: 0.00066[A]. 

Conventional 5.7% ethanol; 
coefficient: 0.01362; 
standard error: 0.00333[A]. 

Conventional 7.7% ethanol; 
coefficient: 0.00219; 
standard error: 0.00139. 

Conventional 7.7% ethanol RVP 9.0; 
coefficient: 0.00945; 
standard error: 0.00260[A]. 

Conventional 10% ethanol; 
coefficient: 0.00349; 
standard error: 0.00071[A]. 

Conventional 10% ethanol RVP 7.0; 
coefficient: 0.00798; 
standard error: 0.00208[A]. 

Conventional 10% ethanol RVP 7.8; 
coefficient: 0.00350; 
standard error: 0.00094[A]. 

Conventional 10% ethanol RVP 9.0; 
coefficient: 0.00602; 
standard error: 0.00094[A]. 

Low sulfur; 
coefficient: 0.00389; 
standard error: 0.00181[B]. 

Low sulfur RVP 7.0; 
coefficient: 0.00582; 
standard error: 0.00199[A]. 

Low sulfur RVP 9.0; 
coefficient: 0.00083; 
standard error: 0.00724. 

RFG 10% ethanol; 
coefficient: 0.02147; 
standard error: 0.00916[B]. 

RFG MTBE; 
coefficient: 0.02142; 
standard error: 0.00910[B]. 

Fuel types interacted with % supply reliance: 

CBG interaction with % supply reliance; 
coefficient: -0.00024; 
standard error: 0.00014[C]. 

CBG with 10% ethanol interaction with % supply reliance; 
coefficient: -0.00044; 
standard error: 0.00016[A]. 

CARB with 5.7% ethanol interaction with % supply reliance; 
coefficient: 0.00022; 
standard error: 0.00013[A]. 

CARB with 2% MTBE interaction with % supply reliance; 
coefficient: -0.00007; 
standard error: 0.00037. 

Conventional RVP 7.0 interaction with % supply reliance; 
coefficient: -0.00008; 
standard error: 0.00010. 

Conventional RVP 7.8 interaction with % supply reliance; 
coefficient: -0.00010; 
standard error: 0.00004[A]. 

Conventional RVP 9.0 interaction with % supply reliance; 
coefficient: -0.00004; 
standard error: 0.00005. 

Conventional 5.7% ethanol interaction with % supply reliance; 
coefficient: 0.00143; 
standard error: 0.00506. 

Conventional 7.7% ethanol interaction with % supply reliance; 
coefficient: -0.00002; 
standard error: 0.00021. 

Conventional 7.7% ethanol RVP 9.0 interaction with % supply reliance; 
coefficient: -0.00004; 
standard error: 0.00028. 

Conventional 10% ethanol interaction with % supply reliance; 
coefficient: 0.00006; 
standard error: 0.00010. 

Conventional 10% ethanol RVP 7.0 interaction with % supply reliance; 
coefficient: 0.00045; 
standard error: 0.00048. 

Conventional 10% ethanol RVP 7.8 interaction with % supply reliance; 
coefficient: -0.00001; 
standard error: 0.00005. 

Conventional 10% ethanol RVP 9.0 interaction with % supply reliance; 
coefficient: 0.00014; 
standard error: 0.00007[A]. 

Low sulfur interaction with % supply reliance; 
coefficient: 0.00040; 
standard error: 0.00031. 

Low sulfur RVP 7.0 interaction with % supply reliance; 
coefficient: 0.00004; 
standard error: 0.00009. 

RFG 10% ethanol interaction with % supply reliance; 
coefficient: 0.00015; 
standard error: 0.00005[A]. 

RFG MTBE interaction with % supply reliance; 
coefficient: -0.00013; 
standard error: 0.00005[A]. 

R-squared: 0.997. 

J-test for over-identification (signif. level): 19.2%. 

Observations: 26325. 

Number of cities: 75. 

Source: GAO analysis of various data sources. 

Notes: The standard error estimates are robust to heteroskedasticity 
and autocorrelation. The regression model included fixed effects for 
cities and time dummies for each week of data. The model is estimated 
using two-stage least squares, treating the spot market HHI as 
endogenous. The fuel dummy variable for CARB with MTBE, and the 
interaction variable between the fuel dummy for low-sulfur with 9.0 RVP 
and outage variable was not estimated, due to collinearities. 

See table 3 for a list of data sources used in this table. 

[A] significant at the 1 percent level. 

[B] significant at the 5 percent level. 

[C] significant at the 10 percent level. 

[End of table] 

* We found unplanned outages were significantly associated with an 
increase in unbranded gasoline prices. We found this impact is 
generally positive with respect to the price of all fuels. We further 
found this impact is significantly greater than the comparative or base 
effect (measured relative to the effect on conventional clear) for 
several special fuels. 

* In addition, we found unplanned outages were significantly associated 
with an increase in branded gasoline prices but the effect was smaller 
than for unbranded prices. This impact is generally positive with 
respect to the price of all fuels. There is also evidence that the 
impact is greater for some special fuels although in fewer cases 
compared to the unbranded price results. 

* Our results using planned outages to explain prices found no general 
statistically significant impact on gasoline product prices, either 
branded or unbranded. 

* We found no substantive difference in our results for outage effects 
when we estimated the model (1) without those observations where 
waivers were in effect. 

Limitations of Our Econometric Model and Data: 

Cities included in model. Our selection of cities was based on data 
availability from the Baker & O'Brien data. There are in excess of 350 
city wholesale racks in the U.S. however: 

* The Baker & O'Brien data contain data on only 89 cities, and only 75 
of those had complete series data that we could use for our model. 

* The Baker & O'Brien cities comprise the most important city racks. 

* Treating each of the 350 city racks as independent rack markets may 
not be appropriate. Rather, we can obtain a national picture by 
selecting the most important cities as per the list of cities in the 
Baker & O'Brien data. 

Time period of analysis. We selected January 2002 through September 
2008 because we deemed the IIR data to provide reliable information 
from 2002 onward. 

Gasoline type. The gasoline data from OPIS were selected so as to 
generally reflect the type of gasoline that would be sold in a city, 
given its local fuel regulations. In most cases, we were able to assign 
prices accordingly but in some cases other types of fuel were used in 
the data. However, in the regression model, we control for whatever 
fuel type we did use. 

Outages data. We believe the outages data from IIR are fairly 
comprehensive but there are no federal requirements for refineries to 
report outages or an effort by the federal government to collect these 
data on a national basis. Consequently, some outages may not appear in 
the data, though it is unlikely that any major outages were missed 
during our study period. Further, we limited inclusion of outages to 
those that were at least 3 days in duration and ranked in the top 60 
percent in terms of recorded capacity offline for a refinery's market 
region (as defined by IIR). Thus, we do not include every single outage 
but we have a broad geographic range of the largest outages in the US. 

Linkage data between refineries and cities they serve. The Baker & 
O'Brien data has the following limitations: 

* They are quarterly estimates of product flows and costs. These data 
are intended to be reasonably reflective of actual product dispersion 
across the United States. However, in the course of our analysis we had 
to interpolate some missing data and to extrapolate our data beyond the 
end-points of the available data. 

* The Baker & O'Brien data did not always contain complete data for the 
particular fuel that regulations required be used in that city. In some 
cases, seasonal variations in fuel requirements, such as RVP or 
oxygenate blending specifications, meant a precise match was not 
possible. However, in general, we were able to match the Baker & 
O'Brien fuel with these regulations. 

Frequency of data. Except for our weekly wholesale gasoline price data, 
our other data were either monthly or quarterly, so we had to parse out 
the lower frequency observations accordingly. 

Geographic level of analysis. Our analysis was performed at the city 
level, but some of the data we used were at a more aggregated 
geographic level. We used capacity utilization and inventory-sales 
ratio data at the PADD level. We did not have a measure of city-level 
sales data to determine the size of inventories relative to a local 
market, nor is there a relevant measure of capacity utilization at city-
level, therefore, PADD-level data were used.[Footnote 41],[Footnote 42] 

Economic indicators. Employment growth, personal income growth rate, 
and the unemployment rate were available at the state-level only. 

Market concentration. Our measure of market concentration was an HHI 
measured using corporate refinery capacity at the spot market level. It 
is possible in some cases these measures were too highly aggregated and 
the control variables were less precise than would be ideal. 

Number of outages. We did not take account of multiple outages at the 
same refinery on the same day-we simply established whether an outage 
occurred in a particular week, at a particular refinery. Although the 
size of the outage determined whether it was included in our analysis, 
the impact is treated the same regardless of how large an impact on the 
refinery the outage had. 

Effects of an outage over time. We did not attempt to include dynamic 
effects of outages on prices in our model. We assigned an effect of 
outage in the same time period (week), after which time our model 
implicitly assumed that the product was supplied from an alternate 
source. 

Planned outages. We did not model planned outages in any detail. These 
planned events by definition, did not generally give rise to surprise 
reductions in product supply. Hence, vendors had the opportunity to 
plan ahead and make arrangements to receive alternative sources of 
product. However, we did estimate an analogous model to equation (1) 
and found no significant impact on prices. 

Inventories. Inventories included those domestic and customs-cleared 
foreign stocks held at, or in transit to, refineries and bulk 
terminals, and stocks in pipelines. 

Gasoline sold outside the city rack: Our analysis does not account for 
gasoline that is not sold at the city rack. It is possible that 
significant transactions occur elsewhere that may affect the general 
wholesale market for a particular city. 

Examining wholesale prices, not retail prices. Our analysis is at the 
wholesale price level and the ramifications for retail prices are 
unclear. The effect on retail prices would depend upon the extent to 
which wholesale price changes are passed onto the retail sector. 

Seasonal effects. Our model included of a set of time dummy variables, 
which account for variation in prices due to seasonal effects. A more 
complete model might have contained specific seasonal effects such as a 
set of monthly dummy variables, interacted with the outage effect, and 
also with each special fuel type. This would have allowed us to 
determine whether outages had a differential impact on prices, 
according to the time of year and the fuel type. However, data 
limitations precluded a comprehensive evaluation of such effects; 
specifically, this would have required us to include more than 200 
additional explanatory variables (number of seasonal dummies times the 
number of special fuel types). 

[End of section] 

Appendix III: GAO Contact and Staff Acknowledgments: 

GAO Contact: 

Frank Rusco, (202) 512-3841 or ruscof@gao.gov: 

Staff Acknowledgments: 

In addition to the individual named above, Shea Bader, Divya Bali, 
Benjamin Bolitzer, Dan Haas, Michael Kendix, Rob Marek, Michelle Munn, 
Alison O'Neill, Rebecca Sandulli, Benjamin Shouse, and Barbara 
Timmerman made key contributions to this report. 

[End of section] 

Footnotes: 

[1] GAO, Energy Markets: Refinery Outages Can Impact Petroleum Product 
Prices, but No Federal Requirements to Report Outages Exist, 
[hyperlink, http://www.gao.gov/products/GAO-09-87], (Washington, D.C.: 
Oct. 7, 2008). 

[2] Department of Energy, Energy Information Administration, Refinery 
Outages: Description and Potential Impact on Petroleum Product Prices 
(Washington, D.C., March 2007). 

[3] Although these commercial data are estimates, we have a large 
number of observations across a large number of cities, and therefore 
believe our results are a reasonable estimation of the averaged effects 
across our study period. 

[4] We analyzed about 1,100 observations of unplanned outages and about 
1,200 observations of planned outages. 

[5] We recognize that in many cases, the price effect may extend beyond 
the one-week period analyzed in our study. The amount of time the price 
effect, if any, can be expected to endure depends upon a number of 
factors, including the length of time it takes to re-supply the market 
from an alternative source. 

[6] The results we report in the body of this report are all 
statistically significant at the 10 percent level or better. 

[7] According to EIA, finished gasoline includes conventional gasoline, 
all types of oxygenated gasoline, including gasohol and reformulated 
gasoline, and gasoline comprised of a blendstock with an oxygenate such 
as ethanol, which has been blended to satisfy emissions and other 
federal standards. 

[8] Ethanol corrodes pipelines and may attract water, which renders 
pipelines unable to transport petroleum products without investing in 
costly upgrades. 

[9] [hyperlink, http://www.gao.gov/products/GAO-09-87]. 

[10] In some situations, a planned outage may last longer than 
expected, which EIA sources stated might cause the planned outage to, 
in essence, become an unplanned outage. A planned outage that extends 
beyond the announced or expected window is different from an unplanned 
outage. For example, experienced market operatives will know that such 
announcements are in general, the best estimates of the duration of an 
outage, and will take the uncertainty of timing into account. This is 
distinct from a situation where an apparently random unplanned event 
occurred, for example, caused by a sudden explosion. Therefore, it may 
be inappropriate to treat these as the same types of events. Further, 
without more information, we do not know to what degree the extended 
planned outage came as a surprise. Therefore, we do not consider these 
outages unplanned. 

[11] It may be noted that in some circumstances, a production facility 
may have been operating below capacity prior to the start date of an 
unplanned outage or have a gradual comeback after the outage end date. 
EIA officials suggested a more ideal way to define the length and 
extent of an outage is to use EIA-810 data on monthly capacity 
utilization and production combined with the IIR data, which identifies 
when a refining unit goes completely down. However, the fact that the 
EIA-810 data do not identify which day or even week the outage or the 
gradual slow-down of a refining unit begins, makes this approach 
incompatible with the structure of the model we used. 

[12] EIA-800 "Weekly Refinery and Fractionator Report"; EIA-801 "Weekly 
Bulk Terminal Report"; EIA-802 "Weekly Product Pipeline Report"; EIA- 
803 "Weekly Crude Oil Stocks Report"; EIA-804 "Weekly Imports Report"; 
EIA-805 "Weekly Terminal Blenders Report"; EIA-810 "Monthly Refinery 
Report"; EIA-811 "Monthly Bulk Terminal Report"; EIA-812 "Monthly 
Product Pipeline Report"; EIA-813 "Monthly Crude Oil Report"; EIA-814 
"Monthly Imports Report"; EIA-815 "Monthly Terminal Blenders Report"; 
EIA-816 "Monthly Natural Gas Liquids Report"; EIA-817 "Monthly Tanker 
and Barge Movement Report; EIA-819 "Monthly Oxygenate Report"; EIA-820 
"Annual Refinery Report." 

[13] EPA officials told us that the ability of EPA to grant waivers for 
certain fuel requirements may be limited because some fuel 
specifications may damage vehicle emissions control equipment. For 
example, gasoline vehicle catalysts can be compromised by using 
gasoline with too high a sulfur content. 

[14] The Clean Air Act allows other states and counties to opt-in to 
the Reformulated Gasoline (RFG) program as part of their State 
Implementation Plans for air quality. Once states opt-in to the 
program, they are required to use reformulated gasoline unless they 
receive a waiver from EPA. 

[15] GAO, Strategic Petroleum Reserve: Issues Regarding the Inclusion 
of Refined Petroleum Products as Part of the Strategic Petroleum 
Reserve, [hyperlink, http://www.gao.gov/products/GAO-09-695T] 
(Washington, D.C.: May 12, 2009). 

[16] [hyperlink, http://www.gao.gov/products/GAO-09-87]. 

[17] In the absence of such arrangements and under normal market 
conditions, unbranded wholesale prices tend to be lower than branded 
wholesale prices, in part because unbranded distributors are able to 
shop around for the lowest wholesale price. 

[18] We believe the under-and over-estimation of results to be small 
because the bulk of the outages we evaluated were related to refining 
units that primarily would have an impact on gasoline production. One 
could use a case study approach and examine the dynamics of price 
effects experienced based on the type of equipment experiencing the 
outage--and such a case study may even look at multiple products. 
However, this was not possible given the model we chose, which was 
developed based on the data available and our intent to cover the 
geographic United States and determine the immediate and average 
effects of outages. EIA's analysis of the impacts of outages on 
production, depending on the type of equipment experiencing the outage, 
was published in its March 2007 report Refinery Outages: Description 
and Potential Impact on Petroleum Product Prices. We believe that 
building a body of work and analyses will contribute to our collective 
knowledge of the effects of outages on gasoline prices. 

[19] The refinery-to-market link would be useful in understanding 
relationships between refineries and the markets they serve, including 
which refineries produce special fuel blends designed to meet federal, 
state, or local requirements. Refiners have an interest in producing 
products that meet a market need, and therefore can be expected to know 
which markets they intend to serve. In addition, jobbers and retailers 
have an interest in knowing where their supply typically originates, so 
they may monitor market developments and price their gasoline 
accordingly. However, the refinery-to-market link may be necessarily 
imprecise, and in some cases the linkage cannot be tracked because the 
pipeline distribution system is designed to enhance fungibility of 
product shipped. For example, the Colonial pipeline system ships 
product in batches that are not segregated by shipper--a single batch 
of gasoline may be comprised of the intermingled production of several 
refineries. While it may not be possible to track the "molecules" of 
production that enter and exit the system, the market participants in 
the system have a strong incentive to know the flows that affect their 
operations. 

[20] In addition to our current work, we reported in December 2007 that 
it was difficult to assess the extent of supply infrastructure 
constraints, or the impacts of these constraints on product prices, as 
there is no central source of data which tracks the entry, flow, and 
exit of petroleum products or the capacity at which pipelines operate 
or the location of system bottlenecks. See GAO, Energy Markets: 
Increasing Globalization of Petroleum Products Markets, Tightening 
Refining Demand and Supply Balance, and Other Trends Have Implications 
for U.S. Energy Supply, Prices and Price Volatility, GAO-08-14 
(Washington, D.C.: Dec. 20, 2007). 

[21] Federal agencies, as required under the Pub. L. No. 106-554 § 515 
(2000), known as the Information Quality Act and related guidelines, 
are generally required to meet high standards in collecting data from 
industry and disseminating information to the public, including having 
sufficient internal controls in place to assure its accuracy and 
reliability. 

[22] "America's Energy Pipeline Network: Assessing Current Strengths 
and Identifying Future Challenges" was issued in January of 2009. The 
study was conducted in response to Section 8 of the Pipeline 
Inspection, Protection, Enforcement and Safety Act, Pub. L. No. 109-468 
(2006). The study was to identify where shortages of pipeline capacity, 
pipeline reliability concerns, or unplanned losses of pipeline 
facilities might contribute to price disruptions or petroleum product 
shortages, as well as to determine whether the current level of 
pipeline regulation is sufficient to minimize future capacity 
constraints. 

[23] A DOT official noted that pipeline size is not the only limiting 
factor for pipeline utilization capacity. Specifically, using drag- 
reducing agents or adding pumps to a pipeline may increase its product 
flow capacity thereby eliminating potential constraints. Moreover, 
storage and other inventory data are generally under EIA's purview. 

[24] In interviews with us, pipeline and other industry officials noted 
that reporting product flow data on a monthly basis--which would 
include volumetric entry and exit information--might not be overly 
burdensome as equipment is already in place to track such flows and 
bill shippers. However, product flow information is considered highly 
sensitive for competitive reasons and would need to be treated as 
confidential by agencies. 

[25] Our analysis of state fuel and biofuel requirements, as reported 
by OPIS, shows that one state, Illinois, specifies conventional 
gasoline and has a so-called "label law" at gasoline stations' pumps to 
advise consumers the gasoline being sold may contain ethanol. Other 
states that specify various types of reformulated gasoline may also 
have label laws advising the possible inclusion of ethanol--a common 
component of reformulated gasoline--in the retail station's gasoline, 
but our analysis identified areas in which conventional, not 
reformulated, gasoline with ethanol content was being sold. 

[26] The 32 states are: Alabama, Alaska, Arkansas, Colorado, Florida, 
Georgia, Idaho, Illinois, Indiana, Kansas, Kentucky, Maine, Maryland, 
Michigan, Mississippi, Nebraska, New Mexico, New York, North Carolina, 
North Dakota, Ohio, Oklahoma, Pennsylvania, South Carolina, Tennessee, 
Texas, Utah, Vermont, Virginia, Wisconsin, West Virginia, and Wyoming. 

[27] This survey, Form EIA-22M, collects company and plant information, 
operating status, annual production capacity, stock changes at 
production facilities, feedstock and other inputs, resale sales of 
biodiesel, and end-use sales of biodiesel data from registered U.S. 
producers of biodiesel. 

[28] Therefore, in our analysis, there can only be one outage event 
during one time period at a given refinery. 

[29] Refer to appendix II GAO's Quantitative Methodology for 
Determining Impacts of Refinery Outages on Wholesale Prices, for more 
information on the econometric techniques used. 

[30] See EIA, John Zyren and Michael Burdette, Gasoline Price 
Passthrough, (Washington D.C. January 2003) and GAO, Motor Fuels: 
California Gasoline Price Behavior, RCED-00-121 (Washington D.C.: Apr. 
28, 2000). 

[31] State-level personal income data were quarterly. 

[32] We do not need to include a price of crude deflator in order to 
model the ratio of the price of gasoline to the price of crude--"the 
crack ratio." Both the price of crude oil and the producer price index 
do not vary across cities, only over time, and would therefore be 
perfectly collinear with the time dummy variables. Therefore, given 
that we include these time dummies, deflating our dependent variable by 
the price of crude oil, would have no effect on our results. While our 
model controls for these time-varying-only effects, it does not 
estimate the contribution of any of these variables to explaining city 
gasoline prices. 

[33] See Im, Kyong So, M. Hashem Pesaran, and Yongcheol Shin. "Testing 
for Unit Roots in Heterogeneous Panels," Journal of Econometrics, vol. 
115, 53-74 (2003). 

[34] In general, the outcome of stationarity tests on gasoline prices 
and crude oil prices varies according to the periodicity of the data, 
the sample period of estimation, and the model specification. While 
some authors have found evidence of stationarity in levels of these 
prices, others have found this not to be the case. Our data are not a 
single time series of the national rate but a panel. Therefore it is 
not unreasonable that our results are consistent with the results of 
some researchers but at variance with others. 

[35] We take the standard type of gasoline as conventional clear, which 
contains no oxygenate additives and does not have a special RVP. 

[36] We recognize that the production of the affected refinery may not 
be zero in some situations, namely, when a refinery supplying 20 
percent of a city's product may reduce supply to only 10 percent in the 
event of an outage. Ideally, we would have preferred accurate estimates 
of production lost due to the outage. This might have been possible 
using EIA data, but likely only for major outages. Regardless, this was 
not feasible for the approximately 1,100 unplanned outages used in our 
analysis. 

[37] The xtiverg2 procedure in STATA implements Instrumental Variable/ 
General Method of Moments estimation of the fixed-effects and first- 
differences panel data models with possibly endogenous regressor. 

[38] See Fumio Hayashi. "Econometrics," Princeton University Press, 220-
221 (2000). This test can be used to test for the endogeneity of a 
subset of the regressors. 

[39] See, for example, W. N. Evan et al. "Endogeneity in the 
Concentration-Price relationship: Causes, Consequences, and Cures." The 
Journal of Industrial Economics, vol. XLI, no. 4, December 1993. 

[40] For a detailed derivation of this test statistics, see Fumio 
Hayashi "Econometric" Princeton University Press, 217-218 (2000). 

[41] State-level data were available, but we were concerned that these 
data were unsuitable for our purpose. In particular, state boundaries 
may not provide good measures of market boundaries, especially for 
cities located at the juncture of multiple states. 

[42] Other researchers have found that inventories might have a 
significant impact on prices. However, in our analysis, the inventory- 
sales variable was not significant. This may reflect the fact that 
these data are measured at the PADD level, which does not adequately 
measure the impact of inventory-sales at the city level. 

[End of section] 

GAO's Mission: 

The Government Accountability Office, the audit, evaluation and 
investigative arm of Congress, exists to support Congress in meeting 
its constitutional responsibilities and to help improve the performance 
and accountability of the federal government for the American people. 
GAO examines the use of public funds; evaluates federal programs and 
policies; and provides analyses, recommendations, and other assistance 
to help Congress make informed oversight, policy, and funding 
decisions. GAO's commitment to good government is reflected in its core 
values of accountability, integrity, and reliability. 

Obtaining Copies of GAO Reports and Testimony: 

The fastest and easiest way to obtain copies of GAO documents at no 
cost is through GAO's Web site [hyperlink, http://www.gao.gov]. Each 
weekday, GAO posts newly released reports, testimony, and 
correspondence on its Web site. To have GAO e-mail you a list of newly 
posted products every afternoon, go to [hyperlink, http://www.gao.gov] 
and select "E-mail Updates." 

Order by Phone: 

The price of each GAO publication reflects GAO’s actual cost of
production and distribution and depends on the number of pages in the
publication and whether the publication is printed in color or black and
white. Pricing and ordering information is posted on GAO’s Web site, 
[hyperlink, http://www.gao.gov/ordering.htm]. 

Place orders by calling (202) 512-6000, toll free (866) 801-7077, or
TDD (202) 512-2537. 

Orders may be paid for using American Express, Discover Card,
MasterCard, Visa, check, or money order. Call for additional 
information. 

To Report Fraud, Waste, and Abuse in Federal Programs: 

Contact: 

Web site: [hyperlink, http://www.gao.gov/fraudnet/fraudnet.htm]: 
E-mail: fraudnet@gao.gov: 
Automated answering system: (800) 424-5454 or (202) 512-7470: 

Congressional Relations: 

Ralph Dawn, Managing Director, dawnr@gao.gov: 
(202) 512-4400: 
U.S. Government Accountability Office: 
441 G Street NW, Room 7125: 
Washington, D.C. 20548: 

Public Affairs: 

Chuck Young, Managing Director, youngc1@gao.gov: 
(202) 512-4800: 
U.S. Government Accountability Office: 
441 G Street NW, Room 7149: 
Washington, D.C. 20548: