Massachusetts Institute of Technology
Department of Economics
Working Paper Series
IS
THE "SURGE"
WORKING? SOME NEW FACTS
Michael Greenstone
Working Paper 07-24
September 14,2007
Revised: Sept. 18, 2007
Room
E52-251
50 Memorial Drive
Cambridge, MA 02142
paper can be downloaded without charge trom the
Social Science Research Network Paper Collection at
This
http://ssrn.com/abstract=1014427
Digitized by the Internet Archive
in
2011 with funding from
IVIIT
Libraries
http://www.archive.org/details/issurgeworkingso0724gree
Is
The "Surge" Working? Some New
Facts^
Michael Greenstone
Massachusetts Institute of Technology
Brookings Institution
National Bureau of Economic Research
Previous Version: September 14, 2007
This Version: September 18, 2007
am grateful
comments from David Autor, Dana Calvo, David Card, Eric Chaney,
Janet Currie, Andrew Dailey, Mark Duggan, Hans Fless, Daniel Greenstone, Lawrence Katz,
Jeffrey Kling, Enrico Moretti, James Poterba, Matthew Rudolph, Adam Sohn, Frank Vannerson,
Vivek Varma, and Annette Vissing-Jorgensen. I especially thank Eric Chaney for alerting me to
the existence of Iraqi government bonds. Abigail Friedman, Elizabeth Greenwood, and Henry
Swift provided outstanding research assistance.
Finally, I thank Katherine Ozment for her
*
I
for valuable
insights, support
errors.
and encouragement
at
each stage of
this project.
I
am
responsible for any
Is
"The Surge" Working? Some New Facts*
ABSTRACT
There is a paucity of facts about the effects of the recent mihtary "Surge" on conditions in Iraq and
whether it is paving the way for a stable Iraq. Selective, anecdotal and incomplete analyses abound.
Policy makers and defense planners must decide which measures of success or failure are most important,
but until
now
paper applies
reliable
few,
if
modem
any, systematic analyses were available on which to base those decisions.
statistical
techniques to a
and widely-cited sources
new
data
to assess the Surge's
file
This
derived from more than a dozen of the most
impact on three key dimensions: the functioning of
the Iraqi state (including civilian casualties); military casualties; and financial markets' assessment of
Iraq's future.
The new and unusually rigorous findings presented here should help inform current
evaluations of the Surge and provide a basis for better decision making about future strategy.
The
analysis reveals
mixed evidence on
has improved insofar as civilian
the Surge's effect
on key trends
in Iraq.
The
security situation
have declined without any concurrent increase in casualties
among coalition and Iraqi troops. However, other areas, such as oil production and the number of trained
Iraqi Security Forces have shown no improvement or declined. Evaluating such conflicting indicators is
fatalities
challenging.
There is, however, another way to assess the Surge. This paper shows how data from world financial
markets can be used to shed light on the central question of whether the Surge has increased or
diminished the prospect of today's Iraq surviving into the fiature. In particular, I examine the price of
Iraqi state bonds, which the Iraqi govermnent is currently servicing, on world financial markets. After the
Surge, there was a sharp decline in the price of those bonds, relative to ahemative bonds. This decline
signals a
40%
date, the
Surge
increase in the market's expectation that Iraq will default.
is failing to
pave the way toward a stable Iraq and
Michael Greenstone
Massachusetts Institute of Technology
Department of Economics
50 Memorial Drive, E52-359
MA
02142
Cambridge,
and Brookings Institution
and NBER
Email: mgreenst@mit.edu
may
This finding suggests
in fact
be undermining
it.
that, to
"We need
Tony Snow,
war
Is the
in Iraq
a surge of facts."
Press Secretary for President George
worth
its
The
cost?
rationales for the
W. Bush,
war include
July 12, 2007
that
it
plays a central role in
protecting the nation against threats from al-Qaeda, ensuring a reliable supply of oil to the world market,
preventing a humanitarian
crisis in Iraq,
and ensuring the future of a sovereign
soldiers
providing an environment that allows for a stable Middle East,
As of September
Iraqi state.
have been wounded and 3,735 have been
killed in the course
V 2007,
approximately 27,662
US
of the war's prosecution. The
US
has spent about $403 biUion (Sunshine 2007) to date on the war and related
estimated that
its
ultimate costs could exceed $2 triUion (Bilmes and Stiglitz 2006).
about the war in Iraq
There
and whether
is
it is
activities,
centered on whether the Surge
is
is
and
The
has been
it
current debate
working.
a paucity of facts about the effects of the recent military "Surge" on conditions in Iraq
paving the way for a stable
Selective, anecdotal
Iraq.
and incomplete analyses abound.
Policy makers and defense planners must decide which measures of success or failure are most important,
but, until
now, few
paper applies
reliable
if
modem
any systematic analyses were available on which
statistical
techniques to a
and widely-cited sources
new
data
to assess the Surge's
file
to
base those decisions.
This
derived from more than a dozen of the most
impact on three key dimensions: the functioning of
the Iraqi state (including civilian casuahies); military casuahies; and financial markets' assessment of
Iraq's fliture.
The new and unusually rigorous findings presented here should help inform
evaluations of the Surge and provide a basis for better decision making about
The
central challenge for this analysis
absence of the Surge.
In the ideal, there
Surge took place and another
Surge's causal impacts.
As
in
which
it
to
is
identical Iraqs
compares outcomes
in the period after the Surge's initiation.
variables were trending upwards or
downwards
in
— one
These two Iraqs could be compared
a feasible alternative, this paper
Surge to the same outcomes
fiiture strategy.
determine what would have happened in Iraq in the
would be two completely
did not.
current
In
to
in
which the
determine the
in the period before the
some cases
the
outcome
advance of the Surge. In order to avoid confounding
these pre-existing trends with the effect of the Surge, this paper tests for changes in these trends.
The
state.
analysis
shows mixed evidence on
Perhaps the crudest expression of the security situation
a substantial reduction in civilian fatalities in
and there
may
be an increase
On
not in Baghdad).
weeks
the Surge's effect on the current functioning of the Iraqi
after the
in the
Baghdad and
in Iraq is the rate
the rest of Iraq.
number of hours of electricity
of civilian
fatalities.
find
Further, there isn't a decline
available to Iraqi
consumers (although
the less positive side, the size of the Iraqi Security Force w^as about
Surge began than would have been predicted by trends prior
the analysis finds that after the Surge oil production didn't increase and
I
to the Surge.
may have
10% lower 25
Additionally,
decreased. Oil
is
Iraq's
primary source of income.
The second category of outcomes examines
coalition soldiers
and
number of casualties suffered by
of a decline
in the
troops in Iraq
The
These
first
Iraqi state.
The
Iraqi Security Forces.
coalition
number of non-fatal
was increasing
and
analysis indicates that the Surge had
Iraqi Security
casualties per
was
decisions about
extraordinarily challenging.
how
to
diminished the prospect of a functioning
is
Since the number of
US
emerge.
on the current functioning of the
proceed in Iraq based on such conflicting indicators
many of
however, another way to assess the Surge.
government
even some evidence
is
For example, does the positive news on security outweigh the negative
financial markets can be used to shed light
the Iraqi
impact on the
noteworthy.
benchmarks)? In the face of this uncertainty, analyst judgment and
is,
soldiers.
the Surge's effect
findings about Iraqi Security Forces (or the failures on
There
US
is
little
to create the conditions necessary for a stable Iraq to
two categories provide mixed evidence on
Making sound
Force troops. There
day among
in this period, these findings are
stated purpose of the Surge
by
the impact of the Surge on casualties suffered both
the
politics
18 congressionally
may play too
This paper shows
how
mandated
large a role.
data from world
on the central question of whether the Surge has increased or
Iraq.
In particular,
I
examine the price of Iraqi
currently servicing on world financial markets.
'
state
bonds
that
Importantly, these bonds trade
on world financial markets; for example, hedge funds hold a substantial fraction of them and
their price
In a novel paper. Chancy (2007) uses variation in prices of these Iraqi bonds during the January 2006 through
August 2006 period to assess the most effective pacification policies in Iraq. This paper has several important
findings but perhaps the most important ones are that sectarian violence is associated with declines in the prices of
these bonds, while signs of political progress in Iraq are associated with increases in bond prices.
'
on the secondary market
markets
is
quoted
is
in
that traders' only concern
There
correct projections.
that these markets
isn't
Bloomberg's financial data
is
room
to
make
profitable decisions
and
for personal biases in this setting.
have a good track record
The appeal of using
services.
financial
this necessarily requires
Consequently,
it
making
isn't surprising
predicting uncertain future events (Wolfers and Zitzewitz
at
2004).
Overall, the
way
for the
bond market findings
fail to
emergence of a stable Iraq and may
support the view that the Surge
be undermining
in fact
40%
increase in the expected aimual probability of default.
helping to pave the
After the Surge, the Iraqi
it.
bonds' yield increased by 115 to 170 basis points, relative to comparison bonds.
approximate
is
This translates into an
In other words, the
world
financial markets are saying that Iraq's prospects declined after the Surge.
Examining bond yields
to learn
about the effectiveness of the Surge requires several assumptions,
including that financial markets efficiently and accurately aggregate public and private information from
diverse sources.
movements
even
in
Further,
bond yields
the
meaning of
reflect
domestic
this
US
exercise
politics.
Democrat
after controlling for the probability that a
does not eliminate the domestic politics explanation, but
There are several caveats to
that predates the invasion
paper
is
I
this analysis that
would be undermined
if the
find that the increases in default risk remain
will be elected President in 2008.
it
documented
substantially
weakens the case
bear highlighting.
First,
due
to the
This finding
for
it.
absence of data
of Iraq, the analysis can only inform the question of the Surge's efficacy. This
not directly informative about the broader question of the impact of the Iraq war.
data on the current functioning of Iraq
fail to
cover
many outcomes
that
would be both of
and potentially more informative than the available ones. Third, even among the available
legitimate questions about quality and reliability.
This
is
Second, the
great interest
data, there are
especially the case with the civilian casualty
data.
The remainder of
Surge. Section
III briefly
the paper proceeds as follows.
describes the data
file
Section
II
provides a brief background on the
compiled for the analysis (there
discussion in the Data Appendix) and IV outlines the statistical models.
Section VI concludes.
is
Section
a
much more
V
discusses the results.
extensive
II.
This section has
initiation
14,
tiiree goals.
of the military strategy that
2007 as the
starting date
implementation on the ground
On
Background on the Surge
First,
is
known
provides a brief review of the factors that led to the
Second,
as the Surge.
of the Surge.
Third,
provides
it
it
motivates the choice of February
some
details regarding the
Surge's
in Iraq.
January 10, 2007, President Bush announced that the number of troops deployed in Iraq
would be increased by roughly 30,000. The
manpower necessary
Baghdad.'
it
so that the
The theory being
US
that
stated short run goal of this troop increase
was
was centered
military could reduce the sectarian violence that
by improving
security, the
Surge could achieve
to provide the
in
ultimate goal of
its
providing the Iraqi government the opportunity to grow and govern more effectively, ultimately leading
to a stable Iraq.
(See Figure
1
for a
map of
Iraq and
Baghdad's central physical location and large
population share.) The President enunciated this goal and
priority for success in Iraq
is
security, especially in
occurs within 30 miles of the capital.
shaking the confidence of
cemented by the
all
Iraqis"
motivation
(Bush 2007).
is
splitting
violence of
said,
"The most urgent
into sectarian enclaves
shift in
policy that
10,
was
2007.
sectarian conflict that the Surge aims to control has an extensive history in Iraq.
Saddam Hussein's Sunni dominated government
would focus on attacking Shi'a
and
followed as General David Petraeus replaced General George
manifestation. In late 2003, following the
"
Baghdad
This focus represented a
Casey as commander of Multi-National Forces-Iraq on February
The
when he
Baghdad. Eighty percent of Iraq's sectarian violence
This violence
shift in leadership that
its
US
to leverage political control (al-Khalidi
euphemism
just
one
was growing
that
against Shiites and Kurdish Iraqis
invasion in that year, a Sunni insurgency
The
is
& Tarmer 2006).
The Kurds are not part of the
problems involving sectarian violence in Kurdish
areas - as well as sectarian violence against Kurds in South and Central Iraq - are not to be discounted, the main
source of sectarian displacement are attacks on Surinis and Shi'a by Shi'a and Sunni radical groups respectively" (al"In Iraq today, the term 'sectarian violence'
current explosion in sectarian
Khalidi
& Tanner, 2006).
strife.
And
is
a
while past and
for Shi'a-Surmi violence.
fiiture
Over
violence.
few
the next
On March
2,
years, there
were several key events
in
Baghdad
wounded
100.
By mid-2005, Sunni and
regularly being discovered throughout Iraq (al-Khalidi
bomb
in
Kadhimiya, a predominantly Shi'a
October, an attack
in
at
mosque
a Shi'a
Baghdad Attacks," 2005;
al-Khalidi
mostly Sunni prisoners held
in the
&
Shi'a Islam.
The bombing
is
in
day of Ramadan killed 87 people ("Scores killed
first
Interior Ministry building
levels.
in
Samarra on February 22, 2006 raised
The Golden Mosque
viewed by some observers
where they had "apparently
& Tanner, 2006).
"Golden Mosque"
new
concerns about the sectarian violence to
September of the same year, a
Tanner, 2006). In November, American soldiers found 173
basement of an
the Al-Askariya
In
of Baghdad, killed more than 100 people; while,
been abused by the largely Shi'a police force" (al-Khalidi
The bombing of
at the
Shi'a victims of execution-style killings were
& Tanner, 2006).
district
on the
in Hilla
Almost simultaneously, suicide bombers
and wounded 200 (Pacific Disaster Management
killed at least 58 people
Information Network, 2004).
car
underscored the increasing sectarian
2004, during the Shiite festival of Ashoura, two bombings near shrines in Karbala,
a city holy to Shi'a Muslims, killed 85 and
Kazimiya shrine
that
is
one of the holiest
as a tipping point that initiated a
new
sites in
equilibrium
with higher levels of sectarian violence.
The bombing and
at that
aftermath also helped to crystallize the view that the
at clearing insurgents fi^om difficult areas
over to the Iraqi security forces. Over time,
it
became
effectively protect these neighborhoods after the
services that
It is
would improve
that the
^
it
The
military's strategy
in
were not able
clear that the Iraqi security forces
to
Further, the ISF's efforts to build local
raids.
Bush increased
number of US and Coalition Troops
number of US troops
was 162,000
US
and neighborhoods and then turning those areas
the quality of life within the neighborhoods were also failing.
in this context that President
the time series of the
that
US
time was not succeeding. Until then, the US's strategy in Baghdad largely consisted of conducting
aimed
raids
its
in January
August 2007.^
vertical line in Figure 2 is
drawn
vertical line occur after the initiation
the
number of troops
in Iraq since
May
in Iraq.
Figure 2 reports
2003. The figure documents
2007 was 132,000 and has increased every month since then so
The
figure also
makes
to intersect the January
of the Surge.
clear that the
2007 data
points.
US
accounts for the vast
All points to the right of the
majority of the coalition: the
US
share ranges from a low of
82%
in
December 2003
to a
high of
93%
in
August 2007.
These additional troops allowed the military
to initiate a
new
achieving security in Baghdad. This strategy has two major components
and Operation Phantom Thunder.
strategy that
was focused on
—Operation Enforcing
Together, these operations aim to stabilize areas by
insurgent groups and stopping violence, then maintaining this
initial stability
first
the Law"*
removing
over a period of time before
giving the Iraqi Security Forces complete responsibility for them.
Operation Enforcing the
Law became
operational on February 14, 2007.
Multi-National Division- Baghdad, Major General Joseph
77^/5
new plan
Fil,
The commander of
described this operation:
involves three basic parts: clear, control
within each of the security districts in the Iraqi capital
and
is to
retain.
The first objective
clear out extremist elements
neighborhood by neighborhood in an effort to protect the population. And after an area
is cleared, we 're moving to what we call the control operation. Together with our Iraqi
counterparts, we 'II maintain a full-time presence on the sti-eets, and we '11 do this by
building and maintaining joint security stations throughout the city. [ ... J. An area
moves into the retain phase when the Iraqi security forces are fully responsible for the
day-today security mission. At this point, coalition forces begin to move out of the
neighborhood and into locations where they can respond to requests for assistance as
needed. During these three phrases, efforts will be ongoing to stimulate local economies
by creating employment opportunities, initiating reconstruction projects and improving
the infrastructure. (Press Conference with Maj. Gen. Joseph FilJr., 2007)
This plan requires not just the raids calculated to break
characteristic
down
insurgent groups and the resources that were
of military operations pre-2007, but also a force
that
remains in the cleared area
until
it is
secure and the Iraqi Security forces are sufficiently large and well-trained to control the area on their own.
The second piece of the Surge
15, 2007.
This operation's goal
is to
strategy. Operation
Phantom Thunder, became operational on June
stem the flow of weapons and insurgents
in
and out of Baghdad
(Kagan, July 2007). According to General Odiemo, Operation Phantom Thunder and the smaller
operations that
fall
under
it
"are intended to eliminate accelerants to
Baghdad violence from enemy
support zones in the belts that ring the city" with the goal of enabling the Iraqi government to function
and move towards
"*
stabilizing the country politically (Press
Operation Enforcing the
Law
is
the official
Baghdad Security Plan (BSP) and has been
Officially Named," 2007).
name
for Operation
Conference with
Lt.
Gen. Ray Odiemo,
Fardh Al-Qanoon which is also known as the
Law and Order ("Baghdad Security Plan
referred to as Operation
2007). Regions within these belts^ are targeted because of their established insurgent strongholds,
placement along major roads and
make them
In
lines
of communication, proximity
vulnerable to destabilizing forces that might
summary,
this
move
Baghdad, and other factors
to
into or out
of the capital
that
city.
paper defines the Surge as the addition of the 30,000 extra troops and the
implementation of Operations Enforcing the
the Surge constitutes a major shift in
US
Law and Phantom
Thunder. As
this section
has documented,
military policy in Iraq that aims to reduce the sectarian violence
centered in Baghdad with the hope that this will allow the Iraqi government to grow, govern more
effectively,
and ultimately become a functioning independent country.
The remainder of
paper
this
assesses whether the Surge's initiation on February 14, 2007 affected the functioning of the Iraqi state,
coalition casualties,
and the world financial market's prediction about the prospects for a stable
Iraq.
>
III.
The data
about
file
used
many of them from
to
Data and
Summary
Statistics
conduct the analysis comes firom more than a dozen data sources.
the Brooking Institution's Iraq Index,
which has played an instrumental
providing valuable information about the war in Iraq for several years.
Weekly
the Department of State's Iraq
my
own.
In general,
I
Status Report
(IWSR).
I
grouped
The
number of
civilian fatalities, the size of Iraqi
involves functioning of the Iraqi
discovered some other data sources on
coalition fatalities, non-fatal
The outer
casualties,
and
Tarmiyah
(one), Buhriz (two),
Yusifiyah (eight), Fallujah (nine) and
Karmah
(ten)"
All of the original
end of this paper.
form the basis of the subsequent analysis
state.
The
variables in this category include the
is
military casuahies.
Iraqi Security Forces fatalities.
ring [of the belt around Baghdad] runs
position), clockwise to
al
US
by
Security Forces, crude oil production, and hours of
The second category
electricity available to Iraqis.
'
at the
into three categories that
categories.
first
role in
collected the data series used in this analysis from their original sources
sources are detailed in the extensive Data Appendix
are
learned
Other data series were found in
following up on the citations found in the Iraq Index and the Weekly Status Report.
The data
I
These variables include
The
third category is the
from Taji (north of Baghdad between the 11 or 12 o'clock
Besmaya (three), Salman Pak (five), Mahmudiyah (six), Sadr
(Kagan, July 2007).
world financial markets' assessment of the future of the
Iraqi state.
bond issued
are the prices and yields of the Iraqi government
in
The key
variables for this category
January 2006 that pays coupons twice a
year and matures in 2028.
Table
presents
1
summary
statistics
on a number of outcome variables.
values of several dependent variables at key dates.
over three time periods.
continues through the
samples begin
1
.5
In the "All"
and
B
presents
In the "1.5 Years Prior" and "1
variables
earliest observation
and
Year Prior" columns, the
year before the initiation of the Surge and continue through the
1
reports the
means of some important
column, the sample begins with the
one.
last available
Panel
A
Panel
last
available
observation.
Finally, the collection of this data file has taken hundreds
writing documentation for these files and will post
them on
of person-hours.
my web
site as
I
am
in the process
soon as possible to
of
facilitate
future research on the Iraq war.
IV. Methodology
The
central challenge for this analysis
functioning of the Iraqi
state.
another Iraq in which the
US
what would have happened
is
The hypothetical
to
ideal for this study
did not initiate a Surge.
in Iraq
determine the causal impact of the Surge on the
without the Surge.
would be a
parallel universe with
This parallel would provide a counterfactual for
It
would then be straightforward
to
compare the
variables of interest in the current Iraq and in the parallel Iraq to assess the causal impacts of the Surge.
In the absence of this impractical ideal, this paper compares outcome variables before and after
the Surge began.
Surge in Iraq
is
The
basic idea
simple: the best available counterfactual for what happened after the
is
what was happening
variables were trending upwards or
in Iraq before the initiation
downwards
in
of the Surge. In some cases the outcome
advance of the Surge. To avoid confounding these pre-
existing trends with the Surge, the paper tests for changes in these trends.
*
The appeal of
classical experiments
without a parallel universe.
and control groups are
is
By randomly
that they reveal the causal
The key idea
impact of an intervention
is
(e.g., a
that
new
it
is
drug)
assigning the treatment, classical experiments guarantee that the treatment
identical so that ex-post
comparisons reveal casual impacts.
possible to estimate the Surge's causal impacts only
Surge
indeed a vaUd counterfactual for post-Surge Iraq
is
This paper implements two
"mean
to as the
the Surge. For example, in an examination of the
tests
y,
where y
is
= ao + P
mean
l(Surge),
+
an indicator variable equal to
change
in the
outcome
The subsequent
civilian fatalities per day, this
after the
Surge
is
equal.
after
approach
assumption
Its
is
after the Surge.
following equation:
1
t
indicates a
on or
for days
day or week. The variable of interest
is
1
(Surge) which
February 14, 2007. The parameter p measures the
after
Thus
relative to the period before the Surge.
fatality rate, a positive estimate
referred
£„
outcome variable and
the
fitting the
first is
whether the variable's mean changed
tests for
number of Iraqi
model involves
shift
that Iraq prior to the
after controlling for pre-existing trends.
of fatalities per day, besides the Surge, are constant before and
Specifically, the
(1)
—
of the variables examined. The
whether the mean number of fatalities per day before and
that all determinants
is
— holds
statistical tests for all
model." This approach simply
shift
untestable assumption
if tiie
of P would indicate that
this rate
in the
case of the daily Iraqi civilian
was higher
in the
period after the Surge.
analysis reports on estimates of P and a heteroskedastic consistent standard error.'
A shortcoming of the mean shift model is that in the presence of pre-existing trends, the estimated
A
P will reflect the trend and any independent impact of the Surge.
trend break
model allows
existing trend in the variable and then tests whether this trend changes. Consequently,
method than
the
mean
Specifically,
(2)
y,
t
The
2007) + P((/- February
14,
The
variable
variable
{{t
'
I
and
it
(t
- February
- February
trend and an indicator variable equal to
interest is p
1
14,
for days
14,
14,
2007)
on or
2007)* l(5wrge),) +
2007)
*
after
Newey-West standard
is
a time trend
\(Surge)
,)
February
measures the average slope change in the outcome
also experiment with
a more robust
estimate the following equation:
are as before.
14, 2007).**
is
model.
= a + 5 (r- February
where y and
February
I
shift
it
for a pre-
is
£„
(i.e.,
equals
it
on
the interaction of the time
14,
2007.
The parameter of
that occurred after the
errors that allow for unspecified heteroskedasticy
Surge began.
and autocorrelation
with 3 lags. These standard errors were very similar to the purely heteroskedastic-consistent ones.
^
For example, it equals -1 on February 13, 2007 and -2 on February
14, 2007, 1 on Febmary 15, 2007, 2 on Febmary 16, 2007, etc.
Febmary
12,
2007. Going forward,
it
equals
on
The key assumption needed
for this
method
controlHng for the pre-existing trend
determinants of the outcome.
As
outcome
the
in
to identify the causal effect
a robustness check,
I
variable, the Surge
will also report
of the Surge
is
the only
is
that after
change
in
the
on models that add month fixed
effects to allow for seasonality in the outcomes.
The parameter 8 provides an opportunity
equation
Specifically, a rejection
(1).
rejection of the
fail to
mean
to test the validity
support the validity of the
mean
model specified
shift
in
of the null hypothesis of a zero pre-period trend would lead to a
model. The subsequent analysis uses
shift
of the mean
shift
cases where the data
this test to identify
model. In these cases, the trend break model's results are the
preferred ones.
Another reason
months for the Surge
time.'
that this
to
become
model
fully operational
In the subsequent analysis,
available data
file,
which
is
appealing in the context of the Surge
is
I
war
is
it
typically at least 5
months
no exception. There appears
to
after
Wars
February
may
precise.
On
14,
(t
- February
treat observations
took several
14,
2007)
grows
in the
2007.
be a trade-off between bias and precision.
the other hand, however, conditions have
be inappropriate to
it
and the
are not conducted in stable environments,
the set of days prior to the Surge used as a counterfactual
more
that
allows the impact of the Surge to increase over
report p times the largest value of
Finally, one other issue merits attention.
Iraq
and
is
On
the one hand, as
larger, the resulting estimates
become
changed quickly throughout the war, and
from long before the Surge as useful
in
it
forming a valid
counterfactual for what happened during the Surge.
I
The
the
experimented with several
first is
1
start dates for the analysis
year before the Surge's initiation and the second
bombing of the Askariya
shrine in Samarra on February 22,
crystallize the sectarian violence instigated
associated with a
marked increase
Specifically, the
new
is
1.5
in violence in Iraq.
The former predates
2006 by a week. This bombing helped
It
seems plausible
later
(perhaps June).
10
to
assume
that this period
The longer pre-period allows
troops began arriving in February 2007, but the full
weren't stationed in Iraq until several months
years before.
by the insurgents through fomenting a cycle of reprisals;
the correct counter-factual for the period after the Surge.
'
and will present two throughout the paper.
for a
more
to
it is
forms
precise
complement of about 30,000 troops
estimation of the outcomes in a pre-period where the overall level of violence was lower.
bombing of the Askariya
results
from the shorter
Since the
shrine in Samarra changed the political and security dynamics substantially, the
year pre-period are emphasized, but results based on both pre-periods are
1
presented for completeness.
V. Results
This section
is
divided into three subsections.
has affected the current functioning of the Iraqi
coalition
and
The
state.
first
subsection examines whether the Surge
The second examines
Iraqi security forces that are prosecuting the war.
The
human
the
final subsection
cost
on the
examines the impact
of the Surge on the yields and the implied default probabilities of Iraqi bonds.
A. Functioning of the Iraqi State
Table 2 reports the impact of the Surge on the daily number of
Civilian Fatalities.
Iraqi civilians using the Iraq
the entries and their
Panel
shift
model
A
meaning
Body Count
data
set.
of
tables will be structured similarly, so
are described in detail here.
reports the estimated shift in the
in equation (2).
The subsequent
fatalities
Its
standard error
is
reported below
indicates that the null hypothesis that the estimate
** means that the null can be rejected
at the
1%
mean number of
is
it
per day (P) from the
fatalities
in
parentheses.
A
*
equal to zero can be rejected at the
level.
mean
next to the entry
5%
For the entries that don't have either a
level,
while
* or a **,
it
isn't possible to reject the null hypothesis at conventional significance levels.
Panel
B
reports results from fitting the trend break
with available civilian casualty data
began. So, the
first
entry
in
equation
that the
The
latest
day
the corresponding standard
Consequently, the entry reports the estimated change in the number of civilian
per day by July 17, 2007 due to the change in trend break after the Surge's
means
(2).
July 17, 2007 and this corresponds to the 153"* day after the Surge
two rows present 153 multiplied by the estimate of p with
error in parentheses.
fatalities
is
model specified
Surge
is
associated with a reduction in fatalities.
11
initiation.
A negative
The
row
entry in the "Pre-trend"
reports the estimate of the coefficient 5 with
estimated
its
standard error below in parentheses. This provides an opportunity to assess whether there was a trend in
A
daily fatalities in the period leading up to the initiation of the Surge.
positive parameter estimate
As noted above, a
indicates that fatalities per day were increasing before the Surge.
rejection of the null
hypothesis of a zero pre-trend means that results from the trend break model are more reliable.
The
Column
table has three sets of columns.
when
in
column (lb)
is
that the
sample begins
1
interest since the Surge's activities
and (3b) report on the remainder of Iraq.
Now
turning to the results for
These
all
results indicate that there
the Surge than there
judged
to
overall
mean of
be
civilian
would have been
statistically different
were concentrated
in
Baghdad
only. This
Baghdad. Finally, columns (3a)
of
Iraq,
shift
it is
was
evident that there
model, leading
me
to concentrate
were approximately 56-61 fewer
if the
pre-trend had continued.'"
from zero by conventional
criteria.
a strong
fatalities
upward
mean of the dependent
pre-trend.
on the trend break
per day 153 days after
These estimates would easily be
This
is
a very high fraction of the
the dependent variable, but daily fatality counts were quite high in the last
before the Surge. For example, the
may
-
mean
This undermines the validity of the
results.
is all
365 days, prior to the Surge. Columns
year, or
(2a) and (2b) conduct the analogous analysis with data on fatalities per day from
be of especial
the sample
per day and the sample begins 1.5 years, or 548 days, before the Surge's initiation. The
fatalities in Iraq
only difference
(la) reports results
variable in the last
month before
few weeks
the Surge
was
92.9 civilian deaths per day. Notably in the case of this dependent variable, the choice of start date for the
sample doesn't affect the
The second and
It is
noteworthy
results in a
meaningful way.
third sets of
columns reveal
that the decline outside
the "squirt" theory that the Surge
'°
The
inclusion of
month fixed
of Baghdad
is
would push violence
60%
in civilian fatalities.
and -64.9
It is
of this decline occurs
in
Baghdad.
also statistically significant. This result contradicts
to other parts
effects in the trend break specification in
impact on the predicted decline
(43.9), respectively.
that roughly
of Iraq.
columns (la) and (lb) has no substantive
In particular, the coefficients (standard errors) are -76.5 (28.0)
evident that the inclusion of
estimates.
12
month
fixed effects decreases the precision of the
Figure 3 provides an opportunity to better understand the whole country resuhs from the sample
that begins
1
independent 7-day periods
The
The squares
year before the Surge.
figure also presents the fitted values
line at t= 0,
which
is
most square
(e.g., the left
is
initiation
the average over days
evident.
is
Most
Table
in
notably, there
is
2.
unlikely that that this
a sharp
is
ft is
certainly possible that
Baghdad and
Iraq
In
the entire story.
more
many
some of
downward break
in trend after the
Finally, the source for these data
civilian casualties, so their
in civilian fatalities after the
numbers
(i.e.,
is
associated with an important
generally, but the sharpness of the results
makes
respects, this result confirms the popular theory
it
of
rate.
IraqBodycount.org) uses only the most reliable reports on
likely understate the true
number of civiUan
casualties, perhaps
Consequently, this section's estimates likley understate the impact of the Surge on total
recent cross sectional survey conducted in Iraq suggests that the Iraq
substantially understate total civilian casualties substantially.'"
casualties revealed
Importantly, the trends in civilian
unlikely to reveal the impact of the Surge on the exact
" The website www.icasualties.org also publishes
is
Body Count numbers
from the survey match up well with the trends from Iraq Body Count.
a data file
methodologies suggests that the Iraq Body Count data
icasulties data file
in
these results are due to the reported
domestic policing that putting more cops on the beat reduces the crime
is
a vertical
is
,
increases in migration out of
this analysis
-359).
,
decline in civilian fatalities. "
may
=
upward trend
Overall, the civilian fatality results demonstrate that the Surge
A
t
Specifically, the
of the Surge. The figure provides compelling evidence of a decline
Surge's initiation.
fatalities.
=-365 through
over
the date that the Surge began.
pre-period
substantially so.
t
fatalities
from the estimation of the trend break model. There
This figure explains the source of the results
fatalities in the
number of
represent the average daily
that
more recent data
collected the icasualties data set for
all
of
is
available:
Iraq,
and
it
it
file
is
on
civilian
likely to
number of
fatalities.
be more
reliable.
An
So afthough
fatalities,
this
examination of the
One advantage of
the
includes data through 191 days after February 14, 2007.
produces results that are qualitatively very similar
I
to those in
t =191 and its standard error is -52.6 (23.7) and -67.8 (27,1) from
and (lb) specifications, respectively.
'^
A nationwide household survey of 1,849 households conducted between May and July 2006 suggests that about
655,000 more Iraqis died than would have been predicted by pre-War trends between March 2003 and the time of
the survey (Bumham, Lafta, Doocy, and Roberts 2006). This estimate is roughly 10 times larger than the estimate
from Iraq Body Count. Importantly, the percentage increase in fatalities between a similar survey conducted in 2004
and the 2006 survey closely follow the trends in the Iraq Body County data.
Table
2.
Specifically, the post-trend evaluated at
the All of Iraq
column
(la)
13
measurement
error should not
undermine the paper's
impact of the Surge on trends in
ability to detect the
civilian fatalities.
Size
of Iraqi Security Forces. Another important indicator of the functioning of the
the nimiber of members of the Iraqi Security Force (ISF). For this paper's purposes, the ISF
the
number of the personnel of the
Army, Air Force, and Navy.
police, national police, other security forces
This variable
Table 3 reports the regression
trend,
I
is
measured weekly, although
results.
Since there
is
it is
Iraqi state
is
is
defined as
working for the Ministry,
missing for several weeks.
of a pre-existing upward
clear evidence
focus on the trend break model results. Specifically, this model indicates that ISF personnel were
increased by about 2,000 per
week
The
prior to the Surge.
results
show
that there
was a downward trend
break after the Surge such that there were 25,000 - 28,000 fewer ISF troops 25 weeks
would have been had
the pre-trend continued.
number of ISF personnel was about 10%
'^
These estimates suggest
less than
that
later
than there
25 weeks after the Surge the
would have been predicted by
pre-existing trends.
'''
Figure 4 displays this result graphically.
Crude Oil Production. Table 4 and Figure
production per day, which
in this period.
There
reported in the table.
is
is
5 present the results for millions
measured weekly. The mean production
little
evidence that the Surge affected
However, versions of these same
indicate declines in oil production of roughly
20%,
oil
is
of barrels of crude
oil
about 2.1 million barrels per day
production from the specifications
specifications that include
relative to pre-existing trends.
month
fixed effect
Specifically, the point
estimates (standard errors) from versions of the columns (1) and (2) trend break specifications that
include
month fixed
effects are -0.4454 (0.0708)
Hours of Electricity per Day. Most
war began,
electricity
and -0.3707 (0.0697).
electricity generation takes place outside
producing regions have been reluctant to fully share
electricity
of Baghdad; and, since the
with the capital
borne out in the means of the dependent variables. The average of hours of electricity per day
in
city.
Baghdad
is
This
is
around
'^
The point estimates (standard errors) from versions of the columns (1) and (2) trend break specifications that add
month fixed effects are -30,997 (4,619) and -14,648 (6,397), respectively.
'"
Figure 4 is an excellent example of how the mean shift model can confound pre-existing trends with the true
impact of the Surge. A simple comparison of pre and post mean ISF personnel levels would lead to the conclusion
that the Surge increased troop levels (in fact, this is the conclusion from the mean shift model). The figure helps to
underscore that this conclusion is unlikely to be valid because of the sharp upward pre-period trend.
14
6 to 6.5 relative to about
unavailable).
1
1
in
the country as a
Before describing the electricity
whole (separate data
results,
non-Baghdad region alone
for the
is
note that the reliability of these data have been called into
I
question on several occasions and that they are missing for
1 1
of the 25 weeks in the period
after the
Surge began
(Levey and Zavis 2007).
Table 5 and Figure 6 repoil the association between the Surge and average hours of electricity
available per day in
Baghdad and
The
the pre-Surge period.
meaningful change
that the
in
from the preferred trend break model show
results
Surge was associated with a
weeks
11 of the 25 post-Surge
weaker
little
hours of electricity per day in Baghdad. However, the results for
Baghdad. The meaning of these
Iraq are
Both dependent variables are trending downwards
the entire country.
for versions
20%-25%
(i.e.,
May
8,
evidence of a
of Iraq indicate
increase in hours of electricity for Iraqis living outside
tempered by the
results is
all
in
fact that the electricity data is unavailable for
2007 through July
of columns (2a) and (2b)
Further, the results for all of
31, 2007).
that include
month
fixed effects.
Specifically, the
point estimates, with standard errors in parentheses, are 1.74 (0.73) and 0.56 (1.00) respectively.
B. Military Casualties
Coalition Fatalities.
186 days after the Surge's
day
in the longer
outside of
Table 6 reports on the impact of the Surge on coalition
initiation.
sample and 2.77
For
all
of
Iraq, there
in the shorter sample.
Baghdad were roughly 0.85 and
Surge.
This
trend break results find
is
partially
However, the estimates
little
The mean numbers
1.85, respectively.
in the period preceding the Surge, so the trend break
The
was an average of 2.66
model
is
It is
more
fatalities
coalition fatalities per
for daily fatalities inside
appropriate.
meaningful evidence of a change
in daily fatalities after the
it
difficult to detect
are also sensitive to sample choice. For example, the point estimate
1.5 years
before the Surge, while
15
and
evident that fatalities were increasing
because the estimates are imprecise, which makes
zero in the sample that begins
through
it is
-0.9 in the
is
change.
essentially
sample that begins
1
year
Notably, the overall result
before.'^
outside of Baghdad.
Overall,
Iraq.
it
Many of these
is
due
findings, especially the noisiness
appears that the Surge had
However, since the number of troops
(e.g., fataUties
to a statistically significant decline in coalition fatalities
little
of the data, are evident
impact on the number of coalition
in this
in
Figure
7.
per day in
fatalities
period increased substantially, the daily fatality rate
per 100,000 soldiers) declined.
US Non-Fatal
Casualties.
There are separate results for
all
Table 7 reports on the number of wounded soldiers per day in Iraq.
non-fatal casualties, casualties that prevent soldiers
action within 3 days, and casuahies that allow soldiers to return within 3 days.
from returning
Over the samples,
to
there
are 18-19 non-fatal casualties per day, with slightly less than half resulting in soldiers' inability to return
For
to action within 3 days.
casualties
was increasing
The trend break
per day
among US
three dependent variables, there
all
results suggest that there is a
accompanying Figure
by
'^
As
in the
number of non-fatal
the less serious casuahies.
the
indicates a smaller decline
number of troops
casualties
is
shown
and one
20%
graphically in
that is statistically
if the
outcome
in Iraq.
Table 8 and Figure 9 present results on the estimated impact of
number of Iraqi Security Force
trends in this variable, so neither
This
with the fatality data, these findings would be stronger
Iraqi Security Force Fatalities.
the Surge on the
among
The longer sample
8.
indistinguishable from zero.
variable were normalized
modest decline
In the shorter and preferred sample, the point estimate implies an almost
reduction that occurs, just about entirely,
the
evidence that the incidence of these
advance of the Surge.
in
troops.
is
Fatalities per day.
model can be rejected
in advance.
There
is little
evidence of pre-Surge
Both models and samples
fail to
find
evidence of a meaningful change in the number of fatalities per day. The inclusion of month fixed effects
does not
C.
''
alter this conclusion.
World Financial Markets' Assessment
The point
of Iraq's Future
estimates (standard errors) from the trend break versions of columns (la) and (lb) are 0.564 (0.543) and
-0.405 (0.744), respectively.
'*
The inclusion of month fixed effects causes the point estimate (standard error) to increase to -3.06 (1.31) in
column (la). In (lb), the estimate becomes -2.76 (1.31). Thus, this richer model finds a decline in the rate at which
soldiers are
wounded
in
both samples.
16
The ultimate purpose of
Thus
emerge.
far,
the paper has
the Surge
is
documented
to create the conditions that will allow for a stable Iraq to
that the
Surge has been effective
situation in Iraq without an appreciable increase in casualties
some
among
other indicators of the current functioning of the Iraqi state
Iraqi Security Forces)
mixed evidence on
have
the impact of the Surge
no readily available model
achieving
its
show improvements. Thus,
failed to
and
coalition
(e.g., oil
improving the security
at
Iraqi troops.
However,
production and the number of
produced
to this point, this exercise has
on the current functioning of the
Further, there
is
whether the Surge
is
Iraqi state.
that synthesizes these conflicting results to determine
ultimate goal of improving the prospects for an autonomous Iraq.
This subsection shows that data from world financial markets can be used to transparently shed
light
on the central question of whether the Surge has altered the prospects
exist in the fiiture.
by the
Iraqi
In particular,
I
examine the prices of Iraqi
state
bonds
that are currently
government and of credit default swaps for those bonds. " The goal
changed the world financial market's predictions about the probability
Thus,
bonds.
that a functioning Iraq will
if
taken
is
that Iraq
to infer
would
being serviced
how
default
Iraq's future prospects
from the many variables analyzed above and the numerous ones
amenable
methodological approach.
Background. Before analyzing these time
bonds, some details on
assume
that
how bonds
on these
subsection solves the problem of trying to draw an inference about
literally, this
to this paper's
the Surge
series,
it
is
.
informative to
know
the history of the Iraqi
on secondary bond markets, and why
are priced
world financial markets can
,
that are not
it
efficiently aggregate the available information
is
reasonable to
on Iraq's
future.
Prior to Iraq's invasion of Kuwait in 1990, Iraq issued about $130 billion in debt. After the Gulf war, the
country defaulted on this debt.
When
were spread around the world.
The
"
the
Paris
US
led coalition invaded Iraq in 2003, the holders of this debt
Club held claims of approximately S40
billion; Persian
Gulf
Exchange but decided that this market was a local one with poor
any results. For example, the market only meets twice a week and
trading appears to be very thin. As evidence of this the total dollar value of all trading in February 2007 was less
than $5 million and only 45 of the 94 had any shares traded. Further, I was able to obtain data on a time series for
an index but 1 was unable to determine how the index was constructed, which companies were included, and
I
also considered using data from the Iraqi Stock
liquidity
and
whether
it
that
it
would be
was ever
reconstituted.
government
failure
of the
on the
Iraqi bonds).
Iraqi
difficult to interpret
Finally, in principle,
(in contrast, a failure
it
is
possible that shareholders wouldn't be
damaged by
of the Iraqi govermnent would almost certainly lead
17
to a
a
defauh
creditors
had another $65
billion;
and the remainder was
in the
hands of commercial creditors (Chaney
2007).
After the end of combat operations in
$1000
in the existing
in Iraqi
bonds so
May
2003, the
US
government brokered a deal
bonds for $200 worth of new bonds for those creditors who held
that the
new
Iraqi
govenmient would not be hamstrung by
US
debt relief agreement, the Iraqi government issued roughly $2.66 billion in
in
this debt.
dollar
to
$35 million
at least
As
exchange
a result of this
denominated notes
These new bonds pay fixed coupons of 2.9% twice a year and have a maturity of
January 2006.
January 15, 2028.
Once
there
is
these bonds were issued, they entered the world financial markets. All indications are that
a liquid and competitive market for them.
For example, they began
markets on January 20, 2006, and their prices and yields are reported
A trader at JP Morgan said that
As much as 90% may have gone from
in
to trade
on world financial
Bloomberg's financial services
database.
engineering firms
etc.,
i.e.,
the original holders (banks, trading companies,
not investment fiinds) to investment fijnds.
And of that
[the
bond are] are about evenly split between hedge funds and "real money" accounts
(maybe slightly more in "real money "). [Quoted in Chaney (2007), page 4]
Further there has been substantial trading of these bonds; about $4.1 billion of these bonds were traded in
the first quarter of 2006,
when
they were issued, and roughly $1.6 billion in the second quarter (Chaney
2007). Finally, the average bid-ask spread as a percentage of the average of the bid and the ask price has
been just 0.6% over the course of the bonds' trading
Why
would
history.
the world financial market's assessment of the value of these bonds be informative
about the Surge's impact on Iraq's prospects? The reason
is
that
owning
a
bond
entitles the
holder to a
stream of payments. In the case of the Iraqi bond, each $100 efface value of these bonds entitles holders
to receive $2.90
on January
on the
15"'
of every January and July through January
15,
2028 plus an additional $100
15, 2028.
In a liquid market, the current value or price of a bond's stream of
payments depends on the
perceived default risk and the yield curve. The former captures the probability that a bond's issuer will
stop
making
the scheduled
payments before the maturity
18
date.
An
increase in the probability of default
would reduce a bond's
measures the
value, because the bondholder's expected
riskless interest rate (e.g.,
present and the scheduled payments.
US
Treasury
bills) at
payments
will
have declined. The
latter
each of the time intervals between the
Here, an increase in the yield curve would reduce the bond's value
because the future payments would be more heavily discounted and therefore
less valuable today.
'*
Secondary bond markets aggregate the available information on the probability of default and
future values of the yield curve to determine the value of
bonds
(or their scheduled set of payments).
Since the initiation of trading, the price of a $100 face value Iraqi bond has ranged between $56.18 and
$73.98.
The lower
prices reflect periods
higher or the yield curve was higher.
participants to reach an agreement
Prices
and
Yields
where the market's perception of the probability of default was
Thus, the secondary bond market provides a venue for market
on the expected probability of defauh and the yield curve.
of Iraqi Bonds. With
this
background,
I
now
turn to a comparison of
how
the
difference or spread between the yields of the Iraqi state bonds and three alternative bonds changed after
the Surge began.
Each of the
alternative
bonds
is
intended to serve as a control to avoid confounding the
impact of the Surge with other drivers of defauh risk or
Qatari
bond due
Qatar's
economy
in
is
2030 with a 9.6% coupon
alternative
for changes in yields due to
generally.
least
The appeal of this
The
first
as an alternative
alternative is a
is that, like Iraq,
heavily dependent on oil revenues, so, in addition to having a similar maturity date,
this will at least partially adjust for
The second
rate.
shifts in the yield curve.
The maturity date
one dimension.
changes in the probability of default due to changes in
bond
is
the
Lehman Emerging Market Bond Index
macroeconomic
for this index
Yields,
which adjusts
factors that affect default risk of developing countries
was 2020 on August 27, 2007, so
The Qatari bond and Lehman Index
partially coincides with the
oil prices.
subprime mortgage lending
it
isn't a perfect
match on
at
are also appealing, because the Surge period
crisis that
caused a "flight
to quality" for
US
treasuries.
'*
Since the Iraqi bonds are denominated in
US
currency, fluctuations in the value of the Iraqi currency do not affect
prices.
19
The
third alternative
bond
is
the
US
Indexed Treasury Bond due
Inflation
a nearly identical maturity date to the Iraqi bond and
impact of the Surge with changes
is
2028. This bond has
in
used as an alternative to avoid confounding the
in the yield curve.
Figure 10 graphs these three spreads daily where days are again normalized so that day zero
February
The spread between
2007.
14,
Figure 10, since
its
One key
yield
is
difference between
it
bond
every point in time.
is
Thus,
this
all
relevant information
Bush announced
February
14,
the Surge
2007. Further, there
December, and
is
it
is
on January
possible that
the rest of the paper,
line is
drawn)
The
I
is little
will
bond
is
that
is
is
reflected in the asset's price
traders
t
=
is
and are predictable.
no perfect
start date
for the analysis.
2007, but the Surge's activities didn't begin until
that the military
were aware of
-50.
It is
had begun planning for the Surge
To account
this plarming.
apparent that the choice of
start
emphasize
results that
compare spreads
to the last available yields in our data file at
figure reveals that the spread
It
t
between the
at
t
=
(which
is
in
for all of
date isn't
where the
Iraqi
bond and
increased by 153 basis points
Interestingly, the increase in spread
is
vertical
=191 (August 24, 2007).
all
three alternatives
when
the Qatari
bond
and 114 and 170 basis points when the comparisons are the Lehman index and
respectively.
" There
outcome variables
variation in the spreads during this 50 day period.'^ For consistency with
substantially in the period after the Surge.
alternative
10,
some evidence
these potential start dates, the graph begins with
important because there
the right axis of
theory implies that pre-existing trends in prices do not predict future
Since financial markets are forward looking, there
President
shown on
is
not necessary to control for pre-existing trends in prices. This
Further, the current prices reflect any future events that affect prices
prices.
late
US bond
prices (or yields) and the other
because financial markets are forward looking and
at
the
substantially lower than the others.
markets theory implies that
efficient
bond and
the Iraqi
is
US
grew
is
the
bond,
almost entirely driven by an increase in the yields of
t = -50 (December 26, 2006) and t = -35 (January
two day's after President Bush's speech announcing the Surge. It may be appropriate to
attribute this decline to the Surge. However, even if this is the case, it is swamped by the increase in spreads that
occur after the Surge is initiated; so attributing this to the Surge would not affect the qualitative results.
12,
is
a decline in spreads of about 20 basis points between
2007) where the
latter is
20
the Iraq bonds,
which increased by
1
74 basis points
in the yields
of the
bonds began trading.
alternatives."" Notably, these are easily the largest spreads since the Iraqi
The
changes
in this period, rather than
increase in the spreads appears to be due to genuine increases in the market's expectation of
a default (and not due to changes in oil prices, the supply of capital to developing countries or changes in
With a few assumptions,
the yield curve).
market's expected default
If
rate.
is
it
we assume
possible to back out an estimate of the change in the
that the
market for these bonds
that the expected default rate is constant across periods"', then following
following relation:
the
is
coupon
rate as a percentage
and g
asset,
(1+0*
number of periods
the
left-hand side
is
The right-hand
asset's yield.
(i.e.
of the face value,
^°
and
An
/-
et. al
(1989)
we
derive the
is
the probability of not defaulting in each period,
is
/
(!+'•)"
+'")')
the risk-free interest rate, r
coupon recovered
is
C is
the yield of the risky
in the case
of a default. The
the expected present discounted value of a dollar's worth of risky asset held to maturity.
side
is
Under
the certainty equivalent of the payoffs of the risky asset, discounted
perfect competition, these
The software Mathematica
i,
coupons),/?
y^c^
(1+0"
J
the proportion of the face value plus the
is
Ma
perfectly competitive and
<
\^
Here, n
is
for each day.^"
The
is
by the risky
must be equal.
used to solve
this
yield series of the Iraqi
equation for p, given different observations of
bond
is
used for
r
n,
and the yield of the 20-year
is to conduct a cash flow analysis of the value of the Iraqi bond with
and t = 191 yield curves. This exercise demonstrates that changes in the yield curve have little impact on
the value of the Iraqi bond in this period. This finding is not surprising in light of the flat and roughly constant yield
the
t
alternative approach to inferring the role
=
curve over this period.
"'
This method for inferring the expected default risk requires the assumption that the probability of default is
This assumption is likely to cause the numbers in Figure 1 1 to understate the expected
constant in each period.
default rate in the next
surviving until 2028
given what
we know
is
few years (because
if the Iraqi
government survives
until
2027, then the probability of
surely higher than the probability that Iraq survives from today until one year from
today).
Thus, the probability of default approximated using
this
fonnula
is
now,
likely to be
an
underestimate of the probability that Iraq survives from today to one year from today or other periods closer to the
present.
"'
There are a variety of ways to account for the fact that each day that passes reduces the number of days until the
next coupon, and thus increases the probability that the next coupon will be redeemed (as well as reduces the
amount by which we should be discounting the coupon). For example, in the above formula, / can be replaced by
(d-124.5)/182.5, where d is the number of days until the bond's due date and 182.5 is half of a year. Such a
substitution is computationally unfavorable. I created similar series by rounding the expression (d-J 82.5)/] 82.5 and
substituting it for t in the numerator or both the numerator and denominator. One problem that arises is that when
21
Treasury bond
used for
is
i.
The bond pays coupons every 6 months
at
a rate of 2.9%. After the invasion
of Iraq, the Paris Club decided to reduce Iraq's sovereign debt by 80%. Assuming that a similar deal will
be offered to future governments of Iraq should the current one
Figure
payments per
plots the annual probability
1 1
After the Surge begins,
market didn't believed
it is
that
it
never
5.65% and 6.10%
this
rises to
that the probability that Iraq will default
Credit Default
from a
relatively
new
on
Swap Premiums for
in default risk
the seller
on
its
and the yield curve
provided by
Series,
in the
premium each period
bond increased
Iraqi Bonds.
risk
The
results are striking.
even
in the early
is
is
=
0.2.
two coupon
The annual
5.75%
at
t
=
days of the Surge the
Perhaps even more notable, the
the end of the period.
clear conclusion
This
is
an approximately
world financial markets believe
that the
after the Surge's initiation.
To supplement
these results,
I
now
turn to data
These instruments provide a direct
and therefore mitigate concerns about confounding change
above analysis. In a credit default swap, the purchaser pays the
until either
defauh occurs or the swap contract matures. In exchange,
In essence, these credit default
obligation.
premiums)
that
obligated to buy back from the buyer the defaulted
is
To
its
1-p' (because there are
days before the Surge and
financial instrument, credit default swaps.
measure of the market's expected default
seller a fixed
is
use a recovery rate of g
Iraq's future prospects.
8.14% by
The
increase in the expected default rate.
in the
low again implying
would improve
expected annual default probability
40%
of default, which
we
year), against the date relative to the Surge's initiation.
probability of a default ranges between
0.
fail,
bond
at its
par value
if the issuer defaults
swaps are insurance against default
for buyers that are
sellers.
further explore the impact of the surge
for 5-year credit default
which
is
swaps for the
Iraqi
on default
bond and
risk,
the
I
CDX
collected data on prices
(i.e.
Emerging Market Diversified
an index of 5 -year credit default swaps for an index of bonds from emerging markets."
About 75% of the constituent
credit default
swaps are sold on government issued debt and the remainder
made, the equation has multiple solutions for p that are real and between and 1 (one solution
is very close to what we get making no substitution). In light of this finding, I decided not
to make any adjustment for the number of days to the next coupon (of course, I adjust for the number of coupons
remaining). However, none of the adjustments change the basic message of the graph.
"^
The index is reconstituted every 6 months, which can cause relatively minor jumps as bonds are added and
subtracted at the reconstitution dates. I focus on 5-year CDS prices because a private conversation with a Lehman
Brothers trader indicated that these are the most liquid CDS products.
such a substitution
is
is
implausible, the other
22
from private companies located
are
LLC,
The
in
emerging countries. The index
a consortium of 16 investment banks, and Markit
constructed by
is
Group Limited,
CDS
a provider of independent data."'*
Market
year costs
1%
full line in
Figure 12 plots the spread in the prices between the Iraqi
1
The
CDS
index.
It is
of the face value of the bond.
apparent that the spread
CDS
and the Emerging
For example, the price of a
is substantial.
CDS
varied between 514 and 530, while the index's varied between 73 and 88 basis points between
=
CO
indices are priced in basis points per year, so a price of 100 basis points implies that insurance against
default for
t
Index
t
for Iraq
= -50 and
This underscores that the expected default rate of the Iraqi bonds was very high, even by emerging
0.
market standards.
The high
price of the Iraqi
CDS
poses a challenge for determining the impact of the surge,
especially in the context of the subprime mortgage crisis that unfolded over part of this period and raised
the expected default rate of all risky bonds.
prices similar to the Iraqi
CDS
I
was unable
price prior to the Surge.
Venezuela) had idiosyncratic shocks to default
strategy
CDS
is
to use the
prices.
It is
it
Further,
some of
risk in this period.'^
Thus,
evident, however, that the perfect counterfactual for Iraq
increases from
is
falls
below the
14,
2007
Both time
series
I
is
believe the most defensible
as a counterfactual for Iraqi
unavailable.
to
676 basis points on August 24,
level at the initiation of the Surge.
the evolution of the spread and the implied annual defauh rate in Figure
'*
CDS
a dramatic increase in the spread after the Surge began.
436 basis points on February
2007. Further, the spread never
bonds with
the higher priced ones (e.g.,
Emerging Market index (which includes Venezuela)
Returning to Figure 12, there
Specifically,
to identify individual sovereign
came from Markit which aggregates
1 1
is
The
similarity
between
striking."^
Markit's data on GDP
from several investment banks and then clean
data from over 85 dealing firms.
prices are the most reliable available, because they collect trading data
the data to create this time-series.
'^
aimed at nationalizing large sectors of the oil sector in this period. This has led to
Exxon and Conoco Phillips, which have refused to agree to Venezuela's terms. The outcome of this
dispute is currently unknown but expensive litigation- in international courts and the seizing of Venezuelan owned
refineries in the US are possible outcomes. The bond market has interpreted these events negatively and the price of
Venezuela
instituted legislation
a dispute with
CDS
"*
for
It is
Venezuelan bonds has increased dramatically.
possible to infer the expected default probabilities from the
few assumptions
CDS
premiums
for the Iraqi state
recovery rate upon a default) (O'Kane and Tumbull 2003).
bonds with a
These calculations indicate
that the expected defauh rate increased by about 50% from 6.68% on February 14, 2007 to 10.13% on August 24,
1997. This increase is similar in magnitude to the estimated 40% increase in the expected default rate from the
yields on the Iraqi bond yields. It is noteworthy that during this same period the expected default rates for Qatar and
(e.g., the
23
What
The defauh
factors determine the financial market's expected probabihty
risk
is
a function of at least two factors.
The second
in Iraq.
is
the continued presence of the
US
projected stability in Iraq and domestic
are due to domestic
US
politics,
The
US
which are unrelated
Iraqi
bonds?
the current and projected future stability
first is
military,
If the
politics.
of default for the
which
movements
to the current
is likely
a function of current and
in default risk in
and projected
Figures
1 1
stability in Iraq,
and 12
then this
exercise cannot be used to judge the Surge.
With
mind,
this issue in
spreads occurs around
t
=
important to highlight that a substantial portion of the increase in the
it is
which
150,
is
the Administration released their "Initial
July 14, 2007.
Two
big events occurred right around then. First,
Benchmark Assessment Report" on
Surge had led to substantial progress toward meeting
that the
benchmarks
for progress in Iraq.
Iraqi state.
While
it
may have
that the causal factor here
was
This event provided
new
many of the
new
1
1
2*.
This failed to find
8 congressionally
mandated
information on the current functioning of the
strengthened the case for withdrawing
the
July
US
troops from Iraq,
it is
evident
information, not a shift in domestic politics.
Second, the Senate Democrats tried to force a vote on a resolution that would have withdrawn
troops from Iraq. Supporters of the resolution failed to overcome the filibuster, but the cloture motion did
receive 52 votes (on July 18*), enough to pass the resolution if it had
vote
is
a political event.
However,
it
decided along party lines and did not
the market concluded that the
default risk
is
alter policy surprised the market.
US would
if
it
did,
was almost
it
is
unclear whether
way
to assess
whether internal
This graph
is
US
politics
elected president in 2008.''
Egypt increased from 0.23%
to
The estimated
coefficient
0.26% and 0.98%
may not form the
to
on the contract price
1.04%, respectively.
caused the
is
15.9 and
if
a Democrat
would
easily be
These two comparisons were the ones
best counterfactual for Iraq.
is offered through https://www.intrade.com/v2/ which is a marketplace for numerous prediction
These prediction markets have an impressive track record of aggregating disperse information into
forecasts about future events (Surowiecki 2004; Wolfers and Zitzewizt 2004).
The contract
efficient
on
derived from a regression of the
spread against a constant and the price of a prediction market contract that pays $100
markets.
entirely
seems unclear.
provided to me. However, they
"'
Even
The outcome of this
stay for a shorter or longer period of time so the impact
increase in the implied annual default risk of Iraqi bonds.
is
to a vote.
unclear whether the outcome of a vote that
Figure 12 provides a more quantitative
CDS
come
24
judged
to
be
statistically different
from zero by conventional
positively associated with this variable."**
The dashed
difference between the full line and dashed line
component of the spread
that cannot be explained
Since Democrats
a proxy for the probability that the
this analysis,
due to changes
The
was
from
t
US
to
t
=
by the contract
seems reasonable
US
I
cannot
assume
the conservative assumption that
CDS
and the
This difference
is
the
market contract
that the prediction
military presence in Iraq.
its
movements
is
For the
in this variable are entirely
spread increased dramatically but the predicted spread
CDS
about the 2008 Presidential election.
88%
191 accounts for about
how
worth considering
of the increase
the Surge could
test the different possibilities.
combatants in Iraq
(e.g., the
US
One
in the
military's
spread in this period
new
in the
largely unrelated to
expected default
expected default risk
is
in the residual
rate.^°
This result
due to changes in
for this explanation.^'
worsen expectations about
alliances with the Sunnis
Over the relevant sample, t =-50 through t = 191,
on t = was $56.3 and it was $58.0 on t = 191.
is
For example, the increase
possibility is that the Surge
"
^'
to
US politics, but it substantially weakens the case
It is
is
price.
will substantially reduce
does not eliminate the possibility that the increase
domestic
so the spread
line plots the predicted default rate
the estimated residual.
Put another way, the increase in the
flat.
3),
domestic issues."'
in expectations
=
make
it
striking finding is that the actual
essentially
changes
in
I
of
general and the leading Democratic presidential candidates in particular are
in
aggressively calling for leaving Iraq,
purpose of
is
criteria (t-statistic
Iraq's future
even though
hardened the positions of the
may deepen
the value of the contract ranges
Shi'a's resentments)
from $53.0
to $59. 0.
Its
value
In practice, this assumption implies that the possibility that the public's assessment of the degree of stability in
Democrat will be elected President in 2008. This assumption is almost certainly
for Democrat Presidential candidates seems based on dissatisfaction with the
allows me to stack the deck in favor of the hypothesis that domestic political factors are
Iraq affects the probability that a
incorrect, because
War
in Iraq.
some of the support
However,
it
the source of movements in the expected default rate.
^^
This result
is
nearly identical with alternative models
(e.g.,
estimating the parameters with data from the period
before the Surge only, including the current price of the contract and three lags, and modeling the contract price with
a cubic polynomial).
of the increase
''
I
If
t
= -50
is
used as the starting date, then the increase
in the residual
accounts for about
66%
in the spread.
also investigated using data on contracts that
Democrats
will hold a majority in the Senate
and House
in 2008.
For example, there were a number of periods lasting several
days with zero trades. Overall, mean daily volume for the senate and house contracts are 2.4 and 0.3 contracts,
respectively, compared to 68.4 for the presidential contracts.
The point is that the thinness of these markets
These contracts are
undermines the
thinly traded during this period.
credibility of
an exercise using these data (although using them doesn't affect the qualitative
findings).
25
and
that this has
situation
lowered the prospects for a stable
have not been followed by meaningful
the depth of the political problem in Iraq
revelation of the magnitude of this problem
President will remove
The
first
US
troops sooner than
of these possibilities
is
is
Iraq.
Another
is
that the
Thus, the Surge
political gains.
greater than previously
may have
improvements
in the security
may have
was recognized.
revealed that
Further, the
increased the probability that Congress or a
new
was expected.
consistent with the Surge directly undermining the prospects for
a stable Iraq, while the other suggests that the Surge allowed for a clearer picture about the future.
Regardless of which explanation
is
correct, the
Surge the spread between the prices for
declined.
The same
CDS
most noteworthy finding from
for the Iraqi
true for the implied annual probability
is
this analysis is that after the
bonds and the emerging markets index never
of default
in Figure
1 1
.
Thus
to date,
financial markets have never believed that the Surge will increase the prospects for a stable Iraq
in fact
and may
be undermining them.
Before proceeding to the conclusion,
thus
world
it
is
important to
make
clear a link that has
been unstated
This subsection's analysis implicitly equates a default on the Iraqi bonds with a failure of the
far.
Surge to achieve
complete
its
goals.
state failure.
extraordinary.
It is
even as a stable
Indeed, governments have defaulted on sovereign debt before without
However, the
difficult to
political
political
and security
imagine a scenario
and security environment
in
is
situation in Iraq
which the
Iraqi
and
its
neighborhood
government defaults on
its
is
bonds
achieved.^"
VI. Conclusions
There
and whether
it
is
is
a paucity of facts about the effects of the recent military "Surge" on conditions in Iraq
paving the way for a stable
Iraq.
Selective, anecdotal
and incomplete analyses abound.
Policy makers and defense planners must decide which measures of success or failure are most important,
but until
^'
now few
if
any systematic analyses were available on which
to
base those decisions. This paper
A so called "soft" or "hard" partition of the current Iraq doesn't imply a default.
include an oil revenue sharing agreement that would commit to servicing the debt.
26
Such a scenario could well
8
applies
modem
reliable
and widely-cited sources
The analysis
situation has
of trained
Iraqi Security
to
coalition
may
There
and
civilian fatalities
concern
is
room
to
effect
on key trends
areas, such as oil production
is
challenging and uncertain.
in
and the number
profitable decisions
on the
and
for personal biases in this setting
it
how
appeal of using financial markets
this necessarily requires
and
this
data from world
central question of whether the Surge has increased or
An
Iraq.
Moreover,
role.
however, another way to assess the Surge. This paper has shown
make
security
have declined without any concurrent increase
allow analyst judgment and politics to play too large a
is,
The
in Iraq.
making
is
that traders' only
correct projections.
There
therefore isn't surprising that financial markets have a
track record of predicting future events.
I
examine the price of
Iraqi state bonds,
world financial markets. After the Surge, there
to alternative bonds.
toward a stable Iraq and
assessments that
is
which the
fail to
may
This finding suggests
in fact
that, to date, the
be undermining
new
data
it.
is
that
In
many
40%
Surge
respects,
in the market's
is
it
failing to
is
facts
available.
expected
pave the way
consistent with recent
many of
the
1
(GAO 2007).^^
on the Surge's impact.
Further, the
new
facts
It is
straightforward to
can be used to aid decision
the future of the Surge.
Most of these benchmarks
progress in Iraq
becomes
new
on
currently servicing,
is
find that the Surge has led to substantial progress in meeting
This paper has produced a series of
making about
government
a sharp decline in the price of the Iraqi bonds, relative
congressionally mandated benchmarks for progress in Iraq
update these facts as
Iraqi
This decline signaled an increase of approximately
annual probability of default.
point
on three key dimensions.
Forces have shown no improvement or declined. Making sound decisions about
diminished the prospect of a functioning
^^
derived from more than a dozen of the most
However, other
Iraqi troops.
financial markets can be used to shed light
good
file
proceed in Iraq based on such conflicting indicators
uncertainty
isn't
data
to assess the Surge's impact
improved insofar as
among
new
mixed evidence on the Surge's
reveals
casualties
how
techniques to a
statistical
is
are measures of political progress.
Notably, Chaney (2007) found that political
associated with increases in the price of the Iraqi bonds in the January
world financial markets appear
- August 2006
to believe that political progress is crucial for a stable Iraq to
27
period.
The
emerge.
More
broadly, the paper
shows
that
even
in
unconventional wars,
analyze important outcomes to learn about the war's success.
that
Finally like
world financial markets may be a useful new source of information
28
it
is
feasible to
Chaney (2007),
measure and
it
highlights
to judge a war's success.
REFERENCES
Ashraf and Victor Tanner, "Sectarian Violence: Radical Groups Drive Internal
Displacement in Iraq," The Brookings Institution (2006).
al-KJialidi,
.
"Baghdad Security Plan
Officially
Named," American Forces Press Service
.
1
8 Feb. 2007.
Bilmes, Linda, and Joseph E.
Stiglitz. "Encore: Iraq Hemorrhage." Milken Institute
Review, December 2006: 76-83.
Bumham,
Gilbert,
Riaydh
Lafta,
Shannon Doocy, and Les Roberts. "Mortality
invasion of Iraq: a cross-sectional cluster sample survey." The Lancet
2003
after the
-
Vol. 368, Issue 9545, 21
October 2006, Pages 1421-1428
Bush, George
(Pres.), "President's
Address to the Nation," January
from
10, 2007. Retrieved
<wvvrw.whitehouse.gOv/news/releases/2007/0 1/20070 1 1 0-7.html>.
Chaney,
Eric.
"Assessing Pacification Policy in
Iraq:
Evidence from
Iraqi Financial
Markets."
Mimeograph, August 2007.
Iraq: Iraqi Government has
Not Mett Most Legislative, Security, and Economic Benchmarks." Testimony of David M.
Walker Comptroller General of the United States on September 7, 2007.
General Accounting Office. "Securing, Stabilizing, and Rebuilding
"Iraq erupts into fresh violence," Al-Jazeera,
1
March 2006.
Kagan, Kimberly, "Iraq Report: Enforcing the Law: The Baghdad Security Plan Begins,"
Institute for the Study of War 15 March 2007.
.
Kagan, Kimberly, "Iraq Report: The Battle
for Diyala," Institute for the
Study of War 7
,
May
2007.
Kagan, Kimberly, "Iraq Report: The Real Surge: Preparing for Operation Phantom Thunder,"
Institute for the Study of War 6 July 2007.
,
Kagan, Kimberly and Fredrick Kagan, "Backgrounder #3: Operation Phantom Thunder
Factsheet" Institute for the Study of War 23 June 2007.
,
Levey,
Noam N.
and Alexandra Zavis. "U.S. Drops Baghdad
Electricity Reports."
Los Angeles Times,
27 July 2007.
Ma, Christopher, Ramesh Rao and Richard Peterson. 1989. "The Resiliency of the High- Yield
Bond Market: The LTV Default." The Journal of Finance 44(4), 1085-1097.
O'Kane, Dominic and Stuart Tumbull. "Valuation of Credit Defauh Swaps." Lehman Brothers
Mimeograph, Lehman Brothers. April 2003.
Pacific Disaster
Management Information Network, "Iraq Humanitarian Assistance Report,"
in Disaster Management and Humanitarian Assistance 22 March 2004.
Center of Excellence
,
Retrieved from <http://www.who.int/disasters/repo/I2293.pdf>.
29
Press Conference with Lt. Gen.
June
Ray Odiemo, Commanding General Multinational Corps
DOD
2007,
22,
News
Iraq,
Retrieved
Transcript.
from
<http://www.defenselink.mil/transcripts/transcript.aspx?transcriptid=3998>.
Press Conference with Maj. Gen. Joseph Fil
Baghdad and
1st
Jr.,
Commanding General
Cavalry Division, February
16,
2007,
Multinational Division,
DOD
News
Transcript. Retrieved
from
<http://www.defenselink.mil/Transcripts/Transcript.aspx?TranscriptID=3891>.
"Scores killed in Baghdad Attacks,"
BBC News
,
14 September 2005.
Smith, Steven D., "Bush: Iraqis Must Help Stem Sectarian Violence" American Forces Press
Service 13 Jan. 2007.
,
Surowiecki, James M., "The
Wisdom of Crowds: Why
How Collective Wisdom
the
Many Are
Smarter Than the
Shapes Business, Economies, Societies and Nations."
Few and
New
York: Doubleday, 2004.
Sunshine. Robert A., "Statement before the Committee on the Budget, House of Representatives:
Estimated Costs of U.S. Operations in Iraq and Afghanistan and of Other Activities Related to the
War
on
Terrorism,"
31
July
2007.
Retrieved
<http://www.cbo.gov/ftpdocs/84xx/doc8497/07-30-WarCosts_Testimony.pdf>.
Wolfers, Justin and Eric Zitzewitz. "Prediction Markets." Journal of Economic Perspectives.
Volume 18(2), Spring 2004.
30
from
DATA APPENDIX
This data appendix provides information on the data sources for
this paper.
It
includes separate
subsections on each of the outcome variables analyzed.
I.
Iraqi Security Forces
Weekly
Troop Strength
makeup of
data on the size and
Weekly
State's Iraq
of
"pillars"
the
Status Report
US
the Iraqi Security Forces (ISF)
(IWSR).
policy
in
comes from
the
Department of
This report provides updates on the progress toward eight
The
Iraq.
http://www.state.gOv/p/nea/rIs/i-pt/iraqstatus/
weekly
were
reports
downloaded
from
.
Each report contains a slide entitled "[2.] Transition Iraq to Security Self-Reliance - Iraqi Security
The slide contains a chart that shows the number of persons trained and equipped in each
division of the ISF. At the bottom of the slide, an "as of date is given (e.g. "Data as of February 5,
2007"). This date is recorded in the variable isf_date. The number of days that this date is later than
February 14, 2007 is given in elec_count.
Forces".
Three parts constitute the Ministry of the Interior's forces: Police, National Police, and Other Ministry of
These are recorded in the variables police, nat_police, and other_moi, respectively.
the Interior Forces.
Before July
7,
2006, National Police and Other Ministry of the Interior Forces were combined in Other
Ministry of the Interior Forces observations.
Army, Air Force, and Navy. These
An
Three parts constitute the Ministry of Defense's forces:
are recorded in the variables army, navy,
important characteristic of the data
is
that
the
and air_force, respectively.
Ministry of the Interior numbers do include
unauthorized absence personnel, whereas the Ministry of Defense numbers do not include unauthorized
Mol numbers
absence personnel.
Thus, the
they include force
members who have been equipped and
are inflated
above the actual strength of the force because
trained but have left the force without
permission.
These numbers do not include what the
IWSR
estimates as
Army numbers
personnel working in 27 ministries.
144,000
facilities
protection services
include Special Operations Forces and Support
Forces.
The
IWSR
lists its
"DoD
source as
Iraq
Weekly
Status Report" or
"DoD
Input to Iraq
Weekly
Status
Report".
Observations that are exact duplicates (including isf_date) were discarded.
What
follows
is
a
list
observations that have changed dates or been eliminated and the reasons:
—
there
were two observations
observation for
1
for 12/28/2005, kept the
one
that
was not
identical to the
2/7/2005 (presumably, they forgot to update the numbers and the second
observation was reported more recently)
~
replaced the
first
observation for 7/7/2006 with the second because the second had
more detail as to how Mol
was reported more recently)
~
replaced the
first
forces
were broken down and was otherwise
identical (and
observation for 9/13/2006 with the second because the second was
more recently reported
31
of other
~ moved
the
first
the second observation for 9/18/2006 to the
~ removed
the observation for 2/20/2007 because
and they were
~
there
week of 9/20/2006 and discarded
observation
in the
it
was
identical to that for 2/19/2007
same week
were two observations
for 3/5/2007, kept the
one
that
was not
identical to the
observation for 2/20/2005 and 2/19/2005 (presumably, they forgot to update the numbers
and the second observation was reported more recently)
~
all
dates 6/13/2007 and
beyond
are extrapolated (one
week behind
the date of the
report)
~ moved
the observation for 6/19/2006 to the
6/26/2006
II.
to the
week of 6/21/2006 and
the observation for
week of 6/28/2006
Oil production
Weekly
oil production data comes directly from the Department of State's Iraq Weekly Status Report
(IWSR), which provides updates on the progress toward eight "pillars" of the US policy in Iraq. The
weekly reports were downloaded from http://www.state.gOv/p/nea/rls/ipt/iraqstatus/
.
Each report contains a slide entitled "[5.] Help Iraq Strengthen Its Economy - Crude Oil Production".
The slide contains a chart with an entry for each of the last nine weeks and the current week. Each entry
is the weekly average of millions of barrels per day of crude oil produced.
The current week's entry is
also reproduced beneath the chart.
When
there was a conflict between the text and the chart, the number given on the chart was used. There
were two instances of this: the observation for 9/10/06 should be 2.37 (from chart in weekly report), but it
is given as 2.47 in the actual weekly report, the observation for 7/22/07 should be 2.10, but it is given as
2.06 in the actual weekly report.
The
IWSR
lists its
source as "Department of State,
NEA-1-ECON, 202-647-9885".
III. Electricity availability
comes directly from the Department of State's Iraq Weekly Status Report
(IWSR), which provides updates on the progress toward eight "pillars" of the US policy in Iraq. The
weekly reports were downloaded from http://www.state.gOv/p/nea/rls/rpt/iraqstatus/
Electricity availability data
.
Each report contains a
slide entitled "[4.]
Help Iraq Build Government Capacity and Provide Essential
- Electricity Overview". This slide contains a line graph of the daily load served and estimated
demanded in megawatt-hours (MWh). It also contains several bullet points that provide additional
figures. One such figure is the average of available hours of electricity in Baghdad and nationwide over
the past week (exact dates for which are given.)
Services
IWSR
name and
phone number. The point-ofThe IWSR
stopped reporting this number after May 7, 2007, though the slide and graph are still present. It was
resumed on July 3 1 2007 with a caveat "Average hours of power via electricity grid after meeting
The
gives as
its
source a point-of-contact consisting of a
contact changes from time to time; in 2007
it
read
,
demand
for essential services."
32
"POC Bob Means
a
(202) 647-9815".
IV.
Weekly United
States Forces
Wounded
The US wounded data were downloaded from http://icasualties.org/oif/woundedByWeek.aspx. From
the
icasualties.org website:
The Department of Defense does not release the names of the wounded, nor do they
provide details on how and when soldiers are wounded. Therefore, we are not able to
provide a list for each and every wounded soldier.
The
DoD does provide a weekly summary,
and returned
We
to action within 72
stating updated totals for the number wounded
hours and the number wounded not returned to action.
arrive at a weekly total by subtracting the
methodology
is
not precise because
it
new
total from the
previous weeks. This
does not account for possible restatements of total
from previous week.
The Department of Defense also maintains a separate list detailing the number of
wounded by month and ser-vice branch. We transfer these totals to our database each
month and provide the same data in html format. There are some inconsistencies in this
reporting - numbers in this chart boimce back and forth from one month to the next.
While compiling items for our news section, we often find articles about wounded
servicemen and we have provided an archive of these articles. Within this list you may
find multiple mentions of the same person, as well as references
since succumbed to their wounds.
to soldiers
who have
The number of wounded who were returned to duty within 72 hours and the number of wounded who
were unable to return with 72 hours are downloaded from the icasualties.org website. These two
variables are also
summed to
obtain the total
This "flow" data comes from differences
a quasi-weekly basis.
may
reflect soldiers
wounded each week.
in the
stock variables reported by the Department of Defense on
One problem with these series is that changes
who were wounded in the previous week and
from week to week
number of wounded
in these variables
revisions in the
from previous weeks. The latter is the variable of interest, while the former adds measurement
The measurement error is unlikely to serve as a source of bias, because it seems likely to have been
soldiers
error.
constant before and after the Surge.
V. Coalition Fatalities
Daily
data
on
coalition
deaths
We downloaded the data from
comes from iCasualties.org.
which lists each death individually. We then aggregated the data by
http://icasualties.org/oif/Details.aspx,
day.
iCasualties follows a rigorous recording methodology.
A US
death
is
recorded
when one of four
events
news story or obituary that reports the death of a US service member is released (this must
also be confirmed by the DoD), (2) Centcom or the Multi-National Force in Iraq (MNF) releases the name
of a soldier who has died (Centcom releases must be confirmed by the DoD), (3) the DoD itself releases
the name of a soldier who has died, (4) the name of a soldier not already recorded appears in the weekly
updated Department of Records listing of Operation Iraqi Freedom (OIF) fatalities.
occurs: (1) a
33
UK
fatalities are
recorded when the Ministry of Defense confirms that a soldier has died.
American, non-British
fatalities, the iCasualties
website
For non-
states:
Deaths for non English speaking countries prove to be the most difficult for gathering
information. Here we rely on our readers to point us in the right direction and try (often
in vain) to translate announcements into the bits of information we need for
documentation.
Complete documentation of iCasualties.org's methodology
http://icasualties.org/oif/Methodologv.aspx
is
VI. Yields for Inflation-Indexed Treasury
Bond Due 2028
The
us2028
for
series
available at
.
was
downloaded
from
The Fed denotes this series
the 30- Year 3-5/8% Treasury Inflation-Indexed Bond, due on
http://research.stlouisfed.org/fred2/series/DTP30A28/downloaddata?cid=47
as
DTP30A28.
It
contains daily yields for
.
4/15/2028.
VII. United States Treasury Yield
Curve
United States yield curve data comes from the U.S. Treasury's website. Nominal rates were downloaded
from http://www.ustreas.gov/offices/domestic-finance/debt-management/interest-rate/yield.shtmI
The
yield rates are constant maturity rates.
yield curve.
According
.
This means that they are interpolated from an estimate of the
to the Treasury's website,
"The Treasury yield curve
is
estimated daily using a
cubic spline model. Inputs to the model are primarily bid-side yields for on-the-run Treasury securities."
In this way, the Treasury provides yields for the following maturities:
10, 20,
and 30 years. Furthermore, we
If the rate for a particular maturity
linearly interpolated rates that
was not
reported,
we
1, 3 and 6 months and 1, 2, 3, 5, 7,
were not reported by the Treasury.
created a line running through the nearest
For example, if the 7and the 10-year rate was given as 7%, then we assigned (2/3) * 4% + (1/3) *
and (2/3) * 4% + (1/3) * 7% = 6% to the 9-year rate.
maturities on either side of the missing maturity and read the rate off of this line.
year rate was given as
7% = 5% to the
4%
8-year rate
and Lehman Emerging Markets Index Bond Price and Yield Time Series Data
from Bloomberg Financial Data Service.
VIII. Iraqi, Qatari,
We downloaded daily prices and yields for an Iraqi state bond due in 2028
(annual coupon of 5.8%) and a
bond (annual coupon of 9.75%) due in 2030 from the Bloomberg financial data service. Both the
and Qatari bonds are denominated in US dollars. The CUSIP for the Iraq bond is EF2306852 and it
Qatari
Iraqi
is
EC2682416
for the Qatar bond.
We
also obtained a times series of prices and yields for the US Lehman Emerging Markets Bond Index
from Bloomberg. The Lehman Index includes fixed and floating-rate US dollar denominated debt fi-om
emerging markets. It had a maturity of March 5, 2020 as of August 27, 2007 when the data used in the
analysis was downloaded.
IX. Coalition Troop Levels
34
The monthly
comes from
data for troop strength
TROOP STRENGTH
IRAQ SINCE
MAY
the Brookings Iraq Index table titled
"COALITION
2003."
The August 27, 2007, version of the Index was
accessed at http://www3.brookings.edu/fp/saban/iraq/index.pdf The table was assembled by Brookings
from a variety of news sources (a list is available in the footnotes of the Iraq Index).
IN
.
X. Civilian Fatalities from IraqBodyCount.org
Our primary source
on civilian fatalities comes from the IraqBodyCount.org (IBC) website. The
Hamit Dardagan and John Sloboda. It is important to note upfront that the
website is partisan that is, anti-war but their methodology is conservative and seeks to provide an
accurate lower bound for the number of civilian deaths reported.
for data
principal researchers are
—
—
The data were downloaded from http://www.iraqbodvcount.org/database/ Documentation of the IBC's
methodology is available at http://www.iraqbodvcount.0rg/background.php#methods and it is briefly
summarized here.
.
which civilian deaths occurred separately. To create a daily time
from incidents occurring on the same day. For some incidents, the date was
uncertain. In such cases we always chose the most recent possible date of the incident that was listed in
The database records each
series,
we summed
incident in
the deaths
the website.
IBC
collects
news
A
updated website,
full
From each
from a variety of trusted news sources with internet access. Currently, they use
news source is one widely cited or referenced by other sources, with a daily
archives of stories, and fully public English- language access.
stories
38 such sources.
trusted
acceptable news story in which Iraqi civilians were reported as killed,
time, and location of the incident, as well as the military target (as stated
IBC
extracts the date,
by military sources), the
weapon(s) used, and the number of civilian deaths (in two measures as described below). IBC records the
deaths from an incident if and only if there are at least two news sources that report the incident.
there are two news sources reporting an incident, two measures are constructed, reportedminimum
and reportedmaximiim. In reportedmaximiim, the highest number of civilian deaths from any of the
Once
reporting
sources
is
recorded.
Unless
a
news source
reports
a
lower number of deaths,
the
reportedminimum is equal to reportedmaximiim. If multiple numbers are reported, the lowest is entered
as reportedminimum. This can be zero if "zero deaths" is reported. However, a wording like "unable to
confirm any civilian deaths" is not a report of zero deaths and thus is not entered in either column.
Moreover, when the report does not mention civilians specifically, this number is entered in the
reportedmaximum variable but zero is entered into the reportedminimum variable unless the proportion of
civilian deaths
is
given or a similar detail
is
given.) If a "family"
is
reported killed, this
is
entered as three
deaths.
IraqBodyCount.org also adds old incidents or hospital counts as much as 4 years old retroactively, which
two issues. First, in order to update the database, we must integrate old and new incidents. This is
done using the identifier provided by IBC. IBC does not specify that this identifier is unique, but a
computer-based check reveals that very few incidents with the same identifier have a different value for
the location field between the two data snapshots that we looked at. Those values that are different are
raises
usually just reworded or more- or less-specific locations.
The second issue is that the fact that IBC adds old incidents retroactively may represent inaccuracies in
the more current data. Comparing data gathered on 7/9/2007 and 8/21/2007 yields the following results.
35
These are the cumulative incidents added between these two datasets
that are hsted as occurring
BEFORE
the date given.
6/1/2003 -> 21
1/1/2004 -> 21
6/1/2004 -> 21
1/1/2005 -> 31
6/1/2005 -> 54
1/1/2006 -> 179
6/1/2006 -> 802
1/1/2007 -> 1312
6/1/2007 -> 1689
analysis is conducted with the mean of reportedminimum and reportedmaximiim.
reportedminimiim over the course of the entire database (which begins on March 21, 2003
The mean of
The
per day, while the mean of reportedmaximum
is
)
is
41.6 deaths
45.4 deaths per day. Thus, the differences are small and
9% of the mean of reportedminimum and reportedmaximiim. Moreover, they
59.6% of the days in the period from the first day of the database (January 1, 2003).
average about
equivalent on
are
coded a variable that differentiates between incidents in Baghdad and those not in Baghdad. This
was coded as 1 if the location field includes "Baghdad" (but not "near Baghdad" or "20 km north
of Baghdad") in its value and otherwise.
I
variable
XI. Civilian Fatalities from Icasualties.org
The
icasualites.org
counting
fatalities
web
site also
publishes a database of civilian casualties.
Their methodology for
Body Count's methodology, so the paper's analysis
Nevertheless, I estimated the mean shift and trend break
appears less rigorous than the Iraq
concentrates on the results using the
IBC
data.^"*
models with these data and this led to results that are qualitatively very similar to the results from the
IBC. These results are described in detail in the section of this paper that examines civilian casualties.
The
icasualties data contains the following disclaimer:
The numbers compiled here are not definitive, but they do provide a good baseline for
measuring daily reportedfatalities. No one should be surprised that monthly deaths tolls
from hospitals and morgues prove to be higher than what we have compiled - it should
be understood by all that the actual level of violence is higher than what the press is able
to report.
Complete documentation of iCasualties. org's methodology
http://icasualties.org/oif/Methodologv.aspx
'''
For example, the icasualties web
Iraqi Body Count."
site
is
available at
.
contains the following disclaimer: "For a detail
The
36
list
of Iraqi Deaths please see
Table
1
:
Summary
Statistics
A. Key Statistics
Number of US Soldiers
Number of US Soldiers
Number of Iraqi
Number of Iraqi
Iraqi
Iraqi
Iraqi
About
in Iraq,
in Iraq,
Iraq
and the Surge
February 2007
August 2007
135,000
162,000
Security Forces, February 14, 2007
Security Forces, August
8,
323,180
2007
359,700
Bond Yield January 10, 2007
Bond Yield February 14, 2007
Bond Yield August 27, 2007
9.60%
9.58%
11.32%
Means Calculated Over Alternative Time Periods
Mean Daily Iraqi Civilian Fatalities
B.
(Baghdad only)
(Outside Baghdad Only)
Mean Number of Iraqi Security Forces
Mean Millions of Barrels per Day Crude
Average Hours of Electricity per Day
Average Hours of Electricity per Day
Mean
in
Production
Baghdad
in All
of Iraq
Daily Coalition Fatalities
Mean
Iraqi
Soldiers (serious
wound)
wound)
Soldiers (not serious
Bond Yield
Standard Deviation of Iraqi
Bond Yield
Yield - Qatari Yield)
Standard Deviation of (Iraqi Yield - Qatari Yield)
Mean of (Iraqi
Notes: Panel
A reports on the value of some of the
Year Prior
1.5
65.26
73.83
26.54
40.93
45.32
18.15
24.33
28.51
279,019
283,474
308,562
2.07
2.07
2.11
6.54
6.52
6.12
1
11.24
Baghdad Only
Outside Baghdad Only
Mean Daily Wounded US
Mean Daily Wounded US
Years Prior
All
44.70
10.77
2.37
2.67
2.77
6.60
0.80
0.89
1.77
1.86
1.88
7.57
7.54
8.24
9.26
10.65
10.94
9.62%
0.59%
9.66%
0.56%
3.63%
0.60%
3.66%
0.59%
subsequent dependent variables
at
some key
dates.
Panel
B
means of some of the
variable over three time periods. In the "All" column, the sample begins with the
earliest observation and continues through the last available one. In the "1 .5 Years Prior" and "1 Year Prior"
calculates
columns, the samples begin
1.5
and
1
year before the initiation of the Surge and continue through the
observation.
37
last
available
u
u
ob
o
c
o 3
CO
a
^
>.
"2
o
-u
-
m
o
o
o
n
CO
o
o
o
o
Ov <N
O
o
O
o
o
o
:
On
<u
oo
3:
w:
^
-a
o
c y
o
CO
(U
n.
c/5
o
o
d
O
fi
ij
•a
oo
dd
ID
T3
•"
O
O
JD
a"
o M
CO
a
S=
•
c
°
o
On On
5r-
— o
(N
>^'
u —
CO
oo
'
ON ON
O
O
O
<N
o o
dd
oo
o
o
d
-^
CO
CN
S
=«
c
00
^- o
w
c
23
'5b
C
O
o
"^ oo
NC On
O
o
1/-1
-
-
o '^
d£
o
d
0)
XI
i B
*- ,S
o
'*•
00
-t3
c
o
o
om
o
o
o
o
d
>
G
*
o
On
O
d
in
'3-
O
•JJ
ss
en
a>
an
o
•"
3
OS
o
u
•S
^
a;
«
O II
en
t«
a>
^ §i
c c
2^
-a
(/I
JJ
IS
O
CO
'C
u
o
CO
>
•73
V5
-a
'5b
u
.2^
Da
"5.
c
s
oJ
ao
H
o
c
c
CO
3
^
Hm
C3
I
a:
oa
i^
0)
T3
-a
_r
— « >
xs
CI,
il;
D
01
U
c«
•
dO t:o
c
C
c
'-S
en
CO
« u
D.
,o
c
>-<
en
CO
3 —
=CO
^ ^ 3
c O O
n. w
1>
n, c
fc
C/3
_
H
-a
!-
frt
O
CB
<
S
2 c
'5
CO
Q
op
'
;
u
to
(U
C/3
J=
s
2
C/]
^
u
a
o
*
^
o
o
lZ
Ui
,o
^4—
«u
w
o
•a
a>
>
'H
T3
.^
t.
n P
J
T3
lU
3
E: £
t^
o
u
g
r5
c«
—u
,
-
o
^
g
a.
u
J3
g £ S
^
3
o
<2
•a "S
«
"
<u
c
i:
<u
'OC
M
—
u
o
C
!=
CO
-
O.
9-
OB-
—
(N
U
i>
OH
---^
ON
^•S
^
(u
^
i:
f<1
aj
c
CQ
u
o
b^
o
li,
v->
CN
oo
r-
u-i
^
o
•^
— ^
^
oo
^
(N OO
—
OO
ON
OO "^
O
-
II
^
° S'S §
a c3
3 C <u
a- c -3
<u .3 "
<^
-5 -s
00 t: 'r
00
>
3 O .£ -S
—
'C
s
o
u
^2
o
c/2
1
t-i
'-^
*
oo
t^
00
g
On
ON
^
ON
r^
ON
oo'
.
1/-,
^
—
^
m
oo
00
ON
o
On
NO
O
t
S
I
a
o
u
3
.S
o
a
Bwe
>
o
M
^«
cd
•c
e
CS
U
55
£
02
«
S-o
^ c
00
cd
a-
< -
[^
oj
H
n
-a
S 3
C
tfi
t-
•2
S
^
^
"^
3
rt
,Si
GO
>>-
o "3 „
S '55
« 3 u ^ g
o O a „
— Uo
3
Ci.'-3
a
.aS
u
w
V-
,p
i£
.3 Qi
"3 "H
c^
^ c/5
a
-a
3
s
u
t--
—
oo
I
^
C
II
-
O a^
U o „
e
3: c Id
O 1) o
B. g C
^° t:5 300
o S 5 i> '^
<u
•a
c
-a
T3
1)
H
2
u
3
O
3
cfl
a-
t? "cS
o >
0.
Q
m
(D
S:
a.
I
2Z
tH
(1>
Table 4: Estimated Impact of Surge on Millions of Barrels per Day of Crude Oil Production Measured Weekly
(Based on Data through 25 Weeks After the Surge)
t
A.
Mean
Shift
B.
-78
t
=
-52
Model
l(Surge)
R-squared
=
Statistic
0.001
-0.066
(0.033)
(0.035)
0.0000
0.0288
Trend Break Model
Post-Trend
Evaluated
at
t
= 25
-0.1827
-0.0063
(0.0973)
(0.1208)
0.002
-0.002
(0.001)
(0.002)
Statistic
0.0465
0.0540
Mean of Dependent Variable
2.07
2.11
103
78
Pre-Trend
R-squared
J^
Notes: The entries report estimation results from the fitting of equations (1) and (2) for crude oil production
measured in millions of barrels per week (so t references a week). Heteroskedastic-consistent standard errors are
reported in parentheses. The Surge
is defined to begin on February 14, 2007. These data are derived from the
Department of State's Iraq Weekly Status Report. Details on the data sources are reported in the Data Appendix.
*
significant at
1
-percent level,
significant at 5-percent level.
40
1
5: Estimated Impact of Surge on Average Hours of Electricity per Day Available
(Based on Data through 25 Weeks After the Surge)
Table
t
Baghdad Only
= -52
= -78
t
1
Mean
Shift
B.
All of Iraq
t
=
-78
t
=
-52
(2a)
-0.814**
-0.364
-0.81*
-0.28
(0.323)
(0.301)
(0.38)
(0.38)
0.0214
0.01
0.0303
0.007
(2b)
Model
(Surge)
R-squared
Baghdad and All of Iraq
(lb)
(la)
A.
in
Statistic
1
Trend Break Model
Post-Trend
Evaluated
at
t
= 25
Pre-Trend
R-squared
Statistic
Mean of Dependent Variable
N
1.584
-0.377
2.47**
2.53*
(0.888)
(0.949)
(0.86)
(0.98)
-0.035**
-0.015
-0.045**
-0.049**
(0.010)
(0.013)
(0.007)
(0.01 1)
0.1553
0.031
0.3650
0.254
6.52
6.12
11.23
10.77
89
64
88
63
Notes: The entries report estimation results from the
fitting
of equations (1) and (2) for hours of electricity per day
in
measured weekly (so t references a week).
Heteroskedastic-consistent standard errors are reported in
parentheses. The Surge is defined to begin on February 14, 2007. Note, these data are unavailable for the weeks t =
12 through t = 22. There are also a few other weeks missing. More details on the data are contained in the Data
Appendix.
Iraq
**
significant at
1
-percent level,
*
significant at 5-percent level.
41
1
CO
•a
*
O
vo 00
o—
oo
^
ts
5to
«
-O
cS
§
o
-2
C3
O
2^
,^
M
«
U
o
o
o
o
00 00
ON f~~
ON "*
dd
d
£
o
d
r00
o
o
CnI
5-2
CO
c «
o —
00 ST
O
o
o
d
m
dd
00
90
d
" S
—
S
01
aj
O
Da
NO
*
^
ITj
*
r-
On
"*
r-
^
d
*
00
rN)
NO
NO <N
dd
d
d d
u-l
00
d
dd
r-
(N
d
NO
ON <N
d
o
NO
'^
00 f^
"^
a
ON
pa
*
— <N
m
O
~: d
o ^^
CNl
o
o
d
§1
^
3
-a
t/J
CS
a-
§^
O
V3
eg
j:
~ c
a-
—
^
O fN
*
On
NO XI
m —
^
5
ON ^^
9 d.
O
*
—
—
^
-^
od
r<-i
o
00
NO
01 "^
o.
^.B
'S 'o
o o
00
'Ti
*
—
NO
—
rtN
(N
00
<N
o
dd
1/^
O NO
O
V~i
CnI
o
d
NO
m
s
o «
o
U
crt
g
,2
!3
a
o
C
00
I
X
o
g
•o
u
J
en
<u
I
•n
NO
00
S
£
es
0^
S
-<
C
1)
T3
_«
C
1
ca
iS
c«
B
>
•0
u
•a
e
C/!)
T3
^-^
a>
00
3
1—
3
a"
C/5
-
(k
w
b
(U
,
oa -0
CS
w
1
ac
H
H
GO
oa a.
_3
^^
§
w
2:
H
"ca
>
W
0.
„M
C
Q.
-a
3
a-
Q
0.
ST
S:
c«
cS
-K
--^
!£
c
^
C
u
o
tH
_;
CO
CO
Q^
0)
„
"H
§ a c c
"
^"S
trt
CS
C
iZ
CO
2z
<
V
'C _o
O
-
c
>
„c
nn.
"S
Z I
S
3
o.
(30
on
H
o
e
o
o
o
o
p —
:
„
8
o O X
O
o
d
.S2
.
X
c
^X
>
o
o
o
d
ON
OO
o
-^
o o
_;
ON <N
vo
r^
O
d
—
~«
o
d
m
00
O
oo —
ro
rn
*
r--
>^ VO
OS
3
1/3
B
O
u
•c
° o S
U-1
<
^?
M
o
d
*
SB
O
d
CM r^
D
£;
-a
"^
<u
"
!§
,2
.3
oo
c
n.
Si
s
oc
^
,o
O
o„
o
d
Z
"^
C
— tN
O—
d
f-^
o
d
in
^
P
D
m
*
3
—
,
.
V3
)^
XU
§ i
'-0
w
^^
r-
NO
O
d
a!
a
o^ <N
irl
d
o
d
rNi
m
o<
u-i
ir~)
3
•S
•a
<U
*-
*.
X>i
O
§
o
(D
° 2
00 "O
c
B
>
o c c
O 0)
^ "1)
g
> ~
;:
o
*
On
•u
00
^
*
r— (^
—
O _;
O
d
t--
On
O
o6
IIP
x-S-^o <u 13
§
.o
X
c
.
§
3
>
t^
3
X)
.2
p.
ca
ae
o
o
s
O
>-
X
c
°
C8
c
o
-a
X
>
<u
'S
a.
0)
ffl-a
O
c
-o
-a
H
c
X
o-
H
-^
9J
a
03
IJ
.
3
D „ C3
C 2 C"
o _,
o «
on
< -
a:
ffl
a. PJ
a.
c3
T3
I
—
o
3
o
-a
_>.
O
'3
"O
2
(U
.
o
^'
g —
a
P
00
ca
U
i
E
a
o
u
00
<
>,
(0
Q
00
J3
3
s
.•£
00
3 3
o
<-^
s
•a
ea
c
CO
^
•*
o.
_c
^ -o
r- 00
o
oo
Ov OO
On
O
O
o
O
—
_: _;
0^ ON
Oo
o
o
o
o
d
o
ON
[
ly^
»n to
S
O o
ca
3
u
O
cr
<u
p
tu
^>
•n
—
oo
^
1
U^ 00
oo
^
ii
II
>0
00 rON
_; _:
o
o
o
—
00
ON
~
"^
o
•"
o
o
d
rj
o
ca
>
o
t
c
NC r- >B -o
1)
ID
ca
-o
0)
s
Cl,
m
e
^
ca
C
« U >< c
tS
•H
^ cau
3
« 'en c
<u C
O o.
2i
O a.
c o < op
o
a
tn
13
^
ca
>
o
c
cd
o
o
QJ
< -
S
-S
en
n.
Ji
-a
S3
(U
Si
H
3
ca
>
-a
1)
.1
ca
c
o
o
ca
en
Ji
ca
o u
u.
^
T3
ca
tn
D ca
O
2
S ca
CO
ci
'w
5
U
aj
o
c
«
K
en
c
T3
la
Q
u
^
co s
C c
aj
eu
c
c/n
-a
3 b
-a
ca
•a
u
s Z Z
0)
c
op
Figure
1:
1997 Population of Iraq and
Its
Govemorates
Population
5.D0D,3DD;o7.CC-: .OCC
2.DDD.DD3;o5.CCC .OCQ
El.D0D,DDD:o2.0C': .OCC
Bec«
P-codes:
I.CCC.OOO
UN OCHA
Population: Center for Humanitarian
Cocperaticn (1997-census)
Source: http://unosat.web.cem.ch/unosat/freeproducts/iraq/aovernorates pcodes population.pdf
.
recent Iraqi Census.
45
1997
is
the
most
—
o
o
o_
o'
as
o
o
o
o
o
o
o
o
o
o
o"
o'
in
00
\
—
o
o
^-
^^
.
* .
o
o
o
o'
o
\
<rj
«,„^
-n.
oo. ^
^""^-^-^
^'-.
*
^if""'
Br
4.
-
"h.
L
o
o
.
J^
,
*'
•*
5
o.
B^
'^^y
-o.^
•^^.
"-~-^
4
.-''
>
'
*
1
•*
*
Z!^
m'
^^'^^
^^.
Jt!^^
*cr^
•
.>
'^--^
"
t
^
^
-.
o
o
•'
o
T3
.
-
' '
^o.
^>
-:=M
"""^
^
p
--*
^-^^
' '
o
\
B^.^
^
o^.
%
.
V"
W"""'^
.
'o.
^,-^
^,,,--»''''^
P"""'^
J
'^
B
z
<u
3
op
E
O
O
OS
O
O
o"
00
1
o
o
o
o
o
o
o
o
in
nI-
sdoojx
^^z-
^
o
o
o
>
*->
1
en
&
n
(U
a
J5
ri
S
O
S
6
IS
o
^9f
\.
^-/
T
1*
D
U
i
T
£
\^
U
%
4o
1!
*•
2
f
~^
"
iy5
-a
(U
ti.
ca
.;-*
^
*•
<2
>
%
f
.
t^
o
o
fN
^
D
^"*~~~~"~-^-f
o
O
O
11
y-
su
i
'
*"
en
a^
,
•s
J*
^
f^'^^
f
V
£?
c-
^^"*--.,^
.
"
'
1/3
o'
o
o
o
»
,'
£P
o'
^
'
«
*-
o
o
o
B,^^
',
u
o
o
o
^
-,"• ^^^^^
.c
1
o
o
(N
o
o
o
o
o
o
^o.
4-
:
Figure
3:
Daily Fatalities of Iraqi Civilians from
1
Year Before
the Surge
Began through 153 Days After
the Surge
Began
-400
-200
Day
Notes: The data points are the average daily
of the Surge through 153 days
fatalities against a constant, a
days after February
14,
200
Relative to Beginning of Surge (February 14)
fatalities
after the Surge.
per
The
week
in the period
beginning one year before the initiation
line is the fitted values
from a regression of daily coalition
time trend, and the interaction of the time trend and an indicator variable equal to
2007. See Table 2 and the text for
fiirther details.
47
1
for
Figure 4:
Number of Iraqi
Security Forces from
1
Year Before the Surge Began through 25 Weeks After the Surge
Began
-60
-40
-20
Relative to Beginning of Surge (February 14)
20
Week
Notes: The data points are the number of Iraqi Security Forces measured weekly in the period beginning one year
before the initiation of the Surge through
the
number of
Iraqi Security
indicator variable equal to
1
1
86 days
after the Surge.
The
line is the fitted
values from a regression of
Forces against a constant, a time trend, and the interaction of the time trend and an
for days after February 14, 2007.
48
See Table 3 and the text for further
details.
Figure
5:
Millions of Barrels of Oil Produced per
through 25
Weeks
Day (Measured Weekly) from
1
Year Before the Surge Began
After the Surge Began
-40
-20
Relative to Beginning of Surge (February 14)
20
Week
Notes: The data points are the barrels (in millions) of
beginning one year before the
oil produced per day (measured weekly) in the period
of the Surge through 1 86 days after the Surge. The line is the fitted values
millions) of oil produced per day (measured weekly) against a constant, a time
initiation
from a regression of the barrels (in
and the interaction of the time trend and an indicator variable equal
Table 4 and the text for further details.
trend,
49
to
1
for days after February 14, 2007.
See
6: Hours per Day (Measured Weekly) of Electricity in Baghdad and All of Iraq from Year Before
Began through 25 Weeks After the Surge Began
A. Baghdad
Figure
D
a
CO
(t)
00
CO
w
"
TO
Q
Q-CD
o
J)
LU
in
3
o
CM -
-60
-40
-20
Relative to Beginning of Surge (February 14)
Week
B. All of Iraq
51-
_
CO
CO
'
Q
L_
0}
Q.
_0)
o
'^
LU
o
CO
o
I
00 -
-60
-40
-20
Relative to Beginning of Surge (February 14)
Week
50
20
the Surge
Figure
7:
Daily Fatalities of Coalition Soldiers from
1
Year Before the Surge through 186 Days After the Surge
Began
-400
-200
Day
200
Relative to Beginning of Surge (February 14)
Notes: The data points are the average daily
of coalition soldiers per week
beginning one year
from a regression of
daily coalition fatalities against a constant, a time trend, and the interaction of the time trend and an indicator
variable equal to 1 for days after February 14, 2007. See Table 6 and the text for further details.
fatalities
before the initiation of the Surge through 186 days after the Surge. The line
51
is
in the period
the fitted value
Figure
8:
Daily
Wounded US
Soldiers from
-400
1
Year Before the Surge through 182 Days After the Surge Began
-200
Day
200
Relative to Beginning of Surge (February 14)
Notes: The data points are the average daily non-fatal casualties per
the initiation of the Surge through 186 days after the Surge.
The
week
in the
period beginning one year before
line is the fitted value
from a regression of daily
and an indicator
coalition non-fatal casualties against a constant, a time trend, and the interaction of the time trend
variable equal to
1
for days after February 14, 2007.
See Table 7 and the
52
text for fiirther details.
Figure 9: Daily Fatalities of Iraqi Security Forces from
1
Year Before
the Surge through 189
Days After
the Surge
Began
-400
-200
Day
Notes: The data points are the average daily
of the Surge through 189 days
200
Relative to Beginning of Surge (February 14)
fatalities
after the Surge.
The
per
week
in the
period beginning one year before the initiation
line is the fitted value
from
a regression
of daily
Iraqi security
force fatalities against a constant, a time trend, and the interaction of the time trend and an indicator variable equal
to
1
for days after
February
14,
2007. See Table 8 and the text for further
53
details.
^
11
en
'm~
^
*-<
^
0)
1
1
^
«
^^•^
T3
M
U
U
>-
>-
oo
X
o
Tt
w
oi
c
»—
0)
(U
hJ
T3
Q
•T3
Da
3
^—
C
O
U5
3
H
o
T3
s
O
tN
nJ
ID
6X)
<U
C
o
Q
tf
u
>
-a
a
3
m
c
o
CQ
S
u
ca
T3
G
CO
O
cs
3
rt
C3
C3
T-5
•n
T5
oa
oa
c«
^
C
t—^
c
o
&
'$
(U
OJ
OJ
CQ
CQ
•o
oa
T3
03
aj
i^
03
i)
O.
1
1
PP?A
n)
u<
)-H
^
C/3
(/5
IT
I—
(1>
Ui
DO
T3
n)
k.
c
C
(l>
S
CT
c
OJ
o>
n)
Iu
c
o
•n
a
T1
2
c
(l>
(1>
pa
c
VJ
J D
rr
Q
52
T3
CQ
T3
c
c
cy
c
o
,g
(Tl
lU
-a
c
00
(N
c
s
w
X
rt
(IJ
ix u.
on in
•1
*
Figure 11: The Implied Annual Default Rate of Iraqi State Bonds
CO
o
If)
ri
%
Q
o
fe
c
c:
<
E
to
o
IT)
O
-50
50
Day
Notes: This graph
is
100
150
200
Relative to Begninning of Surge (February 14)
derived from the yield series of the Iraqi bond due in 2028.
annual default rate of Iraqi state bonds.
A
value of .06 means that there
the text for technical details.
55
is
a
6%
The y-axis
reports the implied
chance of a default per year.
See
Figure 12: The Role of Domestic
Iraqi
US
Politics
on the Spread Between 5-Year Credit Default Swap Premiums for
Bonds and an Index of Emerging Market Bonds
Q)
O
T
r
50
100
150
Day Relative to Beginning of Surge (February 14)
1
CD
-50
Q.
CO
CDS_Spread
Fitted
Notes: The
200
values
full line plots the spread between the credit default swap premiums for the Iraqi bonds and an Index of
Emerging Market bonds. The dashed line reports the fitted value from a regression of the default probability against
a constant and the price of a contract that pays $100 if a Democrat is elected president in 2008. The contract prices
were downloaded from the web site https://www.intrade.com/v2/. Specifically, it is based on the contact with the
name "PRESIDENT.DEM2008" that is described as "Democratic Party Candidate to Win 2008 Presidential
Election". See the text for fiirther details.
56