Download VAR method

Identifying Government Spending Shocks:
It’s all in the timing
By
Valerie A. Ramey
What is the Impact of an Increase in
Government Spending?
Traditional Keynesian
Increase in Government Spending leads to:
↑ Output
↑ Consumption
↑ Hours
↑ Real Wage
↓ Investment
Neoclassical Theory: Impact of Increased Gov.


Aiyagari, Christiano, & Eichenbaum (1992),
Baxter & King (1993)
Key mechanism: Negative wealth effect increases labor
supply
↑ Output
↓ Consumption
↑ Hours
↓ Real Wage
↓ or ↑ Investment
New Keynesian Theory




Rotemberg & Woodford (1992): Countercyclical
markups
Devereux, Head & Lapham (1996): increasing
returns
Gali, Lopez-Salido, Valles (2007): sticky prices,
noncompetitive labor markets, rule-of-thumb
consumers
Ravn, Schmitt-Grohe, Uribe (2006): deep habits
Empirical Evidence on the Impact of ↑ G
•
Papers using VAR techniques for identification
e.g. Rotemberg and Woodford (1992), Blanchard and Perotti (2002), Fatás and Mihov (2001), Perotti
(2004), Montford and Uhlig (2005), Galí, López-Salido, and Vallés (2006)), Perotti(2007)
↑ output
•
↑ hours
↑ consumption
↑ real wages
↓ investment
Papers using Ramey-Shapiro war dates
e.g. Ramey and Shapiro (1998), Edelberg, Eichenbaum, and Fisher (1999), Burnside, Eichenbaum,
and Fisher (2004), and Cavallos (2005)).
↑ output
•
↑ hours
↓consumption
↓ real wages
↓ or ↑ investment
Other Event Studies
e.g. Giavazzi & Pagano’s (1990), Cullen & Fishback (2006), Barro & Redlick (2009)
↓consumption or no effect
Outline of talk
1.
2.
3.
4.
5.
6.
Overview of behavior of government spending
Generate results using the two methods on the
same data and time period
Compare the timing of war dates to the timing of
VAR shocks
Present theoretical model that explains both results
Show that changing the timing on the war dates
produces the same results as the VAR
Introduction of 2 new “news” variables and
extension of sample back to 1939
Real Government Spending Per Capita
1
3
4
1.5
5
2
6
2.5
7
3
8
3.5
Real Defense Spending Per Capita
1950
1960 1970
1980
year
1990
2000
2010
1950
1960
1970
1980
year
1990
Red lines are Ramey-Shapiro military dates + 9/11.
(numbers in thousands of 2005 dollars)
2000
2010
Some Historical Perspective
Real Defense Spending Per Capita
0
2
4
6
8
10
(thousands of 2005 dollars)
1940
1950
1960
1970
1980
year
1990
2000
2010
0
.05
.1
.15
Components of Government Spending
Fraction of nominal GDP
1950
1960
1970
defense
state & local
1980
year
1990
2000
nondefense federal
2010
Two Leading Identification Methods
1. VAR method
Order government spending first in a VAR, use standard Choleski
decomposition to identify shocks to government spending.
2. Ramey-Shapiro Dates
Augment system with a dummy variable that takes the value of
unity at times when a major political event caused Business
Week to begin forecasting large increases in defense spending.
Shocks to the dummy variable rather than actual government
spending are identified as the shock.
Ramey-Shapiro Dates: The dummy variable takes the
value of unity at 4 dates.
1950:3: North Korea invaded South Korea in late
June 1950.
1965:1 Johnson began air strikes against N.
Vietnam in Feb. 1965.
1980:1 The USSR invaded Afghanistan on Dec. 24,
1979.
*2001:3 Terrorists struck the World Trade Center
and the Pentagon on 9/11.
Framework for Comparison
X t  A( L) X t 1  U t
X includes: total govt spending, GDP, Barro-Redlick tax
rate, total hours (including military), nondurable +
services consumption, private fixed investment,
product wage in private business.
Log per capita (except BS tax, product wage)
Quarterly data: 1947:1 – 2008:4, 4 lags, quadratic trend
Identification


Standard VAR: order government spending first
War dates: also include current and 4 lags of RameyShapiro war dates variable augmented with 9/11.
1950:3, 1965:1, 1980:1, 2001:3

Set shock size so that peak response of government
spending is the same across specifications
Comparison of Identification Methods
VAR shocks in top row; War dates in bottom row
68% bands
total hours
-.12
-.2
-.4
0
0
0
.4
.12
.2
.8
.4
.27
gdp
1.2
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
government spending
gdp
total hours
16
20
16
20
-.12
-.2
-.4
0
0
0
.4
.12
.2
.8
.4
.27
quarter
1.2
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
Comparison of Identification Methods
VAR shocks in top row; War dates in bottom row
68% bands
consumption, ndur + serv
real wages
-.18 -.12 -.06
-.9
-.2
-.6
-.1
-.3
0
0
0
.06
.12
.3
.1
investment, nonres + res
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
consumption, ndur + serv
investment, nonres + res
real wages
16
20
16
20
-.18 -.12 -.06
-.2
-.9
-.6
-.1
-.3
0
0
0
.06
.3
.12
quarter
.1
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
Why Do these Two Methods Give Such Different
Results?
It’s All in the timing
I will argue that the shocks identified by the VAR
are mostly anticipated and this explains all of the
difference in the results.
Defense Spending During Vietnam War
.4
0
.5
.5
.6
.7
1
Defense Spending During Korean War
1948 1949 1950 1951 1952 1953 1954 1955
1964 1965 1966 1967 1968 1969
VAR Shocks During Vietnam War
-.1
-.05
-.04 -.02
0
0
.05
.02
.1
.04
.15
VAR Shocks During Korean War
1970
1948 1949 1950 1951 1952 1953 1954 1955
1964 1965 1966 1967 1968 1969
Business Week Defense Forecasts
10
50
20
60
30
70
40
80
50
Business Week Defense Forecasts
1970
1965
1950
1951
1952
fiscal year
actual
Aug 50 forecast
1953
1954
Jun 50 forecast
Sep 50 forecast
1966
1967
fiscal year
actual
Aug 65 forecast
Jan 67 forecast
1968
1969
Jan 65 forecast
Sep 66 forecast
May 67 forecast
.15
.2
.2
.3
.25
.4
.3
.5
.35
.4
Defense Spending During 9/11
.6
Defense Spending During Carter-Reagan
1978
1980
1982
1984
1986
1999
2001
2002
2003
2004
VAR Shocks During 9/11
0
.05
-.05
-.1
-.04 -.02
0
.02
.1
.04
VAR Shocks During Carter-Reagan
2000
1978
1980
1982
1984
1986
1999
2001
2002
2003
2004
OMB Defense Forecasts
300 350 400 450 500
100 150 200 250 300 350
OMB Defense Forecasts
2000
1980
1982
1984
1986
2001
2002
2003
2004
2005
2006
fiscal year
actual
Jan. 1980 forecast
Oct. 1981 forecast
Jan. 1979 forecast
Jan. 1981 forecast
actual
Feb. 2002 forecast
Feb. 2004 forecast
Apr. 2001 forecast
Feb. 2003 forecast
Survey of Professional Forecasters
0
.02
.04
.06
Forecasts of Real Federal Spending Growth from
Quarter t-1 to t+4
1999
2000
2001
2002
2003
2004
Are VAR shocks anticipated? Yes.
Hypothesis Tests
p-value in parenthesis
Do War dates Granger-cause VAR shocks?
Yes (0.017)
Do one-quarter ahead Professional Forecasts
Granger-cause VAR shocks? 1981:3 –
2008:4
Yes (0.025)
Do four-quarter ahead Professional Forecasts
Granger-cause VAR shocks? 1981:3 –
2008:4
Yes (0.016)
Do VAR shocks Granger-cause War dates?
No (0.148)
A simple model shows how getting the dates
wrong can lead to faulty inferences.



Very simple model with government spending
and lump-sum taxation.
Simulate a defense spending process similar to
data.
The key is that the new spending is announced
two quarters before it happens, similar to what
we saw in the data.
Model
Yt  ( Z N )0.67 Kt0.33
t t
U  log( Ct )    log( 1  Nt )
t
Yt  Ct  I t  Gt
Kt 1  I t  (1  0.023) Kt
ln Z t  .95  ln Z t 1  ezt ,  ez  0.01
ln t  .95  ln t 1  et ,  e  0.008
ln GFt  constant  1.4 ln GFt 1  0.18 ln GFt  2  0.25 ln GFt 3  eg t ,
ln Gt  ln GFt  2
 eg  0.028
Theoretical Effect of an Increase in Government Spending
(announced two quarters in advance)
hours
.04
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
quarter
consumption
investment
real wages
20
16
20
0
16
-.005
-.01 wg
-.015
-.2
-.04
-.1
-.03
0
ig
-.02
cg
.1
-.01
.2
0
.3
0
0
0
.01
.01
0
.25
gg
.5
yg
.02
.02ng.03
.03
1
.75
.05
gdp
.04
government spending
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
Estimation of VARs on Simulated Data
Actual Spending Identification in top row; News in bottom row
68% bands
gdp
hours
0
-.1
-.05
0
-.05
.4
0
.8
.05
.05
.1
1.2
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
quarter
government spending news
gdp
hours
16
20
16
20
0
-.05
-.1
0
-.05
.4
0
.8
.05
.05
.1
1.2
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
Estimation of VARs on Simulated Data
Actual Spending Identification in top row; News in bottom row
68% bands
real wages
8
12
16
-.04
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
quarter
consumption
investment
real wages
.4
20
16
20
-.04
-.06
-.8
-.05
-.025
-.4
0
-.02
0
0
.025
16
.02
4
.05
0
-.06
-.05
-.8
-.025
-.4
0
-.02
0
0
.025
.4
.02
investment
.05
consumption
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
If the argument is right, then delaying the military dates
should produce the standard VAR results
Procedure: delay Ramey-Shapiro dates to time of first
big VAR defense shock.
Actual
1950:3
1965:1
1980:2
2001:3
Delayed Dates
1951:1
1965:3
1981:4
2003:2
Empirical Results from Delaying the Timing
of the Ramey-Shapiro Military Dates
Estimates on Actual Data
total hours
-.1
-.2
-.5
0
0
0
.1
.5
.2
.2
1
.4
.3
gdp
1.5
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
quarter
quarter
consumption, ndur + serv
investment, nonres + res
real wages
20
16
20
0
4
8
12
quarter
16
20
0
-.1
-.4
-.05
-.2
0
0
.05
.1
.2
.1
.2
.4
.15
16
.3
quarter
.6
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
A New Measure of Defense Shocks
The simple dummy variable incorporates only a small part
of the information available in the narrative record.
Thus, I created a new variable: the present discounted
value of the forecasted changes in defense-related
spending. This is what matters for the wealth effect.
I created the variable by reading mostly Business Week, but
also the New York Times, Washington Post, and Wall Street
Journal from January 1939 to December 2008.
Business Week 5/25/40, p. 60: “The German drive to the English
Channel this week assured quick adoption of the President’s program to
speed up war preparations. But the proposed expenditure of less than
$3.5 billion in the coming fiscal year is only a small beginning; of that,
business men can now be certain… In the 1919 fiscal year costs ran to
$11 billion. A major war effort in the ‘40s would come higher… since we
have started six years behind, a vast outlay is required if we are to attain
military parity with Hitler’s industrial machine. In a major war at least
four times the $3.5 billion we plan to spend in 1941 would be needed,
and quite conceivably five to six times that – or anywhere from 20% to
30% of the peacetime national income. However, it is not possible to
jump immediately up from a $3.5 billion to a $14 billion military effort. It
takes time to shift a nation from a peace economy to a war-preparation
economy and thence to a war economy. Right now we are at the very
beginnings of a war-preparation economy.”
Business Week 10/25/41: “Expect dramatic developments in the
defense program in the next few weeks. Plans were under way before
the Kearny incident and the sinking of two more American ships but they
have been speeded by this week’s shocks and by the heartening reports
on Russia’s capacity to hold out, brought home by the Harriman mission.
Beginning this week, war production –and it’s ‘war,’ not ‘defense’ …becomes the No. 1 item on the business docket.” p. 7
“Already, Washington is taking cognizance of the imminence of a
shooting war…A year ago, the government thought of armament
expenditures of $10 billion a year; six months ago the goal was $24
billion; as recently as last month $36 billion was regarded as a desirable
but hard-to-achieve outlay; but now an annual expenditure of $50 billion
is begin seriously discussed – not as the desirable goal, but as an
inescapable necessity.” p. 13
-20
0
20
40
60
80
Defense News:
PDV of Expected Change in Spending as a % of lagged
GDP
1940
1950
1960
1970
1980
year
1990
2000
2010
Explanatory Power of New Defense Shock Series
for Growth of Real Per Government Spending
1
2
3
R-squared
F-statistic
1939:1 – 2008:4
0.410
37.58
Marginal Fstatistic
9.00
1947:1 – 2008:4
1955:1 – 2008:4
0.518
0.037
51.15
1.60
19.65
0.429
Columns 1 and 2 from regression on current and 4 lags of shock.
Column 3 is from full VAR.
Framework for Defense News VAR
X t  A( L) X t 1  U t
X includes: defense news variable (as a % of lagged
GDP), total govt spending, GDP, Barro-Redlick tax
rate, 3-month T-bill rate
6th variable is rotated in.
Newly constructed Quarterly data: 1939:1 – 2008:4,
4 lags, quadratic trend
VAR with Defense News Variable: 1939-2008
(red lines: 68%; green lines: 95%)
total hours
.1
-.1
-.1
-.5
0
0
0
.1
.5
.2
1
.3
.2
gdp
1.5
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
16
20
quarter
quarter
manufacturing wage
3 month Tbill rate
average marginal income tax rate
.1
.05
-.05
-6
-.2
-4
-.1
0
-2
0
0
.1
2
.2
4
.15
quarter
.3
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
VAR with Defense News Variable: 1939-2008
(red lines: 68%; green lines: 95%)
services consumption
.1
0
-.05
-.2
-40
-.15
-20
-.1
.05
-.05
0
0
20
.15
nondurable consumption
.05
real baa bond rate
0
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
16
quarter
quarter
consumer durable purchases
nonresidential investment
residential investment
20
0
4
8
12
quarter
16
20
-1
-2
-1
-1
-1.5
-.5
-.5
-.5
0
0
0
.5
.5
.5
quarter
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
Effect of Defense Shocks with and without WWII
(Blue: 1939 - 2008; green: 1947 – 2008)
total hours
-.5
-.05
-.05
0
0
0
.05
.05
.1
.5
.1
.15
.2
.15
gdp
1
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
16
20
quarter
quarter
manufacturing wage
3-month Tbill rate
average marginal income tax
.1
quarter
-.05
-.1
-4
-.3
-3
-.2
-2
0
-.1
-1
0
.05
0
.1
1
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
Effect of Defense Shocks with and without WWII
(Blue: 1939 - 2008; green: 1947 – 2008)
real baa bond rate
nondurable consumption
.02 .04 .06 .08
-40
-.1
-.02
0
-20
-.05
0
0
20
.05
services consumption
0
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
16
quarter
quarter
consumer durable purchases
nonresidential investment
residential investment
.2
-1
-.8
-.5
-.6
-.4
-.5
0
-.2
0
0
.5
20
.5
quarter
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
Alternative Measure for Later Period
•The defense news variable has little predictive power for
the post-Korean War Period.
•As an alternative, I use government spending shocks
constructed using data from the Survey of Professional
Forecasters.
1968:4 – 1981:2: Forecast nominal defense spending,
GDP deflator
1981:3 – 2008:4: Forecast real federal spending
Shock(t) = Δ ln(gt) - Et-1Δ ln(gt)
VAR with Shock Based on Professional Forecasters:
1968-2008
(red lines: 68%; green lines: 95%)
-.6
-1
-.5
-.4
0
-.5
-.2
.5
0
0
1
.2
.5
total hours
.4
gdp
1.5
government spending
4
8
12
16
20
0
4
8
12
16
20
0
4
8
12
16
20
quarter
quarter
quarter
business wage
3 month Tbill rate
average marginal income tax rate
0
4
8
12
quarter
16
20
.2
.1
-.1
-.3
-.4
-60
-.2
-40
-.2
0
-20
0
0
.2
20
.4
0
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
VAR with Shock Based on Professional Forecasters:
1968-2008
(red lines: 68%; green lines: 95%)
services consumption
0
4
8
12
16
-.2
-.4
-.8
-100
-.6
-50
-.4
-.2
0
0
0
.2
.2
nondurable consumption
50
real baa bond rate
20
0
4
8
12
16
20
0
4
8
12
16
quarter
consumer durable purchases
nonresidential investment
residential investment
20
0
4
8
12
quarter
16
20
-2
-4
-6
-3
-3
-2
-2
-1
-1
0
0
0
2
1
4
quarter
1
quarter
0
4
8
12
quarter
16
20
0
4
8
12
quarter
16
20
Conclusions

Standard VAR methods appear to identify shocks that
are anticipated.

Anticipation effects are very important from the view
point of a neoclassical model.


Both theory and empirical evidence suggests that
difference in standard VAR results and Ramey-Shapiro
dates is due to faulty timing of VAR shocks.
Bottom Line: An increase in (nonproductive)
government spending lowers consumption and has a
multiplier of unity or less.