GOVERNMENT SPENDING ULTIPLIERS IN GOOD TIMES AND...

GOVERNMENT SPENDING MULTIPLIERS IN GOOD TIMES AND IN BAD:

EVIDENCE FROM U.S. HISTORICAL DATA

Valerie A. Ramey

UC San Diego & NBER Sarah Zubairy

Texas A&M University

Discussion by

Yuriy Gorodnichenko UC Berkeley & NBER

MOTIVATION Key policy questions

What is the size of the government spending multiplier? o Previous work: multiplier 1 (wide “confidence” bands)



What is the size of the government spending multiplier IN RECESSIONS?




Challenges of estimating state-dependent multipliers

A handful of recessions in the post-WWII data & relatively little variation in G o RZ: construct long, quarterly time series: 1880-2013.

Post-WWII data: standard Pre-WWII data: many sources + interpolate annual series into quarterly







Identification of exogenous, unanticipated shocks to government spending o RZ: News shocks (extend Ramey (QJE 2011)) about military gov’t spending







Identification of exogenous, unanticipated shocks to government spending o RZ: News shocks (extend Ramey (QJE 2011)) about military gov’t spending

Nonlinear models: sensitive estimates + how to model feedback/dynamics? o RZ: Use Jorda (2005) projection method as in AG (2012)

RESULTS

Output responds more strongly in “slack times” (unemployment rate > 6.5%)

RESULTS


Government spending responds more strongly in “slack times”

o Multipliers ≡ ∑∑ are similar in “slack times” and “no-slack times”

RESULTS




Little variation/weak identification in post-WWII data

Multipliers are similar at the zero lower bound (ZLB) and outside ZLB

o More challenges for multipliers at ZLB

RESULTS







A GREAT PAPER!

RESULTS







A GREAT PAPER!

Why are the RZ results different from the results in Auerbach-Gorodnichenko and others?

Measurement Specification Estimation Identification

RZ APPROACH

…

RZ APPROACH ⇒

⇒

RZ APPROACH ⇒

⇒

Multiplier at horizon : ≡,

,,,

RZ APPROACH ⇒

⇒


,,,

Instrumental variable interpretation: Regress on and use as an IV.

RZ APPROACH ⇒

⇒


,,,


The logic extends to state-dependent multipliers

and as IVs.

RZ APPROACH ⇒

⇒


,,,


The logic extends to state-dependent multipliers

and as IVs.

Single equation approach

:

:

FIRST STAGE FIT: FULL SAMPLE

Note: controls are included. F-stat in the figure is capped at 45.

Include WWII0

510

1520

2530

3540

45fir

st-s

tage

F-s

tatis

tic

0 2 4 6 8 10 12 14 16 18 20horizon h

RecessionExpansion

FIRST STAGE FIT: EXCLUDE WWII

Note: controls are included. F-stat in the figure is capped at 45.

Exclude WWII0

510

1520

2530

3540

45fir

st-s

tage

F-s

tatis

tic

0 2 4 6 8 10 12 14 16 18 20horizon h

RecessionExpansion

FIRST STAGE FIT: RECESSION

Horizon 8

1939.5

1940.25

1940.751941

1941.25

1941.75

1942.25

b = 0.77 (0.11)

b = 0.45 (0.39)

0.1

.2.3

.4.5

.6(G

t+h-G

t-1)/Y

t-1

0 .1 .2 .3 .4 .5 .6 .7 .8shock

WWII observationsExclude WWIIFit all observationsFit excl. WWII obs

FIRST STAGE FIT: RECESSION

Horizon 8

Question: which shocks should one use to design/assess the fiscal stimulus in 2009?

1939.5

1940.25

1940.751941

1941.25

1941.75

1942.25

b = 0.77 (0.11)

b = 0.45 (0.39)

0.1

.2.3

.4.5

.6(G

t+h-G

t-1)/Y

t-1

0 .1 .2 .3 .4 .5 .6 .7 .8shock

WWII observationsExclude WWIIFit all observationsFit excl. WWII obs

RAMEY-ZUBAIRY VS. BLANCHARD-PEROTTI Ramey-Zubairy:

use military spending shocks as the instrument



Blanchard-Perotti

use as the instrument



Blanchard-Perotti


o First-stage fit for 0 is perfect 1



Blanchard-Perotti


o First-stage fit for 0 is perfect 1 Alternative IV (Auerbach-Gorodnichenko):

≡ a professional forecast as of time 1 of government spending at time



Blanchard-Perotti


o First-stage fit for 0 is perfect 1 Alternative IV (Auerbach-Gorodnichenko):

≡ a professional forecast as of time 1 of government spending at time

Strength of 1st stage: RZ vs. BP BP (AG) instrument is nearly impossible to beat over short horizons. RZ can perform better over longer horizons b/c it measures present values.

CHALLENGES IN CONSTRUCTING AND ANALYZING LONG-TIME SERIES

Data quality is likely to vary o Linear interpolation ⇒ Attenuate differences between recession/expansion



Regime changes o Balanced budget provisions o Gold standard




Structural changes o Changes in the volatility of government spending

% CHANGE IN REAL PER CAPITA GOVERNMENT SPENDING

-.50

.5dl

og(g

over

nmen

t spe

ndin

g p.

c.)

0.1

.2.3

st.d

ev. d

log(

G p

.c.),

5 y

ear m

ovin

g w

indo

w

1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010




Structural changes o Changes in the volatility of government spending o Secular trend in the size and composition of the government

SHARE OF GOVERNMENT SPENDING IN GDP

0.1

.2.3

.4.5

Gov

ernm

ent s

pend

ing

shar

e in

GD

P

1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010




Structural changes o Changes in the volatility of government spending o Secular trend in the size and composition of the government ⇒ avoid using variables in levels, use differences or/and growth rates

RZ: ∑ ln ∑ ln ∑

Alt.: ∑ Δ ln ∑ Δ ln ∑

NORMALIZATION

Typical approach: Δ log Δ log ⇒

NORMALIZATION


Alternative approach: ⇒

NORMALIZATION



Δ log

NORMALIZATION



Δ log

Potential concerns

and are correlated because shows up in the denominator

varies systematically over the business cycle

NORMALIZATION

Notes: post 1960 data; potential GDP is from the CBO.

-.015

-.01

-.005

0.0

05G

/Y

0 5 10 15Quarters since the start of a recession

Normalized by actual GDP Normalized by potential GDP

MULTIPLIERS: RAMEY-ZUBAIRY

Spec: baseline, IV implementation

-.50

.51

1.5

22.

5M

ultip

lier

0 1 2 3 4 5 6 7 8 9 10 11horizon

ExpansionRecession

MULTIPLIERS: BLANCHARD-PEROTTI

Spec: IV implementation, include more lags, normalize by potential GDP, controls

include variables in growth rates rather than levels.

These estimates are similar to the Auerbach-Gorodnichenko results.

-.50

.51

1.5

22.

5M

ultip

lier

0 1 2 3 4 5 6 7 8 9 10 11horizon

ExpansionRecession

EQUALITY OF MULTIPLIERS OVER THE BUSINESS CYCLE

0.1

.2.3

.4.5

.6.7

.8.9

1p-

valu

e(R

eces

sion

=Exp

ansi

on)

0 1 2 3 4 5 6 7 8 9 10 11horizon h

Blanchard-Perotti Ramey-Zubairy

CONCLUDING REMARKS We need more variation/data to identify G shocks and estimate their effects Cross-state variation (e.g., Nakamura and Steinsson 2014) Natural experiments (e.g., Joshua Hausman 2013) Asset prices and high frequency data (e.g., Johannes Wieland 2012)


We need to study responses of other variables to understand the workings of G shocks Consumption, investment, durables/non-durables Prices, wages, interest rates Employment, capacity utilization Export, import, exchange rates



We need better theory to guide our empirical analyses RZ: Michaillat (2014) is the only modern macro model with state-dependent multipliers Most models are linearized or close to linear



We need better theory to guide our empirical analyses RZ: Michaillat (2014) is the only modern macro model with state-dependent multipliers Most models are linearized or close to linear

“The problem with QE is it works in practice but it doesn’t work in theory.” – Bernanke

GOVERNMENT SPENDING ULTIPLIERS IN GOOD TIMES AND...

Documents

Transcript of GOVERNMENT SPENDING ULTIPLIERS IN GOOD TIMES AND...