Systematic Monetary Policy and the Effects of Oil Price Shocks

BEN S. BERNANKE

Princeton

University

MARK

GERTLER

New

York

University

MARK

WATSON

Princeton

University

Systematic

Monetary

Policy

and

the

Effiects

of Oil Price

Shocks

THE PRINCIPAL OBJECTIVE of this

paper

is

to increase our

understanding

of

the role of

monetary policy

in

postwar

U. S. business

cycles. We take

as our

starting point

two common

findings

in

the

recent

monetary policy

literature based on

vector autoregressions

(VARs).' First, identified

shocks

to

monetary policy explain relatively

little

of

the

overall varia-

tion

in

output

(typically,

less than

20

percent).

Second,

most of the

observed movement

in

the instruments

of

monetary

policy,

such

as

the

federal funds rate or

nonborrowed

reserves,

is

endogenous;

that

is,

changes

in

Federal

Reserve

policy

are

largely explained

by

macroeco-

nomic conditions,

as one might expect, given

the Fed's

commitment to

macroeconomic stabilization.

These

two

findings

obviously

do

not

sup-

port

the view that erratic

and

unpredictable

fluctuations

in

Federal Re-

serve policies are

a

primary

cause

of

postwar

U.S. business

cycles;

but

neither do they rule out the

possibility

that

systematic

and

predictable

monetary policies-the Fed's

policy

rule-affect the course of

the

economy

in an

important way.

Put more

positively,

if

one

takes the

VAR evidence on

monetary policy seriously (as

we

do),

then

any

case

for an

important

role

of

monetary policy

in

the business

cycle

rests on

Thanks to Benjamin Friedman, Christopher Sims,

and

the

Brookings

Panel

for help-

ful comments.

Expert

research assistance

was

provided by

Don

Redl and Peter

Simon.

The financial support of the

National

Science

Foundation is

gratefully acknowledged.

1.

See,

for

example, Leeper, Sims,

and

Zha

(1996).

91

92

Brookings Papers

on

Economic Activity, 1:1997

the argument that the choice

of the

monetary policy

rule (the "reaction

function") has significant

macroeconomic effects.

Using time-series evidence to uncover

the

effects of monetary policy

rules on the economy is, however, a daunting task.

It is not possible to

infer the effects of changes

in

policy rules from

a

standard identified

VAR system, since this approach typically provides

little or no struc-

tural

interpretation

of the coefficients

that make

up

the

lag structure of

the

model.

Large-scale

econometric

models,

such

as

the

MIT-Penn-

SSRC model, are designed

for

analyzing

alternative

policies;

but

criti-

cisms

of

the

identifying assumptions

of these models have

been

the

subject

of

a

number of

important papers,

notably, by

Robert Lucas

and

Christopher

Sims.2

Particularly

relevant

to the

present paper

is

Sims's

point

that

the many overidentifying

restrictions

of

large-scale models

may

be both

theoretically

and

empirically suspect,

often

implying spec-

ifications that

do not match

the basic

time-series properties

of

the data

particularly well. Recent progress

in

the

development

of

dynamic sto-

chastic

general equilibrium

models overcomes much of

Lucas's

objec-

tion

to

the traditional

approach,

but the

ability

of these

models to fit

the

time-series

data-in

particular,

the

relationships

among money,

interest

rates, output, and prices-seems,

if

anything,

worse

than that of

tra-

ditional

large-scale

models.

In this paper we take

some modest

(but,

we

hope, informative) first

steps

toward

sorting

out

the effects

of

systematic

monetary policy

on

the

economy,

within

a

framework

designed

to

accommodate the

time-

series

facts

about the

U.S.

economy

in a flexible

manner. Our

strategy

involves

adding

a little

bit

of

structure

to

an

identified

VAR.

Specifi-

cally,

we assume

that

monetary policy

works

its effects on the

economy

through

the medium of

the

term

structure

of

open-market

interest

rates;

and

that, given

the

term

structure,

the

policy

instrument

(in

our

appli-

cation,

the federal funds

rate)

has

no

independent

effect

on

the

econ-

omy.

In

combination

with

the

expectations

theory

of

the

term

structure,

this

assumption

allows

one to summarize the effects of

alternative ex-

pected

future

monetary policies

in

terms

of their

effects on the

current

short

and

long

interest

rates, which,

in

turn,

help

to determine

the

evolution

of

the

economy. By comparing,

for

example,

the historical

behavior

of

the

economy

with

its behavior

under

an

hypothesized

alter-

2. Lucas

(1976);

Sims

(1980).

Ben S.

Bernanke, Mark

Gertler, and

Mark Watson

93

native

policy

reaction

function, we

obtain a

rough

measure

of the

im-

portance of

the

systematic

component of

monetary

policy.

Our

approach

is

similar

in

spirit to a

methodology due to Sims

and Tao

Zha;

however,

these authors do

not

attempt

to sort out

the effects of

anticipated

and

partially

unanticipated

policy

changes.3

While our

proposed

method-

ology is

crude,

and

certainly

is not

invulnerable to the

Lucas

critique,

we

believe

that it

represents

a

commonsense

approach

to the

problem

of

measuring the effects of

anticipated

policy,

given

currently

available

tools.

To

be able

to

compare historical and alternative

hypothesized

re-

sponses of

monetary

policy

to

economic

disturbances,

one needs

to

select some

interesting

set

of

macroeconomic shocks

to which

policy

is

likely

to

respond. We focus

primarily

on

oil

price

shocks, for

two

reasons.4

First,

periods

dominated

by oil price

shocks are

reasonably

easy to

identify

empirically,

and

the case for

exogeneity

of

at

least the

major

oil

price shocks

is

strong

(although,

there is

also

substantial

controversy

about how these shocks and their

economic

effects

should

be

modeled).

Second,

in the view of

many

economists,

oil

price

shocks

are

perhaps

the

leading

alternative

to

monetary

policy

as

the

key factor

in

postwar

U.S. recessions: increases

in

oil

prices preceded the

reces-

sions of

1973-75,

1980-82,

and

1990-91,

and

James

Hamilton

pre-

sents

evidence that increases

in oil

prices

led

declines

in

output

before

1972 as well.5

Further,

one of

the

strongest

criticisms of

the

neomo-

netarist claim that

monetary

policy

has

been a

major

cause of

economic

downturns is that it

may

confound

the

effects of

monetary

tightening

and

in

oil

prices.

The

rest

of

the

paper

is

organized

as

follows. We

first

document

that

essentially

all the

U.S.

recessions of the

past

thirty years

have

been

preceded

by

both oil

price

increases and a

tightening

of

monetary pol-

icy,

which raises the

question

to what extent the

ensuing economic

declines

can

be

attributed

to each factor.

Discussion of

this

identifica-

tion

problem

requires

a

digression

into the

parallel

VAR-based

literature

3. Sims and

Zha (1995).

4. Hooker

(1996a)

also studies the effects

of oil

price

shocks and

their

interaction

with

monetary

policy

in a VAR framework.

However,

he

does not

explicitly

attempt to

decompose

the effect of oil

price

shocks on the

economy

into a

part

due to the

change

in oil

prices

and a

part

due to the

policy

reaction.

5. Hamilton

(1983).

94

Brookings Papers on

Economic

Activity,

1:1997

on the

effects of oil price

shocks;

one

main

conclusion is

that it is

surprisingly difficult

to find an indicator

of

oil

price

shocks

that pro-

duces the

expected responses

of

macroeconomic and

policy variables

in

a

VAR setting. After

comparing

alternative

indicators,

we

choose as

our

principal measure

of oil

price

shocks the

"net oil

price increase"

variable

proposed by Hamilton.6

We

next introduce

our

identification

strategy,

which

summarizes

the

effects of an

anticipated

change

in

monetary

policy

in

terms of

its

impact

on

the current

term structure

of

interest

rates

(specifically, the

three-month and

ten-year government

rates).

We show

that

this

ap-

proach

provides

reasonable

results

for the

analysis

of shocks

to

mone-

tary

policy

and

to oil

prices;

and,

in

particular,

we find

that the

endog-

enous

monetary policy response

can account for a

very

substantial

portion

(in

some

cases,

nearly all)

of the

depressing

effects of oil

price

shocks

on

the real

economy.

This result is

reinforced

by

a

more

dis-

aggregated

analysis,

which

compares

the

effects

of

oil

price

and mon-

etary

policy

shocks on

components

of GDP.

Looking

more

specifically

at individual

recessionary episodes

associated with oil

price shocks,

we

find that both

monetary policy

and other

nonmoney,

nonoil

disturbances

played

important roles,

but that oil

shocks, per

se,

were

not a

major

cause

of these downturns.

Overall,

these

findings

help

to

resolve the

long-standing puzzle

of the

apparently disproportionate effect of

oil

price

increases on the

economy.

We also show

that our method

produces

reasonable results

when

applied

to

the

analysis

of

monetary policy

reactions to other

types

of

shocks,

such

as

shocks

to

output

and

to

commodity prices.

After

presenting

the basic

results,

we look in

more

detail at

their

robustness and

stability.

Regarding robustness,

we find

that the

broad

conclusion that

endogenous

monetary policy

is an

important component

of

the

aggregate impact

of oil

price

shocks

holds across a

variety

of

specifications, although

the exact

proportion

of

the effect

due to mon-

etary

policy

is sometimes

hard to determine

statistically.

We

also find

evidence

of

subsample

instability

in our

estimated

system.

To

some

extent,

however,

this

instability helps

to

strengthen

our main

conclu-

sions about the role

of

endogenous monetary

policy,

in

that the

total

effect of oil

price

shocks

on the

economy

on

output

is

found

to be

6.

Hamilton

(1996a,

1996b).

Ben S.

Bernanke, Mark Gertler,

and Mark Watson

95

strongest during

the Volcker

era-when the

monetary response

to in-

flationary shocks

was

also the strongest.

Our analysis

uses interpolated

monthly data

on GDP and its

com-

ponents. Appendix

A documents the construction

of

these data,

and

appendix B describes

all

of

the

data

that we

use.

Is It Monetary

Policy

or

Is It Oil? The Basic

Identification

Problem

The idea that

monetary policy

is

a

major

source of real fluctuations

in

the economy

is

an

old one;

much

of

its lasting appeal reflects

the

ongoing

influence

of the seminal

work of

Milton Friedman and

Anna

Schwartz.7 Obtaining

credible measurements

of

monetary policy's

con-

tribution

to

business

cycles

has

proved difficult,

however. As discussed

above,

in

recent

years

numerous authors

have

addressed

the

measure-

ment

of

the

effects

of

monetary policy by

means

of

the VAR

method-

ology,

introduced

into economics

by

Sims.8

Roughly speaking,

this

approach

identifies

unanticipated

innovations to

monetary policy

with

an unforecasted shock

to some

policy

indicator,

such as

the

federal

funds

rate or

the

rate

of

growth

of nonborrowed reserves.

Using

the

estimated

VAR

system,

one can trace

out

the

dynamic responses

of

output, prices,

and

other macroeconomic

variables to this

innovation,

thereby obtaining

quantitative

estimates

of how

monetary policy

inno-

vations

affect the

economy.

As John Cochrane

notes,

"this

literature

has

at last

produced

impulse-response

functions

that

capture

common

views about

monetery policy";

for

example,

in

finding

that a

positive

innovation to

monetary policy

is

followed

by

increases

in

output,

prices,

and

money,

and

by

a

decline

in the short-term nominal interest rate.9

In

addition,

despite ongoing

debates

about

precisely

how

the

policy

innovation

should

be

identified,

the

estimated

responses

of

key

mac-

roeconomic

variables

to

a

policy

shock are

reasonably

similar across a

7.

Friedman and

Schwartz

(1963).

8. Sims

(1980);

more recently,

see Bernanke

and Blinder (1992),

Christiano

and

Eichenbaum

(1992),

Sims (1992),

Strongin

(1995),

Bernanke

and Mihov

(1995),

Sims

and Zha (1995),

and

Leeper,

Sims,

and Zha

(1996).

9.

Cochrane

(1996,

p.

1).

96

Brookings Papers

on

Economic

Activity,

1:1997

variety

of studies

and suggest

that

monetary policy

shocks

can have

significant

and persistent

real

effects.

The VAR literature

has focused on

unanticipated

policy shocks

not

because

they are quantitatively

very important-indeed,

the conclusion

of this

literature

is that policy

shocks are

too

small to

account

for much

of

the overall variation

in

output

and

other

variables-but because

it is

argued

that cause

and

effect can

be

cleanly

disentangled only

in the

case

of

exogenous,

or

random, changes

in

policy.

However,

looking

only at

unanticipated

policy

changes

begs

the

question

of how

system-

atic,

or

endogenous,

monetary policy

changes

affect

the

economy. '?

Earlier

work on

the

effects of

monetary policy

often does

not make

the distinction between

anticipated

and

unanticipated

policy

changes.

"

I

These

studies frequently

find a very

large

role for monetary

policy

in

cyclical

fluctuations.

An

important

recent

example

of this

genre

is an

article by Christina

Romer and

David Romer. 12

Following

the

narrative

approach

of Friedman

and

Schwartz,

Romer and Romer use

Federal

Reserve records to

identify

a series

of dates at

which,

in

response

to

high

inflation, the

Fed changed policy

in

a

sharply

contractionary

di-

rection.

Their dates presumably

correspond

to

policy

changes

with

both

an

unanticipated

component

(because

they

were

large,

or

decisive) and

an

anticipated

component (because

they

were

explicit

responses

to in-

flation);

indeed,

Matthew

Shapiro

shows that

these dates are

largely

forecastable.

'

Romer

and Romer

find that their dates

were

typically

followed

by large

declines

in

real

activity

and conclude that

monetary

policy

plays

an

important

role

in fluctuations.

But

as

several

critiques

of

Romer and

Romer's

article and

the earlier

work

on

anticipated

monetary policy point

out,

studies

that blur the

10. Cochrane

(1996)

has

emphasized

that

even

identification

of the effects of

un-

anticipated

policy

changes

may

hinge

on

distinguishing

between

anticipated

and

unan-

ticipated

changes, since

an innovation

in

policy

typically

also

changes

the anticipated

future path

of

policy.

The analyst

thus faces

the

conundrum of

determining how

much

of the

economy's response

to

a

policy

shock

is due to the shock,

per se,

and how

much

is due

to the change

in

policy

anticipations

engendered

by

the shock. The focus

of

this

paper

is different from that

of

Cochrane,

in that

we

emphasize

the effects

of nonpolicy

shocks,

such as oil shocks,

on

anticipated

monetary

policy;

but

our methods could

also

be used

to address

the

specific

issue

he raises.

11. Nor,

for that matter,

between

changes

in

the

money

stock induced

by policy

and

those

induced by

other factors.

See,

for

example,

Andersen

and

Jordan (1968).

12.

Romer and Romer

(1989).

13. Shapiro

(1994).

Ben S.

Bernanke, Mark

Gertler,

and

Mark Watson

97

distinction

between

anticipated

and

unanticipated

policies

suffer from

precisely the

identification

problem that

the VAR

literature

has at-

tempted to

avoid;

namely, that

it is

not obvious how to

distinguish the

effects of

anticipated

policies

from

the

effects

of

the

shocks to

which

the

policies are

responding.

This is

not

merely

methodological

carping,

but

is

potentially

of

great practical

importance

in

the postwar

U.S.

context, since

a

number of the most

significant

tightenings

of

U.S.

monetary

policy have

followed on the heels of

major

increases

in

the

price of imported oil.

4

This point

is

illustrated

in

figure

1,

which shows

the

historical be-

havior of the

federal

funds

rate

(here,

taken

to be an

indicator of mon-

etary policy)

in the

upper

panel

and

the

log-level

of

the

nominal

price

of

oil in the lower

panel.

Recessions,

as dated

by

the

National Bureau

of

Economic

Research,

are shaded. The

upper

panel

also

indicates the

five

dates

identified by

Romer

and

Romer that fall

within our

sample

period. The lower

panel

shows,

in

analogy

to the

Romer

dates,

seven

dates

at

which there were

major

disruptions

to the oil

market,

as

deter-

mined

in

part

by

Kevin

Hoover

and

Stephen

Perez. '5

The upper

panel of

figure 1,

taken

alone, appears

to

support the

neomonetarist

case

that

tight

money

is

the cause

of

recessions:

each of

the

first

four

recessions

in the

figure

was

immediately preceded

by

a

sharp

increase

in

the federal

funds

rate,

and the

1990

recession followed

a

monetary

tightening

that

ended

in

late

1989. Peaks in

the

federal

funds rate

also

tend to coincide

with

the Romer dates.

However,

the

lower panel

of

figure

1

shows

why

it would be

premature

to

lay

the

blame for

postwar

recessions

at the door of the Federal

Reserve: as

was

first

emphasized by

Hamilton,

nearly

all

of

the

postwar

U.S. recessions

have

also followed

increases

in

the nominal

price

of

oil, which,

in

turn,

have been associated with

monetary

tightenings.

16

Further,

many

of

these oil

price

shocks were

arguably

exogenous,

reflecting

a

variety

of

developments

both

in

the

Middle East and

in

the

domestic

industry,

as

indicated

by

the Hoover-Perez

dates. Thus

the

general

identification

problem

is here cast in a

specific

form: what

portion

of

the

last five

14.

See Dotsey

and Reid (1992) and Hoover

and

Perez

(1994).

15.

Hoover and

Perez (1994), in their

critique

of

the

Romer

and Romer

approach,

introduce six dates,

which

are,

in

turn,

based on a

chronology

due to

Hamilton

(1983).

We

have

added August

1990,

the month

when

Iraq

invaded

Kuwait.

16.

Hamilton

(1983).

In

ON

I/N

00

ON

tC)

0

C

ON

Ct

Cu

0

Ct

I/N

z

0

>#

C

"0

0

- 0

Cu

4)

"0

4)

-O

CO

4) CO

0

o

Cl Cl -

-

S

_

o

-

o'

t

x~~~~~~~~O

X

'

0

E

<.o

o

ON

0

-)'

CC3

C..-

Ce

0u*

U

0

C

OOC

V

0-

C- ,O

_

___

_

________

V

.iS

*

\~~~~~~~

"S

O

0

..

-

CU

~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~1

_7 ~~~~~~~~~~~~~~~~~~~~~~~~~~13 t E

<T

h8

5-o

"0

~ ~ ~ ~ ~ ~ ~~~~~~~~~~~

0~~~~~~~~

0

o0

C

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~0

&r-

?C's

C'sw

100

Brookings Papers

on

Economic

Activity,

1:1 997

U.S. recessions, and

of

aggregate

output

and

price fluctations

in gen-

eral,

was

due to oil price

shocks, per se,

and

what

portion was

due to

the Federal Reserve's response

to

those

shocks? To answer this question

requires a means of

measuring the

effects of

anticipated

or

systematic

monetary policies.'7

Measuring Oil

Price

Shocks and

their

Effects

We

propose

to

identify

the

importance

of

the

monetary policy

feed-

back

rule in a modified VAR framework.

In order to do

that,

however,

one needs to find an

appropriate

indicator

of

oil

price

shocks

to

incor-

porate

into

the

VAR

systems.

This is

a more

difficult

task

than it

may

appear

at

first. The

most

natural indicator

would

seem

to

be

changes

in

the

nominal oil

price;

and

indeed,

in

an

article which

helped

to initiate

the

literature on

the

effects

of oil

price shocks,

Hamilton shows

that

increases

in

the nominal

price

of oil

Granger-cause

downturns in

eco-

nomic

activity.'8 However,

the arrival

of

new

data

has

shown

this

simple

measure to

have

a

rather

unstable

relationship

with

macroeco-

nomic outcomes, leading

subsequent

researchers to

employ

increasingly

complicated specifications

of the "true"

relationship

between

oil and

the

economy.

II

In

particular,

Hamilton

argues

in his

more

recent work

that

the

correct

measure

of oil shocks

depends very

much

upon

the

precise

mechanism

by

which

changes

in

the

price

of oil

are

supposed

to

affect the

economy,

a

question

for

which

many

answers

have been

proposed

but on which

there is little

agreement.20

For

our

purposes,

the

exact

channels

through

which oil affects

the

economy

are

not

crucial.

17.

In this

paper,

we

take as

given

that

anticipated

as well

as

unanticipated

monetary

policies

influence the real

economy,

owing

to the existence of various nominal

rigidities.

Our objective

is to

provide

an

estimate of the

real

impact

of the

systematic

component

of monetary policy,

as opposed

to

testing

the null

hypothesis

that this

component

is

neutral.

18.

Hamilton

(1983),

to the

surprise

of

many,

also

demonstrates that

there appears

to have

been a

close

relationship

between oil

price

increases

and recessions

even

before

the major OPEC

shocks

of the

1970s.

19. See,

for

example,

Mork

(1989),

Lee, Ni,

and Ratti

(1995),

Hamilton

(1996a),

and

Hooker

(1996a,

1996b).

20. Possibilities

discussed by

Hamilton (1996a)

include

aggregate supply

effects

operating through

costs

of

production

and

the

indirect effects of

wage rigidity; aggregate

demand

effects;

effects

arising

from the interaction

of

uncertainty

about future

energy

prices

and

the irreversibility

of

investment;

and

asymmetric

sectoral

impacts that

force

costly

reallocations

of resources.

Ben S. Bernanke, Mark Gertler, and

Mark

Watson 101

What matters

is that one can identify

an

exogenous

movement in the

price of oil that

has a significant

and a priori plausible

reduced-form

impact on the economy.

Figure 2 illustrates

the effects

of some alternative

measures of oil

price shocks on

selected variables,

as indicated by

estimated impulse

response functions (IRFs).

Each

IRF

is based on a

five-variable VAR

that

includes,

in

this

order:

(1)

the

log

of real

GDP;

(2) the log

of the

GDP

deflator; (3)

the

log

of an index of

spot

commodity prices; (4)

an

indicator of the

state of the

oil

market;

and

(5)

the

level

of the federal

funds rate. Data

are

monthly;

the

VAR is

estimated

using

a constant

and

seven

lags,

as determined

by

the Akaike

information criterion

(AIC); and the

sample period

is 1965-95.2

Only

the impulse responses

of real

GDP,

the GDP

deflator,

and

the

federal

funds

rate are

shown,

in

each case over

a forty-eight-month

horizon and

for an

oil price shock

normalized to

correspond

to a

1

percent

increase in

the current nominal

oil

price.

Dashed

lines

correspond

to

one standard error

bands.

As is

standard

in

the

VAR

literature

on the effects of

monetary policy,

the

index of commodity

prices is added

to the

VAR to

control for infor-

mation

that the Fed

may

have about

future

inflation which is not

cap-

tured

by

the other

variables

in

the

system.22

The federal

funds rate

is

included as an

indicator

of

monetary policy.23

The

ordering

of

the

oil

indicator after

the macroeconomic

variables

imposes

the reasonable

21 Appendix A

describes the

construction of

monthly

data for

GDP

and the

GDP

deflator.

The logarithm

of real GDP

is

detrended with a cubic

spline with

three equally

spaced knot

points

imposing equality

of the levels and first

two derivatives

at the knot

points.

The resulting

estimated trend

component

is

essentially piecewise

linear, with

a

break in

the early 1970s

reflecting

the

productivity

slowdown. Other

data

are from the

CITIBASE

electronic

database,

available

from

Citicorp

Database Services

(see appen-

dix

B).

The CITIBASE

labels

for the series are:

FYFF

(federal

funds

rate), PSCCOM

(commodity

price

index),

and

PW561

(nominal

oil

price

index, Producer

Price Index

for crude oil

and

products).

We focus

here

on

full

sample

results;

we

discuss possible

subsample

instabilities

below.

22.

The inclusion

of the

commodity price

index is

suggested

by Sims

(1992) as

a

way of

eliminating the

so-called price puzzle

in

monetary

policy VARs.

In

the present

context,

it is important

to note that,

for

most

of

its

history,

the

commodity price

index

appears

to have excluded

oil and other

energy prices

(a

little

uncertainty

remains because

of

the

poor

documentation

of the

series).

Since

1987,

an oil

price

has been included

in

the

index. As we report

below,

however,

there is little evidence that its

inclusion

has

any

substantive effect

on our results.

23.

Results from

Bernanke and

Blinder

(1992),

Bernanke and Mihov (1995), and

Friedman

and Kuttner

(1996)

suggest

that

it

is reasonable to

use

the funds rate

as a

policy

indicator,

except possibly

during

the 1979-82

reserves-targeting

period.

co

_4l

kr)

_4l

c"I kr)

I

II I

_4

I

o

~~~~~~~~~~~~~I I

Cm

~~~~I

I

I II

I

le

_

0

I

/ I

lOPC

_~~~~~~~~l

X

6

f

i

=

6ci

l

= ~ ~

_-

- , -

ii~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Z

C; C; C

99

I, Ioq"

4

O~~~~~~~

I I

~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~

I I

0~~~~~~~~~~~~~~~~~~

oz

loI

p

o

dU

Ben S.

Bernanke,

Mark

Gertler,

and Mark

Watson

103

assumption

that oil price shocks do not

significantly

affect the economy

within

the month. Similarly,

ordering the funds rate

last follows the

conventional assumption that

monetary policy operates

with at least a

one-month

lag. The results

are

not sensitive to these

ordering assump-

tions,

as

we document below

in

the

context of a

larger

system.

In

figure 2 we report results for four

alternative

indicators of the

state

of

the

oil market; one

is a

slight

variation

of

the

original Hamilton

indicator, the other three are more exotic

indicators

that have been

developed

in

ongoing attempts

to

identify

a

stable

relationship between

oil

price

shocks and the

economy:

-Log

of

the

nominal Producer

Price

Index

(PPI)

for

crude oil

and

products; the nominal oil price, for short.

Hamilton

employs the log-

difference

of the

nominal oil

price, which, given

the

presence

of

freely

estimated

lag parameters,

is

nearly equivalent

to

using

the

log-level.

Given the

other

variables included

in

the

VAR,

this indicator is

also

essentially

the same

as

that

used

by

Julio

Rotemberg

and

Michael

Woodford.24

-Hoover-Perez. These

are the

oil

shock

dates identified

by Hoover

and Perez

plus August 1990,

as

discussed

in

regard

to

figure

1

.25 To

scale these dates

by

relative

importance,

for

each

month we

multiply

the

Hoover-Perez

dummy

variables

by

the

log change

in

the nominal

price of

oil over the three

months

centered

on

the

given

month.

-Mork.

After

the

sharp

oil

price

declines of

1985-86

failed to lead

to

an

economic

boom,

Knut

Mork

argued

that the effects of

positive

and

negative

oil

price

shocks on the

economy

need

not be

symmetric.26

Empirically,

he

provided

evidence that

only positive

changes

in

the

relative

price

of

oil

have

important

effects

on

output.

Accordingly,

in

our

VARs

we

employ

an indicator that

equals

the

log-difference

of

the

relative

price

of

oil

when

that

change

is

positive

and otherwise is

zero.2

24. Hamilton (1983); Rotemberg

and Woodford

(1996).

25. Hoover and Perez (1994).

26. Mork (1989).

27. We measure the relative price

of oil as

the PPI

for

crude

oil

divided by the GDP

deflator. Mork (1989) argues

that the PPI for crude oil is a

distorted measure of the

marginal cost of oil during certain periods

marked

by

domestic

price controls; he there-

fore measures oil prices by refiner acquisition

cost

instead, for the period for which

those

data are

available.

We choose

to stick with the crude oil PPI

for

simplicity,

and

because

we feel

that

there are also

problems

with

the

refiner

acquisition

cost

as

a

measure

of the marginal cost of crude.

104

Brookings Papers

on Economic

Activity,

1:1997

Hamilton. In response to

the

breakdown of the

relationship

be-

tween output and simpler

measures

of oil

price shocks, Hamilton

has

proposed a more complicated

measure of oil

price changes: the

"net

oil price increase.

"28

This measure

distinguishes between oil price

in-

creases that establish

new

highs

relative

to

recent

experience and

in-

creases that simply reverse

recent

decreases.

Specifically,

in

the context

of

monthly data,

Hamilton's

measure

equals

the maximum of

(a)

zero

and

(b)

the difference between the log-level

of

the crude oil

price

for

the current month and

the

maximum value

of

the

logged

crude oil price

achieved

in

the

months.

Hamilton

provides

some evi-

dence

for the

usefulness

of this

variable, using semiparametric

methods,

and Hooker also

finds it to

perform

well,

in

the sense of

having

a

relatively

stable

relationship

with

macroeconomic variables.29

The

deficiencies

of the

simplest

measure

of

the state

of

the oil

market,

the

nominal

price

of crude

oil,

are

apparent

from

figure

2.

In

particular,

for our

1965-95 sample period,

a

shock to

the nominal

price of

oil

is

followed

by

a

rise

in

output

for

the

first

year

or so and

by

a

slight

short-

run

decline

in

the price

level. Both

of

these results

(which

have been

verified

in

the recent literature

on

oil

price shocks)

are

anomalous,

relative to the conventional

wisdom

about

the

effects

of oil

price

shocks

on

the economy.

As

indicated

in note

29,

other

simple

measures,

such

as the

relative price

of

oil, give

similarly unsatisfactory

results.

The three more complex

indicators

(Hoover-Perez, Mork,

and Ham-

ilton) produce

"better

looking" IRFs,

in

that

output

falls

and

prices

rise following an oil price

shock, although generally

neither

response

is statistically significant.

The

point

estimates

of the

effect of an

oil

price

shock on

output suggest

a modest

impact

from an

economic

per-

spective.

For

example,

in

the case

of the

Hamilton

indicator,

the sum

28. Hamilton

(1996a, 1996b).

29. Hamilton

(1996b);

Hooker (1996a).

We also

experimented

with

VARs including

the log-difference

of

the

nominal price

of oil

(the

indicator

used

by

Hamilton, 1983);

the log of the

real price of oil (the

nominal

oil

price

divided

by

the

GDP

deflator); the

log-difference

of the real price

of oil; and the log

of

the nominal price

of oil weighted

by the

share of

energy costs

in

GDP (as suggested by

William Nordhaus at the

Brookings

Panel meeting).

As the results

obtained

were

very

similar to

those

using

the log nominal

price

of

oil,

we

do

not report

them

here.

The

literature

provides yet

additional

indicators

of oil

price

shocks.

Those

proposed

by

Ferderer

(1996)

and

Lee, Ni,

and

Ratti

(1995),

for

example,

focus on the

volatility

of

oil

prices

rather than the level. For

simplicity,

we ignore

these second-moment-based

measures and concentrate on

measures that

are

functions of

the level of oil

prices.

Ben S.

Bernanke,

Mark

Gertler,

and Mark

Watson

105

of the impulse

response

coefficients

for

output

over the

first forty-eight

months

is

-0.538, implying

that a

1

percent (transitory)

shock to oil

prices leads to a

cumulative loss of about

0.5

percent of a month's real

GDP, or 0.045 percent of a

year's

real

GDP,

over

four years. As is

touched

on

below,

more

economically

and

statistically significant ef-

fects of oil price shocks are

estimated

(a)

when the

latter

part of the

sample,

which contains the

somewhat

anomalous 1990

episode,

is

omit-

ted;

and

(b)

when the

VAR

system

is

augmented

with

short-term and

long-term market

interest rates.

Figure

2

also

shows that for

all

four

indicators of

the oil

market, a

positive

innovation to oil

prices

is followed

by

a rise

in

the

funds rate

(tighter monetary

policy),

as

expected,

and the

response

is

generally

statistically significant.

This

funds

rate

response

illustrates the

generic

identification problem: without further

structure,

it is

not

possible

to

determine how much

of the decline

in

output

is

the direct

result of the

increase in oil

prices,

as

opposed

to the

ensuing tightening

of

monetary

policy.

This

brief

exercise demonstrates

a main result of

the recent

literature

on the macroeconomic effects

of oil

prices,

that

finding

a

measure

of

oil

price

shocks that

"works"

in

a

VAR context is

not

straightforward.

It is also true that the

estimated

impacts

of these

measures on

output

and

prices

can

be

quite

unstable

over different

samples,

as

discussed

below. For

present

purposes, however,

based on the

evidence of the

literature and our

own

analysis (including figure 2),

we choose

the

Hamilton net oil

price

increase

measure of oil

price

shocks for

our basic

analyses.30

As we

discuss further

below,

we have checked

the robust-

ness of our exercises

to the use

of alternative

oil

market

indicators;

in

general,

we find that

when a

given

oil-market

indicator

yields

reason-

able results

in exercises like

those shown

in

figure 2,

our alternative

simulations also

perform

reasonably.

Measuring the

Effects of Endogenous Monetary Policy

Figure

2

shows

that,

at least

for some more

complex-some might

argue,

data-mined-indicators

of oil

prices,

an

exogenous

increase in

the

price

of oil has the

expected

effects

on the

economy: output falls,

30. In particular,

Hooker

(1996a)

finds that the Hamilton measure is the

most

stable

across subsamples.

106 Brookin2s Pavers on Economic Activity. 1:1997

prices rise, and

monetary policy tightens

(presumably

in

response to

the

inflationary pressures

from the oil

shock).

Since

James Tobin's

Brookings paper,

however,

it has been

argued

that

oil

and

energy costs

are

too small relative to total

production

costs to

account for

the entire

decline

in

output that, at

least

in

some

episodes,

has

followed

increases

in

the price

of

oil.31 A

natural

hypothesis, therefore, is that

part of the

recessionary impact of oil

price

increases arises from

the

subsequent

monetary

contraction.

Sims and

Zha

attempt

to

provide rough

estimates of

the

contribution

of

endogenous

monetary policy changes

in

a VAR

context.32

Their

approach

is to "shut down" the

policy

response

that would

otherwise

be

implied by the VAR

estimates;

for

example, by

setting

the

federal

funds rate

(the

monetary policy

indicator)

at its

baseline level

(the

value

that it would have taken in the absence

of

the

exogenous

nonpolicy

shock).

The difference between

the total

effect

of

the

exogenous

non-

policy

shock on

the

system

variables

and

the estimated

effect when

the

policy response

is

shut down

is then

interpreted

as a

measure of

the

contribution of the

endogenous policy

response.

As

Sims and Zha

correctly point out,

this

procedure

is

equivalent to

combining

the

initial

nonpolicy

shock with a series of

policy

innova-

tions

just

sufficient to off-set

the

endogenous

policy response.

Implic-

itly, then,

in

the Sims-Zha

exercise,

people

in the

economy

are

repeat-

edly "surprised" by

the failure

of

policy

to

respond

to the

nonpolicy

shock in

its accustomed

way.

The authors

argue,

not

unreasonably,

that

it would take some time for

people

to

learn

that

policy

was

not

going

to

respond

in

its usual

way;

so

that,

for

deviations of

policy

from its

historical

pattern that

are neither

too

large

nor

too

protracted,

their

estimates

of the

policy

effects

may

be

acceptable approximations.

This

justification

is

similar

to

the one that

Sims

uses

in

earlier articles for

conducting policy

analyses

in a VAR

setting, despite

the

issues raised

by

the

Lucas critique.33

31.

Tobin (1980) .See also Darby

(1982),

Kim

and

Loungani (1992), and Rotemberg

and

Woodford (1996). Rotemberg

and Woodford

argue

that a

monopolistically compet-

itive market structure, which leads to

changing markups over the business cycle, in

principle can explain the strong

effect

of

oil

price

shocks.

32. Sims and Zha (1995). Counterfactual

simulations

in

a VAR

context have also

been

performed by

West

(1993)

and Kim

(1995);

neither

paper distinguishes anticipated

from unanticipated movements

in

policy.

33.

See,

for

example,

Sims

(1986).

Ben S.

Bernanke,

Mark

Gertler,

and Mark

Watson

107

Rather than

ignoring Lucas's

argument

altogether, however,

one

might try to

accommodate it

partially

in

the VAR

context, by

acknowl-

edging that it may be more

important

for some

markets

than for

others.

In

particular, the

evidence for the

relevance of the

Lucas critique

seems

much

stronger for financial markets-for

example,

in

the

determination

of

the term structure

of interest rates-than

in

labor

and product

mar-

kets, which has led

some economic

forecasters and

policy analysts

to

propose and

estimate models

with

rational expectations in

the

financial

market

only.34

In

that

spirit,

we

modify

the Sims-Zha

procedure

for

measuring

the

effects

of

endogenous

policy by

assuming

that interest

rate

expectations are formed

rationally (and

in

particular,

that

financial

markets

anticipate

alternative

policy

paths),

but that the

other

equations

of

the VAR

system

are invariant to the

contemplated

policy change.

The

latter assumption

can be rationalized

by

assuming either that ex-

pectations

of

monetary policy

enter the

true structural

equations

for

output, prices,

and so

forth

only

through

the term

structure of

interest

rates; or,

if

other

policy-related expectations

enter

into those

structural

equations,

that

(for

policy changes

that are not too

large)

these

respond

more

sluggishly

than financial

market

expectations,

as

proposed

by

Sims.35

Although

our method

is

obviously

neither

fully structural nor

immune to

the

Lucas

critique, it provides

an

interesting alternative to

the Sims-Zha

approach.

More

specifically,

we consider small

VAR

systems

that

include stan-

dard macroeconomic

variables,

short-term and

long-term

interest

rates,

and

the federal funds

rate

(as

an indicator of

monetary policy).

We

make the

following

assumptions:

First,

that the

federal funds

rate does not

directly

affect

macro-

economic variables

such

as

output

and

prices;

a

reasonable

assumption,

since the funds rate

applies

to

a

very

limited

set

of

transactions

(over-

night borrowings

of

commercial

bank

reserves).

Hence the

funds rate

is

excluded

from the

equations

in the

system

determining

those

varia-

bles.

However,

the funds

rate

is

allowed to affect

macroeconomic

var-

iables

indirectly,

through

its

effect on short-term and

long-term

interest

rates, which,

in

turn,

are allowed to enter

every

equation

that deter-

34. See Blanchard (1984)

on the

comparative

relevance of the Lucas

critique.

See

Taylor (1993) for an example of

a model with rational

expectations limited to

the

financial market.

35.

Sims

(1986).

108 Brookings

Papers

on

Economic Activity,

1. 1997

mines

a macroeconomic

variable. Note that the

assumption

that mon-

etary policy

works strictly through

interest rates

is

conservative, as it

ignores other

possible channels,

such as the

exchange rate and the

"credit channel." In

this

sense,

our estimates should

represent

a lower

bound on the

contribution

of

endogenous monetary policy.

-Second,

following many previous authors,

that

the

macroeco-

nomic variables in

the system are Wold-causally prior

to

all interest

rates. That is,

in our

monthly data,

we assume

that

interest

rates

respond

to

contemporaneous

developments

in the

economy,

but that

changes

in

interest rates do

not affect "slow-moving"

variables

such

as

output and

prices within the

month.

This is

a

plausible assumption, given planning

and

production lags.36

-Third, that the

funds rate

is

Wold-causally prior

to the other mar-

ket interest rates.

That

is,

the covariation between innovations in

the

funds rate and

in

other

interest rates

is

caused

by

the influence of

monetary policy

changes

on interest

rates,

rather than

by

the

response

of

the policymakers

to market

rates

within

the month.

This is a

strong

assumption,

although

it

appears

to

give fairly

reasonable results in

the

context

of the

expectations theory

of

the term

structure. It

may

be

justified

if

the

term

premium

contains

no information

about the

econ-

omy

that is not also contained

in

the

other

variables seen

by

the

Fed.

Below, we briefly discuss

an alternative

ordering assumption

that al-

lows for considerable

reaction

by

the

Fed

to current

market

interest

rate

movements.

Formally,

let

Y,

denote

a set

of

macroeconomic

variables, including

the

price

of

oil,

at date

t.

Similarly,

let

R,

=

(Rs,

RI)

represent

the

set

of

market interest

rates; specifically,

the three-month

Treasury

bill rate

(the

"short

rate,"

RS)

and

the

ten-year Treasury

bond

rate

(the "long

rate," R,).

Finally,

the

scalar

FF, is

the

federal funds

rate. Under the

assumptions

above,

the

restricted

VAR

system

is

written

p

(1)

Yt (I'v,Yt-i

+

FvriR,t-)

+

GN'l'Et

36. As Sims points out, however,

the

assumption

is less

plausible

for the

commodity

price index, which is included

in the

nonpolicy

block as an

information

variable;

see

Leeper, Sims, and Zha (1996).

Ben

S.

Bernanke,

Mark

Gertler, and Mark Watson

109

p

(2)

FF, =

,

a,jYt_j

+

nr,jR,t_

+

Trr0jFF,t_)

+

Er,

+

G

+ GfE,,

(3)

R,

(_rr-',iYt-i

+

_r,.,.,jR,_j

+

Trl,if;ft,_)

+E>,t

+

Gr,!

e! + G,E;,

where the rr and

G

terms are matrices

of

coefficients of

the appropriate

dimensions, the

E

terms are vectors

of

orthogonal

error

terms, and

constant

terms have

been

omitted for notational

convenience. For equa-

tion

1,

the exclusion

of

FFt_i

follows

from

the first

assumption above,

that the

funds

rate does not

directly

affect

macroeconomic

variables;

and the

exclusion

of

Er,

and

E11,

is

implied

by

the

second

assumption,

that innovations to interest rates

do not

affect

the

nonpolicy variables

within

the period.

In order to

apply the expectations

theory

to

identify a relationship

between

the funds rate

and the market

interest

rates,

and

to implement

our

policy

experiments,

it

is useful to

decompose

the

market

rates into

two

parts:

a

part

reflecting expectations

of

future values of

the

nominal

funds

rate,

and

a

term

premium.

We define

the

following

variables:

tIs-

I

(4)

Rs

=

E,

(

O FF,+)

i

0

(5)

RI=

E,

(

O

WFF+)

i

0

(6)

Ss= RS-R

(7)

Si RI RI

where ns

= 3

months and

nl

= 120

months are the terms

of the

short-

term and

long-term

rates, respectively;

the

weights, w,

are

defined

by

ts-

I

,11-

I

(S.=

i

X

P

and

w,

=

i

>

E

1;

and

E

is

the

expectations

j=O

J=0

110

Brookings Papers

on Economic

Activity,

1:1997

operator.

We set the

monthly discount

factor, E

equal to 0.997,

so that

112

is

equal

to 0.9637.

The

R

variables

defined

in

equations

4

and 5 are

the "expectations

components"

of the

short

and

long market

interest

rates, and

the residual

S terms

in

equations

6 and 7 are

time-varying

term-cum-risk

premiums

associated

with rates at the two maturities.

Note that

the time series

of

the two components

of

short

and long

interest

rates are easily

calculated

from current and

lagged values

of Y,

FF, and

R, using the

estimated

rr parameters

in

equations

1-3. In

particular,

finding the

estimated expectations

components

of

short and

long rates

is purely a

forecasting

exercise and does not

require

structural

identifying

assumptions.

With these

definitions,

it

is useful to rewrite the

model of

equations

1-3

as

p

(8)

Yt

[1

T7.iY,-i

+

1Tvri(R,-Ji

+

S,_)]

+

GVyE

t

p

(9)

FF,=E (TrjY,_;

+

Tfr,jRt_;

+

1T0jFF,t;)

+Ef,

+

GE,,lt

+

GSES,

p

(10)

S,

=

>

(XA,.!jYt_

+

_

srjR

+

XS,IFF,t;)

+

Es,t

+

G,VE,,t

+

GsfEft,

Equation

8

is

identical

to

equation

1, except

that the two market

interest

rates have

been

broken

up

into their

expectations

and

term

premium

components.

Equations

9

and

10

correspond

to

equations

2

and

3,

with

the interest

rates, R,

replaced by

the

corresponding

term

premiums,

S.

Since

the difference

between

R

and

S is

the

expectations

component

of

interest rates,

which

is constructed

as a

projection

on

current and

lagged

values of observable

variables, equation

10 are

equivalent

to

equations

2

and

3.

In

particular,

the

coefficients

in

equations

9

and

10 are

simply

combinations

of the coefficients

in

equation

3

and

the

projection

coef-

ficients

of

the

federal

funds

rate on current

and

lagged

variables.

37.

This weighting

function

and the

value of

I3

are suggested by Shiller,

Campbell,

and Schoenholz

(1983).

Ben

S. Bernanke,

Mark Gertler, and Mark Watson 111

We work with the system of equations 8-10 because it simplifies the

imposition of some alternative identifying restrictions. Our main iden-

tifying assumption, discussed above,

is that the

federal funds rate is

Wold-causally prior to the other interest rates

in

the model; this corre-

sponds to the assumption that G1,

= 0

in

equation

9.

However, an

alternative assumption, which

allows for

two-way causality between

the

funds

rate and market

rates,

is that shocks to the federal

funds rate

affect

other interest rates

contemporaneously only through

their

impact

on

expectations

of the future

funds rate

(that is,

funds rate

shocks do

not affect term premiums contemporaneously);

this

corresponds

to

the

restriction that

Gs,

=

0

in

equation

10. Note

that

this

alternative as-

sumption allows the funds

rate to

respond

to

innovations in

term pre-

miums.

In

both

cases,

we assume that

GVV

is

lower-triangular (with

ornes

on

the diagonal),

as

in

conventional

VAR

analyses employing

the

Cho-

leski decomposition.

In

most

of

our

applications,

the

"macro block"

consists

of real

GDP, the

GDP

deflator,

the

commodity price index,

and

Hamilton's

net oil

price

increase

variable,

in

that

order;

as we

show

below, our results

are robust

to the

placement

of the

oil market

indicator.

To illustrate how we carry

out

policy experiments,

consider

the

scenario of greatest

interest

in

this

paper:

a shock

to the oil

price

variable. The base case,

which

incorporates

the

effects

of

the

endoge-

nous policy response,

is calculated

in

the

conventional

way, by simu-

lating the effects

of an

innovation

to

the oil

price

variable

using

the

system of equations 8

to

10.

Among

the results

of

this

exercise are the

standard impulse response functions, showing

the

dynamic impact

of

an oil

price

shock

on

the

variables

of the

system, including

the

policy

variables.

To simulate the

effects of

an oil

price

shock under

a counterfactual

policy regime,

we

first

specify

an alternative

path

for the

federal funds

rate-more

specifically,

deviations

from the baseline

impulse response

of the funds rate-in a manner

analogous

to the

approach

of

Sims and

Zha.38

However,

we

assume

that

financial markets understand

and an-

ticipate

this alternative

policy response; by assuming

"maximum cred-

ibility"

of

the

Fed's announced

future

policy,

we stand

in

direct con-

trast to

Sims and

Zha,

who assume that

market

participants

are

purely

38.

Sims and

Zha (1995).

112

Brookings

Papers

on

Economic

Activity,

1:1997

backward-looking.

To

incorporate

this

assumption

into

the

simulation,

we

calculate the

expectations

component of

interest

rates,

R,+,,

i

=

0,

1,

..., that

is

consistent with

the

proposed

future path

for the

federal

funds

rate. We

then

resimulate the effects of

the oil shock in

the

system

of

equations

8-10,

imposing

values of R,

consistent with

the

assumed

path of

the

funds

rate,

and also

choosing

values of

E11, such

that the

assumed future

path

of the funds

rate

is

realized.

Note

that this

method

can be

used to construct

alternative

impulse

response functions

based

on

full-sample

or

subsample

estimates and

to

simulate

counterfactual

economic

behavior for

specific

episodes,

such as

the

major

oil

price

shocks. We

use it

in

both

ways

below.

Some

Policy

Experiments

With the

methodology

described

above,

we are

able

to

perform a

variety

of

policy

experiments, using

estimates from

our

sample

period,

January

1965

through

December 1995. The VAR

is

estimated

using

a

constant

and

seven

lags,

as

determined

by

AIC.

A

Monetary

Policy

Shock

To check

on the reasonableness

of

the basic

estimated

system,

we

begin

with the conventional

analysis

of a

monetary

policy shock,

mod-

eled

here as a 25 basis

point

innovation

to the

federal

funds

rate. The

effects of an innovation to

the

federal funds rate are

traced

out in a

seven-variable

system

that

includes

output,

the

price

level,

the com-

modity price

index,

the

Hamilton oil

measure,

the funds

rate,

and the

short and

long

term

premiums.

Figure

3

presents

the

resulting

impulse

response

functions.

As described

above,

the

values of

the short and

long term

premiums

at each date are calculated

by

subtracting

the ex-

pectations

component

of short and

long

rates

(based

on

forecasts

of

future values of the funds

rate)

from

the

short

and

long

rates

themselves.

In

this

base case

analysis,

equivalent

results are

obtained

by

directly

including

the short and

long

rates

in

the

VAR

(ordered

after

the

funds

rate),

and the

implied

responses

for short and

long

rates are

included

in

figure

3. In

the

data,

there are

large

low-frequency

movements

in

the

term

premium

of the

long

rate,

with trend

increases of

about 1

per-

I

Cl

I

/

I

/

0

CC)

/

I

0,1

/

C)

I

'Cl-

Cl

o /

/

0

,

-

0

I,

C)

, /

Cl

--

II

If) 0

If)

0

If)

666

0 c

I

0

I

/ r

II

Ii

/

Ii

Cl

//

Ii

II

'Cl-

/ /

Ii

/ /

II

C)

/

0 / /

C)

0

','

-

u

-

/

Cl

Q

LL

I

oo

II

I I

/

CI

-

I

Ig

1

I

II

I

-

-,

Cu

-

0 'f)

0

Cu

0

1-

0 C-

0

1-

0

0 0

-

0

'I

I

00

*0

I

Cu

I

Cl

'Cl-

Cu

o

Cu

0

0I

o

.

-

0

C)

'C1

C,,

C)

0

'I

0o

Cu

0

7

-

C)

/

Cl

I

/

--

/'I

C)

e#

Cl C-

Cl

0

0 0

-

0

I/

6oo

666

II

114

Brookings

Papers

on

Economic

Activity,

1:1997

centage point in

both the 1970s and the 1980s. We

remove

this trend

variation with a

cubic spline

(specified

as

described in

note 21). As

we

report

in the section on robustness

below, leaving

the

long premium

undetrended

does not

significantly

affect the

results

.9

Impulse

response

functions to

the funds rate

innovation

in

figure

3 are

shown

with

one-

standard-error bands.

The

results of this exercise

will

look

quite

familiar

to those

who

know

the recent

VAR literature on the effects of

monetary

policy. The

innovation to the

funds rate

(initially

25 basis

points, peaking

at about

35 basis

points) is

largely

transitory,

mostly dying

away

in the

first nine

months.

Output

declines

relatively

quickly,

reaching

a

trough

at

about

eighteen to

twenty-four

months and then

gradually

recovering.

The

price

level

responds

sluggishly,

but

eventually

declines,

nearly two

years after

the

policy

innovation.

Commodity prices

also

decline,

and

do so

much more

quickly

than

does the

general price

level.

The

model's

only

exclusion

restriction,

that the

funds rate

does not

belong in the

"upper

block"

(which

includes the oil

indicator,

output,

prices,

and

commodity

prices),

conditional on the

presence of

short-

term

and

long-term

interest

rates in that

block,

is

marginally

rejected:

the

p

values for the

exclusion

of the funds

rate from

the

upper

block

are,

respectively,

0.01

for the

output

equation,

0.06 for

the

price

level

equation,

0.23 for the

commodity price

equation,

and

0.18 for the oil

equation.

However,

the effects of this exclusion

do not

seem

to be

economically very

significant.

For

example,

if

we

compare

the

effects

of

a funds

rate

shock on

output

in the

restricted,

seven-variable

system

with the

analogous

effects

in the

conventional,

unrestricted,

five-

variable

system

(excluding

the market interest

rates),

we

obtain

vir-

tually

identical results.

An

interesting

new feature

of

the seven-variable

system

is

that it

allows

one to examine the

responses

of market

interest rates

to

monetary

policy

innovations,

and in

particular,

to

compare

these

responses

to the

predictions

of the

pure expectations

hypothesis.

Looking

first at

short-

39. Fuhrer

(1996) shows that

the large movements

in

the

long rate

can be explained

in

a way

consistent

with

the expectations

hypothesis

if the

market

was

making

rate

forecasts at each

date based

on

a particular

set of beliefs

about how the

Federal Reserve's

objective function

has varied

over time.

However,

there is

nothing

in

Fuhrer's

analysis

that connects these

hypothesized

beliefs with the actual time-series

behavior of

the funds

rate.

Ben S. Bernanke, Mark Gertler,

and Mark

Watson

115

term (three-month)

rates,

a

25 basis

point

innovation to

the funds rate

implies about a 15

basis point

increase in the short rate, and the two

rates then decline synchronously.

This seems quantitatively reasonable.

To

check

the consistency

of

this

response

with

the

expectations hypoth-

esis, one can look at

the behavior

of the short

rate

term

premium, which,

by construction,

is the difference

between the actual short

term rate

and

the short

term

rate

implied by

the

pure

expectations hypothesis.

The

short

rate term

premium

is

significantly negative

immediately

following

a

funds

rate

innovation,

implying

that

in the

first

month

or

two

after

an

innovation

to

the

funds rate,

the short-term interest

rate is

estimated to

respond

less

than

would be predicted by

the

expectations hypothesis.

However,

the short

rate term

premium quickly

becomes

statistically

and

economically insignificant,

suggesting

that

the

expectations hypothesis

is a reasonable description

of

the

link between the funds rate

and

the

short-term

interest

rate

after

the first

month.

The

long-term

interest

rate

is a different

story.

As

shown

in

fig-

ure 3,

the

long

rate responds by

about

5

basis

points

to

the

impact

of

a

25 basis point

innovation

in the

funds

rate,

and the

response

remains

above

zero for some three

years,

which

again

does not seem

unreason-

able.

However,

comparison

of

the

responses

of the

long-term

interest

rate and

the

long rate

term

premium

reveals that

they

are

very close,

the

latter being slightly

less than

the former. The

implication

is

that

the

expectations theory

explains relatively

little of the

relationship

between

the

funds

rate

and the

ten-year government

bond rate. This

finding

is

not

so

surprising, given

the

transitory

nature

of funds rate

shocks

com-

pared

with

the duration

of

these bonds.

The

estimated behavior

of the

long term premium

thus constitutes

some

evidence that

long

rates

"overreact"

to

short

rates,

a

phenomenon

that

has

frequently

been

documented

in the

term structure

literature

(although,

we

appear

to

find

less overreaction than

is

typically reported

in

the

literature).40

Simulations

of

the

Effects of

an Oil

Price Shock

Since our

expanded

model seems

to

perform

reasonably

in

the

case

of an innovation to

monetary policy,

we now

turn

to

the exercise

of

40. An alternative

explanation

for the overreaction of the

long

rate is that the policy

shock is imperfectly

identified.

Note, for

example,

the slight "output

puzzle"-output

increases in

the first few

months after the

policy shock.

Possibly a

better identification

scheme

would eliminate

the overreaction.

116

Brookings Papers

on Economic

Activity,

1:1997

greatest

interest,

which is to

use the model

to decompose

the effects

of

an

oil

price

shock

into direct

and

indirect (that is,

through endogenous

monetary

policy)

components.

Figure

4 shows impulse

responses

fol-

lowing

a shock to

Hamilton's

net oil price

increase

measure under

three

scenarios.

The first scenario,

which we

label

"base,"

shows the impulse

re-

sponses

of the variables

to a

1

percent

innovation

in

the nominal

price

of

oil

in

the seven-variable

system.

This is a normal

VAR simulation,

except

that the funds

rate

does not

enter

directly

into the

equations

for

output,

prices, commodity

prices,

or

the

oil indicator.

This

case is

intended

to show

the effects

on

the

economy

of

an oil

price

shock,

including

the

endogenous

response

of

monetary policy,

in contrast with

the next two simulations,

which

involve alternative methods

of

shutting

off the policy response.

The second

scenario

we label

"Sims-Zha"

(with

some

abuse

of

terminology).

In this

case

we

simply

fix

the funds rate at

its base

values

throughout

the simulation,

in the manner

of Sims and

Zha.41

However,

recall that

in contrast

to

the original

Sims-Zha

exercise, in our

system

the funds rate

does not enter directly

into the block

of

macroeconomic

variables.

Rather,

the funds

rate

exerts its macroeconomic

effects

only

indirectly,

through

the

short-term

and

long-term

interest rates included

in the system.

Thus in

this

exercise,

we are

effectively

allowing the

change

in the funds

rate to act

through

its

unconstrained,

reduced-form

impact

on market

interest rates

(which

are ordered

after

the

funds rate).

The

third

scenario,

which we

label

"anticipated

policy," applies

our

own

methodology,

described

above.

We

again

set the

funds rate

equal

to

its baseline

values;

that

is,

we shut off the

response

of

monetary

policy

to

the oil shock

and the

changes

induced

by

the oil shock in

output,

prices,

and so forth.

But

in

this

case,

we

let the

two

components

of short-term

and

long-term

interest

rates be determined

separately.

The

expectations

component

of both

interest

rates is set to

be

consistent

with

the future

path

of the funds

rate,

as assumed

in

the

scenario.

The

short

and

long

term

premiums

are allowed

to

respond

as estimated in the base

model. (Below,

we also consider

a

case where the

term

premiums

are

kept

at their baseline values.)

For

the

simple,

constant funds rate

case

being

examined

here,

the Sims-Zha

and

anticipated

policy approaches

41. Sims and

Zha

(1995).

Ben

S.

Bernanke,

Mark

Gertler,

and Mark

Watson

117

Figure

4. Responses

to a

Hamilton

Oil Price

Shock,

Seven-Variable

Systema

Output

Prices

0.00875

0.035

-

0.00250

0.020

-0.01375

0.005

6 12 18

24

30 36

42

6

12 18

24

30

36 42

Commodity

Oil

prices

0.09 ~~~~~~~~0.4

0.02

i

&

;'

-

1 0.4

6

12

18

24 30

36

42

6 12 18 24

30 36,

42

/\ ~~~Federal

Short

rate

5.5

fud

4 .5l;

-

3.0

.

0.5

-.

6 12

1

18

1

24

30

36

42

6 12

18

24

30

36

42

Months

1.255

-0.50

l

- Base

scenario

-

Sims-Zha

scenario

V

.

._______________________

-

Anticipated

policy

scenario

6

12 18 24

30

36

42

Months

Source: Authors'

VARs, using

data described

in

appendix

B.

a.

Graphs

show

forty-eight

month

response

of variables

to a

I

percent

Hamilton

oil

price

shock.

Sims-Zha and

anticipated

policy

scenarios

eliminate

the

normal response

of monetary

policy.

Vertical

axis

scales represent

percent

deviations

of

variables

(basis point

deviations

of

interest

rate

variables).

Sample

period

is 1965-95.

118

Brookings

Papers

on

Economic

Activity,

1:1997

show

roughly

similar

departures

from baseline.

Note, however, that the

former cannot distinguish between

policies

that

differ only in the ex-

pected future values

of the

funds

rate, whereas,

in

principle, the latter

approach

can make that distinction.

The results of figure 4 are

reasonable, with all variables exhibiting

their

expected qualitative behaviors.

In

particular, the absence of an

endogenously restrictive

monetary policy

results in

higher output and

prices, as one would anticipate.

Quantitatively,

the

effects are large, in

that

a

nonresponsive monetary

policy

suffices to

eliminate most of

the

output

effect of an oil

price

shock, particularly

after the first

eight to

ten months. The conclusion that

a

substantial

part

of the

real

effects of

oil

price shocks

is due to the

monetary policy response helps

to

explain

why the effects of these shocks

seems

larger

than can

easily be ex-

plained

in

neoclassical (flexible

price)

models.42

The anticipated policy

simulation results

in

modestly higher output

and

price responses

than the Sims-Zha simulation

in

figure

4. The

differences

in

results

occur

largely

because the

anticipated policy

sim-

ulation involves a

negative

short-run

response

in

both the

short and

long

term

premiums,

and thus

lower interest

rates in

the

short run.

Figure

5

repeats

the

anticipated policy

simulation

of

figure 4,

but with

the re-

sponse of the term premiums

shut

off;

that

is,

the

funds rate

is

allowed

to affect the macroeconomic

variables

only through

its

effects on the

expectations component

of market rates.

This

alternative

simulation

attributes somewhat

less

of

the recession

that follows an oil

shock

to

the

monetary policy response,

but

endogenous monetary policy

still

accounts for two-thirds

to three-fourths

of

the total

effect

of

the oil

price

shock

on

output.

As

another exercise

in counterfactual

policy simulation,

we exam-

ine

the three

major

oil

price

shocks followed

by

recessions:

OPEC

1,

OPEC

2,

and the

Iraqi

invasion

of Kuwait.

Figure

6

shows

the

results,

focusing

on

the behavior

of

three

key

variables

(output,

the

price level,

and the funds

rate)

for

the

five-year

periods surrounding

each

of these

episodes (respectively, 1972-76,

1979-83,

and

1988-92).

Each

panel

shows

three

paths

of the

given

variable. One line

depicts

the

actual

historical

path

of

the

variable. The line

marked "federal

funds

endog-

42. It should be emphasized

that

we

are not

arguing

that the policies actually fol-

lowed by the Fed

in

the face of

oil shocks were

necessarily

suboptimal; the usual output-

inflation trade-off is present in our simulations,

and

we

do not

attempt

a

welfare analysis.

Ben S.

Bernanke,

Mark

Gertler,

and

Mark

Watson

119

Figure

5.

Responses

to a Hamilton

Oil

Price

Shock,

No Premium

Term

Responsea

Output

Prices

0.00875

0.02

-0.00250

0.01

-0.01375

0.00

6 12 18 24

30

36 42

6

12 18 24

30

36 42

Commodity

Oil

prices

0.06

0.7

0.00

0.4

-0.06

0

X

< ~

1

0.1

6

12

18

24

30

36 42

6

12 18

24

30

36 42

Federal

Short rate

5.5

4.25

3.0

2.50

0.5

0.75

6

12 18

24

30 36

42

6

12

18 24

30

36 42

Months

1.8

1.2

0.6

- Base

scenario

-

Anticipated

policy

scenario

6

12

18 24

30

36

42

Months

Source:

Authors'

VARs,

using

data described

in

appendix

B.

a.

Graphs

show

forty-eight-month

response

of variables to

a I

percent

Hamilton oil

price

shock. Scenarios

are as

those

shown

in

figure

4,

except

that

the

responses

of term

premiums

are shut

off.

Vertical

axis scales

represent

percent

deviations

of

variables

(basis

point

deviations

of

interest

rate

variables).

Sample period

is

1965-95.

I I>

i

00~~~~~~~~~~~~~0

It

0~~~~

00~~~~~~~

Y~~~~~~~~0

'~

"0

00

o

,a

O 0ol _ sX,

N

<,

col

cO

00 ~ ~ ~ ~~~~~"

r

I

o

,

*E

M

I I

00

I00

0 0 0

0~~~~~~~~~~~~~~~~~~

00~~~~~~~0

00>

pouad

aldwuIS

Ben S. Bernanke,

Mark Gertler, and Mark Watson

121

enous"

shows the behavior

of the system when

the oil variable is

repeatedly shocked, so that

it traces out its actual

historical path; all

other

shocks in the system are set to

zero;

and

the

funds rate is allowed

to

respond endogenously to

changes

in

the oil

variable and the induced

changes

in output, prices, and other variables. This

scenario is intended

to

isolate

the portion

of

each

recession

that results

solely from the oil

price

shocks and the associated

monetary policy

response. Finally, the

line

marked "federal funds

exogenous"

describes the

results of an

exercise

in

which oil prices equal their

historical

values,

all

other shocks

are shut

off,

and the nominal

funds rate is

arbitrarily

fixed

at a value

close to its initial value

in

the

period. (Term premiums

are

allowed to

respond

to the oil price

shock.)

This last scenario

eliminates the policy

component of the effect

of the oil

price shock,

leaving only

the

direct

effect of the

change

in

oil

prices

on the

economy.

Several observations

can be made from

figure

6.

First,

the

1974-75

decline

in

output

is

generally

not

well

explained by

the oil

price shock.

The pattern

of

shocks

reveals, instead,

that the

major

culprit

was

(non-

oil)

commodity prices.

Commodity prices (not

shown)

rose

very sharply

before this recession

and

stimulated

a

sharp

monetary policy response

of their

own,

as can be seen

by comparing

the

historical

path

of

the

funds

rate

with

its

path

in

the

federal funds

endogenous scenario,

in

which

the commodity price

shocks are set to zero. The

federal

funds

exogenous

scenario,

in which

the funds rate

responds

to

neither

com-

modity

price

nor oil

price

shocks,

exhibits

no

recession at

all, suggest-

ing that

endogenous

monetary policy (responding

to both oil

price

and

commodity price shocks)

did, indeed, play

an

important

role in

this

episode.

The results for 1979-83

generally

conform

to

the

conventional wis-

dom. The decline

in

output

through

1981

is well

explained by

the

1979

oil

price

shock and the

subsequent response

of

monetary policy.

After

the

beginning

of

1982,

the main

source of

output

declines

(according

to this

analysis)

was

the

lagged

effect

of

the autonomous

tightening

of

monetary policy

in

late 1980

and 1981. Note that

if

one

excludes both

the

monetary policy

reaction to the

oil

price

shocks and

the

autonomous

tightening

of

monetary

policy by

Federal

Reserve

Chairman Paul

Volcker

(as

in

the

federal funds

exogenous

scenario),

the

1979-83

period

exhibits

only

a modest

slowdown,

not a

serious recession.

The

experiment

for 1988-92

similarly

shows

that

shutting

off

the

122

Brookings Papers

on

Economic

Activity, 1:1997

policy

response to oil

price

shocks

produces a

higher

path of

output and

prices

than

otherwise;

again, compare the

paths of the

endogenous

monetary policy

and

exogenous

monetary policy

scenarios.

One puzzle

that

emerges

is

why the

substantial

easing

of

actual policy from

late

1990

did not move

the

actual path

of

output

closer

to

the

alternative

policy

scenario. It

is

possible

that

special

factors,

such

as

credit prob-

lems,

may have

been at

work.

Oil,

Money, and the

Components

of

GDP

The

application of

our

method

for

separating

the

direct

effects of oil

price

shocks and the indirect effects

operating

through the

monetary

policy

response

leads to a

rather

strong

conclusion:

the

majority

of

the

impact of an oil

price

shock on the real

economy

is

attributable

to the

central

bank's

response

to the

inflationary pressures

engendered

by the

shock.

A

check on

the

plausibility

of this

result, using

a

different

identifying

assumption

and

more

disaggregated

data,

is

provided

by figure

7. This

figure is

based

on

the seven-variable

VAR

system

employed

above (real

GDP, the GDP

deflator,

commodity

prices,

the

Hamilton

oil

market

indicator, the funds

rate,

and short-term and

long-term interest

rates),

with the funds

rate

excluded from the first four

equations.

To

this

system

we

add,

one at a

time

and without feedback into

the main

system, eight

components

of

GDP:

consumption,

producer

durables

expenditure,

structures

investment,

inventory

investment,

residential

investment,

government

purchases,

exports,

and

imports.43

With these

systems

we

conduct two

experiments.

First,

we

examine the

impulse

responses

obtained

when

the Hamilton

oil

price

variable

is

shocked

by

1

percent

and the federal funds rate is allowed to

respond

endogenously (these

responses

are shown

by

dashed

lines in

figure 7).

Second,

we

examine

the

impulse responses

to

an

exogenous

federal funds

rate

shock

of

equal

maximum

value to the

endogenous

response

of

the funds

rate

in

the

first scenario

(shown

by

solid

lines).

We

think

of this

exercise

as

a

comparison

of

the total

effect

of

an oil

price

shock,

including

the

43. Except

for

consumption,

which is available

at

the

monthly

frequency, monthly

data

for

the

GDP

components are interpolated by state

space methods;

see appendix A.

Components

are measured relative to the

exponential

of

the

trend

for the

logarithm of

real

GDP, as calculated from

the

spline regression

described

in

note

21.

1~~~ i

..

-

LD

02

00

*w

? ?

?

O O

O

c

O

C)

'

\

C

X ~ ~ ~ ~

N

Ct 0

-L

'c

E?f

0

((B)(f

0

()

CL

4

*

CD CD

O

?O O O

C,

I I

I

I ~ I

C

6

C's

>-N)

Q

0j

--

(

4(N)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~l

~~~~~~Cu

-~~~~~~~~~~~~~~~~~~~~~~~~~

-

U

-~~~~~~~~~~~~~~~~~~~~~~~~~~~l

0

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~0C

I-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~I

Cu~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Cu~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'

Cu

/

-

~~~~~~~~~~~~~~~~~~~~~~~~~~-1

.~ ~ ~ ~~~~~~~~~~~

'

C.)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'

-~~~~~ -I--.-

(N)

C)

-

I~~~~~~~~~~~~~~~~~~~~~~~~~~~~l

_ _ _ _ _ _

-

I~~~~~~~~~~~~~~

124

Brookings Papers on

Economic

Activity,

1:1

997

endogenous

monetary response,

with the effect of a

monetary tightening

of similar magnitude

but not

associated with an oil

price shock. To the

extent that the two

responses are quantitatively

similar, it seems rea-

sonable

to

attribute

most of the total effect

of

the oil

price shock

to

the

monetary policy

response. Note, however, that we are

using a different

identification

assumption here than above; that is, we

implicitly assume

that the

economy

responds

in

the same

way

to

endogenous

and

exoge-

nous tighenings of

monetary policy.

The results

of

shown

in

figure

7

provide

substantial

support for the

view that the

monetary policy response

is the

dominant

source of the

real effects of an oil

price

shock.

In

particular,

the

response of output

is

virtually identical

in

the

two

scenarios, implying

that it

matters little

for real economic outcomes whether

a

change

in

monetary policy

of a

given magnitude is

preceded by

an oil

price

shock or

not.

Very similar

responses across the

two experiments

are also found

at the disaggre-

gated level,

especially

in

equipment

investment

(producers' durable

equipment),

inventory investment,

and residential

investment. Slightly

greater effects for the

scenario

including

the oil

price

shock are

found

for

consumption

and

structures

(although

the latter

difference is

quan-

titatively

small and

statistically insignificant).

Government

purchases

responds

more

strongly

in

the scenario

that

includes the oil

price shock,

for

reasons

that

are

not obvious.

The differences

between

the

two scenarios are

also

instructive. The

experiment

that includes

the initial

oil

price

shock does

show a

sub-

stantial

inflationary

impact

in the

short

run,

which

gives

some

indication

as to

why

the

Fed

responds

so

vigorously

to

such shocks. On

the

margin,

the oil

price

shock also raises

commodity prices

and

the

long-term

interest rate

(presumably, reflecting

an

increased risk

premium)

and it

leads to increased

real

exports

and

decreased real

imports (net

of

terms-

of-trade

effects).

These

responses

are as

expected.

Some

Alternative

Experiments

Although

we have focused

on the role

of

systematic

monetary policy

in

propagating

oil

price shocks,

our

methodology

applies equally

well

to other sorts

of

driving

shocks.

As a further check on

the

plausibility

Ben S.

Bernanke,

Mark

Gertler, and Mark

Watson

125

of

our

method, we

briefly consider

two alternative

cases: a

shock to

commodity prices and a

shock to

output.

A

COMMODITY PRICE

SHOCK.

Figure

8

looks at

the effects of

a shock

to the

commodity price index

in

our

original seven-variable

system. As

with

the

oil price shock

studied

in

figures

4 and

5,

we

consider three

scenarios.

First,

in

the base scenario we

calculate

the

impulse

responses

resulting

from a

1

percent

innovation in

commodity prices,

allowing

monetary

policy (as

represented by

the

federal

funds rate) to

respond

in its

normal way.

Second, we examine

the

effects of

shutting off the

policy

response, using

the Sims-Zha

methodology described

above.

Finally,

we shut

off

the

monetary

policy response

by means of

our

anticipated

policy approach.

For

simplicity,

in

the

anticipated

policy

simulation we set the

responses

of the term

premiums

to zero

(as

in

figure 5),

so

that

both

short-term

and

long-term

nominal

interest

rates

are

effectively

assumed not to

respond

to the

shock to

commodity

prices.

Figure 8

shows that

a

1

percent innovation

in

commodity

prices has

an

ambiguous

effect on

output:

real

GDP

rises

for

the first

year

but

declines thereafter. Prices rise

unambiguously.

One

explanation

for

these

results is that what

we are

labeling

a

positive

shock to

commodity

prices is,

in

fact,

a

mixture

of an adverse shock

to

aggregate

supply

and an

expansionary

shock

to

aggregate

demand.

The

federal funds

rate

rises

sharply

in

response

to

an increase in

commodity prices,

which

we

interpret

as

the

Fed's

response

to

the

inflationary

surge;

other

interest

rates

also

rise. The

oil

price

indicator

responds

very

little

in

the

short

run to a

commodity price

innovation,

which is

reassuring,

in

the sense

that

it confirms that the

commodity

and oil

price

variables

are not

excessively

collinear.

Shutting

down the

monetary

policy response

to the

commodity price

shock, by

either

the

Sims-Zha

or the

anticipated

policy

method,

leads

to the

expected

response.

Analogous

to the case of oil

price

shocks,

the

recessionary impact

of a

commodity

price

shock

is

eliminated and

the

inflationary impact

is

magnified.

Although

it

may

well

be the

case that

the innovation

in

commodity prices

is

not a

cleanly

identified

supply

shock,

there

is

no

evidence that

an increase in

commodity prices

de-

presses

real

activity

in

the

absence

of a

monetary

policy response.

AN

OUTPUT SHOCK.

Figure

9

shows

analogous

results

when

the

driving

shock is a shock to

output.

As with

the

commodity shock,

we

compute

126

Brookings Papers

on Economic

Activity,

1:1997

Figure 8.

Responses to a

Commodity Price Shock, No

Term Premium Response"

Output

-Prices

0.09

0.455/

-0.01

0.15

6 12 18 24 30

36 42 6

12 18 24 30 36 42

- -

-

- -

-

~Oil

prices

1.8

0.1475

1.2

0.0450

Commodity

lli t.

0.6

prices

-

0.0575

6

12

18

24

30

36 42

6 12 18 24 30 36

42

15

10

6

12

18

24 30 36 42

6

12 18 24 30 36

42

Months

6

-

Base scenario

3

_

- -

Sims-Zha scenario

-

Anticipated policy

scenario

6 12 18 24 30 36 42

Months

Source: Authors'

VARs, using

data described

in

appendix

B.

a.

Graphs

show

forty-eight

month

response

of variables to

a

I

percent

commodity price

shock

when the

responses

of

term

premiums

are shut off. Sims-Zha and

anticipated

policy

scenarios eliminate the normal

response

of

monetary policy.

Vertical

axis scales

represent percent

deviations

of variables

(basis point

deviations

of interest rate

variables). Sample

period

is

1965-95.

Ben S.

Bernanke,

Mark

Gertler,

and

Mark

Watson

127

Figure

9. Responses

to an

Output

Shock,

No

Term

Premium

Responsea

i

Output

Prices

0.65

~~~~~~~~~~1.0

0.3006

-0.05, ~~~~~~~~0.2

6 12 18 24

30

36 42

6 12

18

24 30 36 42

Commodity

O

.

p

_

-

35 5

prce

0.125pie

2.0

0.050

0.5

-0.225

6

12

18

24

30 36

42

6

12 18

24 30 36 42

52.5

|

/

X~~~~ederal

Short

rate

l

52.5

fns37.5

35.0

.

17.5

1.

6

12

18

24

30 36

42

6 12

18

24 30

36

42

Months

21

,

Long

rate

14

'/

t

Base

scenario

7

F X

- -

Sims-Zha

scenario

-

~ ~

-

_

Anticipated policy

scenario

6 12 18

24 30 36.

42

Months

Source: Authors' VARs,

using

data described

in

appendix

B.

a.

Graphs

show forty-eight

month response

of

variables

to a

I

percent

output

shock when the responses

of

term

premiums

are shut off. Sims-Zha and

anticipated

policy

scenarios

eliminate

the normal response

of monetary

policy.

Vertical axis

scales represent

percent

deviations

of variables

(basis point

deviations

of

interest

rate variables). Sample

period

is 1965-

95.

128

Brookings Papers

on

Economic

Activity,

1:1

997

the

impulse

response functions for three cases: a

base

case

in

which

monetary policy is allowed

to

respond

in its

normal

way to

the output

shock,

and cases

corresponding

to the Sims-Zha and

anitcipated

policy

methods for

shutting

down the

policy

response.

As

before,

we assume

no

response of

the term

premiums.

Admittedly,

like a shock to

commodity prices,

an

output

shock does

not

have a clear

a priori

economic

interpretation;

it is

an

amalgam of

various random

factors

affecting

output, holding

constant

the other

variables included in

the

system.

However,

based on

figure 9

it seems

reasonable to

interpret output

shocks

in

this

system

as

being

dominated

by

aggregate

demand fluctuations:

a

positive

output

shock is

followed

by increases

in

oil

prices,

commodity

prices,

and

the

general price

level,

as well

as

in

all three interest rates.

Because the

historical

tendency

of

monetary policy

is to "lean

against

the

wind,"

when

the

normal

policy

response is

shut

off, the

effects of the

aggregate

demand

shock

(as we

interpret

the

output shock)

are all the

greater.

Figure

9 shows

that

in

the Sims-Zha and

anticipated

policy

scenarios,

the

output

effect of

the

shock is much more

persistent

and

prices

rise

by

more

than in

the

base

case.

Interest rates are

lower,

reflecting

easier

monetary

policy. Note

that in this

analysis,

the Sims-Zha and

anticipated

policy

approaches

give

almost identical results.

These

experiments demonstrate

that

our methods for

shutting down

the

response

of

monetary

policy

are

applicable to,

and

give

reasonable

results

for,

shocks other than

oil

price

shocks. It

would be

interesting

to combine our

methodology

with

identified VAR

techniques

that

could

give

a

sharper

structural

interpretation

to

innovations

estimated in

the

macro block of the model.

Robustness

and Stability

We

return to

our

main

theme,

the role of

systematic

monetary policy

in

amplifying

the real

effects of oil

price

shocks,

to

consider the ro-

bustness and

stability

of our results.

Robustness

of the Results

We perform a

variety

of checks

for

robustness,

some of which

(such

as

shutting

down the term

premium response)

are alluded

to above.

To

Ben S.

Bernanke,

Mark

Gertler, and Mark Watson

129

provide more

systematic information, table

1

reports

some summary

statistics from

alternative specifications of

our VAR

system. We con-

sider (a) three

alternative oil-market

indicators; (b)

three alternative

orderings of

variables within

the

VAR;

and

(c)

two

alternative

detrend-

ing

assumptions. We

also calculated

results for

alternative measures of

output (for example,

industrial

production), alternative

measures of the

price level (for

example,

the

personal

consumption

expenditure deflator

and the consumer

price

index),

and alternative

interest

rate

maturities;

but

since none of these

variable

substitutions

have

important

effects on

our

findings, they are omitted

from the

table.

The first row of

table

1

reports

results

for

the Hamilton

oil indicator

(our

base

specification),

whereas the second and third

rows

substitute

the Mork and Hoover-Perez

indicators,

respectively (see

figure 2).

The

fourth

row

corresponds

to

ordering

the federal funds

rate

after,

rather

than

before,

the

two

open

market

interest rates. The fifth

row

orders

the

Hamilton oil

market indicator first

in

the

system,

and the

sixth

row

orders the oil market indicator third-after

output

and

prices,

but before

the

commodity price

index. The seventh row

is

for

a

specification

in

which

output

and the

long

rate term

premium

are not

detrended,

and

the

eighth

row

reports

results

when all

variables

in

the

system

are

detrended

by

a

cubic

spline (as

described

in

note

21).

For

each

of

the

eight

alternative

specifications,

table

1

reports

the

effects on

output

and

prices

of a 1

percent

oil

price

shock,

under

(a)

a

standard

simulation,

allowing

for

the

endogenous response

of

policy

to

the oil

price shock;

(b)

the Sims-Zha

simulation,

in

which the

federal

funds rate is

fixed

at

its baseline

value;

and

(c)

the

anticipated policy

simulation. Under

the

heading

"output,"

we

report

the sum

of the

impulse response

coefficients for

output

for the first

twenty-four

months

after the

oil

price

shock,

which

we

employ

as

a

measure of the

output

loss associated with the shock.

Under the

heading

"prices,"

we

report

the

twenty-fourth

impulse response

coefficient for

prices,

divided

by

two,

which

can be

interpreted

as the

increment

in

the annual

average

inflation rate over

the first two

years

following

the

shock.

Standard

errors,

calculated

by

Monte Carlo methods

employing

500

draws

per

specification,

are shown

in

parentheses.

The

table also

shows the dif-

ferences between the baseline

(endogenous

policy) specification

and

the

results obtained under the

Sims-Zha and

anticipated

policy assump-

tions, again

with

the

associated standard errors.

-r t-o e-

i-O o- nn

n- oc

Q' o- 6

6 6 66

6

66 6

W

QNflo

o

ocr

o

J~

Co ~

o

r-r

o

t

6666666

666

X~~~~~ r-.

-

ONm

-Pm - ct -

O

r--L

-

D

o j- - r

--

X

- -

Q N t

-

X ~ ~ ~ ~~r

-

Lfo

-

.

66

..

,o- I'-

o

oI

0

N~~~~~~~~~

o 0 0

m

0

.

-S

cc

1 1 2 x

0e

A P_

oN -

S o o

o o o o

o

o0

I

O

- _ _

_

Cq o t

N O ' > '

>0

X o - .

o - . - - .

o

_ _

_

_ _

_ :)

(

- e)

--c

00 rq

1- xD

o

m

00

oo

tr

ct

C)-

o

- -

(=

-

-- b#

'IC

I~~~~~~t

lt

00 rq

M

r~~~~~~~~~.n c: I :

_

VI

_

0

I _ ON

rq

m~~~~~~~~~~~~

CC

_-r

C4:

N

O r

u

~ ~~~~~~~~~~~~~~~~~~~~ u

OC CZ'

132

Brookings Papers

on

Economic

Activity, 1:1 997

The point

estimates reported

in

table

1

are

consistent with

the find-

ings

discussed above (in figures

4

and

5, for example). In

particular,

the baseline

simulations show that

an oil

price

shock

depresses

output

and increases

inflation, by magnitudes that

are

reasonably

comparable

across all

specifications. The Sims-Zha method

of

shutting

off

the mon-

etary policy

response tends to eliminate

all or

most

of

the

negative

effect

of

the

oil

price

shock

and,

in

almost

all

cases,

increases the

inflationary

impact, as expected. The

anticipated policy method of

elim-

inating the

policy response has even

larger effects, fully

eliminating

the

recessionary impact of the

oil

price

shock

in

all cases.

The

standard

errors for

most entries

in

table

1

are

quite

high, reflecting

the

fact that

the

standard

error

bands

on

the impulse

response

functions

spread

out

rather quickly.44

However, the differences

in

the

output

responses be-

tween

the baseline

and

alternative simulations are

statistically

signifi-

cant

in a

number

of

cases,

in

particular,

when the

policy response

is

shut down by the Sims-Zha method.45

In

general, our results

appear

to be

qualitatively robust,

although

they

are not

always precisely

estimated.

In

particular,

a

view

that

as-

cribes most

or

even

all

of

the real effects

of

an oil

price

shock to the

endogenous

monetary response

does

not

seem

inconsistent

with the

data.

Stability of

the

Results:

The Role

of

a

Changing Policy Response

We take

up

the issue of

subsample

stability

not

only

as a

qualification

of

our

results,

but

also because it

appears

that at least

some

of

the

observed instabilities

of our

system

can be

given

an

interesting

eco-

nomic

interpretation. Indeed,

we show

that

variations in

the Federal

Reserve's

reaction

function

have

something

of

the flavor

of

a

natural

44. The standard errors are particularly high

for the

anticipated policy simulations,

apparently reflecting, in part, the uncertainty

associated with

the long-term interest rate

forecasts required by this method.

45. We also considered alternative models estimated

with twelve lags, rather than

the seven chosen by AIC.

In

this case, the finding

that

shutting

off

the monetary policy

response eliminates the effect of the oil shock obtains at

short horizons but not at the

twenty-four-month horizon.

The reason

is

that with

twelve

lags,

the

funds

rate

is esti-

mated to rise in response to an

oil

price shock,

but then to

fall

quickly

below

trend. Our

alternative policy,

which assumes no

response throughout,

is

thus

not

effectively

easier

than

the baseline policy

over

the twenty-four-month

horizon.

Ben S. Bernanke, Mark

Gertler, and Mark Watson

133

experiment, which may help to improve the identification of

the endog-

enous policy effect.

Some tests

of

the stability

of the coefficients

in

our

seven-variable

base VAR, with lag lengths

chosen

by

the

Bayes

information

criterion,

are reported in table 2. For simplicity, the funds rate is allowed

to enter

all

equations. The upper panel,

labeled

"Quandt tests," gives

asymp-

totic

p

values for

the

hypothesis

that

the

coefficients

of

the variable

listed in the column heading, together with the regression constant

term,

are stable over the sample period

in

the

equation given

by the row

heading. Thus, for example, the Quandt tests

show

that the

hypothesis

that the

coefficients

on

the

price

level

in

the

oil

equation

are

stable over

the

entire sample

can be

rejected

at the

0.016

confidence

level.

In

a

similar

format, the Chow split-sample

tests

reported

in

the lower panel

of

table 2 tests

each set

of

coefficients

for

stability

across

the two halves

of the sample. These

tests are

included

because,

unlike

the

Quandt

tests, they are robust

to

heteroskedasticity.

There is substantial evidence

of

instability

in

the VAR

system. The

equation for

the

price

level is

clearly quite unstable,

with

p

values near

zero for most

blocks

of

coefficients.

The

Quandt

tests also

suggest

that

there is instability

in the coefficients

relating

the

funds

rate

and the

short-term

and

long-term

interest

rates.

Nevertheless,

stability

of the

output equation cannot

be

rejected.

It

appears, however, that

at least some

of the

instability

in

the

link

between oil and the macroeconomy may

be due

to

a

shift

in

the

policy

response. Figure 10

illustrates this

point.

The

figure

shows the

output,

price level, and federal

funds

rate

responses

to an

oil

price

shock,

as

implied by systems

estimated

over

the whole

sample

and over

each of

the three decades of

the

sample (1966-75, 1976-85,

and

1986-95).

The

full sample estimates

of

the effects of an

oil

price

shock

are as

seen above.

Note, though,

how

the

responses vary

over

subsamples

(keeping

in mind that

ten-year subsamples

are short

for

this

purpose).

The output response

across different

periods

is

inversely

correlated with

the funds rate

response.

The

sharpest

decline

in

output

occurs in the

period 1976-85,

which

also exhibits

the most

aggressive

rise

in the

funds rate.

The

strong response

of

monetary policy

during

this

period

presumably

reflects

the Federal

Reserve's

substantially

increased

con-

cern with

inflation

during

the Volcker

regime.

The

output

response

is

weakest

in

the

1986-95 subsample.

In

this

case,

there

is

virtually

no

00

>

Ic

tr,

oo

t o

om

o o

o

--E

OC

C

t

O)

'

)

N0

r0-~f

--

z~~~~~~ m s o m

W')

00

)

"C

O)

OC

ON

O OOOOr)

C L

,,C

ON C) ON

CA

-

tr

--

-OC-

ca

NN

c

-

xD

O -

O 0

O

m

o00L 0

r

=

o-

E

N

0

o

oo-o

o

U Cq

00

0

m -O

00tr

00

t

r

o>>

~~ ~. 6666666 66

6666

%

0u

> t

- o)

O O

O

2

-

4O

t1 -

r

t

;

0

u

Ot O

OOOOOO

U ~ ~ ~ ~~~~~~~~~~~~-

U

C)

> U>

0

0 U

0

U~~~~~~~~~~o

U CU'

'U O

0.-

>~~~~

CEc;

o

ooo

o

c,

S~~~X

o

~~~~~~~~~I

I

42

I

C

C C C)

C

)

I

C

c C

-

C

'0

I I

I

ii

.

I

~I

I I

I

C)

,C

C)

;~~~~~~~~~~~~~~~~~~~~~~~~~~

0:

_

O

~~ONOOO

O?O?

ON

ONOOa

@

I

II

III

I I

*- n

O~~~~~~~~

I

K.

.I

X

~ ~~

~~ ~~

In

....

I

0

5 e

Y

J

/0

.-t

pouad

oldwuxS

136

Brookings Papers

on

Economic

Activity, 1:1997

response

in

the funds rate. The atypical

behavior of

the funds

rate during

this period may reflect the presence

of

confounding factors,

such as the

weakness of financial sector

balance sheets and the decline in

consumer

confidence that depressed

the

economy

at

the

time

of

the

one

major

oil

shock of

that subsample,

the 1990 increase

in

prices.

In

any

event,

the

subsample evidence

is

highly

consistent

with

the view that

the reduced-

form impact of

oil on the

economy depends significantly

on

the mone-

tary policy reaction function.

Conclusion

This paper offers both

methodological

and

substantive

contributions.

Methodologically, we

show

how to

modify

standard

VAR

systems to

permit simulations

of

the

economy

under

alternative endogenous poli-

cies. Since our focus is

on quantifying

the economic impact of historical

feedback

policies,

the

alternative policy

that we consider is very simple;

a

virtue of our

approach

is that

it would not be

difficult

to

extend the

analysis

to

consider

more

interesting alternatives,

for

example, "Taylor

rules." It

would

also be interesting

to

compare

our results with

those

obtained

from alternative

(possibly,

more

structural) methodologies.

Substantively,

our results suggest

that

an

important

part

of the

effect

of

oil

price

shocks

on the

economy

results

not from the

change

in oil

prices, per se, but

from the resulting tightening

of

monetary policy.

This

finding may help

to

explain

the

apparently

large

effects

of

oil

price

changes

found

by

Hamilton

and

many

others.

APPENDIX

A

Interpolation

of

Monthly

NIPA Variables

IN THIS PAPER

we use interpolated

monthly

values

of GDP, the com-

ponents of

GDP, and the

GDP deflator.

This

appendix

describes the

interpolation

process.

The

data and additional detailed

estimation

re-

Ben S. Bernanke,

Mark Gertler, and

Mark

Watson

137

sults are available on a distribution diskette from the authors, upon

request.

We designate quarterly series by capital letters and monthly series

by lower-case letters. Quarters are indexed by

T

=

1, 2, ..., N, and

months by t

=

1, 2, ..., n. Let QT be an (observed) quarterly variable

that is

to

be interpolated-for example, real GDP-and let ST be a

scaling variable such that

YT

QTIST

is

nontrending. Similarly, let q,

be the (unobserved) monthly series corresponding to QT-for example,

montly

real

GDP-and

let

s, be

a

scaling

variable such

that y,

q,/s,

is

nontrending. QT and q, are related by the identity

2

1

QT

3

-

q3T-i,

and hence

YT

and

y,

are related

by

the

identity

= 1

2

YT 3 Y3T_ i(S3T i/ST)

Interpolation

is

by state space

methods.

Suppose

that

there

is

a

vector

of

(observable) interpolator

variables at the

monthly level, x,; industrial

production, for example,

is a

monthly variable

that

provides informa-

tion about

within-quarter

movements of real GDP.

We assume

that

the

unobserved

monthly

variable

y,

is related to

the

interpolator

variables

according

to the

"causal,"

or

"transition," equation

y,=

x,' +

u,,

where

u,

=

pu,_,

+

E,, E,

-

N(O,&2).

In

our application,

all transition

equations

include a

constant term.

When

one or more

of

the

interpolators

becomes available

midsample,

all

interpolators (including

the

constant

term)

are interacted

with

dummy

variables and

the

possibility

of a

shift

in

the

value

of

.2

is

allowed

for.

Let

z,

be

a

monthly

"indicator" variable

that

equals Y,,3

in

the third

month of each

quarter

and

is zero

otherwise. Then the

indicator,

or

measurement,

equations

are

given by

138

Brookings Papers

on

Economic

Activity,

1:1997

2

z,=

E

y,ti(s,tJlS3,),

t

=

3, 6, 9,

12, ...,n

and

z=0

x

y,,

for all other

values of t.

The parameters

p,

p, and

02

are

estimated by

maximum likelihood,

assuming Gaussian

errors.

Conditional

on

the estimated

parameters, let

ytl,

=

E,yt,

where E is the

expectations operator.

The

interpolated

values,

given

the

full information

set,

are

thus

given

by

qt,E

=

ytl,,st.

This

method

is

similar to that

proposed by

Chow

and Lin

(1971),

although

it

allows for a more

general

treatment of

the serial

correlation

in

u,.

To

estimate the

accuracy

of

the

interpolation,

one can

use R2 mea-

sures

of

fit.

In

levels,

the measure of

fit

is

R,eis =

var(y2,,)/var(y2),

and

in

differences

it is

R

=

var(zAy2,,)/var(zAy2).

Table Al

lists the

quarterly

series

that

we

interpolate,

the

corre-

sponding monthly

interpolators,

and

the

measures of fit

(corresponding

to the

scaled values

of

the

variables).

Variables

are listed

by their

CITIBASE

mnemonics,

which are defined

in

appendix

B. The

scale

variables

used for real

flow

variables

are

personal

consumption expen-

ditures

(GMCQ),

at both the

quarterly

and

monthly

levels. The

personal

consumption expenditure

deflator

(GMDC), monthly

and

quarterly,

is

used as

the scale variable

in the

interpolation

of

the

GDP

deflator.

Consumption

data

(disaggregated

to

durables,

nondurables,

and

ser-

vices)

exist at a

monthly frequency

and

thus

do

not have

to be

inter-

polated.

Monthly

GDP is calculated

as

the

sum

of the

monthly GDP

components (we ignore

the

slight

deviations from

that

relationship

caused

by

chain

weighting).

The

R2

values

suggest

that the

interpolators explain

nearly

all of

the

variability

in

the levels of

the

scaled series. With the

exceptions

of

government

consumption

and the GDP

deflator, they

also

explain nearly

all

of the

implied

month-to-month variation

in

the

series.

Ben S.

Bernanke,

Mark

Gertler,

and Mark Watson

139

Table Al.

Interpolators and

Goodness of

Fit

Quarterly

series

Monthly

R2,

by

specification

interpolateda interpolatorsa

Levels

Differences

GDPD

PWFSA

0.997

0.489

PWFPSA

PWIMSA

PWCMSA

GIPDQ

IPE

0.999

0.775

MSNDFb

MSMAEb

GIRQ

IPIC

0.999

0.894

MMCON

CONFRC

HSF

GISQ IPIC 0.999

0.807

MMCON

CONIC

CONCC

GVQ A

IVMFGQ

0.970

0.929

A

IVRRQ

A

IVWRQ

GGEQ

CONQC

0.999

0.633

IPH

FBOb

GEXQ

FSE602

0.999

0.919

FTE71

FTEF

GIMQ

FSM612 0.998

0.861

FTM333

FTM732

Source: Authors'

calculations

based

on data described

in

appendix

B.

a.

Series identified

by

CITIBASE

mnemonics,

see

appendix

B.

b.

Available beginning

in

January

1968.

APPENDIX

B

Data

THIS APPENDIX describes the data series used

in the

paper.

All

data are

from the

CITIBASE electronic

database,

available

from

Citicorp

Da-

tabase Services.

Series are identified

by

their

CITIBASE

mnemonic

codes.

140

Brookings

Papers

on

Economic Activity,

1:1997

Quarterly series

GDPD GDP

deflator, index, 1992

=

100.

GEXQ

Exports of goods and services, chained

1992 dollars.

GGEQ

Government consumption expenditures

and gross investment,

chained

1992

dollars.

GIMQ Imports

of

goods

and

services,

chained

1992

dollars.

GIPDQ

Investment, producers' durables,

chained

1992 dollars.

GIRQ

Investment, residential,

chained

1992 dollars.

GISQ

Investment,

nonresidential

structures,

chained 1992

dollars.

GVQ

Change

in business

inventories,

total,

chained 1992 dollars.

Monthly series

CONCC Construction

put

in

place,

commercial, seasonally adjusted,

1987

dollars.

CONFRC

Construction

put

in

place, private

residential

building,

season-

ally

adjusted,

1987 dollars.

CONIC

Construction

put

in

place,

industrial

building, seasonally

ad-

justed,

1987

dollars.

CONQC

Construction

put

in

place, public,

seasonally adjusted,

1987

dollars.

FBO

Federal

budget,

net

outlay,

not

seasonally adjusted;

deflated

by

interpolated government purchases

deflator

(GDFGEC),

seasonally adjusted by

the authors

by

means of

a

regression

on

monthly

dummies.

FSE602

Exports, excluding military

aid

shipments, seasonally adjusted;

deflated

by

the PPI

for

finished

goods

(PWF).

FSM612

General

imports, seasonally

adjusted;

deflated

by

the PPI for

finished

goods (PWF).

FTE71

U.S.

merchandise

exports,

nonelectrical

machinery,

season-

ally

adjusted;

deflated

by

the

PPI

for

machinery

and

equipment

(PWME).

Ben S.

Bernanke,

Mark Gertler,

and

Mark

Watson

141

FTEF

U.S. merchandise

exports,

agricultural

products,

seasonally

adjusted;

deflated

by the

PPI for

farm products,

processed

foods,

and

feeds

(PWFPF).

FTM333

U.S. merchandise

imports,

petroleum,

and

petroleum

prod-

ucts,

seasonally

adjusted;

deflated by

the

PPI for crude

petro-

leum

(PW561).

FTM732

U.S.

merchandise

imports,

automobiles and parts,

seasonally

adjusted;

deflated by

the

PPI for motor

vehicles and equipment

(PWAUTO).

FYFF

Federal funds

rate,

percent.

FYGM3

Interest

rate,

three-month

Treasury

bills

from the secondary

market,

percent.

FYGT5

Interest

rate,

five-year

Treasury

bonds,

constant

maturity,

from

the secondary

market, percent.

FYGT10

Interest

rate,

ten-year

Treasury

bonds,

constant maturity,

from

the secondary

market, percent.

GMCQ

Personal consumption

expenditures,

seasonally

adjusted,

chained 1992

dollars.

GMCDQ

Personal consumption

expenditures,

durables,

seasonally

ad-

justed,

chained

1992

dollars.

GMCNQ

Personal consumption

expenditures,

nondurables,

seasonally

adjusted,

chained

1992

dollars.

GMCSQ

Personal consumption

expenditures,

services,

seasonally

ad-

justed,

chained 1992

dollars.

GMDC

Implicit

price

deflator, personal

consumption

expenditures,

in-

dex,

1987

=

100.

HSF Housing

starts,

new

private

housing

units,

seasonally

adjusted.

IP Industrial production

index,

total,

seasonally

adjusted,

1987

=

100.

IPE

Industrial

production

index,

business

equipment,

seasonally

adjusted,

1987

=

100.

142

Brookings Papers

on

Economic Activity, 1:1997

IPH Industrial production index,

defense and space equipment, sea-

sonally adjusted, 1987

=

100.

IPIC Industrial production index, construction

supplies, seasonally

adjusted, 1987

=

100.

IVMFGQ Inventories, manufacturing,

seasonally adjusted, chained 1992

dollars.

IVRRQ Manufacturing and

trade inventories, retail trade, seasonally

adjusted, chained

1992

dollars.

IVWRQ Manufacturing and

trade inventories, merchant wholesalers,

seasonally adjusted,

chained 1992 dollars.

MMCON

Manufacturing shipments,

construction materials

and

supplies,

seasonally adjusted;

deflated by

the

PPI

for

materials and com-

ponents

for

manufacturing

(PWIMSM).

MSMAE Manufacturing shipments,

machinery

and

equipment, season-

ally adjusted;

deflated

by

the

PPI

for

machinery

and

equipment

(PWME).

MSNDF Manufacturing shipments,

nondefense

capital goods indus-

tries, seasonally adjusted;

deflated

by

the PPI for

capital equip-

ment

(PWFP).

PSCCOM Spot

market

price index,

all

commodities,

from

Commodity

Research

Bureau,

not

seasonally adjusted,

1967

=

100.

PUNEW

CPI-U,

all

items,

seasonally adjusted,

1982-84

=

100.

PW561 PPI,

crude

petroleum,

not

seasonally adjusted,

1982

=

100.

PWFPSA PPI, capital equipment,

seasonally adjusted,

1982

=

100.

PWFSA

PPI,

finished

goods,

seasonally adjusted,

1982

=

100.

PWIMSA

PPI,

intermediate

materials, supplies,

and

components,

sea-

sonally adjusted,

1982

=

100.

PWCMSA PPI,

crude

materials,

seasonally adjusted,

1982

=

100.

Comments and

Discussion

Christopher A. Sims: The broad

aim of this

paper

is

to

go

beyond the

result, now widely

confirmed in the empirical time-series

literature on

monetary policy, that

surprise changes

in

monetary policy

are

a minor

source of economic

fluctuations.

The nature of

systematic

reactions of

monetary policy

to the state of the

economy

could be

a major

determi-

nant of the character of

fluctuations, even though erratic

disturbances

to

monetary policy are

not. The

paper concludes that the

evidence

is

consistent

with

a

major

role for

monetary policy; so large

that,

for

example, most of the

observed output effects of oil price

shocks would

disappear

with a different

monetary policy.

I

agree with the main

conclusion

of

the paper, but only

because the

authors

have

been so careful in

stating

it.

I

would

emphasize

more than

they do

how

much

uncertainty

remains

about

the

size

of

the real effects

of

monetary policy.

It remains

possible

for

a

skeptic

to

maintain the

view that the effects of both

systematic

and random shifts in

monetary

policy

are

negligibly

small.

My

comments

therefore

emphasize

the

reasons to doubt that the

effects

of

systematic monetary policy

are large,

despite

the

paper's

evidence

to

the

contrary.

The authors pursue their aim

by focusing

attention

primarily

on

the

reaction

of

the economy

to

surprise changes

in oil

prices. On

the

face

of

it,

this focus

is

appealing,

because

most

economists believe

that

they

know

roughly

when

large

surprise changes

in

oil

prices

have

occurred

and have little doubt that

these

changes

were

distinct

from

surprise

changes

in

monetary

policy. Identification-separation

of

the

inter-

pretable disturbance

from other sources

of

variation

in

the data-there-

143

144

Brookings

Papers

on

Economic Activity,

1:1997

fore promises

to be

easier

than

it

would

be

with other

types

of

private

sector

disturbances.

This

idea,

it

seems

to

me,

has not turned

out

as

well as

one

might

have

hoped.

In the

first

place,

the

intuition

that

historical

oil

price

"shocks"

are

well understood

and

easily

identified

is incorrect.

Although

Hamilton's

original

work

did

not

require

elaborate

filtering

of the

data,

it

appears

that to

extend

it to

the

current

time

does

require

such

filtering.

In

the

present

paper,

four

different

measures

of

oil price

shocks

are shown

in

figure

2

to deliver four quite

distinct estimated

effects

on the

economy.

The authors

choose

to proceed

with

Hamilton's

filtration

of the

oil price

data to

generate

their

oil price

shocks.

As the paper

notes,

the

estimated

effects

of the oil

price

shock

are

small:

a

1

percent

oil

price

shock-which,

by the

definition

of

the

variable,

is

expected

to lead

to

a

fairly

persistent

change

in the actual

level

of

oil prices-leads,

in figure

4,

only

to

a 0.02

percent

response

of the price

level and

a

0.025 percent

output

response

at

the peak

of

the

responses.

This

is the size

of the

pure

supply-side

effect on GDP

that

one would expect

if

oil-related energy

inputs

had a 2

percent

factor

share,

and

most economists

would

expect

estimated reduced-form

ef-

fects of

oil

price

increases

to

be

larger

than that.

(This

assumes

that

domestic

oil

is treated

correctly

as a

primary

input

and that

imports

of

foreign

oil

are

treated

correctly

as

intermediate inputs

in

GDP account-

ing, a

perhaps

dubious

assumption.)

It would be useful

in assessing

these

results to

know

both

the

response

of the

oil

price

level,

as

opposed

to the

filtered

variable,

to this

shock

and the

size of a one standard

deviation shock

to the

filtered

oil

price

measure.

Furthermore,

though

taken

from different

models,

both the first

row

of table

1

and the error

bands

in

the

bottom

row of

figure

2 show

that

the

responses

of the variables

to an oil

shock could

easily

be zero

and

yet

still consistent

with

the

data;

one-standard-error

bands

about

the

responses

barely

clear

zero.

It

is

true

that

table

1

shows that the

differ-

ence

in the

response

of

the

economy

in the

case where

monetary policy

responds

according

to

historical

norms and the

case where

it

pegs

the

interest

rate

is

fairly

sharply

defined

by

the

data

and

is in

the

direction

expected

by

the

authors.

But

since the

oil shock

itself

has turned out

to

be

something

of a

will-o'-the-wisp,

the

idea that

economists'

intuitive

knowledge

of

the size

and

nature

of

oil shocks would

help

with identi-

fication

ends

up

not

having

contributed

much.

Ben

S. Bernanke,

Mark

Gertler,

and Mark Watson

145

The

paper

also

shows

some

results

for "output"

and

"commodity

price"

shocks.

These

are derived

from

the

statistical

model

and

are

harder

to interpret

than

oil shocks.

The

model

gives

them

no

interpre-

tation,

except

that

they

are different

from and

independent

of

monetary

policy

shocks. But

while

these model-based

shocks

probably

mix con-

ceptually

distinct

non-monetary

policy

influences

on the

economy,

they

do

have the advantage

of

having

large

effects

and accounting

for

much

of the

observed

variance

in

the

data.

It is

encouraging

to

see in

figures

8

and

9 that the

effects

of

systematic

monetary policy

as

measured

with

the

oil shocks

seem to

be confirmed

with the

output

and

price

shocks,

but it is

disappointing

that all

of the careful analysis

of

robustness

and

statistical strength

centers

on

the less

sharply

defined

oil

price

shocks.

The

authors

point

out that previous

experiments

with analyzing

the

effects

of

systematic

changes

in

monetary

policy

in

identified VAR

models

have stuck to

replacing

the estimated

policy

rule in

the

model

with

something

else.

This kind

of

exercise

implicitly

assumes

that

in

forming expectations

of

future policy

actions, private

agents

treat

all

deviations

of

policy

variables

from their historical

patterns

of

behavior

as

unsystematic

deviations

from the

historical

policy

rule.

The

Lucas

critique

warns

that this

can

lead

to error.

My

own view

of the

Lucas critique

is

that

it

explains

that it is

always

a mistake to

imagine

that one

can

implement

changes

in

policy

that

have probability

zero

according

to the

model

of

policy

underlying

pri-

vate

sector

behavior.

The

implication

is

that

if

one

can

contemplate

changing

the

coefficients

of the

"rule,"

or "reaction

function,"

those

coefficients

should

have

been

modeled

as

stochastic

in the first

place.

There

is an internal

contradiction

in

pretending

that one

can

change

the

coefficients,

even

though

the

public

is modeled

as

absolutely

certain

that

they

can

never

change.

While

this

point

is

correct

in

principle,

it

is difficult

to

implement

in

practice.

Especially

for

policy

changes

quite

different from

any

that

have

been observed

historically,

estimation

of

an

appropriate

stochastic

model

that allows

for

such

changes

will be difficult

and

may

need

to

rely

heavily

on

guesswork

and

a

priori

knowledge.

It is therefore

a

good

idea,

where

possible,

to focus

attention

on

policy

changes

that

are not

too

dramatic,

which can

reasonably

be modeled

as

sequences

of random

disturbances

to the

policy

behavior

that

is

explicit

in

the

model.

This

applies

even

when one

is

generating

variations

in

policy

by

changing

146

Brookings

Papers

on

Economic Activity, 1:1997

coefficients that in the

model

are treated

as

nonstochastic. The changes

in coefficients are best

chosen so as to correspond

to not too dramatic

sequences of shocks to

the model's original policy

rule.

The type of rule change

studied

in this

paper-a

shift to an exoge-

nously fixed funds rate

from a historical

policy

that, by contrast, made

the funds rate react very sharply

to

inflationary

disturbances-is

dra-

matic.

As is

made clear

in the recent

literature on

the interaction of

monetary

and

fiscal

policy,

in

particular,

the

seminal

paper by

Eric

Leeper, a fixed interest

rate as

policy

rule

(contrary

to some discussions

elsewhere

in the literature)

is consistent

with a

uniquely determined

price

level.

'

However,

this

is true

only

if the

fixed interest rate rule is

accompanied by

an appropriate

fiscal

policy,

and the

appropriate

fiscal

policy

in this case is

quite

different

from that

consistent with a

deter-

minate

price

level in

the context of an

"anti-inflationary" monetary

policy.

Since

in this authors'

model

fiscal

policy

has

to

be

thought

of

as

wrapped

into the

"non-monetary policy"

sector,

one

would

expect

to find that

changing

the

monetary policy

rule alone

to

a

fixed interest

rate form would imply

unsustainably explosive

behavior

of

prices; and

indeed, figures 4, 5,

8,

and 9

show that

this is

exactly

what

emerges.

Private agents are

likely

to

recognize

that such a shift

in

the

monetary

policy

rule

is

unsustainable

and

therefore

to

expect

it to

end,

or to be

followed

by

a shift in

fiscal

policy.

This

makes

interpreting

the

effects

of the authors'

exercise

rather difficult.

Their

paper

in

places

reads as

if

a different

monetary policy might actually

have

eliminated the

output

effects

of oil

price

or

even

output

shocks. But since the alternate mon-

etary policy

considered

is

not

sustainable,

this

interpretation

does not

seem

to me

correct.

The simulations

suggest

instead

only

that

by delay-

ing or dampening an

interest rate

response

to

inflationary pressures,

the

monetary authority

can

trade

delay

or

dampening

of the

output

effects

for increased

inflationary

effects. It would

also

have

been

interesting

to

see

an

analysis

of effects

of

less

extreme

shifts in the

policy

rule that

would

have

been sustainable;

for

example,

smaller or

slower,

rather

than

zero,

interest

rate

responses.

The authors

attempt

to

respond

to

the

Lucas

critique by building

into

the model

one

particular

form of

endogenous

adjustment

of

private

sector

expectations

to

the

change

in

policy

rule.

They impose

the the-

1. Leeper (1991).

Ben S. Bernanke, Mark Gertler,

and

Mark

Watson

147

oretical term structure

relationships

between

the

federal

funds rate,

another short rate,

and a long

rate. Then

they

attribute to those private

agents doing

interest rate

arbitrage perfect

foresight

of

the new policy

fixing the federal funds

rate. It

is

apparent

from the figures that this

modification of the model

does

nothing

to

correct

the

fundamental prob-

lem that the change

in policy rule

is

unsustainable.

Indeed, one might

think that the sector

most likely

to realize

that fixing the federal funds

rate

is

not a sustainable

policy,

in the absence

of

a

change

in

fiscal

policy,

is the

bond market.

Requiring

that the bond

market,

but

no

one

else,

treat the

policy

as

firmly

in

place

forever

therefore seems

exactly

backward

from what might

be

plausible.

Furthermore,

this

adjustment

to the model

is not

in

fact

very large,

as

is

made

clear

by

the closeness

of

the

simulation

paths

for

many

variables in cases

where this

adjust-

ment is

imposed

and

in

those

where

it is not. The

estimated

statistical

model

already captures

the

strong tendency

of

the

federal

funds rate

and

other short rates

to move

together-a

relation

not

very different

from

the theoretical

term

structure relationship.

And the connection

of

long rates to short

rates, although

it

differs

more between

simulations,

appears

not to be

of

great

importance

for

predicting

the

effects

of shocks

on

prices

and

output.

Thus the exercise

undertaken

here

is a

step

toward

modeling private

sector learning

behavior that

might,

in

principle,

be

useful. But because

the

term

structure

relationships

are

simple

and well

approximated

in

the

original

estimated

model,

it

does

not

seem to me

likely

that

this

partic-

ular

aspect

of

private

sector

expectations

is

of

central

importance

in

this

endeavor.

The entire identified

VAR

literature

on

the

effects of

monetary policy

runs

the risk of overestimating

the

real effects of

monetary policy.

It is

not hard to

construct

a stochastic

equilibrium

model in

which

monetary

policy

is

neutral

and

certain

types

of

technology

shocks

raise

real in-

terest rates

and,

later,

lower real

output.

The essential

ingredients

are

conventional Solow-residual

technology

shocks and

increasing

costs in

the

investment

goods

industry (or

within-firm

adjustment

costs to

in-

vestment).

If

the

monetary authority

did not react

to

such

shocks, they

would be a

source

of movements

of interest rates and

output

in

opposite

directions that

was not related

to

price

behavior or to

money

stock

behavior. One

might

think of the identified

VAR literature on the effects

of

monetary policy

as a search

for restrictions

on

a macroeconomic

148 Brookings Papers

on Economic

Activity, 1:1997

time-series

model in which

some shock,

labeled "monetary

policy"

and orthogonal

to other shocks,

moves

interest

rates up, money

down,

output down,

and prices down,

with possible

delays

in

all these

effects

except the interest

rate movement.

If

the data

are generated

by a model

in which there are

real shocks

connecting

real

rates and future

output

movements,

as

I

suggest, this

identified

VAR research strategy

can

easily end up

confounding

the real

shocks with

monetary policy.

The

variety of real

effects found

in

this

literature, and

the

tendency

of real

effects to

be smaller in

models estimated

for

countries other

than the

United

States,

gives

me

genuine

concern that

this

may

have

happened.

Let me

conclude by saying

again that,

despite

the

skeptical

tone

of

my comments,

I

find this paper

useful evidence

on the

effects

of sys-

tematic

changes in monetary

policy that,

on the

whole, does

weigh

in

favor of those effects being

substantial.

It

is

quite unlikely

that mone-

tary policy

could come

close to

eliminating

the

output

effects

of oil,

"commodity price,"

or

"output" shocks,

despite

the

authors'

apparent

evidence to

the

contrary.

This

strong

conclusion

rests

on the their use

of

an unsustainable

policy

as the counterfactual alternative. But very

substantial delay

or

smoothing

of the

output

effects

via

monetary policy,

at the expense

of more inflation, probably

would

be

possible.

Benjamin

M. Friedman:

This

paper by

Bernanke, Gertler,

and Watson

is a

highly

useful

contribution

to the

empirical

literature of

monetary

policy,

both for its methodological approach

and

for

some of its

specific

findings.

I

suspect

that

it,

like the earlier

paper by

Sims

and

Zha

on

which it

draws,

will

fruitfully spur

further research

following

this kind

of

empirical

strategy.

Indeed,

as

I

suggest

below,

this

way

of

thinking

about

how

monetary policy

affects

the

economy

has at

least one

poten-

tial

application

that

may

help

to inform an issue

of

very great

impor-

tance for the

practical

conduct

of

monetary

policy,

both

in the United

States and

elsewhere.

The best way to place

in context

the

empirical

strategy

taken

by

this

paper

is to recall the

parallel

distinctions,

between what is

systematic

and what

is

unsystematic

and

between what

is

anticipated

and

what is

unanticipated,

that

have stood

behind much

of the literature of

monetary

policy

from the

past

two decades.

At the theoretical

level,

the

argument

made

by

Robert

Lucas,

Thomas

Sargent

and

Neil

Wallace,

and

others

Ben S. Bernanke, Mark Gertler, and

Mark

Watson 149

was that the

only monetary

policy actions

that have real

effects are

those that are

unanticipated.

As is

now

well

understood,

this proposi-

tion rests

on a variety

of

assumptions-for

example, perfect

competi-

tion and perfectly

flexible wages

and

prices-that

few actual economies

of practical

interest

satisfy.

Nevertheless,

because achieving analytical

precision about

the failure

of those

assumptions and

about the macro-

economic consequences

of that failure

is

highly problematic

(it

is

dif-

ficult

to

spell

out precisely

how

competition

is

imperfect

and

why wages

and

prices

are

sticky),

the

presumption

that

only unanticipated

monetary

policy

actions have real

effects

has continued to

underlie-sometimes

explicitly

but nowadays

more often

implicitly-much

of

modern

re-

search

in

the field.

Further,

as the standard

assumption

of rational

expectations

is

usually applied,

any part

of the conduct

of

monetary

policy

that is systematic (for

example,

the central bank's

always raising

interest rates

following

a

decline

in unemployment

or a

surge in infla-

tion)

is

assumed to

be

anticipated,

and

so

in this line

of

thinking

it is

also assumed to be

without real effects.

At the

empirical level,

the

parallel argument

has been

that even

if

such systematic

monetary

policy

actions did affect

real

economic activ-

ity, it would

be impossible

to

distinguish

those effects from the inde-

pendent

consequences

of

the

events

to which

monetary policy

was

reacting. (For

example,

to the extent

that the central

bank

simply moves

interest rates

in

response

to prior

observed

inflation,

any subsequent

effect

on

real

output

could

just

as

well be

attributed

to

the inflation

itself as

to the

consequent

movement

in interest

rates.)

Hence the

appeal

of the vector

autoregression

approach

in this context is

that it focuses

only

on those

monetary policy

actions

determined to be

unsystematic,

in the sense

that the

VAR cannot

explain

them in

terms

of

prior

move-

ments

in other variables. One

danger

of

this

approach

is

that a

VAR

that includes too much

information

may

overexplain

the

movement

of

monetary policy

in terms

of

prior

movements in other variables. Such

a

VAR will

erroneously

shrink the

remaining component,

which is

taken

to be

unsystematic

and

therefore

also

unanticipated,

to

the

point

that

it

then

appears

to have

only

trivial

economic

consequences.

But the

main

point

is that

the

empirical

rationale

for

assessing

the effects of

monetary

policy by

looking only

at its

unsystematic

variation,

which

continues

to be

in

widespread use,

resonates

closely

with the

now

outdated

the-

150

Brookings

Papers

on

Economic

Activity,

1:1997

oretical

presumption that, at least for

purposes

of

effects on

real vari-

ables, only

unanticipated

policy actions matter.

There is

an inherent

congruence

between the two lines

of

thinking.

The principal

thrust of the

approach taken

by Bernanke,

Gertler, and

Watson is to

sever that connection

by

designing a way to

use the em-

pirical VAR

methodology

to

investigate specific

aspects of

systematic

monetary

policy.

To

be sure, the

paper

simply presumes,

rather than

shows, that systematic and therefore

anticipated

monetary

policy ac-

tions can have real effects. But for

readers

who

accept

that there are

reasons why this

may be

so

and

who

do not

require that the

empirical

model used

to

investigate

these effects be

explicitly

tied

to a

theoretical

model

detailing

how

they

come

about,

the

resulting

advance

is

clear.

And

indeed,

the authors find that

the

specific

aspect

of

systematic mon-

etary policy

on which

they

choose to focus-the central

bank's

response

to oil

price

shocks

and to the

consequences

of

those

shocks for

prices

and

output-does

have sizable real effects. This

finding

is

both inter-

esting

and

important.

(To

be

clear,

the within

month

response

of mon-

etary policy to

an oil

price

shock would

be

unanticipated

and

therefore

presumed

to have real

effects,

even in

a

Lucas-style

model.

Although

the

paper

is not

specific

on this

distinction,

I

assume

that the

bulk

of

the real effects that the authors attribute to the

monetary

policy response

to oil

price

shocks results from movement

in

the

policy

variable

occur-

ring after the

month

in which the oil

price

moves.)

As

indicated at the

outset,

I

suspect

that this

methodology

has an

immediate

application

of

potentially great

importance.

A

question

that

has

rightly

attracted

widespread

attention,

among

industrial as well

as

developing

countries,

is how

price

inflation

affects

a

country's ability

to

maintain real

economic

growth.

Evidence

shows

that above

some

modest

level

(the

high

single-digit range),

inflation does

reduce the

average pace

of real

growth

over time. A familiar

view,

however,

is

that

inflation

negatively

affects real

growth

not

because

inflation, per

se,

matters in this

context,

but because the central bank

acts to

resist

inflation; and

in a

world

in which the

Lucas-Sargent-Wallace

assump-

tions do not

obtain,

it can

only

do

so

by

slowing

("sacrificing")

real

output.

The

methodology

used

in this

paper

seems

potentially

able

to

address

this

question

too.

If

so,

the

findings

would be

very

valuable.

Although

both the

methodology

and the

findings

of

Bernanke,

Gertler,

and Watson's

paper

are

highly

useful,

three

specific

aspects

Ben S. Bernanke, Mark Gertler, and

Mark

Watson 151

give cause

for

reservation. First,

as

they

are at some

pains to emphasize,

there is substantial evidence

of

instability

in their

results

across the

three

decades

of their

sample.

In

particular,

as

figure 10 clearly shows,

the

"systematic" response

of

monetary policy

to

oil

price shocks in the

Volcker

period

was

far

greater

than either earlier or

later.

A

question that this instability immediately

raises is

whether it is

reasonable to view the more energetic anti-inflationary monetary policy

of

the

Volcker

era

exclusively

as

a

response

to an oil

price

shock.

I

believe

that the Federal

Reserve

System

under

Paul Volcker

adopted a

policy broadly

aimed at

reducing

the

U.S.

inflation

rate, and that the

rise in oil prices in 1979 and 1980

was

only

one

element

in

the inflation

process against which it directed

its

policy.

The

results plotted in the

middle right-hand panel

in

figure 6, showing

that the

simulated

response

to the historical oil shock

accounts for

only

a

small

part

of the

increase

in the federal funds rate during 1981-82,

are

certainly consistent with

this

view. Because

of the

post

hoc

ergo propter

hoc character

of VAR

analysis,

the

Bernanke-Gertler-Watson

paper may

attribute to

the

spe-

cific

response (here

and

in other

subperiods)

of

monetary policy

to oil

price shocks what was actually the

more

general conduct of monetary

policy,

based

on other considerations.

The

findings

of

subsample instability

also

highlight

the

difficulty

of

identifying

what

"systematic" policy

means in

the

first

place.

For

purely empirical purposes

of

extracting impulse responses

and

variance

decomparisons

from

past data, systematic simply

means whatever

hap-

pened

on

average

across the

arbitrarily

chosen

sample

under

study.

But

as is

the case

in this

paper,

researchers often seek to connect

this

purely

empirical

notion

of

systematic

behavior

with the

concept

of

policy

"rules,"

so as

to go

on to draw

inferences about the

consequences

of

the

central

bank

following

one rule rather than another. As a

number

of

people (Sims,

John

Taylor,

and

I, among many others)

have

argued

in

one context

or

another,

it

is

not

clear that in

practical settings

the

central bank is ever

following

a

rule,

in

the crucial dual

sense that

its

actions are not

only systematic

but also

perceived

to

be

so

and therefore

properly anticipated by

the relevant

public.

The fact that

estimating

the

authors'

VAR over the 1976-85

sample

delivers the

federal

funds rate

response

shown in the

right-hand panel

of the

third

row

in

figure

10

does not

necessarily

make

this

response

a

characterization of

systematic

monetary policy

in

any

substantive

sense.

152

Brookings Papers on

Economic

Activity,

1:1997

A

second

set of

reservations

stems from

the

authors' use of

oil price

shocks as the

principal

empirical

vehicle

for their

study of

systematic

monetary

policy. To put

it

bluntly, does the

Hamilton

idea

really make

sense? For

example,

should one

really

think of the

1957-58

recession

in

the United

States as

a ripple

from the

1956

Suez affair?

To take

Hamilton's

idea

seriously would

require a

major

rethinking of

most of

post-World

War

II

U.S.

business

cycle

history-which

clearly

has not

happened in

the decade

and a half

since

Hamilton's

intriguing paper

appeared. The

authors

of the present

paper

are

perhaps more

secure in

that

the role

of oil

prices is more

plausible

in

at

least two,

possibly

three, of the

five recessions covered in their

sample, which

mostly

postdates

Hamilton's. Even

so,

I

suspect

that their

difficulty

in

finding

a

measure

of

oil

price

shocks that

satisfactorily

fits

the oil

facts

to the

macroeconomic data

is a

warning

of

just

this

problem.

Finally, several

aspects

of the authors'

treatment of

interest

rates

also bear

closer attention.

The

assumption

that interest rate

movements

are a

sufficient statistic

for the channels

by

which

monetary

policy

affects macroeconomic

activity

is, by

itself,

not

unusual.

Indeed,

the

authors

may

well

overemphasize

its

limitations.

Costs of

financing (in-

cluding

opportunity

costs)

are

an

important

factor in

many kinds

of

spending

decisions,

and

for this

purpose

interest

rate

fluctuations

may

also

plausibly

stand in for at least

part

of the

relevant

movement

in

either

exchange

rates

or broader asset

prices.

While

the

strong

rejection

of the restriction

excluding

the federal funds rate from

the

output

equa-

tion is somewhat

surprising,

the

authors

are

presumably

correct

that the

practical effects of

imposing

this

restriction are small.

Further,

it

is

my

conjecture that if a stock

price

index were

included

in the

VAR,

the

data

would

accept

this restriction.

(Because

the

analysis

in

this

paper

depends

so

crucially

on the role of short- and

long-term

interest

rates,

however,

there is

probably

much to be learned from

examining

the

coefficients

of

these

interest rates in the

output

equation,

as well

as the

impulse

responses

relating output

to the

independent

components

of

the

two

interest rates. It would therefore

be useful

to show

explicitly

these

key

elements

of

the

analysis.)

The

potential

problem,

however,

is the

strong

implied rejection

of

the

expectations

hypothesis

of the term

structure

of

interest

rates,

which

the authors

use as the

organizing

principle

for

this

part

of their

model.

Normally,

within this

framework,

the

"term

premium"

included

in

any

Ben S. Bernanke,

Mark

Gertler,

and Mark

Watson

153

specific interest

rate is a substantive reflection of

borrowers' and lend-

ers'

attitudes

toward

such features as the

risk

and

liquidity

of

the un-

derlying debt

instrument. But

in this

paper,

the term

premium simply

serves to undo

the behavior that

the

built-in

expectations hypothesis

implies that

interest rates should be

following (see,

for

example, fig-

ure

3).

Moreover,

the results

plotted

in

figure

4 for

prices and

the

long-

term rate are

dramatically

at variance with

standard notions of how

inflation

expectations affect

nominal

interest rates. In this

experiment,

not surprisingly,

moving

from

the

base simulation to

either the

Sims-

Zha simulation

or the anticipated policy simulation results in

far higher

prices and hence

much greater

inflation. But in the

Sims-Zha

simulation

the long-term interest rate

is

uniformly

below its

level in

the base

simulation, and in

the

anticipated policy

simulation it

even

declines

absolutely. So much

for the notion that investors

rationally anticipate

the

consequences

of

monetary policy

for future

inflation

and

incorporate

the

resulting inflation

expectations

into current

bond

prices!

These three sets of

reservations

notwithstanding,

I

applaud the

broader

methodological direction

taken

by

Bernanke, Gertler, and Wat-

son

and retain

my

sense that

their

finding

of

quantitatively significant

effects from

systematic

monetary policy

is

both correct

and

important.

General discussion:

Participants generally

accepted

the

authors'

con-

clusion that the

output

declines

following

oil

price

shocks

had come

mainly

from the

responses

of

monetary

policy

to the

shocks. Several

also

discussed

the

plausible magnitude

of oil shock

effects themselves.

One

issue

was

how

much

an oil

price

increase,

or a

decrease in oil

supply, should

affect

potential output;

a

second

was

whether oil

price

increases

reduce

demand and

lead to

lower

levels of

utilization of

pro-

ductive

capacity.

Robert Hall observed

that,

for

infinitesimal

changes

in

oil

prices,

the

ability

of the

United

States

to

produce

should not be

impaired by

a

rise

in the

price

of

imported

oil,

even if it

reduces

oil

use;

the

derivative

of

real GDP

with

respect

to the

price

of

oil is

zero

no matter how

large

the

adjustment,

with Division

GDP.

However,

he

and

William

Nordhaus

agreed

there could be

effects on

potential

GDP

as

the

equilibrium

supply

of

domestic

factors

adjusted

to

the

change

in

oil

prices.

George Perry

added

that

some

estimates from earlier

studies,

such as a

reduction

of several

percentage points

of

GDP

from

OPEC

1,

were too

large

to

be

viewed

as a

supply-side

effect.

However, taking

154

Brookings

Papers

on

Economic

Activity,

1:1997

into account the effect of an oil price increase on aggregate demand,

where

the

price increase could

be

analyzed approximately

like

an increase in

excise taxes with high-saving foreigners getting

the

revenue, a large short-

run impact on GDP was believable.

He

added that the allocation of such

an

impact

between a "fiscal" and a

monetary

effect

would depend, some-

what

arbitrarily,

on how baseline

monetary policy

was

defined.

Nordhaus raised several issues

about the

appropriateness of the var-

ious measures of

oil

shocks used by

the authors. He

suggested that

almost any theory,

whether

Perry's

that the short-run

impact of in-

creases could

be

regarded

as

a

tax

paid

to

foreigners

or

Sims's

that it

should

be

treated simply

as

an increase in

input prices, should lead to

some measure involving

oil

purchases

relative to the

size of

the econ-

omy.

This

scaling

makes an enormous difference.

For

the last

three oil

shocks in

the sample,

he calculated

the

increased

costs

of

imported oil,

with

quantities fixed,

were 1.8

percent

of

GDP

in

1973,

1.0

percent

of

GDP

in

1979,

and 0.2

percent

of GDP

in

1990.

Using

this

measure

would

preserve

the

peaks

of

the

Hamilton

series,

but

the shocks

would

be

progressively

smaller. Nordhaus

also

noted that the

paper ignores

the

negative

oil shock of

1986,

when the

price

decline

corresponded

to

a

negative

shock of 0.5

percent

of GDP. He reasoned that

the

failure

to

scale the

shocks, along

with the fact that the

positive

shocks of

1986 and

1990

were

quickly reversed, may explain why

the

responses

in

the two

subperiods

look so different

in the

authors'

analysis.

William

Brainard

agreed

with Nordhaus's

argument

for

scaling

the shocks

and

added that it

might

be

useful

to construct

a similar measure

indicating

the

magnitude

of

the redistribution

between

domestic

producers

and

consumers.

Robert Shiller observed

that the

stochastic

properties

of the oil

price

series

seemed

to have

changed

after the

Organization

of

Petroleum

Exporting Countries

broke

up

in 1986.

Before

that,

the

oil

price

was

a

series

of

plateaus separated by

sudden

jumps,

so that

changes

seem to

have

a lot of information.

But

afterward,

the

oil

price

looks like a

mean-

reverting process,

so

the

movements

have less

information. He

reasoned

that the

public may

realize

this

difference,

which

would

explain why

oil

price changes

are no

longer big

news.

Reflecting

on

the

widespread

concerns about

oil in

the

1970s and

1980s,

Shiller

suggested

that

the

long

view

is

important

in

economics

and

the best

way

to

deal with an

anomaly

is

to

wait it

out

until it

disappears.

He

suggested

that

may

have

happened

with

oil.

Ben

S. Bernanke,

Mark

Gertler,

and Mark

Watson

155

References

Andersen, Leonall C., and

Jerry L. Jordan. 1968. "Monetary and

Fiscal Ac-

tions: A Test of Their

Relative Importance in Economic

Stabilization."

Federal Reserve Bank of St.

Louis

Review

(November): 11-24.

Bernanke, Ben S., and Alan

S. Blinder. 1992. "The Federal Funds

Rate and

the

Channels of Monetary

Transmission."

American Economic

Review

82(4): 901-21.

Bernanke,

Ben

S.,

and

Ilian

Mihov.

1995.

"Measuring Monetary

Policy."

Working Paper 5145.

Cambridge,

Mass.: National

Bureau of

Economic

Research (June).

Blanchard, Olivier J. 1984.

"The Lucas

Critique

and the

Volcker

Deflation."

American

Economic

Review, Papers

and

Proceedings 74(2):

211-15.

Chow, Gregory,

and An-loh

Lin.

1971.

"Best

Linear

Unbiased

Interpolation,

Distribution,

and

Extrapolation

of

Time Series

by

Review

of Economics and Statistics

53(4):

372-5.

Christiano,

Lawrence

J.,

and Martin Eichenbaum.

1992. "Identification and

the

Liquidity Effect

of a

Monetary Policy

Shock." In

Political

Economy,

Growth, and

Business

Cycles,

edited

by

Alex

Cukierman,

Zvi

Hercovitz,

and Leonardo Leiderman. MIT Press.

Cochrane,

John H.

1996.

"What Do the VARs

Mean?

Measuring

the

Output

Effects

of

Monetary

Policy" Unpublished paper, University of

Chicago

(August).

Darby,

Michael R.

1982.

"The Price of Oil and World

Inflation and

Reces-

sion.

"

American Economic

Review

72(4):

738-51.

Dotsey, Michael,

and Max Reid. 1992.

"Oil

Shocks, Monetary

Policy,

and

Economic

Activity."

Federal Reserve

Bank

of

Richmond

Economic Review

78(4):

14-27.

Ferderer, J.

Peter.

1996.

"Oil

Price

Volatility

and the

Macroeconomy." Jour-

nal

of

Macroeconomics

18(1):

1-26.

Friedman, Benjamin M.,

and

Kenneth N. Kuttner.

1996.

"A

Price

Target

for

U.S.

Monetary Policy?

Lessons

from the

Experience

with

Money

Growth

Targets." BPEA, 1:1996,

77-125.

Friedman, Milton,

and Anna Jacobson Schwartz. 1963.

A

Monetary

History

of

the United

States,

1867-1960.

Princeton

University

Press.

Fuhrer, Jeffrey C. 1996.

"Monetary Policy

Shifts and

Long-Term

Interest

Rates."

Quarterly

Journal

of

Economics

111(4):

1183-209.

Hamilton,

James D. 1983.

"Oil and the

Macroeconomy

Since World War

II."

Journal

of

Political

Economy 91(2):

228-48.

.

1996a.

"Analysis

of

the Transmission of Oil Price Shocks

Through

the

Macroeconomy."

Unpublished paper. University

of

California,

San

Diego (August).

156 Brookin2s Pavers on

Economic Activity.

1:1997

.

1996b.

"This

is What

Happened

to the

Oil Price-Macroeconomy

Relationship."

Journal of

Monetary Economics

38(2): 215-20.

Hansen, Bruce

E.

1997. "Approximate

Asymptotic

P Values for

Structural-

Change

Tests."

Journal of

Business

and Economic

Statistics

15(1): 60-67.

Hooker, Mark

A. 1996a.

"Exploring

the Robustness

of

the

Oil Price-

Macroeconomy

Relationship:

Empirical Specifications

and

the

Role of Mon-

etary

Policy." Unpublished

paper. Wellesley

College (September).

.

1996b.

"What Happened

to

the Oil

Price-Macroeconomy

Relation-

ship?" Journal

of Monetary

Economics 38(2):

195-213.

Hoover, Kevin

D., and Stephen

J.

Perez. 1994. "Post

Hoc

Ergo

Propter Hoc

Once More:

An

Evaluation

of

'Does

Monetary Policy

Matter?' in

the Spirit

of James Tobin." Journal

of Monetary

Economics 34(1):

47-73.

Kim,

In-Moo, and Prakash

Loungani.

1992. "The

Role of

Energy

in

Real

Business Cycle

Models." Journal

of

Monetary Economics 29(2):

173-89.

Kim, Soyoung.

1995.

"Does

Monetary Policy

Matter

in the

G-6

Countries?

Using

Common Identifying

Assumptions

About

Monetary

Policy

Across

Countries."

Unpublished paper.

Yale

University (January).

Lee, Kiseok,

Shawn

Ni,

and Ronald

A.

Ratti.

1995.

"Oil

Shocks and the

Macroeconomy:

The

Role of

Price

Variability." Energy

Journal 16(4):

39-56.

Leeper,

Eric M.

1991. "Equilibria

Under

'Active' and 'Passive'

Monetary and

Fiscal Policies."

Journal of

Monetary

Economics

27(February):

129-47.

Leeper,

Eric

M., Christopher

A.

Sims,

and Tao

Zha. 1996. "What Does

Monetary

Policy

Do?"

BPEA, 2:1996,

1-63.

Lucas,

Robert

E.

Jr.

1976.

"Econometric

Policy

Evaluation:

A

Critique."

Carnegie-Rochester

Conference

Series

on Public

Policy

1:

19-46.

Mork,

Knut Anton.

1989. "Oil

and

the

Macroeconomy

When Prices Go

Up

and

Down:

An

Extension

of

Hamilton's

Results."

Journal

of

Political Econ-

omy 97(3):

740-44.

Quandt,

Richard E. 1960.

"Tests of

the

Hypothesis

that

a

Linear

Regression

System Obeys

Two

Separate

Regimes."

Journal

of

the

American Statistical

Association 55(290):

324-30.

Romer,

Christina D.,

and

David

H. Romer. 1989.

"Does

Monetary Policy

Matter?

A

New Test

in the

Spirit

of

Friedman and Schwartz." In NBER

Macroeconomics

Annual

1989,

edited

by

Olivier Jean Blanchard and

Stanley

Fischer.

MIT

Press.

Rotemberg,

Julio J.,

and

Michael

Woodford. 1996.

"Imperfect

Competition

and

the Effects

of

Energy

Price

Increases

on

Economic

Activity."

Journal

of Money,

Credit, and

Banking 28(4, part 1):

549-77.

Shapiro,

Matthew

D.

1994.

"Federal Reserve

Policy:

Cause and Effect." In

Monetary

Policy,

edited

by

N.

Gregory

Mankiw.

University

of

Chicago

Press.

Ben S. Bernanke, Mark

Gertler, and

Mark

Watson

157

Shiller, Robert J., John

Y.

Campbell,

and Kermit

L.

Schoenholtz.

1983. "For-

ward Rates and Future Policy: Interpreting

The Term

Structure of Interest

Rates." BPEA, 1:1983, 173-223.

Sims, Christopher

A.

1980.

"Macroeconomics

and

Reality." Econometrica

48(1):

1-48.

. 1986. "Are Forecasting

Models Usable

for

Policy Analysis?" Federal

Reserve Bank of Minneapolis Quarterly

Review

10(1): 2-16.

.

1992. "Interpreting

the Macroeconomic Time

Series Facts: The Ef-

fects of Monetary Policy." European

Economic Review

36(5): 975-1011.

Sims, Christopher A.,

and

Tao Zha. 1995.

"Does

Monetary Policy

Generate

Recessions?" Unpublished paper.

Yale

University

and Federal

Reserve Bank

of Atlanta

(December).

Strongin,

Steven. 1995.

"The Identification

of

Monetary Policy

Disturbances:

Explaining the Liquidity

Puzzle." Journal

of Monetary

Economics

35(3):

463-97.

Taylor,

John

B.

1993.

Macroeconomic

Policy

in

a

World

Economy:

From

Econometric Design to Practical Operation.

W.W.

Norton.

Tobin,

James. 1980. "Stabilization

Policy

Ten Years After."

BPEA, 1:1980,

19-71.

West,

Kenneth D. 1993.

"An

Aggregate Demand-Aggregate Supply Analysis

of

Japanese Monetary Policy,

1973-1990."

In

Japanese Monetary Policy,

edited

by

Kenneth J.

Singleton. University

of

Chicago

Press.