SfcCffflte,
orTti
D£WEY
HB31
.M415
working
department
of economics
MONEY, OUTPUT, AND PRICES:
EVIDENCE FROM A NEW MONETARY AGGREGATE
Julio J. Rotemberg
John C. Driscoll
James M. Poterba
No.
585
May 1991
Rev. July 1991
massachusetts
institute of
technology
50 memorial drive
Cambridge, mass. 02139
tttj
•
!
MONEY, OUTPUT, AND PRICES:
EVIDENCE FROM A NEW MONETARY AGGREGATE
Julio J. Rotemberg
John C. Driscoll
James M. Poterba
May 1991
No.
585
Rev. July 1991
NOV
J
4 1991
MONEY, OUTPUT, AND PRICES:
EVIDENCE FROM A NEW MONETARY AGGREGATE
Julio J.
Rotemberg
MJT&NBER
John C. Driscoil
Harvard University
James M. Poterba
MIT&NBER
May
1991
Revised July 1991
ABSTRACT
This paper develops a new utility-based monetary aggregate which we label the currencyThis aggregate equals the stock of currency that would be required for
equivalent aggregate.
households to obtain the same liquidity services that they get from their entire collection of
monetary
assets.
It
equals
rb,t
'
where
rb
t
is
the rate of return
on a zero
and n\, respectively
the ability of the
and M2, and other indicators of
liquidity asset, while
represent the yield and stock of monetary asset
i
at
time
t.
r- ,
We compare
aggregate and conventional aggregates, such as Ml
monetary policy to forecast real activity. The CE aggregate has more predictive power for
output and prices than standard aggregates, and the time path of the estimated output
response is more consistent with broad classes of theoretical models.
new
We are
Michael
Woodford and seminar participants at the Federal Reserve Board and the NBER 1991
Summer Institute for suggestions, as well as to William Barnett, Anil Kashyap, Richard
Porter, Brian Reid and Piyu Yue for providing us with data. This research is part of the
NBER Programs in Economic Fluctuations and Financial Markets and Monetary Economics.
grateful to the National Science Foundation for research support, to
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium
Member
Libraries
http://www.archive.org/details/moneyoutputprice00rote2
1
How
monetary shocks
affect prices
and real activity are two of the central questions in
macroeconomics. The implications of various theoretical models addressing these issues
have been explored in
in
literally
what money does, there
hundreds of empirical papers.
which
studies use relatively arbitrary rules in deciding
By choosing
not.
to study
how
activity, researchers implicitly
Narrow
definitions of
liquidity services.
the monetary base, or
money, such
A
GNP
more
is.
Most previous empirical
assets are monetary, and
Ml,
made judgments about the
or
M2,
which are
affects prices
identity of
monetary
and
real
assets.
as the base, exclude a variety of assets that provide
Broader definitions, such as
with arguably quite different liquidities.
measure of
money
consensus on what
is little
Despite the substantial interest
This
M2,
is
give equal weight to a variety of assets
hardly more defensible than constructing a
by adding together the physical volume of output
attractive approach involves weighting different assets
in different industries!
by the value of the
monetary services they provide. This principle underlies Barnett's (1980) derivation of
Divisia monetary aggregates.
The continued widespread use of conventional aggregates
particularly surprising, since research has repeatedly
as
good
at predicting
GNP.
In this paper
we
currency-equivalent (CE) aggregate, which
aggregate
is
is
shown Divisia aggregates
to
be
is
at least
propose a new monetary aggregate, the
related to the Divisia aggregates.
The
CE
a time-varying weighted average of the stocks of different monetary assets, with
weights which depend on each asset's yield relative to that on a benchmark "zero liquidity"
asset.
One
it is
attractive feature of the
CE
aggregate
the stock of currency that yields the
is that it
has a straightforward interpretation:
same transactions
services as the complete
2
constellation of
monetary
Thus, currency gets a weight of one in our aggregate.
assets.
Other assets with higher rates of return get a smaller weight because their marginal liquidity
services are lower.
After deriving our aggregate and summarizing
Divisia aggregates,
We
output.
we
first test
use
it
to develop
its
relation to both conventional and
new evidence on
the correlation between
whether, in post-1959 U.S. data, there
money and
any correlation between
is
changes in various monetary aggregates, including CE, and subsequent movements in output.
We
then study whether these correlations suggest that changes in
money cause changes
in
output.
Most models where exogenous changes
this effect will
be short lived and
in
money have an
that, eventually,
effect
on output predict
that
a one percent increase in money leads to a
one percent increase in prices. Thus the impulse response function of output with respect
monetary shocks ought to exhibit pronounced, short lived
that fail to display this property, so that output
not causal.
due
to
effects. Indeed,
monetary shocks
responds permanently to them are probably
Rather, money-income correlations of this type can easily be interpreted as being
monetary endogeneity with respect to either income or other variables which
income.
to
Our evidence suggests
that shocks to the
CE
affect
aggregate are both more strongly
correlated with subsequent changes in output, and have a shorter lived "effect" on output,
than shocks to
more
traditional
monetary aggregates.
Several recent papers, for example Bernanke and Blinder (1990) and Friedman and
Kuttner (1990) have suggested that interest rates and interest rate spreads are better indicators
of monetary policy than the monetary aggregates themselves.
We
therefore compare the
3
predictive
power of
comparison
CE
is
aggregate.
these variables to the predictive
CE
CE
aggregate.
This
of particular interest because some changes in interest rates directly affect the
We
find that the correlation
and subsequent movements
the
power of the
On
aggregate.
between
interest rates,
and
in output is stronger than the correlation
interest rate spreads,
of these variables and
the other hand, the impulse responses of output to these variables do
not correspond to the pattern one would expect if the economy were responding to a
monetary disturbance.
This paper
is
divided into five sections.
monetary aggregate and compares
it
The
first
derives the currency-equivalent
with existing Divisia aggregates.
the data inputs needed to construct our aggregate, and reports
univariate time series properties.
The
causes real output, and compares
its
monetary policy.
It
third section tests
Section two describes
summary
whether the
statistics
new
on
its
aggregate Granger-
predictive performance to that of other indicators of
also tests the long run neutrality of
money with
respect to output.
Section four presents a parallel analysis of the effect of monetary shocks on prices,
comparing the findings for the currency equivalent aggregate with those for more
measures.
There
is
traditional
a brief conclusion which sketches the implications of our findings for
competing theories of the monetary transmission mechanism.
4
1.
The Currency-Equivalent Monetary Aggregate and
1.1
Its
Properties
Derivation
Some
assets can readily
be used for transactions. Individuals pay for the
these assets offer by foregoing the higher expected returns that are available
liquid assets.
account, and
liquidity that
on
other, less
Holding one dollar in currency costs more than holding one dollar in a
it
presumably generates greater liquidity services.
assuming that individuals derive
utility
from holding
We
certain assets.
NOW
formalize this idea by
Our
results could also
be
obtained by assuming that individuals and firms incur transactions costs which depend
negatively on asset-holdings and positively on the volume of transactions.
We consider an
individual
whose expected
V=E
t
lifetime utility in period
t is
given by
±VUN
(D
/-o
where E, takes expectations
as of time
the level of instantaneous utility at
and the levels of n monetary
currency holdings at
t.
We
t.
assets,
assume
is
t,
an intertemporal discount factor, and U, gives
This instantaneous
m
it .
that
We let
U
t
i
utility
range from
depends on consumption, C„
to n-1,
equals U(C„L,), which
is
and define
m<, it as
concave in both
arguments, where
Lt'f(mot ,mu ,... m„. u).
(2)
t
The
function f
Any
is
an aggregator function for liquidity services.
desirable monetary aggregate
under two assumptions about
this
must in some way correspond
to L,.
We
show
that
aggregator function, L, equals the currency equivalent (CE)
5
Our
monetary aggregate introduced by Rotemberg (1991).
aggregator function
is
homogeneous of degree one.
separable in currency and other monetary assets:
f(mot ,mu ,m^ t ,...) =
1
Second,
assets
by holding
h(mot)
sufficient currency.
might be controversial,
assumption
we assume
the
that f is additively
k(mu ,m^ t ,...)
+
it is
While the
(3)
possible to obtain any level of
inessentiality
consistent with the fact that circulating
it is
is that
2
This assumption gives a central role to currency, since
liquidity services
first
of other monetary
media of exchange
appear to predate the introduction of other liquid assets.
Our assumption
that the aggregator function f has constant returns to scale implies that
Lt =
E
fu
mu
<4)
/-o
where f
i>t
is
the partial derivative of f with respect to the quantity of monetary asset
evaluated in period
t.
i
Since (3) implies fo,i=h „
'The key part of this assumption
is
that f
be nomothetic since
it
can then be converted into a
homogeneous function by transforming U. As shown by Samuelson and Swamy(1964), homotheticity
considerably simplifies the construction of index numbers.
It is
required for Divisia aggregates to be
path independent as well as to ensure that the true economic index lies between the Paasche and the
Laspeyres index.
*We would
obtain identical results
if
we
treated currency as a collection of disparate assets
(denominations) which provide different types of liquidity. Letting cjtt denote the holding of currency
of type
j at t,
so that f
is
f would have to be given by
additively separable in a function that depends
function that depends
on the other monetary
assets.
on the various currency
assets
and a
h(mot )
The
must be equal, so the second and
This implies that h
constants.
is
all
(5)
and of
h^nio,, with respect to
m^,
higher derivatives of h with respect to m^, must be
a linear function, and that the derivatives ho and
,
f0> , are
3
Our assumptions
thus imply that currency does not lose
the quantity of currency rises.
appear, because
function
= f r^,r
0>
equality in (5) implies that the derivatives of h
first
zero.
hot mot
=
it
does not imply
U is concave in
We also assume
we
and t+1
known with
certainty at time
rbt .
not as disturbing as
total liquidity so that, indirectly, it is
focus on a particular asset of
nominal return equal
is
might
it
that overall utility is linear in currency.
that there are assets that
simplicity,
is
This implication
to provide liquidity as
its ability
4
t.
concave
Rather, the
in currency as well.
do not provide monetary
this
at first
services.
For
form whose nominal return between time
We call
In our empirical work,
this the
benchmark
we measure
rb
,
asset
and
t
let its
by the return on prime
'Interpreting traditional simple-sum monetary aggregates as equalling total liquidity services also
on f but these are much stronger than those implies by our separability
and homogeneity assumptions. In particular, the function f must be linear in all its arguments with
equal coefficients for each. But this implies that all monetary assets provide the same return.
requires linearity assumptions
'Even
if all
actual assets without liquidity services had stochastic returns, the analysis
However, the benchmark return would have to be the return on an imaginary
liquidity services and with a nonstochastic payoff in period t+1.
apply.
would
asset with
no
7
grade commercial paper.
The benchmark
5
return has the standard property that
UAC„L,) - fl.frj qSgMMAfl?
where Pt measures the price of a unit of consumption
Equation (4) says that the consumer
profile and a profile in
is indifferent
which he increases
in terms
(6)
of currency at date
t.
between his current (optimal) consumption
his holdings of the
benchmark
asset
by reducing
current consumption and uses the proceeds to increase future consumption.
Monetary
assets
do not share
their stock raises the level
utility
by
this property
of liquidity services received by the consumer.
from lowering current consumption
further liquidity services.
is offset
UL(Cv L
Substituting the expectation^ term
from
t)
f
(1
u
Thus
the loss in
by both additional future consumption and
In particular, for monetary asset
UJLC„Lt) -
*We chose
of the benchmark asset because increasing
+
i
r,,)E,^^M
(6) into (7),
we
(7)
obtain
the commercial paper rate because the market for prime grade commercial paper
We also tried the Federal
is
funds rate, which
even safer (though less widely held) instrument. The results were identical, so we only report the
results for commercial paper. We did not use Treasury bill returns because during much of the 19601980 period, these yields were below the interest rates on passbook saving accounts. Since saving
broad and the instrument
is
fairly safe.
accounts would appear to be
puzzling.
Our own
more
pertains to
useful for transactions purposes than T-bills, this behavior
tentative explanation
is
an
is
that Treasury bills provide other types of services.
They
are well understood throughout the world, they can easily be used as collateral and they can be posted
as
margin
services
in financial transactions.
we do
Because
we
not include Treasury bills in L.
our analysis as long as the
utility
function
is
think these services are different from transactions
The
existence of these services does not invalidate
separable in consumption, L, and these services.
1
8
UJU
This says that the consumer
by one
unit,
is
indifferent
his holdings
and increasing
%& U
-
c
(8)
.
between reducing his holdings of monetary asset
of the benchmark asset by (l+rit)/(l+rbl). This
change leaves the value of the individual's portfolio
at the beginning
of period t+
unaffected, so that future consumption can remain unaffected as well.
However,
portfolio change allows the individual to raise his current consumption
by
The
gain in
utility
from
his reduced holdings of
For currency,
this
this
(rbt-rit)/(l
consumption increase must therefore equal the
monetary asset
i
+rb ,).
fall in utility
from
i.
the analogue of equation (8)
is:
r
UJo,
«
WL U
ii
j*~
c
(9)
'b,t
Since
U
is
an arbitrary concave function,
normalization ensures that
equations (8) and (9)
now
total liquidity,
we
L,
is
measured in units of currency.
This
Combining
yields
fkt =
The
can normalize f so that f0jt equals one.
level of the liquidity aggregate L,
Lf =
[tLJu.
(10)
rb,t
from equation
E^m
(4) then satisfies
tfS
CE
f
.
This expression defines the currency-equivalent monetary aggregate.
(11)
9
Actually, our derivation has considered an individual's holding of monetary assets so
that L» in (11) is the level
However, because (11)
of liquidity held by an individual.
in individual asset holdings, the
to aggregate asset holdings.
sum of
the L's held
by
all
is
linear
individuals is simply (11) applied
Thus CE, provides an accurate measure of the sum of the
individual L's even if the aggregator functions f differ for different individuals.
CE
The
as currency
aggregate has the attractive property that assets which do not pay interest, such
and
travelers' checks, are
added together with weights of unity. Other assets are
added with weights between zero and one, with higher-yield assets receiving lower weights.
This makes intuitive sense since these assets must, given their higher returns, provide smaller
liquidity services.
The benchmark
asset, in particular,
provides no liquidity services.
Therefore, by the same logic, assets whose returns equal or exceed those of the benchmark
asset
do not contribute
CE
The
new
assets,
to the currency equivalent aggregate.
such as interest-bearing
shows
that
is
are introduced, they can be added to the
is
resolved by reference to their interest yield.
interesting interpretation of this aggregate is developed in Barnett (1991).
assuming
static expectations,
expenditures on liquidity services.
which
NOW accounts,
deciding whether such accounts are "liquid enough" to be
included in the monetary aggregate
One
When
aggregate also adapts easily to changes in the financial environment.
The problem of
aggregate.
6
He
CE, equals the expected present discounted value of
These expenditures in period
t
equal
E, (rb
,
-
r^m^,,
the total amount of interest foregone by holding monetary assets instead of the
'Although
many
assets
promise yields greater than those on the benchmark
necessarily evidence that liquidity services are provided
simply reflect differential risk across
assets.
by the benchmark
asset, this is
asset.
Rather,
it
not
could
10
benchmark
asset.
their level divided
1.2
Under
by
static expectations, the present
the rate of return on the
Properties of Divisia and
value of these expenditures equals
benchmark
asset, yielding
CE.
CE Ag gre gates
The currency equivalent aggregate
closely resembles the Divisia monetary aggregates
proposed by Bamett (1980). Both types of aggregates attempt to measure the
derived from monetary assets.
sub-utility, L,,
This sub-section compares these two approaches to monetary
aggregation.
The
principal advantage of Divisia aggregates is that their derivation requires
assumptions only that the aggregator function f be homogeneous of degree one.
weak assumption and
in
only under stronger assumptions,
However,
it
it
would appear
to
be
Since the
D„
this
CE
aggregate equals
strictly inferior in this case.
has five advantages in comparison with divisia aggregates.
First, in discrete time, the Divisia aggregate tends to drift
while the
Under
a continuous time setting without uncertainty, the change in the
Divisia aggregate gives the exact value for the change in L,.
L
weaker
CE
away from
The Theil-Tornquist approximation
aggregate does not.
the level of
L
to the Divisia index,
is
logDM +£-1 (y^Mr
r m
( m
fo.M-'/.f-Mn Vlog^t
'
logD, =
E Kr
V
/
i)
i.t
x
>
,
E
J
(*itM
~r
j.t-i)
m
j.t-i
mu-\)
(12)
11
Diewert (1976) shows
change in
L
if f
change
that the
in this index
becomes unboundedly
Since
large.
The
CE
this error is
eventually
aggregate does not suffer from this problem.
If f does
arguments remain bounded,
its
CE
second advantage of the
a random walk with
drift, it
CE
not satisfy the linearity restriction that makes the
A
to another is equal to the
7
has the translog functional form. Otherwise each change in the index
introduces an approximation error.
with respect to
from one period
aggregate
it
aggregate exact, but the derivatives of f
will not drift arbitrarily far
is that its
level has
a definite meaning.
Divisia indices seek to measure the changes in the sub-utility from assets,
be used
in international
liquidity held
A
third,
by
comparisons of
L,.
They cannot
liquidity, or to evaluate the different levels
of
different individuals.
and perhaps the most important, advantage of the
easily handle situations
where the function
f changes over time.
CE
assets changes constantly.
their liquidity properties,
For example, when
and therefore
that the utility
NOW accounts
their respective
f;,
aggregate
is that it
Such changes
of the most important challenges of monetary aggregation since the
assume instead
from L.
set
can
constitute
one
of available monetary
eliminate charges on checks,
change.
Divisia aggregates
provided by each asset remains immutable.
The
CE
aggregate
deals with changes in asset characteristics by incorporating the idea that, for asset holdings
not to change,
(rb
,-r;
)/rb>t
if
interprets increases in (rb
must
rise as
,-rit)/rb
,
much
as f
;
.
The
CE
aggregate thus correctly
with an unchanged asset stock as an increase in the assets'
liquidity services.
^Because the translog
Divisia aggregate
is
is itself
a quadratic approximation to any
homogeneous function
"superlative" in the terminology of Diewert (1976).
f,
the
12
The
function f changes most dramatically
when new
introductions cannot be addressed using Divisia aggregates, they present
no
CE
is
aggregate.
As
equation (12) shows, the change in the Divisia index
changes in the logarithms of
its
A
new
minus
infinity
when an
assets.
fourth advantage of the
consumers.
is
Barnett and his collaborators therefore use a different index method to
asset is introduced.
incorporate
difficulty for the
based on the
components. Because the logarithm of zero
growth rate of the Divisia aggregate equals
infinity, (12) implies that the
While these
assets are introduced.
CE
aggregate concerns aggregation across different
Equation (12), which defines the change in the Divisia aggregate,
is
a good
approximation to the change in an individual's liquidity. However, the version of (12) that
uses the aggregate value of the components does not generally bear any relationship to an
It
does so
other individuals.
While
individual's L.
paradigm
if
each individual's asset holdings are proportional to those of
this is
is unattractive in
implied by the representative individual paradigm,
money
is linear in asset
Such transactions reduce one
holdings while raising those of someone else.
holdings,
this
the case of monetary assets because monetary transactions
generally involve a payment from one individual to another.
individual's
all
it
Because the
CE
aggregate
remains equal to the sum of individual L's even when
individuals engage in transactions with each other.
A
final
advantage of the
research, is that
it
aggregate, which
simplifies the estimation
to one, equation (9) can
individual's
CE
demand
we hope
will provide the basis for future
of money demand equations.
With
fo, t
normalized
be viewed as a conventional money demand equation relating an
for total liquidity, L,, to the "market" interest rate
and a measure of
his
13
transactions, in this case consumption.
across individuals if either
utility
function
is
all
This equation applies directly to the sum of the L's
individuals have identical L's and C's or if each individual's
quadratic with the same parameters.
often use the commercial paper rate, which to
paper as our benchmark
asset.
other interest rates as well.
interest rates
have no role
In contrast to the
interest rate to include in a
Our
some degree justifies our use of commercial
on f implies the
restriction
of
aggregate, which
money demand
additional unit of the Divisia aggregate
is
testable proposition that other
(9).
makes a simple
prediction about the appropriate
equation, the foregone interest cost of holding an
a complicated function of
Moreover, for a given Divisia quantity aggregate, there are two
price indices.
The
first
by the Divisia
many
interest rates.
different, equally valid,
can be computed using a formula analogous to that for the Divisia
quantity index applied to prices, while the second is obtained
liquidity
money demand equations
However, conventional money demand equations often add
in the estimation
CE
Estimates of
quantity index.
by dividing expenditure on
of demand
Serletis (1991) estimates systems
equations for Divisia aggregates using the latter approach.
Before closing
this sub-section,
it is
worth remembering
that Divisia
and
CE
aggregates
have many things in common. Changes in monetary assets affect both aggregates only to the
extent that their rate of return
broker-dealer
are in
on
M2
is
lower than that of a benchmark
money market mutual
Thus, changes in
funds and in time deposits at commercial banks, which
significant
impact
This contrasts with the conventional wisdom, which holds that
money
but essentially yield the commercial paper
either aggregate.
asset.
rate,
do not have a
market accounts can be used for making payments and so must be "like" checking accounts.
14
The conventional view confuses average and marginal
liquidity services.
money market accounts do provide
dollars invested in broker-dealer
individuals are likely to write checks
The
first
few
liquidity services since
on them. Because these accounts pay
interest rates
equal to those on zero-liquidity, assets, however, they also attract additional funds which will
not be used to
make payments
in the near future.
changes in these asset stocks do not affect
One
difficulty with this
portfolio of all assets.
argument
total liquidity services.
is that
Some consumers
So, starting from the equilibrium holdings,
not
all
consumers
money market accounts do not
consumers are only slowly drifting into such
"regular" checking accounts into
CE
assets,
money market
For these consumers,
from regular checking to money market accounts
1.3
rate,
As
is
these consumers convert their
not too severe
gradual since
whatever form of aggregation
is
it
if the
then has
fall
as the
conversion
little
effect
on
used.
Comparison with Other Methods of Monetary Ag gregation
There are two alternatives
the
is
further
Other
raise their liquidity.
accounts, their liquidity does not
and Divisia aggregates would imply. This problem
money's monthly growth
the optimal
thereby gaining the inframarginal
consumers already enjoy.
liquidity services the optimizing
by holding
are very sophisticated and hold their wealth in assets
that provide liquidity services with negligible losses in return.
increases in their holdings of
start
CE
aggregates.
The
first is
to the sort of aggregation that underlies both the Divisia
and
the estimation of a parametric utility function along the lines
of Chetty (1969) or Poterba and Rotemberg (1987).
Armed
with the
utility function's
parameters, one can obviously construct a measure of the utility derived from liquidity
15
services.
While we are sympathetic
to this approach, Spindt (1985) correctly argues that
copes poorly with changes in the liquidity characteristics of the underlying
A
second approach, introduced by Spindt (1985),
is to
measure the
provided by various monetary assets using data on their turnover.
The
it
assets.
liquidity services
idea
is that assets
with large turnover must be more useful in transactions and so have higher liquidity services.
The problem
is that
turnover need not be a good measure of liquidity services.
For
instance,
relatively small denominations
of currency turn over more quickly than larger ones. Yet,
one would not want
money
outstanding changes.
to say the
The reason
services at the margin.
rarely
large
1.4
by being
sums of
Slow
The
change
supply changed
is that
one expects
when
all
the composition of currency
currency to provide the same
Large denominations make up for the
fact that they are used only
particularly useful in certain circumstances, such as
cash.
when one must
carry
8
Portfolio Adjustment
currency-equivalent aggregate does encounter difficulty if individuals are slow to
their portfolios in response to interest rate changes.
will not always hold
variations in
f<.
and the variations
This problem
in (r^-r^/rfc,,
is likely to
do not necessarily correspond
to
be important, particularly since our benchmark rate
on monetary
is
quite volatile while the rates
is
not very useful in predicting output or prices.
*We do
In this case, equations (6)-(8)
assets tend to
change sluggishly. 9 Because
this
not pursue the empirical properties of this index here because Serletis (1988) shows that
it
'Of course, violations of the consumer's fust order conditions also degrade the approximation
properties of Divisia aggregates. In the case of interest rate changes that do not lead to rebalancing of
individual portfolios it is possible that Divisia's errors are smaller than CE's because changes in
16
sluggish adjustment
is
due in part to regulatory
inertia,
it is
hard to attribute the resulting
changes in (r^-r^J/r^, to changes in the assets' liquidity services.
When
individuals
rebalance their portfolios only occasionally, they choose a portfolio of monetary assets such
that the
each
f/s at any
f is close to the (r b -rj)/rb
s
moment
therefore depend
expected to prevail in current and future periods.
on the
The
current, past, as well as expected future values of
the interest rate differentials.
In implementing the
CE
by constructing averages of
is
aggregate,
(rb
,-r;
,)/rb
,
we
address the problem of slow portfolio adjustment
and using them
to
Our benchmark
measure CE.
constructed using weights corresponding to a 13-month centered moving average.
consider a 37-month centered moving average, as well a "fixed weight"
ri,i)/ rb,t's are replaced
2.
by
Data and Summary
CE
in
case
We also
which the
(rbl-
overall sample averages.
Statistical Properties
Data Construction
2.1
We
include eight component assets in the currency equivalent aggregate.
currency (CUJ, travelers' checks (TQ), and demand deposits (DD,)
The
interest
paying assets are other checkable deposits (OCD,) such as
savings accounts at
money market
interest rates
~
thrift institutions
(SS10, savings accounts
accounts at commercial banks
have a bigger
effect
on the
latter.
(MMCB^
at
Three
do not yield
~
interest.
NOW accounts,
commercial banks (SCBJ,
and money market accounts
at thrift
17
institutions
aggregate
(MMSL,). 10 Using
given by:
is
CE
=
CU
t
t
*
TC
+
t
DD
t
y-^)SCB
t
*
*
(l-^Of)OCD,
For the reasons given
t
weights currency equivalent."
rate spreads with the overall
we
*
(13j
-!™*d)MMSL
(1
rcp,t
rate, are
.
t
our benchmark estimates of the currency
moving averages of
interest rate spreads as
high
also consider a second aggregate, labelled the "fixed
In this aggregate
we
replace the 13-month average interest
sample average differentials.
quantities of various
commercial paper
t
this aggregate's short-term fluctuations are sensitive to
frequency interest rate changes,
Data on the
+
rcp,t
in the previous section,
Because even
rcp,t
{\-!™2*±)MMCB
equivalent aggregate use 13-month centered
11
{\-^d)SSL
+
rcp.t
rcp,t
weights.
CE
these symbols as subscripts to identify returns, the
monetary assets that enter equation
drawn from Federal Reserve Board
Statistical
(8), as
well as the
Release H.6.
The
data are available at a monthly frequency, seasonally adjusted, for the period since January
1959.
Data on
available.
10
M1
deposits.
Our measure of
equals the
M2
monetary
interest rates corresponding to the
Tqcd.i,
sum of currency,
includes
all
the interest rate
on other checkable
travellers checks,
monetary assets
in
Ml
assets are not as readily
demand
deposits, is
an average
deposits and other checkable
plus savings deposits, small denomination time
and Eurodollar deposits, and money-market
components of this aggregate yield market returns
deposits, short-term (overnight) repurchase agreements
mutual fund shares.
We assume that the last two
and consequently give them no weight in our CE aggregate. M3, a still broader aggregate, includes
large-denomination time deposits and term repurchase agreements and Eurodollar deposits.
"We
For each of
its 13-month average centered in month t.
months of the sample, we use the average interest rate ratio for the first Oast) 13
months instead of the centered moving average.
replace, for example, roco,,/^,, with
the first Oast) 13
18
across various types of interest-bearing accounts computed
The
interest rate
accounts,
on savings accounts
r^,, r^c^, and Tmms^u were
constructed the interest rate
1986,
at
when
these accounts, as reported
maximum
occasional
FHLBB
by
also provided
the Federal Reserve Board.
at thrifts,
we
were deregulated,
r^m,
in
two
parts.
use the average interest rate paid on
Our imputation
commercial banks.
allowable interest rates at
thrifts
We
Since mid-
by the Federal Reserve Board. Before 1986, we impute
rates for thrifts using rates at
relating the
the Federal Reserve Board.
commercial banks as well as those on money market
on savings accounts
interest rates at thrifts
by
relies
interest
on regulations
and commercial banks, as well as
surveys of average interest rates paid at these institutions. 12 For part of
our sample, the interest rate on passbook savings accounts exceeded that on commercial
We assume
paper.
that these interest rates
were equal
in these cases, thereby setting the
weight on the associated asset stock to zero.
Figure
1
plots
Ml
and the currency-equivalent aggregate defined
in (13).
The two
aggregates are drawn on the same scale so the Figure shows that their size and growth rate
are generally similar.
In particular,
OCD
t
is
difference between the
much more
periods.
When
12
While
CE
includes
more
given a smaller weight in
two aggregates
sharply than
Ml
in
is that
assets,
CE
Mi's
some periods and
These pronounced changes
in
CE
than in
rate
CE
Ml. The most
of growth
is
striking
smoother.
CE rises
also declines steadily during certain
on regulated monetary
assets often
we assumed r^, = r^,, + .75. This additive factor was .50 for
we assume r^ = 5.24%, and for the 1979-1986 period, we
1970-1973. For the 1973-1979 period
make crude
these are given small weight.
are associated with changes in interest rates.
the commercial paper rate changes, the rates
For the 1959-1970 period,
many of
estimates using occasional surveys of interest rates paid.
19
change
Thus, as an example, declines in the commercial paper
less than proportionately.
on OCD,
rate tend to raise rocD.Acp.t so that the weight
falls.
The
result is that aggregate
itself falls as well.
2.2 Stationaritv Properties of Alternative Aggregates
The
apparently cyclical behavior of
CE
raises the question
traditional aggregates exhibit different stationarity properties.
CE
of whether
This issue
and
important because
is
part of our subsequent analysis will focus on the long-run effects of monetary shocks on the
level of
economic
Table
two
CE
1
activity.
reports standard Dickey-Fuller tests of the null hypothesis that
These
aggregates contain unit roots.
tests
Ml, M2, and
the
are equivalent to testing whether in an
equation of the form
log/If,
the 7i's
sum
to one.
The
trend, to equal to zero
reported in the
first
= a * pf
tests
column, suggests
rejected for any of the aggregates.
aggregates.
that the
<™>
e?
table reports tests constraining B, the coefficient
and also
the hypothesis that there
2^ Y/logM,., +
allowing
it
be unrestricted.
contrast
when
fi
is
unrestricted (the second column),
a unit root can be rejected for CE, though not for the other
is
CE
are
a unit root cannot be
This rejection reflects the cyclical patterns displayed in Figure
dependence of
linear
The former, which
that the hypothesis that there is
By
on the
on high frequency movements
rejection of nonstationarity.
1,
and suggests
in interest rates accounts for the
20
These
to
results leave us with
compare the
properties of
specifications, for
CE
a choice as to
with those of the traditional aggregates.
Our
has one unit root as well. However,
3.
to treat the stationarity
We wish
of CE.
Conventional
example Kormendi and Meguire (1984) and Stock and Watson (1989),
treat the latter as nonstationary.
assume instead
how
that
CE
basic specifications will be therefore assume that
we
will also discuss
how
CE
when we
the results change
is stationary.
Testing Methods
Developing a new monetary aggregate
is
of some intrinsic interest for what
it
about the evolution of the money supply in the United States. The new aggregate
differ
from previous aggregates
real activity.
13
We
in the evidence
it
reveals
may
also
presents on whether monetary shocks affect
follow the tradition of Sims (1972) and perform causality tests to see
whether various measures of money and monetary policy have significant predictive power
for future output realizations.
Under the assumption
exogenous monetary impulses, these
We
measure
for married men.
real activity
The
latter is
tests
by both
M
that the residuals in (14), e „ are
inform us about the influence of money on output.
industrial production
and by the unemployment rate
a particularly useful measure of labor market tightness
because, unlike other unemployment rates which are partly contaminated by decisions to
leave the labor force, very few married
When we
men withdraw from
the labor force.
use industrial production as our measure of activity y, the estimated equations
are of the form
"Blanchard (1990) surveys the voluminous prior
literature
on
this issue.
21
Alny, = o +
The polynomials
/S(L)
M+
M
p(L)Alny
+
and 7(L) are of 12th order which seems natural given
analyzing monthly data.
industrial production has
to
use
its
a unit
When we
root.
level, not its first difference, in (15)
be stationary.
we assume
deterministic trend in this regression even
of money
is
14
This
because the unemployment rate
when we
is
These are changes
in central
changes in the demand for monetary
and
it
would be justifiable
exogenous changes of
argue
that, if these
money
itself
included a
first
difference
has a unit root for, in
this type
it
effect of the e^'s
makes sense
to
on output.
judge monetary aggregates
bank policy, changes
assets.
The
first
two
to treat the correlations
in the financial system
affect the supply of
were the only influence on the stock of money.
were the only disturbances,
theoretically
data, the inclusion of this trend has a relatively
and
money
of money and output as causal
We now
more appropriate monetary
aggregates ought to be more closely linked to output than other aggregates.
"For our
likely
In practice, there exist three separate kinds of monetary
their correlation with output.
directly
we have
more
monetary shocks of (14) are functions only of current and lagged differences in
This brings us to the question of whether
shocks.
is
are explaining the behavior of the change
reasonable if
money. Equation (15) can thus capture the causal
by
of
study the behavior of the unemployment rate
Equation (IS) assumes that only the logarithmic
correlated with output.
this case, the
are
that the logarithm
Following the suggestion of Stock and Watson (1989)
in industrial production.
we
that
Industrial production enters the equation in logarithmic first
differences because, following Nelson and Plosser (1982),
we
(15)
Y(^)Alnmr + <.
minor
effect,
After making
however.
if
22
we
this case,
return to the issues that plague the interpretation of all studies of money-output
correlations.
Both changes in central bank policy and changes in financial institutions affect the
economy only through
their effect
interest rates
and economic
these shocks
on economic
that
the
more
activity.
L
L
which, via the
L
summarize the
(9), affects
effect
of
all
ought to be more closely linked to subsequent changes in
be more closely linked to output than arbitrary
aggregates or even than variables which are
exogenous changes in
L
equation
This means that one should fmd that monetary aggregates
particular, they should
such as the monetary base.
money demand
Therefore, measures of
activity.
closely approximate
economy. In
on
more strongly influenced by
the central bank,
These variables would be particularly poor proxies for
if the
monetary authority chooses
its
policy to neutralize
some of
the
changes in financial institutions.
Any
money
study of the correlation of
is set
endogenously.
money with output
In particular,
faces the problem that the level of
when money demand changes exogenously, money
supply probably accommodates the change in
money demand
cannot be fully resolved without a structural model of the
assets
which
is
well beyond the scope of this paper.
neglect of this endogeneity
this
may
is
some
extent.
demand and supply
This problem
for liquid
While our, and our predecessors',
mainly a matter of expedience
it is
important to point out that
not be a very large problem in practice.
Endogeneity has two components. The
that
to
money demand
rises
when economic
first,
stressed
activity rises
by King and Plosser (1984),
exogenously and
that the
money
is
23
money demand. 13 One expects
supply then accommodates the increased
increases in economic activity to lower prices so this story
rise less rapidly after positive
be the case, so
monetary shocks.
We
these exogenous
would predict
show below
that prices
that this does not
would
seem
to
of endogeneity may not be very important. The second form of
this type
endogeneity arises when there are increases in money demand that are unrelated to
exogenous changes
have the same
supply
is
in output.
effect as
money supply response,
In the absence of any
monetary contractions, reducing output and prices.
endogenous and responds
partially, the resulting increases in
associated with subsequent reductions in output and prices.
associated with both price
and output
With these provisos, we
by
increases,
changes in money affect economic
rigidities
could
activity.
and issues pertaining to the
make monetary shocks
—
and
we
If the
money
money ought
to
be
Insofar as monetary increases are
problem may be minor as
absence of correlation between
test for the
testing the null hypothesis that y(L)
this
these ought to
well.
money and
output
interpret rejections as suggesting that
There are many factors, such as nominal
distribution
of newly supplied monetary
non-neutral in the short run.
It is
much more
assets, that
difficult,
however, to explain permanent, long-run effects of money on output. Insofar as monetary
,
shocks do affect output in the long run, the interpretation that the money-income relationship
represents true causation
more
either
from money
likely that the relationship is
income
"We
itself
a
to
income
statistical
is
called into question.
fluke or
is
Instead
it
the result of responses of
or to third variables which affect income as well.
money
to
This endogeneity
problem to some extent by excluding the contemporaneous change
from our regressions although this exclusion has no material effect on our results.
deal with this
becomes
in
money
24
could well be a serious problem for conventional aggregates
if
non-monetary shocks which
of assets for a given level of overall
affect output also affect the subsequent allocation
liquidity.
In the case where both output and the monetary indicator have a unit root, our
monetary
neutrality corresponds to the
this case, (15)
one proposed
Kormendi and Meguire (1984). In
can be rewritten as
Aln;'f
*
- c 5fn(L)Aln/"'* e{:
<
16
>
.
So, a one permanent one percent increase in
percent.
in
test for
money
raises long-run output
by
(1-/3(1))"
1
7(1)
Testing long-run neutrality therefore requires testing whether the estimated value of
this quantity is significantly different
One
from zero. 16
issue that arises at this point is
given the criticism levelled by
McCallum showed, Lucas'
why
our
McCallum (1984)
(1980) method
coefficients in a regression of output
is
tests
of long run neutrality
at the earlier attempt
equivalent to computing the
this
sum of
is
and Seater (1990), measuring the response to a once and
itself.
the
true of our
method applies only when money
Then, monetary shocks have no long run effect on money
sense
by Lucas (1980). As
on many lags of money and the same
method. McCallum's (1984) criticism of
make
is stationary.
So, as argued by Fisher
for all increase in
money
is
then
impossible because such increases never occur in practice.
l6
They
Note
that long run neutrality
will fail to
shocks do.
be cointegrated
if
money and output are not cointegrated.
monetary impulses have no long run effect on output but other
is
perfectly possible even
if
25
Suppose however that money has a unit
one percent
raises
money
at
t
plus infinity by
root.
Z
Then a shock
percent.
The
that raises
quantity
Z
money
at
t
by
can be estimated
from univariate regressions of money growth on lagged values of money growth using
standard methods.
to
a once and for
The
quantity
neutrality.
From
Then the sum of
increases in
all
Z
money
method works because an event analogous
actually takes place every period.
generally not equal to one but this does not matter for testing long run
is
(15) and the univariate regressions for
things: Z, the long run response
l
coefficients
money growth we can compute two
of money to a unit innovation in money growth and
y(\)Z, the long run response of output to a unit innovation in money.
that
Z
is
nonzero (so that money
(HWtO)
A
is
certain
effects,
If
nonstationary), long run neutrality prevails only if
form of long run neutrality, whether monetary shocks can have permanent
can be tested even when
if
is itself stationary,
one accepts
money
is stationary.
money rose and
then
that industrial
it is
cannot be tested by the
17
all
This
is
not identical to the question
stayed higher forever but
it is
closely related.
If
production has a unit root, then analysis of the long run effects
sum of
is
For the reasons stressed
the coefficients
output have a unit root.
(14) and the equation which
17
impossible for monetary shocks to have long run effects.
of monetary shocks becomes interesting.
money and
Thus, under the null
zero.
of what would happen
output
is
(1-/3(1))"
Instead,
method we used
what
is
required
is
in
McCallum
in the case
(1984) this
where both
a simultaneous analysis of
analogous to (15) and relates output growth to the current and
Fisher and Seater (1990) feel instead that long run monetary neutrality means only that once for
changes in money have no real effects.
This hypothesis can obviously be tested only
shocks have permanent effects on the level of money.
if
monetary
26
past levels of
money. Using both these equations simultaneously one can compute the
impulse response function of output with respect to monetary shocks and analyze the long run
response to these shocks.
We
carry out these tests for our currency-equivalent aggregate, which
stationary.
We also
apply
this test to
two
variables,
which may also be
may be
stationary, that
have
recently been proposed as measures of monetary policy: the federal funds rate and the spread
between the yield on commercial paper and the yield on Treasury
is
bills.
If
monetary policy
indeed the predominant force acting upon these variables, then their changes ought not to
have any long run effects on output.
We
also test
whether money has any
effect
on prices by considering regressions of the
form
Alnp, = a' +
where we measure the price
hypothesis that
there are
6't +
pU)Alnp, + y '(L)A\nm t +
level (p) as the
money has no
no plausible theories
effect
that
Consumer Price Index.
this null hypothesis.
analyze the long run response of prices to monetary shocks.
(1990) this can be done using the same method to the one
run effect on output.
In the case where
hypothesis that permanent increases in
(1-/3'(1))V(1) equals one.
We then
on prices by testing whether 7'(L)
imply
money and
money
we
It is
<
ef
=
0.
thus of
As shown by
17
>
test the null
Admittedly,
more
interest to
Fisher and Seater
used for measuring the long
prices both have a unit root, the
eventually raise prices one for one implies that
27
4.
Money and Output
4.1
Granger Causality Tests
We
first
first
report tests of whether
results
do not
significant effect
monetary
null of
the Divisia
By
matters at
all in
equations such as (15).
The
8 rows of Table 2 reports the rejection p-values for the null hypothesis of no link
between current real
The
money
M3
activity
and past money for a range of
yield a consistent pattern, except for
on
real activity.
When
traditional
monetary aggregates.
M2, which always
the measure of activity
has a
is industrial
neutrality is rejected at high levels for both ordinary
but neither Divisia
M2
nor
unemployment
M3, nor
ordinary
rate,
M3,
we
production, the
and Divisia
aggregate proposed by Bamett and his co-authors, but not by
contrast, in the case of the
statistically
Ml
obtain high rejections for
M2
or
Ml
and
M3. u
and M2,
yield strong rejections of neutrality.
Other conventional monetary aggregates yield even weaker evidence against neutrality.
For currency and the money
multiplier,
either industrial production or the
rejected
when we
we do
not reject the null of monetary neutrality for
unemployment
rate.
For the monetary base, the
focus on the unemployment rate, but not for industrial production.
Our benchmark
CE
aggregate, which
is
based on a 13-month moving average of interest
rate spreads, yields strong evidence against neutrality regardless of
activity
of
this
we
CE
rejections
null is
analyze; this rejection pattern resembles that for
which index of
M2. Because
real
the current value
aggregate depends on future interest rate spreads, the interpretation of these
might be viewed as problematic. In particular,
it
might be thought that these
"The ease with which we can accept that Ml has no effect on industrial production contrasts with
Stock and Watson (1989) who show that Ml does affect output in specifications like (13). The reason
our results are different is that our sample is longer than theirs. We obtain the same results as them
when we restrict attention to the period before 1985.
28
future spreads simply contain information about future output and that this is the cause of our
significant correlations.
We consider three variations which
CE
"fixed weight"
are not subject to this problem.
interest rates but, leaving aside
to 18
months
months. Thus
it
is
As we
to the next affect output
seems implausible
this
CE
influenced only by the change in yields from 18
after the current period.
money growth from one month
This too uses future information
changes in the component stocks, the change in
aggregate from one month to the next
months before
first is the
which uses the overall sample average yield spreads as weights. The
second uses a three-year moving average of yield spreads.
on
The
will see
below, changes in
growth only for approximately 6
that the yield changes in the 3-year
through their inclusion of future information on output.
The
final
CE
CE
is
affect output only
based only on
contemporaneous yield spreads and would be the appropriate aggregate in the absence of
costs to portfolio adjustment.
As can be
The
seen in Table 2 the results are fairly robust to these changes in specification.
"fixed weight"
CE
leads to strong rejections of neutrality for both industrial production
and unemployment. The "3-year"
its
effect
on
CE
industrial production is
also leads to rejections at conventional levels though
weaker.
Finally, the
contemporaneous CE, which
is
very volatile, only affects industrial production.
The remaining rows of Table 2
present Granger causality tests for several recently
proposed alternative measures of monetary policy. Bernanke (1990) argues that
shifts in
policy are captured by the yield spread between commercial paper and Treasury
bills.
correlation of this spread with subsequent output
was
first
The
demonstrated by Friedman and
29
Kuttner (1990).
We too
rejections of 7(L)
=
find strong correlations with output.
when we use
We find
the federal funds rate suggested
similarly strong
by Bernanke and
Blinder (1990) or the commercial paper rate itself as a measure of monetary policy.
One
important limitation of the interest-rate based measures of monetary stance
the estimated effects of these aggregates are sensitive to the time period in question.
the
CE
is that
Using
aggregate, within-sample stability tests for the coefficients in (15) for industrial
production, for example, do not reject the null hypothesis that 7(L) and 0(L) are identical for
the pre- and post- 1980 period.
By
contrast, this hypothesis is rejected at the .01 level for
the commercial paper-Treasury bill yield spread.
for
Ml
and
M2
is
The evidence
against sub-sample stability
also weak, suggesting that these aggregates behave
more
like the
CE
aggregate.
The
results in
Table 2 suggest that the
CE
aggregate provides stronger and more
consistent evidence against the monetary neutrality hypothesis than
measures of the money supply.
predictors of output.
whether
CE
Since
CE
However,
many, but not
all,
other
the interest rate variables are actually better
based on interest rates this raises the question of
is partially
has any incremental explanatory power once the other measures of
money
or
monetary policy are included in an equation for output.
Table 3 addresses
that all coefficients
this question
by reporting p-values for the F-tests of the hypothesis
on the lagged values of
other indicators of monetary stance.
The
CE
are zero in equations which already include
findings suggest that
CE's information content
remains even after the other measures of money or monetary policy are controlled
most surprising finding
in the table is that
CE
for.
The
remains a significant explanator of industrial
30
Ml
production (though not of the unemployment rate) after controlling for both
commercial paper
benchmark
4.2
CE
rate.
and the
This suggests that the variations in interest rates that affect the
are not solely responsible for
its
explanatory power.
Long Run Neutrality
We now
the long run.
whether once and for
We
thus measure (1-/3(1))''7(1), the percentage increase in long run output
when money
increases
measure and
its
The
by one
all
increases in
percent, and test whether
standard error for conventional,
CE
M2
rejected for all these assets
and M3.
19
Not only
Table 4 displays
equals zero.
M3, where
this
and Divisia aggregates.
Ml, M2, and both
the hypothesis that (l-/3(l))" ly(l) equals zero
is
but the point estimates exceed one; they suggest that a one
percent increase in the asset raises real output by
for
it
output significantly in
results provide strong evidence against long-run neutrality for
the Divisia aggregates
Even
money change
test
more than one percent
-1
the hypothesis that (l-/8(l)) -y(l)
is
in the long run.
zero cannot be rejected, the point
estimate exceeds 0.5.
In contrast, the null cannot be rejected for the
contemporaneous weights variants.
CE
aggregate nor for
its
3-year and
For the benchmark CE, the point estimate of the long-
run elasticity of output with respect to
money
is
very small (.05). For the contemporaneous
and three year versions the point estimates are somewhat larger but
still
On
below 0.15.
other hand, the long run response of output to a one percent change in the fixed weight
similar to that of output to
Ml. The
point estimate of this response
is
"Fisher and Seater (1990) also report rejections of long run neutrality for
longer time series.
near one and
Ml
using a
is
much
the
CE
is
31
significantly different
from
So, the
zero.
utility
based aggregates that do
characteristics vary over time, such as the fixed weight
be long run neutral.
20
CE
let asset
and the Divisia aggregates,
fail to
This suggests that changes in the liquidity characteristics of assets
are important.
4.3 Impulse Response Functions
In this sub-section
monetary shocks.
stationary.
do
For these we
estimates of (15).
impulses.
We
we
We
From
present impulse response functions of output with respect to
this first for the
monetary indices which are arguably non-
report the effects of a
these
we
once and for
all
increase in
money
using the
learn the time profile of the effects of these monetary
then study the impulse response functions for the indicators of monetary
policy which appear stationary.
These impulse response functions, which are based on
simulating the analogues to equations (14) and (15) simultaneously, also serve to test the long
run neutrality of these shocks.
Figures 2a through 2c display these impulse response functions for the effects on
industrial production of a
also report
1% once and
for all increases in
one standard-error bands on
Ml,
M2
and CE.
either side of the point estimates.
the cumulative effect of a shock to the monetary variable
on
real output.
The
figures
Each figure shows
Long run
neutrality
therefore implies a zero effect at long horizons.
by simply adding together all the components of
our CE aggregate. We created this aggregate to see whether our results were attributable to the fact
that our menu of included assets does not correspond to any conventional aggregate.
"The same
is
true of an aggregate
we
created
32
The impulse response
M2
do not correspond
money on
virtually
functions (IRFs) of industrial production with respect to
to the predictions of standard models.
output rises for the
no reduction
first
The cumulative
fourteen months after a monetary shock.
in the cumulative output effect thereafter,
which leads
Ml
and
effect of
There
is
to the finding in
Table 4 of strong evidence against the long-run neutrality hypothesis.
The IRF
pattern is substantially different for the currency equivalent aggregate.
innovation in the
money
to increase for seven
stock raises industrial production, and the cumulative
months
after the shock.
shock.
By
is
after the shock, the total
estimated to be only one tenth as large as seven months after the
eighteen months after the shock, the point estimate of the cumulative
effectively zero.
continues
After this point, however, the lagged monetary
shock has a negative effect on output and by fourteen months
increase in real output
IRF
A positive
This suggests not only that the long-run effects of the
CE
IRF
is
aggregate are
consistent with monetary neutrality, but also that any non-neutralities implied
by these
estimates are resolved in a relatively short time span.
We
now
turn to the effect of the stationary monetary impulses.
functions for shocks to the commercial paper-Treasury
rate are given in Figures
2d and
2e.
bill
The impulse response
spread and to the federal funds
Figure 2d shows that a 100 basis point increase in the
spread between the annualized yield on commercial paper and that on Treasury
permanently lowers industrial production by 5%.
a monetary shock since the maximal reduction
months)
is essentially
It is
hard to interpret this as a response to
in output
equal to the long run reduction.
bills
(which occurs after about 10
Figure 2e shows a similar pattern for
the response of output to a 100 basis point increase in the annualized Federal funds rate.
33
Output
falls steadily
and becomes about
5%
This too seems hard to
lower in the long run.
reconcile with typical analyses of monetary contractions.
We also
assuming
and (IS)
M
c,
that
in
CE
which
We did
.
computed the impulse response of
this
The
stationary.
CE
In other words
results are
CE
we
considered the system of equations (14)
suggest that the
CE
aggregate might be
very similar to those obtained when
CE
is
first raise
no
significant long run effect.
5.
Money and
1
assumed
to
also be used to
examine the
effects
level.
shocks and velocity
prices should rise one-for-one with increases in the
This section
is stable,
tests
In the long run, if output
first
column
in
is
Table 5 reports
tests
effects
in the price
of Granger-causality between traditional
correspond to p-values for rejecting the null hypothesis that y'(L)
CE
money
of monetary shocks.
monetary aggregates, other indicators of monetary policy, and the price
of the
of
not affected by monetary
whether monetary impulses Granger-cause changes
and investigates the long-run price-level
The
have a unit
Prices
monetary shocks on the price
level,
to
output significantly and then lower output so that they have
The new currency equivalent aggregate can
stock.
CE
computed the response of both variables
enters only in levels and
because the results of Table
Increases in
root.
is stationary.
industrial production to shocks in
=
level.
The
entries
in equation (17).
All
aggregates except that with time-invariant weights display strong predictive power
for future
movements
traditional
monetary aggregates.
in the price level.
The Granger
The monetary base
is
causality results are
much weaker
for
the only such aggregate for which
we
34
can reject the null hypothesis of no effect.
The
rows of the
last three
Federal Funds rate and the commercial paper rate do
show
table
show
strong predictive
that the
power
for
changes in the price level, while the null hypothesis of no causality cannot be rejected for the
commercial paper-Treasury
The second column
bill yield
in Table 5 presents our estimates
The
monetary shocks on prices.
relative merits
spread.
results
do not permit us
of the various monetary aggregates.
For
of the long-run effects of
to
all
draw sharp conclusions about the
of the monetary aggregates
consider, the hypothesis that in the long run prices rise proportionately to the rise in
cannot be rejected at conventional significance levels.
however, and the point estimates for the long-run
range from .3 to 3.7.
permanent
1
%
money
standard errors are large,
elasticity
of price with respect to money
Figures 3a and 3b report estimates of the CPI's response to a
increase in
Ml
and CE. In the case of
reached three years after the monetary shock.
respond to monetary shocks
6.
The
we
is relatively
CE
half of the long-run effect (.62)
is
This implies that the process by which prices
slow and gradual.
Conclusions
This paper develops a
liquidity services.
We
new monetary
compare
monetary aggregate, other
aggregate based on the amount households spend on
the performance of this "currency-equivalent" (CE)
traditional
monetary policy in forecasting real
monetary aggregates, and various other indicators of
activity.
The
than any of the traditional monetary aggregates.
CE
aggregate has more predictive power
While some other
indicators of monetary
policy based on nominal interest rates can forecast output as well as the
CE
aggregate, they
35
money. This suggests
also suggest long-run non-neutrality of output with respect to
changes in the
assets.
best
CE
that
aggregate are a more plausible indicator of changes in the supply of liquid
This result in turn supports the idea that the monetary aggregate with which one can
measure monetary shocks
is
which measures the
the one
utility (or the services) that
people derive from their liquid assets.
Our
results also contain
disturbances affect output.
rigidities are important,
some information on the channel through which monetary
The
literature
make people hold
lowers investment.
this topic stresses
and whether "money" or "credit"
"money" view, monetary contractions
rise to
on
the
new
According
initially
important. 21
is
money
findings provide
some support
view,
there
for the
situation
where the
A
According to the
a direct reduction in investment
is
to curtail the
"money" view.
stock which has the most plausible effect on output
point of view of holders of money.
whether nominal
This increase in interest rates then
which comes about because monetary contractions force banks
Our
issues:
reduce money balances and interest rates must
constellation of assets.
to the "credit"
two
We have
one
is
that
volume of
shown
loans.
that the
makes sense from the
monetary expansion can thus be thought of as a
level of liquid assets is high, at least for
some consumers.
In the "credit" view, the level of consumers' liquid assets does not matter, except
insofar as
it is
The
correlated with funds available for loans.
correlated with broad aggregates like
M2
than with the
CE
latter are
aggregate.
presumably more
From
the point of
"The distinction between the "money" and the "credit" view has been stressed mainly by authors
working with models of sticky prices, including Bernanke and Blinder (1990), Gertler, Hubbard and
Kashyap (1990), Kashyap, Stein and Wilcox (1990) and Romer and Romer (1990). However, this
distinction is just as important if nominal rigidities are absent and
different people carry out financial transactions at different times.
Grossman and Weiss (1983) and Rotemberg (1984) and Fuerst
money
matters only because
This can be seen by comparing
(1990).
36
view of
credit availability,
whether banks raise
their funds
accounts at zero interest, or through expensive CD's
suggest that this distinction
is
is
through "regular" checking
probably irrelevant.
Because the data
important for the subsequent evolution of output they also
suggests that the "credit" view
is
incomplete.
37
REFERENCES
Barnett, William A., 1980,
Economic Monetary Aggregates: An Application of Index
Number and Aggregation Theory,
,
1983,
New
Journal of
,
Indices of
Bu siness
Money
& Economic
Journal of Econometrics 14. 11-48.
.
Supply and the Flexible Laurent
Demand
System,
Statistics 1, 7-23.
1991, Reply, in Michael T. Belongia, ed. Monetary Policy on the 75th Anniversary
of the Federal Reserve System.
Barnett, William A.,
(Boston:
Edward K. Offenbacher, and Paul A.
Monetary Aggregates, Journal of
Bernanke, Ben S., 1990,
Spreads,
NBER
Bemanke, Ben
S.
Kluwer Academic), 1991.
On
Political
the Predictive
Economy
Power of
Spindt, 1984,
The New
Divisia
92, 1049-1086.
Interest Rates
and
Interest
Rate
Working Paper 3486.
and Alan Blinder, 1990, The Federal Funds Rate and the Channels of
Monetary Transmission,
Blanchard, Olivier
J.,
1990,
NBER
Why
Working Paper 3487.
Does Money Affect Output?
A
Survey, in B. Friedman
and F. Hahn, eds., Handbook of Monetary Economics (Amsterdam: Elsevier Science
Publishers), 779-835.
Chetty, V. Karuppan, 1969,
On
Measuring the Nearness of Near-Monies, American
Economic Review .59. 270-281.
Diewert, E., 1976, Exact and Superlative Index Numbers. Journal of Econometrics .4. 115-45.
Fisher,
Mark
ARIMA
E. and John
J.
Seater, 1990,
Long-Run Neutrality and Superneutrality
in an
Framework, mimeo, Federal Reserve Board of Governors.
Friedman, Benjamin
M. and Kenneth N.
Kuttner, 1990,
Money,Income, Prices and
Interest
38
Rates after the 1980s, Federal Reserve Bank of Chicago Working Paper
Fuerst,
Timothy S, 1990. Liquidity, Loanable Funds, and Real
J.L. Kellogg Graduate School of
Gertler,
Constraints and Investment Fluctuations:
Working paper 3495
mimeo,
Management, Northwestern University.
Mark, R. Glenn Hubbard and Anil Kashyap, 1990,
NBER
Activity,
.
An
Interest Rate Spreads, Credit
Empirical Investigation
.
Grossman, Sanford and Lawrence Weiss, 1983,
A
Transactions Based Model of the
Monetary Transmission Mechanism, American Economic Review 73, 99-122.
.
Kashyap, Anil, Jeremy Stein and David Wilcox, 1990, The Monetary Policy Transmission
NBER
Mechanism: Evidence from the Composition of External Finance
Working Paper
.
King, Robert and Charles Plosser, 1984, Money, Credit and Prices in a Real Business Cycle,
American Economic Review 74, 363-80.
.
Kormendi, Roger and Phillip Meguire, 1984, Cross-Regime Evidence of Macroeconomic
Rationality, Journal of Political
Lucas, Robert E. Jr., 1980,
Two
Economy
Illustrations
.
92, 875-908.
of the Quantity Theory of Money, American
Economic Review 70, 1005-14.
.
McCallum, Bennett T., 1984, On Low Frequency Estimates of Long-Run Relationships
in
Macroeconomics, Journal of Monetary Economics 14, 3-14.
Poterba, James
M. and
Julio J.
Rotemberg, 1987,
Empirical Implementation, in
W.
Money
in the Utility Function:
Barnett and K. Singleton, eds.,
New
An
Approaches to
Monetary Economics (Cambridge: Cambridge University Press), 219-240.
.
Poterba, James
M.,
Julio J.
Rotemberg and Lawrence H. Summers,
A
Tax-Based Test
39
for
Nominal
Rigidities.
American Economic Review 76, 659-675.
.
Romer, Christina D. and David H. Romer, 1990,
New
Evidence on the Monetary
Transmission Mechanism, Brookings Papers on Economic Activity
Rotemberg, Julio
1984,
J.,
Journal of Political
,
A
.
149-98.
Monetary Equilibrium Model with Transactions Costs,
Economy
.
92, 40-58.
1991, Monetary Aggregates and Their Uses, in
M.
Belongia, ed., Monetary Policy
on the 75th Anniversary of the Federal Reserve System (Boston: Kluwer Academic),
224-232.
Samuelson, Paul A., and
Swamy, 1964,
S.
Invariant Index
Numbers and Canonical
Duality:
Survey and Synthesis. American Economic Review 64, 566-593.
.
Serletis, Apostolos, 1988,
The Empirical Relationship Between Money,
Economic
Revisited, Journal of Business and
,
A
1991,
Dynamic
The Demand
Flexible
for Divisia
Money
Demand System,
Statistics . 6,
Prices and
Income
351-358.
in the United States:
Journal of
Money. Credit and Banking
.
23 (February), 35-52.
Sims, Christopher, 1972, Money, Income and Causality, American Economic Review
.
62,
250-257.
Spindt, Paul, 1985,
Money
is
of Exchange, Journal of
Stock, James H. and
What Money Does: Monetary Aggregation and
Political
Economy
Mark W. Watson, 1989,
causality, Journal of Econometrics
.
.
the Equations
93, 175-204.
Interpreting the Evidence of
40, 161-82.
Money-Income
Table
1
:
Unit Root Tests for Various Monetary Aggregates
Aggregate
0=0
Unrestricted
M1
.999
.429
M2
.994
.750
Divisia
M2
.988
.591
Divisia
M3
.977
.669
.993
.012
CE
Notes: Each entry reports the confidence leveKbased on Fuller(1976)) at which the null hypothesis
in the regression
that k =
A lnM =a + p t+ylnM _, +d(L) A In AfM +e"
t
t
can be rejected. The regressions are estimated using monthly data for the 1960:2-
1989:6
period. d(L)
is
a twelfth-order lag polynomial.
6
Table 2: Causality Tests: Monetary Aggregates and Real Activity
Monetary
Aaareaates
Industrial Proi
Traditional
Unemployment Rate
Married
M1
.205
.019
M2
.003
.005
M3
.160
.068
Monetary Base
.174
.007
Money
.223
.179
.517
.473
Multiplier
Currency
Divisia
for
Men
Aaareaates
Divisia
M2
.013
.093
Divisia
M3
.025
.580
<. 00001
.0005
Alternative Indicators of
Monetarv Policv
Commercial Paper
Federal Funds Rate
.001
<. 00001
Commercial Paper-Treasury
Bill Yield Spread
<. 00001
<. 00001
Currency Eauivalent
Aggregates
Currency Equivalent
<. 00001
.0002
Fixed Weight
<. 00001
Currency Equivalent
.002
Three- Year Currency
Equivalent
.042
.0003
Contemporaneous Currency
.031
.721
Equivalent
Notes: Each entry reports the confidence level at which the null hypothesis that
be rejected in the specification
A \ny = a +
t
1+
gamma(L)=0 can
y
p (L) A lny,+ y (/.) A \nm t *e t .
Both £(L) and y{L) are 1 2th order lag polynomials, so the statistics reported above test the joint
hypothesis that Ki = Yi= ••• =Ki2 = 0. Th e sample period for all estimates except Divisia M2 and
is 1960:2 to 1989:6. For the two Divisia aggregates, the sample is 1961:2-1989:6.
M3
Table 3:lncremental Explanatory Power of Different Monetary Aggregates
Challenger
Incumbent
Industrial
Unemployment
Production
Rate for Married
Men
M1
CE
.002
.011
M2
CE
.003
.005
Divisia
M2
CE
.023
.009
Divisia
M3
CE
.015
.006
CE
.002
.003
CE
.002
.004
CE
.009
.097
Commercial PaperTreasury
Spread
Bill
Yield
Federal Funds Rate
Commercial Paper
and M1
Note: Each entry reports the p-value for the null hypothesis that GIL)
=0
in
the specification
Alny =a +6f+p(Z.)Alny,+Y(^)Alnm1 f+e(L)Aln/nif+€{'.
f
m 2 denotes the "challenger"
2th order, so the statistics reported above test the joint
= 0. The sample period for all estimates except those for the
where m, denotes the incumbent aggregate and
,
aggregate. All lag polynomials are
hypothesis that 0,
Aggregates
1989:6.
Divisia
=0 2 =
is
...
=G 12
1
1960:2 to 1989:6. For the
,
Divisia aggregates, the
sample
is
1961:2 to
Table 4: Long-Run Effects of Monetary Shocks on Industrial Production
Traditional
Monetary Aggregates
Estimated Long-Run Effect
^
M1
1.04
(0.44)
M2
1.57
(0.52)
M3
0.53
(0.50)
Monetary Base
1
.42
(0.63)
Money
0.99
Multiplier
(0.61)
0.13
Currency
(0.61)
Divisia
Oivisia
Aggregates
M2
1.31
(0.53)
Divisia
M3
1.15
(0.44)
Currency Eguivalent Aggregates
Currency Equivalent
0.05
(0.10)
Fixed-Weight Currency Equivalent
1
.08
(0.28)
Three-Year Currency Equivalent
0.17
(0.16)
Contemporaneous Currency Equivalent
-0.10
(0.09)
Notes: Each entry reports the long-run effect of a unit shock to the monetary aggregate on the
logarithm of industrial production. The estimates are computed as f05,K) = (1-/?(U)'V(U evaluated at
L= 1, where the lag polynomials are estimated in
A Iny, = a + 6 1+ p (Z.) A lnyr+ y (L) A \nm +e y
t
t
.
Standard errors, which are shown in parentheses, are calculated as in Poterba, Rotemberg and
Summers(1986) as Idf/dfi df/dy)*Vi0,Y)*ldf/d/3 dt/dy)' where V(0,k) is the variance-covariance matrix
of the estimated parameters.
Table 5: Causality Tests and Long-Run Effects: Monetary Aggregates and Prices
Traditional Aggregates
Ml
Causality
.241
Long-Run Effect
2.42
(2.02)
M2
.394
0.86
(0.87)
M3
.693
0.95
(0.70)
Monetary Base
.003
2.49
(1.06)
Money
Multiplier
.439
1
.75
(2.35)
Currency
.159
1.52
(0.55)
Divisia
Divisia
Aggregates
M2
.802
0.94
(1.07)
Divisia
M3
.640
1.21
(1.34)
Currency Equivalent
Aooreoates
Currency Equivalent
.002
0.62
(0.38)
Fixed-Weight Currency
.141
Equivalent
Three-Year Currency
.002
Equivalent
Contemporaneous Currency
3.71
(7.68)
0.81
(0.42)
.046
Equivalent
0.31
(0.17)
Alternative Indicators of
Monetary Policy
Commercial Paper
Federal Funds Rate
Commercial Paper-Treasury
Bill Yield Spread
.0004
<. 00001
.666
Notes: Entries in column 2 are p-values for Granger-Causality tests of the sort reported in table 2,
while entries in column 3 are long-run responses of the price level to shocks in the growth rate of
the monetary aggregates, whose analogue for industrial production was reported in table 3.
Standard errors are in parentheses.
Billions of
Current Dollars
H
I
m
O
c
a
s
m
z
5
s
5
S
Q
3D
rn
s
CD
PERCENT RESPONSE
"0
CD
-cm
0(53
?cc
pi
CO
3
PERCENT RESPONSE
i
p
p
fo
fo
p
p
O
CO
->•
ro
o>
OD
S3
1°
fo
to
-fe.
"0
CD
o
c
m
30
0) jo
(A
>
Z CD
m
|8 c
-9 ozH
JO2
I
CO fO
cr
"Tl
CD
3
PERCENT RESPONSE
I
o
N3
o
o
to
04
o
o
01
—
ojH
*.—
—OH
>H
CO —
01
{£>
~
o~
_l
-• 04
^^
-I Nj"
C0~
o
to
fo*3^
roid_
C CD
fO_
04 .=±1
ro 04
04 0j_
04
04
A
or
o»tr
04 Vl"
00 04
t>.
tO
0"5~]
is C
5
m
O
CO
0)
33
W
js.
CD
Js.
tour
tni±_
*»> 01
01 SZL.
Up
O o>_
>
is*
O
2
5i
®
!
M
m o)
CO
o
mO
D
C
3o
"0
o
(D
o
CD
ro
C7>
"TJ
CD
3
PERCENT RESPONSE
o>
Ol
03
ro
>
8* c
"0
r-
O co
2LC m
0)
o
2
E! "n
5> Sgoo
oi
z
co ro
m Q.
-n
c
z
CO 3 o
H
S o
o
z
a z
o
O 33
o O
D
C
co
CD
en -±_
k>T/T
*> cn_
cnic_
a>
PERCENT RESPONSE
CO
00
en
en
03
ro
-*-
o
8
03
»
O
no
C
-oz r3 Cm
co
'
33
si
°i
z•
CV)
£o m
S° c
Tl
s
z
go
H
o
si
Q.
(A
31
0J
CD
z
ro
cd
PERCENT RESPONSE
TJ
<D
3
0)
3
<d
3
TJ
C
CO
m
J
5"
33
a
°
=•0
3
ii
TJ CO CO
cr
a
o
(D
3
mo
-n
c
i
O
PERCENT RESPONSE
p
b
o
o
p
b>
-o
CD
2
3
3
CO
CD
cen
m
3
m
CO,?
I
g,
3
o
o)
O
% "n
rO
3 TJ
O"
m
CX
«<
CD
-i
O
CD
3
O_
z
mc
-n
c
z
3
o
z
MIT LIBRARIES
3
TDflD
DD?n5Sb