bernanke-measuring monetary policy
TRANSCRIPT
-
8/7/2019 Bernanke-Measuring Monetary Policy
1/34
-
8/7/2019 Bernanke-Measuring Monetary Policy
2/34
and Romer [1989] reintroduced t he na rra tive a pproach to t he
study of monetary policy. Based on a reading of the minutes of the
Federal Open Market Committee, Romer and Romer determined
a set of dates at which policy-makers appeared to shift to a more
a n t i -i n fla t i on a r y s t a n ce . A n a p p ea l in g a s p ect of t h e R om e r s
approach is that i t uses additional informat ionspecifically,
p ol icy -m a k e r s s t a t e m en t s of t h e ir ow n i n t en t i on s t o t r y t o
disentangle money supply from money demand shocks.Adisadvan-
ta ge of th is appr oach, besides inh eren t pr oblems of repr oducibility
and subjectivity, is that it does not clearly distinguish between the
endogenous and exogenous components of policy change, which is
necessary for identifying th e effects of moneta ry policy on th e
economy [Dotsey and Reid 1992; Leeper 1993; Sha piro 1994;
Hoover a nd Perez 1994; Sims and Zha 1995; Bernanke, Gertler,and Wat son 1997]. The Romers meth odology also yields a rat her
limited amoun t of inform at ion: th ey give dat es only for contra ction-
ary changes in policy, not expansionary shifts, and their method
provides no indication of the severity or duration of each episode.
Building on the Romers work, Boschen and Mills [1991] u sed
FOMC documen ts t o ra te m oneta ry policy in each m onth as very
tig h t, t igh t, n eu tr a l, ea sy, or ver y ea sy, dep en din g on t h e
relat ive weights the policy-maker s assigned to redu cing unem ploy-
ment an d redu cing inflat ion. Although Boschen and Mills providea more continuous and possibly m ore informat ive measure of
policy th an do Romer an d Romer, their in dicator likely also suffers
relatively m ore severe problems of subjectivity an d commingling
of endogenous a nd exogenous policy chan ges.
The second general strategy for measuring monetary policy
stancewhich is the focus of the present articleis to use prior
information about centr al ban k operat ing procedures, in conjunc-
tion with vector autoregression (VAR) estimation techniques, to
develop data-based indexes of policy. For example, Bernanke andBlinder [1992] argu ed tha t over much of th e past 30 years th e Fed
has implemented policy changes through changes in the federal
funds rate (the overnight rate in the market for commercial bank
reserves). They concluded that the funds rate may therefore be
used a s a n indicator of policy stance (see also Laur ent [1988] an d
Bernanke [1990]); in particular, they interpreted VAR innovations
to the funds r at e as innovations to the Feds policy. In a similar
vein, Sims [1992] used short-term r at es as m onetary indicators in
a m ulticoun try st udy. However, not a ll researchers working in t heVAR-based l i terature have a dopted sh ort-term interest r ates as
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS87 0
-
8/7/2019 Bernanke-Measuring Monetary Policy
3/34
their preferred indicator of policy. Following a suggestion of
Thornton [1988b], Christian o and E ichenba um [1992] ha ve made
the case for using the quantity of nonborrowed reserves as the
prim ar y mea sur e ofm oneta ry policy (also see Eichen bau m [1992]).
Strongin [1995] proposed as a policy indicator the portion of
nonborrowed reserve growth that is orthogonal to total reserve
g r o w t h . H e m o t i v a t e d t h i s m e a s u r e b y a r g u i n g t h a t t h e F e d i s
constrained to meet total reserve demand in the short run but can
effectively tighten policy by r educing nonborrowed reserves and
forcing banks to borrow more from the discount window. Cosi-
mano and Sheehan [1994] characterized Fed policy after 1984 as
borrowed-reserves tar geting, which suggests tha t borrowed re-
serves might be a useful indicator for the more recent period.
Both the narr ative and VAR-based methods for measuring
monetary policy have been widely used in applied work.2 Unfortu-
na tely, th ere is evident ly little agreement on which of the var ious
measures most accurately captures the stance of policy, leading
man y au thors t o hedge by using a variety of indicators. Eichen-
baum and Evan s[1995] study of the effect of moneta ry policy on
exchange rates is fairly typical in employing three alternative
policy mea sur es: in their case, Strongins m easur e, innovations to
the federal funds rate, a nd the Romer dates. However, although
using altern at ive measu res allows the resear cher t o claim robust-
ness when the results for each indicator are similar, this strategy
provides no guidance for cases when the results for different
indicators are inconsistent. (Indeed, we show below that alterna-
tive indicators can lead to quite different inferences.) Moreover,
simply using a variety of altern at ive mea sur es of monetar y policy
cannot guarantee that some more accurate indicator has not been
excluded; that the best indicat or is not perh aps some combinat ion
of the var ious pur e indicat ors; or t ha t the best indicator is th e
same for all countries or for all periods. Thus, it would be quite
u s e fu l t o h a v e a s ys t e m a t ic m e t h od of com p a r i n g a l t er n a t i ve
candida te indicators of policy.
Eichenbaum [1992, p. 1010] has stressed th e importan ce of
2. There a re dozens of examples. Asa mpling of better known studies includesKashyap, Stein, a nd Wilcox [1993]; Friedma n and Kuttn er [1993]; Ramey [1993];Gertler and Gilchrist [1994]; Cochrane [1994]; Eichenbaum and Evans [1995];Cecchetti [1995]; an d Burnside, Eichenbaum, and Rebelo [1995]. Applications tocount ries other t han the United St ates, which use the VAR methodology almostexclusively, include Sims [1992]; Cushman and Zha [1994]; Gerlach and Smets[1995]; Kim and Roubini [1995]; Tsatsaronis [1995]; Barran, Coudert, and Mojon[1996]; and Clarida and Gertler [1997], among others.
M E A S U R I N G M O N E T A R Y P OL I C Y 87 1
-
8/7/2019 Bernanke-Measuring Monetary Policy
4/34
finding a means of choosing among indicators, noting that in his
part icular application inference depends very sensitively on
which of th e two can didate mea sures [short-term inter est ra tes or
nonborrowed reserves] we work with. He also suggests that
furth er progress on t hese issues can be ma de only by carefully
s t u d yi n g t h e i n st i t u t ion a l d e t a il s of h ow m on e t a r y p ol icy i s
actua lly car ried out in th e different coun tries. . . . Following
Eichenbaum s su ggestion, in t his a rt icle we develop a nd imple-
ment a general, VAR-based methodology in which the indicator of
monetary policy stance is not assumed but rather is derived from
an estimated m odel of the central banks operating pr ocedure.
More specifically, we employ a semi-str uctu ra l VAR m odel t ha t
leaves the relationships among macroeconomic variables in the
system unrestricted but imposes contemporaneous identification
r e s t r ict i on s on a s e t of v a r i a bl es r e le va n t t o t h e m a r k e t for
commercial ba nk reserves.
O u r m e t h od h a s s ev er a l a d va n t a g es ov er p r e vi ou s a p -
proaches. First , because our specification nests the best known
quantitative indicators of monetary policy used recently in VAR
modeling, including all those mentioned above, we are able to
perform explicit st atistical compa risons of these an d other poten-
tial measures, including hybrid measures that combine the basic
indicators. Second, our analysis leads directly to estimates of a
new policy indicator that is optimal, in the sense of being most
consistent with the estimated param eters describing t he central
banks operating procedure a nd the market for bank reserves.
Third, by estima ting t he model over different sam ple periods, we
are able to allow for changes in the structure of the economy and
in operating procedures, while imposing a minimal set of identify-
ing assumptions. Finally, although we consider only the post-1965
U . S . ca s e in t h is p a pe r, ou r m e t h od is a p pli ca b le t o ot h e r
countr ies and periods, and to alterna tive institu tiona l setups.3
A frequent ly hea rd criticism of the VAR-based appr oach is
that it focuses on monetary policy innovations rather than on the
argu ably more importa nt systemat ic or en dogenous component of
policy. We believe this criticism to be misplaced. The emphasis of
th e VAR-based app roach on policy innovat ions a rises n ot becau se
shocks to policy are intrinsically important, but because (as we
d is cu s s fu r t h e r b el ow ) t r a ci n g t h e d yn a m i c r e s p on s e of t h e
3. Bernanke a nd Mi hov [1997] apply t hese met hods t o a st udy of Germanmonet ar y policy.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS87 2
-
8/7/2019 Bernanke-Measuring Monetary Policy
5/34
economy to a monetar y policy innovation provides a means of
observing the effects of policy changes under minimal identifying
assumptions.4 However, although we disagree with the view that
an alysis of policy shocks is unin ter estin g, we also recognize th at it
w ou l d b e u s e fu l t o h a v e a n i n di ca t or of t h e ov er a l l s t a n c e o f
monetary policy, including the endogenous component. An addi-
t i on a l con t r i bu t i on of t h i s a r t i cl e i s t o p r op os e ju s t s u ch a n
indicator, one t hat is both consistent with our underlying a p-
proach and, we believe, intuitively appealing.
The rest of the article proceeds as follows. Section II briefly
describes our general meth odology for identifying innovations t o
monetary policy. Section III lays out a standa rd model of the
market for bank reserves that nests some common alternative
descriptions of F ed operating procedures. E stimation of t his
model by GMM (Section IV) allows us both to evalua te t he lead ing
candidate indicators of policy innovations and to develop an
alternative measure. Section V discusses how the choice of policy
measure affects our conclusions about the impact of monetary
policy on th e economy. Section VI int roduces our total policy
measure, inclusive of both the systematic and random compo-
n e n t s of p o li cy, a n d com p a r e s i t w it h n a r r a t i v e m e a s u r e s of
monetary policy. Section VII concludes.
II. METHODOLOGY
Berna nke an d Blinder [1992] proposed t he following str ategy
for measuring the dynamic effects of monetar y policy shocks.
Su ppose th at th e tr ueeconomic str uctu re is
(1) Yt i0
k
B iYti i0
k
Cip ti Ayv t
y
(2) p t i0
k
D iYti i1
k
g ip ti v tp.
Equa tions (1) an d (2) define an un restr icted linear dynamic model
that allows both contemporaneous values and up to k lags of any
variable to appear in an y equation.5 Boldface letters are used to
4. Bernan ke, Gert ler, a nd Watson [1997] provide a VAR-based method forestimat ing th e effects of systematic or en dogenous policy changes; see also Simsand Zha [1995].
5. Expectations variables a re n ot explicitly included in (1)(2), but th ese canbe accommodated by replacing expected future values of variables occurring in the
M E A S U R I N G M O N E T A R Y P OL I C Y 87 3
-
8/7/2019 Bernanke-Measuring Monetary Policy
6/34
indicator vectors or matr ices of variables or coefficients. In
particular, Y is a vector of macroeconomic variables, and p is a
variable indicating the stan ce of policy. Note t ha t for th e moment
p i s t a k e n t o b e a s ca l a r m e a s u r e , e .g ., t h e fe d er a l fu n d s r a t e .
Equation (2) predicts current policy st ance given current and
lagged values of macroeconomic variables and lagged policy
variables, while equa tion (1) describes a set of structur al relat ion-
ships in th e r est of the economy. The vector v y and the scalar v p
ar e m ut ua lly uncorrela ted prim itiveor str uctu ra lerr or t erm s.
As in Bernanke [1986], the structural error terms in equation (1)
are premultiplied by a general matrix Ay , s o t h a t s h o c k s m a y
enter into more than one equation: hence the assumption that the
elements of v y a r e u n cor r e la t e d i m pos e s n o r e s t r ict i on . T h e
assumption tha t the policy sh ock v p i s u n cor r e la t e d w it h t h e
elements of v y is also not restrictive, in our view; we think of
independence from contemporaneous economic conditions as part
of th e defin ition of an exogenous p olicy sh ock.6
The system (1)(2) is not econometrically identified in gen-
eral. Bernanke and Blinder point out that to identify the dynamic
effects of exogenous policy shocks on the various macro variables
Y, without necessarily having to identify the entire model struc-
tur e, it is su fficient t o assum e t hat policy shocks do n ot affect th e
given macro variables within the current period; i .e., C0 0.7
Under this assumption the system (1)(2) can be written in VAR
format by projecting the vector of dependent variables on k lags of
i t se lf. E s t i m a t ion of t h e r e s u lt i n g s ys t e m b y s t a n d a r d VAR
methods, followed by a Choleski decomposition of the covariance
mat rix (with t he policy variable ordered last) yields an estimat ed
series for the exogenous policy shock v p (s e e B e r n a n k e a n d
Blinder). Impulse r esponse functions for all variables in the
system with respect to the policy shock can then be calculated and
ca n b e i n t e r p r et e d a s t h e t r u e s t r u ct u r a l r e s pon s e s t o p ol icy
shocks.
model by t he l inear proj ect ions of t hese expect at i ons on current and l aggedvariables in t he system . Altern atively, think of some of the component equations of(1)(2) as reduced-form decision rules relating choice variables to observable statevariables.
6. The idea of an exogenous policy shock ha s been criticized as implying thatthe Fed ran domizes its policy decisions. Although t he Fed does not explicitlyrandomize, it seems reasonable to assert that, for a given objective state of theeconomy, many random factors affect policy decisions. Such factors include thepersonalities an d intellectua l pr edilections of the policy-mak ers, politics, da taerrors a nd r evisions, an d various technical problems.
7. This assu mpt ion is not plau sible for all macro var iables, notably for var ioustypes of ass et p rices. See footn ote 18 below.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS87 4
-
8/7/2019 Bernanke-Measuring Monetary Policy
7/34
The Bernanke-Blinder method assumes that a good scalar
mea sur e of policy is available. However, it ma y be the case th at we
have only a vector of policy indicators, P , which conta in informa -
tion about the stan ce of policy but are affected by other forces a s
well. For exam ple, if th e Feds operat ing procedur e is neith er pu re
i n t er e s t -r a t e t a r g et i n g n or p u r e r e s er v es t a r g et i n g, t h e n b ot h
in t e r e st r a t e s a n d r e se r ve s w il l con t a i n i n for m a t i on a b ou t
monetary policy; but in t hat case, both variables m ay a lso be
affected by shocks to the demand for reserves and other factors. In
t h i s m or e g en e r a l ca s e t h e s t r u ct u r a l m a cr oe con om i c m o d el
(1)(2) may be written as
(3) Yt i0
k
B iYti i1
k
C iP ti Ayv t
y
(4) P t i0
k
D iYti i0
k
GiP ti Apv t
p.
Equa tion (4) states t ha t t he set of policy indicators P depend
on cu r r e n t a n d l a gg ed v a lu e s of Y a n d P , a n d on a set of
disturbances v p . We assu me th at one element of th e vector v p is a
money su pply shock or policy distu rba nce vs; the other elements of
vp
m a y i n cl u de s h ock s t o m on e y d e m a n d or w h a t e ve r d is t u r -bances affect the policy indicators. Equation (3) allows the non-
policy var iables Y to depend on curr ent a nd lagged values ofY a n d
on lagged values (only) ofP ; allowing t he nonpolicy var iables to
depend only on lagged values of policy variables (C0 0 ) i s
an alogous t o the identifying assu mption ma de above in t he scalar
case.
As in th e case of a scala r policy indicator, we would like t o find
a w a y t o m e a s u r e t h e d yn a m i c r e s pon s e s o f va r i a b le s i n t h e
system to a policy shock vs . As before, we can rewrite t he system
(3)(4) in VAR form (with only lagged variables on the right-hand
side) and estimate by stan dard methods. Let u tp be th e portion of
th e VAR residuals in t he policy block t ha t are orthogona l to th e
VAR residuals in the nonpolicy block. Then straightforward
calculation shows th at u tp satisfies
(5) u tp (I G0 )
1Apv tp,
where t he var iables on t he right-ha nd side of (5) are as defined in(4). Alternatively, dropping subscripts and superscripts, we can
M E A S U R I N G M O N E T A R Y P OL I C Y 87 5
-
8/7/2019 Bernanke-Measuring Monetary Policy
8/34
rewrite (5) as
(6) u Gu Av.
Equation (6) is a standa rd structural VAR (SVAR) system,
which relates observable VAR-based residuals u to unobserved
structural shocks v , one of which is the policy shock vs . T h i s
system can be identified and estimated by conventional methods,
allowing recovery of the structural shocks, including vs . Th e policy
shock vs is ana logous t o the innovation to the federal fun ds ra te in
the scalar case analyzed by Bernanke and Blinder [1992]. As in
the scalar case, the structural responses of all variables in the
s ys t e m t o a p ol icy s h ock ca n b e m e a s u r e d b y t h e a s s oci a t ed
impulse response functions. Furt her, th e historical sequence of
policy shocks can be recovered from t he VAR residua ls by mea ns of
(6). In the remainder of this article we apply this approach to the
measu rement of U. S. monetar y policy sta nce since 1965.
III. MONETARY P OLICY AND THE MARKET FOR BANK RESERVES
To implement the methodology described in Section II, we
need a specific model r elating t he VAR residuals and the struc-
tural shocks in the policy block. We employ a standard model of
the market for commercial bank reserves and Federal Reserve
operat ing procedures. 8 Although simple, this model is rich enough
to n est all t he VAR-based policy indicators mentioned in the
introduction, as well as other plausible measur es.
Continuing to use u to indicate an (observable) VAR residual
a n d v to indicate an (unobservable) structural disturbance, we
a s s u m e t h a t t h e m a r k e t f o r b a n k r e s e r v e s i s d e s c r i b e d b y t h e
following s et of equat ions:
(8) u T R u FF R vd
(9) uBR (u FF R uDISC) vb
(10) u NBR dvd bvb vs.
Equa tion (8) is th e ban kstotal dema nd for r eserves, expressed in
innovation form. It states tha t the innovation in th e demand for
total reserves u T R depends (negatively) on the innovation in the
federal funds rate u FF R (the price of reserves) and on a demand
8. Similar models are estimat ed by Brunner [1994] and Gordon and Leeper[1994], am ong man y others. A stochast ic general equilibrium model of the federalfunds market is provided by Coleman, Gilles, and Labadie [1993].
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS87 6
-
8/7/2019 Bernanke-Measuring Monetary Policy
9/34
disturbance vd. E quation (9) determines th e portion of reserves
t h a t b a n k s ch oos e t o b or r ow a t t h e d is cou n t w in d ow. A s i s
conventional, the demand for borrowed reserves (in innovation
form), uB R , is taken to depend positively on the innovation in the
federal funds rat e u FF R (the rate at which borrowed reserves can
be relent) and negatively on the discount rate u D ISC (the cost of
borrowed reserves); vb is a disturba nce to the borrowing fun ction.9
T h e i n n ov a t ion i n t h e d e m a n d for n on b or r ow ed r e s er v es , t h e
difference between total and borrowed reserves, is u T R uBR .
Equa tion (10) describes th e beha vior of the Federa l Reserve.
We assume that the Fed observes and responds to shocks to the
t ot a l d em a n d for r e se r ve s a n d t o t h e d em a n d for b or r ow ed
reserves within t he period, with t he str ength of the response given
by th e coefficient s
d
a n d
b
. T h a t t h e F e d ob s er v es r e s er v edeman d shocks with in th e period is rea sonable, since it monitors
total reserves (except vault cash) and borrowings continuously.
However, the case in which the Fed does not observe (or does not
respond to) one or t he other of th ese disturban ces can be accommo-
dat ed by sett ing th e relevant coefficients t o zero. The distu rban ce
term v s is th e shock t o policy th at we ar e inter ested in ident ifying.
Note that the system (8)(10) is in the form of equation (6).
It will also be useful to write the reduced-form relationship
between the VAR residuals u and the structural disturbances v ,a s i n e qu a t i on (5 ). To d o s o, w e fi r s t m a k e t h e s im p li fy in g
assu mpt ion th at th e innovation to the discount ra te u D ISCis zero.10
To solve th e model, we impose the condition that the supply of
n on b or r ow ed r e s er v es p lu s b or r ow in g s m u s t e qu a l t h e t ot a l
demand for reserves. Solving in terms of innovations to t otal
reserves, nonborrowed reserves, and the federal funds rate, we
9. Various sanctions and rest rictions imposed by the F ed on banksuse of thediscount window make the true cost of borrowing greater than the discount rate;hence, banks do not at t empt t o borrow i nfini t e quant i t i es when t he funds rat eexceeds t h e di scount rat e. A borrowi ng funct ion of t he form of (9) i s used i nstandard Federal Reserve models of money markets (e.g., Tinsley et al. [1982]).Goodfriend [1983], P eristiani [1991], and Clouse [1994] explore the empiricalrobustness of the borrowing function along various dimensions.
10. We make t hi s assumpt i on t o conform wi t h t he previ ous st udi es bei ngexamined, all of which ignore the discount rate. The discount rate, which is ani nfrequent l y changed admi ni st ered rat e, may al so not be wel l model ed by t hel inear VAR framework. An a l t ernat i ve t o assumi ng t hat t he i nnovat i on t o t hedi scount rat e i s zero, but whi ch has essent i al l y t he same effect , i s t o t reat t hediscount r ate innovation a s pa rt of the innovation to th e borrowings function. Forestimat es of the m odel with a n onzero discount ra te innovation, see Bernan ke an d
Mihov [1995] or Bernanke and Mihov [1997] (for Germany); their results are quiteconsistent with those reported here.
M E A S U R I N G M O N E T A R Y P OL I C Y 87 7
-
8/7/2019 Bernanke-Measuring Monetary Policy
10/34
have
(11) u (I G)1Av,
where
u [u T R u NBR u FF R ] v [v d v s v b ]
a n d
(I G)1A 3 1 2 (1 d) 1
1
2 (1 b )d 1 b
1 1 2 (1 d) 1
1 1 2 (1 b )
4 .One can a lso invert t he r elationship (11) to determ ine how the
monet ar y policy shock vs depends on the VAR residuals:
(12) vs (d b )u T R (1 b )u NBR (
d b )u FF R .
The model described by equation (11) has seven unknown
p a r a m et e r s (in clu d in g t h e v a r ia n ce s of t h e t h r e e s t r u ct u r a l
shocks) to be estimated from six covarian ces; hence it is under iden-
tified by one restriction. However, as was noted earlier, this model
nests some pr evious at tempt s to mea sur e policy innovations, each
of which implies additional parameter restrictions. We consider
five alternative identifications of our unrestricted model, corre-
sponding to four indicators of policy proposed in the literature
(each of which implies overidentification) and one just-identified
varian t. These a re a s follows.
Model FFR (federal funds rate). The Bernanke-Blinder as-
sumption t hat the F ed tar gets th e federal funds r ate corresponds
to the pa rametr ic assumptions d 1, b 1; i.e., the Fed fullyoffsets shocks to total reserves demand and borrowing demand.
From (12) we see th at the monetar y policy shock implied by these
restr ictions is vs ( )u FF R ; i.e., the policy shock is propor-
tiona l to the innovation to the federal fun ds ra te, as expected.
Model NBR (nonborrowed reserves). Christiano and Eichen-
baum s assu mption is th at nonborr owed reserves respond only to
policy shocks. In our cont ext th is assu mption implies th e r estric-
tions d
0, b
0 in (10). With th ese rest rictions th e policy shockbecomes vs u NBR .
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS87 8
-
8/7/2019 Bernanke-Measuring Monetary Policy
11/34
Model N BR / TR (orthogonal ized nonborrowed reserves).
Strongins key assumption is tha t shocks to total r eserves ar e
purely demand shocks, which the Fed h as no choice in the short
run but to accommodate (either through open-market operations
or th e discount window). His specificat ion a lso ignores th e possibil-
ity that t he Fed responds to borrowing shocks. Hence th e param et-ric restr ictions imposed by Str ongins model ar e 0, b 0. For
Str ongins model innovat ions t o monet ar y policy ar e given by vs
du T R u NBR .
Model BR. Several authors have provided evidence for bor-
rowed-reserves targeting by the Fed during certain periods (see,
e.g., Cosimano an d Sh eehan [1994]), implying th at th e quan tity of
borrowed reserves is another potential indicator of policy. It is
straightforward to see tha t borrowed-reserves targeting corre-sponds to the restrictions d 1, b /. The implied policy
shock is vs (1 /)(u T R u NBR ), which is pr oportiona l to t he
negative of the innovation to borrowed reserves.
Model J I ( 0, just-identification). Ea ch of th e four models
above imposes two restr ictions an d h ence is overidentified by one
restriction (recall tha t the base model is un deridentified by one
restriction). Tests of these models thus take the form of a test of
t h e ov er i de n t ify in g r e s t r ict i on . An a l t er n a t i ve s t r a t e gy i s t o
e s t im a t e a ju s t -i de n t ifi ed m od e l a n d ch e ck h ow w el l t h e p a -
ra meter estimat es correspond t o the predictions of the altern ative
models. Strongin makes plausible institutional arguments for his
i de n t ify in g a s s u m p t ion t h a t t h e d e m a n d for t ot a l r e s er v es i s
i n e l a s t i c i n t h e s h o r t r u n ( 0). Hence we also consider as a
separate case the just-identified model that imposes only that
restriction.11
IV. DATA, E STIMATION, AND RESULTS
An important practical issue in measuring policy stance is
that the preferred indicator of monetary policy may change over
time, as operating procedures or other factors change. A useful
featu re of our appr oach is th at it can a ccommodate su ch cha nges,
by allowing for changes in the values of parameters (e.g., those
describing the F eds behavior). In th e estimat es present ed in th is
11. Altern atively, the model could be identified by imposing a long-run
restr iction, e.g., tha t monetar y policy shocks have only price-level effects in t helong run; see, e.g., Bernanke and Mihov [forthcoming].
M E A S U R I N G M O N E T A R Y P OL I C Y 87 9
-
8/7/2019 Bernanke-Measuring Monetary Policy
12/34
s ect i on , w e t a k e t w o a l t er n a t i ve a p p r oa ch e s t o d e a li n g w it h
possible regime shifts. First , we present estimates for both the
entire 19651996 sample and various subsamples, with sample
break dat es chosen u sing a combinat ion of historical and st at isti-
cal evidence. Second , we estima te a Ha milton [1989]-style regime-
switching m odel, which ba ses inference about t he da tes a t wh ichFed behavior chan ged entirely on t he dat a. As we show, the results
from th ese two approaches ar e qualitat ively consistent.
B eca u s e ou r id en t ifyi ng a s su m p t ion i s t h a t t h e r e i s n o
feedback from policy varia bles to th e economy with in th e period,12
the length of th e per iod is potentially importa nt. For our first
approach, with fixed sample break dates, we report results based
on monthly and biweekly data (to conserve space, results from the
regime-switching model, below, are reported for monthly data
only).13 Estimates using quarterly data generated qualitativelysimilar conclusions, bu t it is m ore difficult t o defend t he iden tifica-
tion assumption of no feedback from policy to the economy at the
quarterly frequency.14
As we discussed in Section II, our procedure accommodates
the inclusion of both policy var iables an d nonpolicy var iables in
th e VARs. At both frequ encies the policy variables we use a re tota l
bank reserves, nonborrowed reserves,15 a n d t h e f e d e r a l f u n d s
rat e. At th e month ly frequency the nonpolicy variables used were
real GDP, the GDP deflator, and the Dow-Jones index of spotcommodity prices. Real GDP and the GDP deflator were chosen
becau se presu mably th ey are better indicators of broad macroeco-
nomic conditions th an ar e more conventional m onthly indicators
12. As Bernanke and Blinder [1992] discuss, there are actually two alterna-tive timin g assu mpt ions tha t can be u sed for identifying the effects of policy, whichmay be appr opriate u nder different circumsta nces: either th at policy-mak ers ha vecontemporaneous information about the nonpolicy variables (implying that thepolicy variables should be ordered last in the VAR), or that policy-makers know
only lagged values of the n onpolicy variables (implying tha t the policy variablesshould be ordered first).13. Weekly dat a are avai labl e pri or t o t he change i n reserve account i ng
procedures i n 1984, but subsequent l y onl y bi weekly dat a are avai labl e. Forcomparability we report only biweekly results for the whole sample period.
14. However, in relat ed work using only reserves-mar ket dat a, Geweke andRunkle [1995] find that time aggregation from biweekly to quarterly intervals isnot a p roblem for t he ident ification of monet ar y policy.
15. We use n onborrowed reserves plus extended credit, in order t o eliminatethe effects of a bulge of borrowings ass ociated with t he Contin ent al Illinois episodein 1984. To induce stationarity, we normalize total bank reserves and nonborrowedreserves by a long (36-month) moving average of total reserves. This normalizationis preferable to taking logs because the model is specified in levels. Strongin [1995]normalized by a short moving average of total reserves. However, we found that
this procedure creates jerky impulse response functions and does not cleanlysepara te the dynamics of total reser ves and nonborrowed reserves.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS88 0
-
8/7/2019 Bernanke-Measuring Monetary Policy
13/34
like indu strial pr oduction a nd t he CPI. Monthly data for real GDP
and the GDP deflator were constructed by state space methods,
using a l ist of monthly interpolator variables a nd assuming th at
the interpolation error is describable as an AR(1) process (see
Bernanke, Gertler, a nd Watson [1997] for details).16 At t h e
biweekly frequency the nonpolicy variables included the BusinessWeek production index and the index of spot commodity prices
(broader weekly price indices are unavailable).17
The index of commodity prices was included as a nonpolicy
variable in order to capture additional informat ion available to
th e Fed a bout th e futur e cour se of inflation. As is now well-known,
exclusion of th e comm odity price index ten ds t o lead t o th e price
puzzle, the finding th at monetary t ightening leads to a rising
rather than falling price level.18 If the commodity price index is
included t o capt ure th e Feds informa tion about futu re infla tion,then a pa rallel argument suggests putt ing an indicator of futur e
output movement s into the system a s well. For this reason, in our
initial estimation we included t he index of leading indicators
(short horizon) in the quarterly and monthly systems. However,
unlike the case of the commodity price index, inclusion of the
index of leading indicators had little effect on model estimates or
implied impulse response functions. For comparability with ear-
lier results, th erefore, we excluded th at variable when deriving
the estimates presented here.For estimat ion of the m odel with fixed break dat es, we used a
two-step efficient GMM procedure (maximum likelihood est i-
16. Ja mes Stock pointed out to us that t he moving average interpolation errorcreated by the interpolation procedure could in principle invalidate our identifyingassu mption, tha t policy shocks do not feed back to th e economy within t he per iod.As a check for robust ness, we r epeat ed our mont hl y est imat es usi ng i ndust ri alproduct ion i n pl ace of real GDP and t he CPI (l ess shel t er) i n pl ace of t he GDPdeflat or. The resul t ing pa ramet er est i mat es were virt ual l y i dent ical t o t h ose
report ed here. Because t he resul t s usi ng real GDP and t he GDP deflat or yi el di mpul se response funct i ons t hat are more easi l y i nt erpret able and useful forpol icy-making, we cont i nue t o focus on t he est i mat es usi ng t h e i nt erpol at edvariables.
1 7. T h e n o n p ol icy v a r ia b le s a r e m e a s u r e d S a t u r d a y t o S a t u r d a y a n d t h epolicy variables are measured Wednesday to Wednesday. We use values of thenonpolicy variables corresponding to the week lying in the middle of the two-weekreserve account ing period.
18. For further discussion see Sims [1992] and Chr istiano, Eichenbaum , andEvans [1994a, 1994b]. In particular, the lat ter sh ow that the pr ice puzzle is largelyan artifact of not controlling for oil supply shocks. Sims and Zha [1995] argue thatt reat i ng commodit y pri ces as predet ermined for monet ary pol icy shocks i sinappropriate, since these prices may well respond within t he period to moneta rysurpri ses. As a check for robust ness, we reest i mat ed t he j ust -i dent i fied model
allowing commodity prices t o be determined simultan eously with policy inn ova-tions; this cha nge did not significan tly affect our conclusions.
M E A S U R I N G M O N E T A R Y P OL I C Y 88 1
-
8/7/2019 Bernanke-Measuring Monetary Policy
14/34
mat es were similar). The first st ep of the procedure wa s equat ion-
by-equat ion OLS estimat ion of the coefficients of the VAR sys-
tem.19 The second step involved matching the second moments
implied by the pa rticular t heoretical model being estima ted t o the
covar ian ce m at rix of th e policy sector VAR r esidua ls. We per-
for m e d t w o t y pe s of t e s t s of t h e v a r iou s m od e ls : ( 1) t e s t s of overidentifying restrictions based on the minimized value of the
sample criterion function (Ha nsens J t e s t ); a n d (2 ) t e s t s of
hypotheses on the estimates of the structural parameters.
Estimates of the model based on monthly data are given in
Table I , estimates from biweekly dat a in Table II . Each table
reports parameter estimates, with standard errors in parenthe-
s es , for t h e fi ve m od e ls i n t r od u ce d i n S e ct i on I I I. P a r a m e t e r
restrictions associated with each model are indicated in boldface.
The final two columns of Tables I and II show, for each of the fouroveridentified models, (1) a p-value corresponding to the test of
the single overidentifying restriction (OIR); and (2) a p-value for
t h e t w o p a r a m e t e r r e s t r ict i on s of t h e m od e l, con d it i on a l on
maint aining the just -identified model (and hence assuming 0).
P-values greater than 0.05 are shown in boldface, indicating that
the particular model cannot be rejected at the 5 percent level of
significan ce. As indicat ed above, to allow for possible r egime
s w it ch e s , w e p r e se n t e s t im a t e s for t h e e n t ir e s a m p le p e r iod
(1965:11996:12) and for selected subperiods, including 1965:11979:9, 1979:101996:12, 1984:21996:12, a nd 1988:91996:12.20
The subperiods were chosen as follows. First, we made a list of
can didate subperiods corresponding br oadly to periods identified
by Federal Reserve insiders and observers as possibly distinct
operating regimes (see, e.g., Strongin [1995]). For example, 1979:
10, when Chairman Volcker announced dramatic changes in the
operat ing procedure, is a conventional br eak dat e; 1984:2 r eflects
b ot h t h e e n d of t h e Vol ck e r e xp e r im e n t a n d t h e b eg in n i n g of
contemporaneous reserve accounting; and 1988:9 roughly marks
19. To determine th e nu mber of lags in the VAR, for ea ch sample we beganwit h fift een l ags an d kept el imi nat i ng t he l ast l ag a s l ong as i t was st a t i st i cal lyi nsignificant . Thi s procedure l ed t o t he use of t hi rt een l ags i n t he ful l sampl e,eleven lags in the 19651979 subsample, twelve lags in the 19791996 subsample,and seven lags in the 19841996 subsample. For the short 19881996 sample wealso tested lags other tha n th e final one, and sett led on using lags 1 to 6, 8, 10, and11. A si mil ar procedure was fol lowed i n t he biweekly dat a, st art i ng wit h amaximum of 29 lags. Results based on a fixed lag structure of twelve or thirteenlags were very similar to what we report here.
20. Bot h t he reduced-form VAR and t he st ruct ural m odel paramet ers arereestimated within each subsample.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS88 2
-
8/7/2019 Bernanke-Measuring Monetary Policy
15/34
TABLEI
PARAMETERESTIMATESFORALLMODELS(MONTHLY)
Sample
Model
d
b
Tests
ForOIR
Restrictions
underJImodel
196
5:11996:12
FFR
0.0
04(0.0
01)
0.0
12(0.0
01)
1
1
0.
119
0.0
04
NBR
0.0
31(0.0
10)
0.0
14(0.0
01)
0
0
0.0
00
0.0
00
NBR/TR
0
0.0
49(0.0
12)
0.8
28(0.0
61)
0
0.0
33
0.0
20
BR
0.0
04(0.0
01)
0.0
41(0.0
08)
1
/
0.
119
0.0
00
JI
0
0.0
20(0.0
06)
0.8
09(0.0
58)
0.6
36(0.2
74)
196
5:11979:9
FFR
0.0
05(0.0
02)
0.0
15(0.0
03)
1
1
0.
148
0.0
48
NBR
0.0
14(0.0
04)
0.0
56(0.0
05)
0
0
0.0
00
0.0
00
NBR/TR
0
0.0
77(0.0
12)
0.7
76(0.1
06)
0
0.
078
0.0
49
BR
0.0
05(0.0
02)
0.0
67(0.0
10)
1
/
0.
148
0.0
00
JI
0
0.0
28(0.0
11)
0.7
49(0.1
02)
0.6
20(0.3
15)
197
9:101996:12
FFR
0.0
02(0.0
01)
0.0
13(0.0
01)
1
1
0.0
13
0.0
00
NBR
0.0
29(0.0
08)
0.0
14(0.0
01)
0
0
0.0
01
0.0
00
NBR/TR
0
0.0
36(0.0
09)
0.7
25(0.0
76)
0
0.
778
0.
753
BR
0.0
02(0.0
01)
0.0
24(0.0
04)
1
/
0.0
13
0.0
01
JI
0
0.0
41(0.0
21)
0.7
30(0.0
79)
0.0
40(0.1
28)
198
4:21996:12
FFR
0.0
07(0.0
05)
0.0
05(0.0
02)
1
1
0.0
41
0.0
31
NBR
0.4
98(0.5
93)
0.0
05(0.0
02)
0
0
0.0
07
0.0
00
NBR/TR
0
0.2
54(0.1
26)
0.8
12(0.0
90)
0
0.
161
0.
202
BR
0.0
07(0.0
05)
0.1
17(0.0
73)
1
/
0.0
41
0.0
02
JI
0
0.0
43(0.0
27)
0.8
10(0.0
78)
0.4
02(0.3
15)
198
8:91996:12
FFR
0.0
21(0.0
05)
0.0
01(0.0
02)
1
1
0.
466
0.
636
NBR
0.1
31(0.0
29)
0.0
05(0.0
02)
0
0
0.0
48
0.0
00
NBR/TR
0
0.1
41(0.1
17)
0.9
04(0.0
35)
0
0.0
01
0.0
00
BR
0.0
21(0.0
05)
0.4
62(1.3
74)
1
/
0.
466
0.0
00
JI
0
0.0
02(0.0
05)
0.9
84(0.0
33)
0.9
80(0.0
71)
Theestimatescomefromasix-variablemonthlyVAR(seetextforexplanations).Thenext-to-the-lastcolumnpresentsp-valuesfromtestsofoveridentifyingrestrictionsbasedonthe
mini
mizedvalueofthecriterionfunction.T
helastcolumngivesp-valuesfromtests
oftheimpliedrestrictionsunderthejust-identifiedmodel(
0).Inthelasttwocolumnsthevalues
inbo
ldfaceindicatethattherestrictionsimp
liedbytheparticularmodelcannotberejectedatthe5percentlevelofsignificance.T
hefiguresinparenthesesaresta
ndarderrors.
M E A S U R I N G M O N E T A R Y P OL I C Y 88 3
-
8/7/2019 Bernanke-Measuring Monetary Policy
16/34
TABLEII
PARAMETERESTIMATESFORALLMODELS(BIWEEK
LY)
Sample
Model
d
b
Tests
ForOIR
Restrictions
underJImodel
196
7:11996:12
FFR
0.0
03(0.0
01)
0.0
13(0.0
02)
1
1
0.
185
0.
110
NBR
0.0
63(0.0
18)
0.0
15(0.0
02)
0
0
0.0
00
0.0
00
NBR/TR
0
0.0
95(0.0
22)
0.9
12(0.0
47)
0
0.
147
0.
117
BR
0.0
03(0.0
01)
0.0
87(0.0
19)
1
/
0.
185
0.0
01
JI
0
0.0
30(0.0
16)
0.9
07(0.0
45)
0.6
18(0.3
94)
196
7:11979:9
FFR
0.0
05(0.0
02)
0.0
17(0.0
03)
1
1
0.
361
0.
212
NBR
0.0
12(0.0
04)
0.0
71(0.0
06)
0
0
0.0
00
0.0
00
NBR/TR
0
0.1
32(0.0
20)
0.8
70(0.0
81)
0
0.
087
0.
069
BR
0.0
05(0.0
02)
0.1
30(0.0
19)
1
/
0.
361
0.0
00
JI
0
0.0
34(0.0
21)
0.8
65(0.0
82)
0.7
27(0.4
00)
197
9:101996:12
FFR
0.0
02(0.0
02)
0.0
13(0.0
02)
1
1
0.0
21
0.0
00
NBR
0.0
71(0.0
29)
0.0
12(0.0
03)
0
0
0.0
00
0.0
00
NBR/TR
0
0.0
84(0.0
29)
0.8
50(0.0
61)
0
0.
944
0.
940
BR
0.0
02(0.0
02)
0.0
47(0.0
14)
1
/
0.0
21
0.0
48
JI
0
0.0
90(0.1
09)
0.8
50(0.0
62)
0.0
11(0.1
43)
197
9:101982:10
FFR
0.0
02(0.0
01)
0.0
11(0.0
03)
1
1
0.
102
0.0
13
NBR
0.0
00(0.0
01)
0.0
61(0.0
23)
0
0
0.
627
0.
905
NBR/TR
0
0.0
58(0.0
22)
0.1
39(0.3
90)
0
0.
657
0.
722
BR
0.0
02(0.0
01)
0.0
47(0.0
16)
1
/
0.
102
0.
075
JI
0
0.0
39(0.0
31)
0.1
19(0.3
98)
0.1
44(0.4
05)
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS88 4
-
8/7/2019 Bernanke-Measuring Monetary Policy
17/34
198
4:21996:12
FFR
0.0
03(0.0
05)
0.006(0.0
03)
1
1
0
.097
0.
059
NBR
0.5
02(0.3
87)
0.010(0.0
02)
0
0
0
.003
0.0
00
NBR/TR
0
1.333(2.4
20)
0.8
98(0.0
63
)
0
0
.596
0.
737
BR
0.0
03(0.0
05)
0.261(0.1
91)
1
/
0
.097
0.
215
JI
0
0.123(0.2
15)
0.8
94(0.0
63
)
0.1
51(0.4
50)
198
8:91996:12
FFR
0.0
07(0.0
07)
0.011(0.0
02)
1
1
0
.226
0.
172
NBR
0.0
19(0.0
30)
0.073(0.0
50)
0
0
0
.000
0.0
00
NBR/TR
0
0.079(0.0
13)
1.0
37(0.0
31
)
0
0
.300
0.
198
BR
0.0
07(0.0
07)
0.076(0.0
13)
1
/
0
.226
0.0
19
JI
0
0.027(0.0
15)
1.0
48(0.0
26
)
0.5
56(0.4
32)
Theestimatescomefromafive-variablebiweeklyVAR(seetextforexplanations).
Thenext-to-the-lastcolumnpresen
tsp-valuesfromtestsofoveridentifyingrestrictionsbasedon
theminimizedvalueofthecriterionfunctio
n.
Thelastcolumngivesp-valuesfrom
testsoftheimpliedrestrictionsunder
thejust-identifiedmodel(
0).Inthelasttwocolumnsthe
valu
esinboldfaceindicatethattherestrictionsimpliedbytheparticularmodelcan
notberejectedatthe5percentlevelofsignificance.T
hefiguresinparentheses
arestandarderrors.
M E A S U R I N G M O N E T A R Y P OL I C Y 88 5
-
8/7/2019 Bernanke-Measuring Monetary Policy
18/34
t h e b eg in n i n g of t h e G r ee n s pa n r e gi m e, e xcl u di n g t h e s t ock
m a r k et cr a s h a n d it s a ft e r ma t h . We t h e n ch e ck e d w h et h e r
i n st a b il it y of t h e s t r u ct u r a l p a r a m e t e r s cou l d b e s t a t i st i ca l ly
rejected for adjacent candidate subperiods (on this basis we chose
not to report results separately for 19651972 and 19721979,
two subsam ples suggested by Str ongin). Finally, using t he t est fors t r u c t u r a l c h a n g e w i t h u n k n o w n b r e a k d a t e s p r o p o s e d b y A n -
drews [1993], we confirmed that our imposed break dates are not
seriously inconsistent with the data (in particular, the Andrews
test found evidence for a chan ge in the structural param eters in
1980 and in 1988 or 1989, corresponding closely to our imposed
breaks in 1979 and 1988).
A useful starting point for reviewing Tables I and II is the
estimat es ofd, th e coefficient th at describes th e Feds pr opensity
to accommodate reserves demand shocks (see equation (10)).Estimates of this coefficient are obtained under two of the five
i de n t ifi ca t i on s , n a m e l y N B R /T R a n d J I . I n m on t h l y d a t a t h i s
coefficient is always estimated to lie between 0.725 and 1.00, with
high statistical significance, implying something close to full
accommodation of reserves demand shocks (d 1). This result is
inconsistent with th e n onborrowed r eserves (NBR) model, which
a s s u m e s t h a t d 0; consequently, the NBR model is strongly
rejected in all sample periods in the monthly data. The biweekly
dat a (Table II) gener ally show an even grea ter degree of accomm o-dat ion of dema nd sh ocks (values ofd between 0.85 and 1.05), and
hence rejection of the NBR model. The important exception to this
finding occurs in the subperiod 1979:101982:10, during which
accommodation of reserves deman d shocks is estimated to be
quite low, around 0.1, and the restrictions of the NBR model are
far from being rejected. This last result is quite int eresting, since
1 97 9 19 82 w a s t h e on l y p e r i od i n w h ich t h e F e d i n di ca t e d
p u b li cl y t h a t i t w a s u s in g a n on b or r ow ed -r e s er v es t a r g e t in g
p r oce d u r e. We t a k e t h i s cor r e s pon d e n ce of ou r r e s u lt s t o t h econventional wisdom as support for our approach.21
A tenden cy by the Fed t o offset sh ocks in t he reserves ma rket
2 1. A n a l t er n a t i ve , a n d p e r h a ps s h a r p er t e s t o f t h e N B R m o de l c a n b eobtained under the identification d b , which nests the FFR model (d 1),the N BR model (d 0), and combination policies in which the Fed puts weight onboth interest-rat e sm oothing and nonborrowed-reserves smoothing objectives(0 d 1). E st i mat i on under t hi s i dent i ficat i on general l y finds val ues of d
much closer t o 1 tha n to 0, especially in the pre-1979 an d post-1988 subsam ples,consi st ent wi t h t he view t ha t t hose were peri ods i n whi ch t he Fed smoot hed
i nt erest ra t es. Est i mat es for 19791982 usi ng weekly dat a , i n cont rast , find t hevalue ofd close to zero, again confirming the relevance of the NBR m odel to the
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS88 6
-
8/7/2019 Bernanke-Measuring Monetary Policy
19/34
is also indicat ed by th e fact tha t th e Feds r esponse t o borrowing
shocks, b , is in almost all cases estimated to be negative (e.g., the
full-sample estimates under the JI model are 0.64 and 0.62 in
monthly and biweekly data, respectively). However, in general,
the absolute values of the estimates of b a r e s m a l l e r , a n d t h e
standard errors larger, than for d. As we will see momenta rily,this makes it difficult in some cases to choose the preferred model
fr om a m on g F F R , N B R/T R, a n d B R, w h ich a r e d is t in g u is h e d
primarily by their predictions for the value ofb .
Estimates for the slope coefficients a n d ar e a vailable for
all models in the latter case, and for all models except NBR/TR
and JI in the former. Although the magnitudes of estimates of
these coefficients differ across models, they seem reasonably
consistent across subsamples and data frequencies for any given
model. Estimates of are often found to be negative (the wrongs ig n ), a l t h ou g h v er y s m a ll i n m a g n it u d e ; t h i s p r ov id e s s om e
support for the identifying assumption 0 made by the NBR/TR
and JI models. The sign of the estimates of is almost always as
predicted (positive), except for a few models in the most recent
subsample. As a check on the reasonableness of the estimated
ma gnitudes of th ese para meters, n ote that th e liquidity effect (the
effect of a 1 percent exogenous increase in nonborrowed reserves
on th e funds r ate) in our en compa ssing model is 1/( ). So, for
example, in biweekly data for the full sample, the FFR modelimplies a l iquidity effect of a bout 100 basis points. Ham ilton
[1997, p. 94] notes t ha t this nu mber is rema rka bly consistent
with his estimate of the short-run l iquidity effect, obtained by
different methods. Admittedly, however, other identifications of-
ten give smaller liquidity effects, and the size of this effect is not
always sh arp ly ident ified stat istically. Est imation of the liquidity
effect is discussed further in the next section.
What do the r esults in Tables I and II indicate about which
model of th e Feds operat ing procedure, an d hen ce which indicatorof policy innovations, is to be preferred? The strongest message is
th at the F eds procedures a ppear t o have cha nged over time, an d
hence no single model is optimal for the 19651996 time period.
The FFR model is found to do well for the pre-1979 period, as
ar gued by Berna nke an d Blinder [1992], an d it does exceptiona lly
well for the Greenspan era, post-1988, which most Fed watchers
early Volcker period. We th ank an anonymous referee for su ggesting th is altern a-tive identification.
M E A S U R I N G M O N E T A R Y P OL I C Y 88 7
-
8/7/2019 Bernanke-Measuring Monetary Policy
20/34
w ou l d n o t fi n d s u r p r is in g . T h e F F R m od e l a l so a p p e a r s t o b e
(marginally) the best choice for the sample period as a whole,
based on both t he m ont hly an d biweekly results. As noted above,
the NBR model does well for the brief period of th e Volcker
experiment, 19791982, but is otherwise strongly rejected.
Evaluation of t he borrowed reserves (BR) model is moredifficult . I t is well-known that operating procedures based on
targeting the funds r ate a nd on t argeting borrowed reserves are
quite similar in practice (see, e.g., Thornton [1988a]), and indeed,
the tests of overidentifying restrictions (OIR) give essentially
identical results, subperiod by subperiod, for the FFR and BR
models. Formally, the models differ only in that the BR model
predicts b /, while the FFR model predicts b 1; and as
we have noted, b is not always well identified in our data. Much
stronger differentiation between the two models is found underth e just -ident ified model, however, in wh ich we im pose Str ongins
assumption of inelastic reserves demand ( 0 ). U n d e r t h e J I
restrictions the FFR model generally outperforms the BR model
for th e full sample, the pre-1979 period, an d (part icular ly) in th e
post-1988 period. However, i t is interesting that the BR model
does noticeably better than FFR in t he sample of biweekly dat a
beginn ing in 1984. This last r esult is consistent with t he common
assert ion th at borrowed-reserves tar geting was part of the t ran si-
tion from the nonborrowed-reserves targeting of the early Volcker
era to the federal-funds-rate targeting of the Greenspan period
(see Cosiman o and Sheeha n [1994]). Comparison of the biweekly
and monthly results, however, suggests that even in the period
beginn ing in 1984, borrowed-reserves ta rgeting was relevan t only
at higher frequencies.
What of the N BR/TR m odel, proposed by Str ongin [1995]?
This model has th e advan ta get ha t, u nlike th e other identifica-
tions, it tr eats t he degree to which t he F ed accommodat es reserves
demand shocks (d) a s a fr e e, r a t h e r t h a n i m pos e d, p a r a m e t e r.
Also, l ike our JI model, the NBR/TR model assume that 0.
This flexibility is probably the rea son why th e NBR/TR model has
by far t he highest p-values for the 19791996 subsample, a period
that appears to mix several different operating procedures. The
NBR/TR model is also not rejected for the 19651979 period in
monthly data (although it is rejected for the sample as a whole and
for the post-1988 subsample), and it is not rejected for any sample
or subsample in the biweekly dataa good performance.
Overall, our methodology seems to do a plausible job of
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS88 8
-
8/7/2019 Bernanke-Measuring Monetary Policy
21/34
ident ifying the F eds operat ing regime in different p eriods. There
is considerable evidence that this regime has changed over time.
I t is a s t r en gt h of o u r a p pr oa ch t h a t t h e se ch a n g es ca n b e
identified and (at least potentially) accommodat ed in th e constru c-
t i on of i n n ov a t ion s t o m on e t a r y p ol icy, i n con t r a s t t o m os t
VAR-based a pproaches curr ently in t he literat ur e.I n t h e e s t i m a t e s r e p o r t e d i n T a b l e s I a n d I I , w e i m p o s e d
sample break points using prior knowledge of recent monetary
history. An alternative approach is to allow the sample breaks to
be determined ent irely by sta tistical procedures. To do this, we
began by looking for evidence for breaks in the reduced-form
param eters of t he VAR system. Next, we applied Hamiltons
[1989] regime-switching m odel to th e estima tion of the str uctur al
VAR model of the bank reserves m ark et, focusing part icularly on
possible switches in t he pa ram eters describing Fed beh avior.Conceptually, the appropriate test for stability of the reduced-
form VAR system should allow for an arbitrary break point or
points, as in Andrews [1993]. However, such a test applied to our
six-variable, thirt een-lag system as a whole would no doubt have
minimal power, given the large number of estimated parameters.
To increas e power, and t o focus at ten tion on qualita tive cha nges in
the dynamics, we tested for breaks in the coefficients on all lags of
each variable in each equ ation; e.g., we tested a ll thirteen lags of
GDP in the nonborrowed reserves equation simultaneously, allow-ing arbitrary break points.
Since ther e are six equations in th e system, each with six sets
of lagged var iables, our procedure involved 36 sepa ra te tests. We
u s ed t h e L M v a ri an t of t h e t e st p r op os ed b y An d r ew s; t h is
involves calculating the LM statistic for every possible break
p oi n t a n d t h e n com p a r i n g t h e h i gh e s t v a lu e i n t h e s e qu e n ce
against the tabulated crit ical values. Of the 36 tests conducted,
o n l y o n e w a s s i g n i fi c a n t a t t h e 5 p e r c e n t l e v e l , a n d t h e n o n l y
marginally so. We did not consider this to be very strong evidenceagainst sta bility of the r educed-form VAR system, an d so for the
purposes of this exercise we proceeded under the assumption of no
breaks in the reduced-from coefficients of the VAR over the
19651996 sample.22
22. We a lso conducted approximate Andrews-type tests for each completeequation in the VAR, finding no evidence against stability. Stability of the reducedform is also accepted for shorter lag lengths (e.g., three or four lags). Bernanke and
Mihov [forth coming] also fail to r eject full-sample st ability in a similar VARsystem.
M E A S U R I N G M O N E T A R Y P OL I C Y 88 9
-
8/7/2019 Bernanke-Measuring Monetary Policy
22/34
Given the residuals from a VAR estimated for the whole
sample period, our second st ep was to app ly Ham iltons r egime-
switching appr oach to t he (just -identified) structu ral VAR model
of t h e m a r k e t for b a n k r e s er v es . W e f ocu s e d o n t h e t w o p a -
ram eters describing the F eds response t o shocks in th e reserves
m a r k e t , d a n d b . We assum ed tha t th ere are two possible stat es,each char acter ized by a differen t (freely estima ted) combina tion of
the two reaction parameters.23 Note tha t, although we key regime
switches t o cha nges in th e reaction pa ram eters, since the model is
ju s t -i de n t ifi ed , w e m u s t a l low for t h e p os s ib il it y t h a t ot h e r
structural para meters change when t he state changes. Estimates
were obtained using the EM algorithm, derived for this approach
by Ha milton [1990].
Figure I shows the results of this exercise graphically. The
solid l ine in the figure shows the probabili ty that the operating
regime is in the first of the two states at each date (the probabili-
t ies are smoothed in the sense that full-sample information is
used in inferring the state probabilities at each date). The results
are striking: there is extremely st rong evidence of there having
been pr ecisely two r egime switches in t he sam ple periodone in
l a t e 1 97 9, t h e ot h e r d u r in g 1 98 2. T h es e s w it ch e s cor r e s pon d
closely to th e period of th e Volcker exper imen twith n onborr owed-
reserves t argeting (the conventional beginning an d end dates of
the experiment, 1979:10 an d 1982:10, are indicat ed by th e vertical
dash ed lines in Figure I). The estima ted values of the two rea ction
parameters are consistent with this interpretation. During the
19791982 regime the estimated values are d 0.183, b
0.216, suggesting a nonborrowed-reserves targeting regime
with just a bit of concern for interest-rate smoothing. Outside of
t h e 1 97 9 19 82 w in d ow, t h e p a r a m e t e r e s t im a t e s (d 0.863,
b 0.778) suggest just the opposite approach by the Fed, an
interest-rate-focused operating regime with minor attent ion tosmoothing reserves or other aggregates. These r esults (using
23. This specification is the simplest possible. In a previous version of thepaper, we reported resu lts in which we a llowed d a n d b to switch independently,for a total of four possible states. We have also considered a specification in whicht he t wo paramet ers must swi t ch t oget her but t here are t hree, rat her t han t wo,possible states. Both of these var iants ga ve results very similar to what we reporthere. In pa rticular, using the model with independent switches, we found t hat theswi t ches occurred at about t he same t i me, suggest i ng t hat swi t ches i n t he t woparameters are linked; and estimates of the three-state model yielded two states
with similar parameter estimates, suggesting that a two-state model is adequateto capture Fed beha vior over th is period.
Q U A R T E R L Y J O U R N A L O F E C O N O M I C S89 0
-
8/7/2019 Bernanke-Measuring Monetary Policy
23/34
monthly da ta) conform closely to the estimates using biweekly
dat a r eported in Table II.
Summ arizing the results of this section, we conclude first th at
th e common pra ctice of using only one policy indicator for th e
entire 19651996 period in the United Sta tes is not justified. In
particular, there is considerable evidence that the Fed did indeed
switch to targeting nonborrowed reserves during the 19791982
period, as many have claimed. During the rest of the sample, in
con t r a s t , t h e F e d w a s l a r ge ly a ccom m od a t i n g s h ock s t o t h e
deman d for r eserves. Treat ing the federal funds r ate as t he policy
i n d ica t or for t h i s g r ea t e r p or t i on of t h e s a m p le i s t h e r e for e
probably a reasonable approximation, although as we have seen
the Strongin model is also useful, particularly for the post-1979
period.
T h e c om p a r i son s w e h a v e d r a w n i n t h i s s e ct i on t h u s fa r
presuppose a choice among the existing simple indicators. How-
ever, as discussed in Section II, our ana lysis also suggests an
F IGURE I
Structural Parameter Estimates and Smoothed Probabilities from a
Regime-Switching Model of Federal Reserve Operating Procedures
M E A S U R I N G M O N E T A R Y P OL I C Y 89 1
-
8/7/2019 Bernanke-Measuring Monetary Policy
24/34
alternative strategy, which is to use the just-identified model to
calculate the policy shock in each period which is most consistent
with our est imates of the Feds operating procedur e in th at period
(see equat ion (12)). This st ra tegy mak es it possible to an alyze the
effects of monetary policy, despite differences or chan ges in
regime, in a unified framework. In addition, this approach canaccommoda te hyb rid as w ell as pu reoper at ing pr ocedu res . In
the next section we show that using our framework may substan-
tially affect the inferences one draws about monetary policy.
V. IMPLICATIONS FOR E STIMATED RE S P O N S E S T O MONETARY
P OLICY SHOCKS
As the Introduction discussed, our reason for estimating the
param eters of the model for bank reserves is n ot our interest intha t ma rket , or in th e Feds opera ting procedures, per se. Rath er,
the goal is to isolate relatively clean measu res of monetar y
policy shocks. Given these shocks, standar d impulse response
functions can be used to provide a quantitative measure of the
dynam ic effects of policy changes on th e economy. As we ha ve
noted, many recent st udies have used this appr oach; an d methods
of this type have been used a s an input to monetary policy-ma king
both in the United States and other countries.
Figure II shows estimated dynamic responses of output , th eprice level, an d t he federal funds rat e to a monetar y policy shock,
as der ived from t he a lterna tive models we ha ve been considering.
For compara bility, in each case we consider an expansionary
shock with an impact effect on t he funds ra te of25 basis points.
The left column shows the impulse responses, with 95 percent
confiden ce ban ds, implied by our just-ident ified model.24 The right
pan el of Figur e II sh ows impulse r esponses a s implied by the four
overidentified models (FFR, NBR, NBR/TR, a nd BR). Standa rd
error bands are omitted from the right-hand panels for legibility.
24. Conditional on chosen values for the monetary policy shock, the impulseresponses for t he J I model depend only on t he r educed-form VAR paramet ers a ndt he est i mat ed val ue of . They do not depend on t he paramet ers descri bing t heFeds operat i ng procedure. (Est i mat es of t he l at t er paramet ers are, of course,needed t o calculat e policy shocks from mea sur ed VAR residua ls; see equat ion (12).)Since we cannot reject stability of the reduced-form coefficients for the 19651996sample, and there is little evidence of a shift in , we consider it reasonable to baset he i mpulse responses i n Fi gure II on t he ful l-sampl e est i mat es of t he JI model(Table I). Similar results are obtained if we use parameter estimates from eitherregime of the regime-switching estimation (see Figure I). Bernanke and Mihov
[1995, Figure 9] show impulse responses for t he J I an d alter nat ive models for t hekey subsamples.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS89 2
-
8/7/2019 Bernanke-Measuring Monetary Policy
25/34
F IGURE II
Responses of Output, Pr ices, and t he Feder al Fu nds Rat e to Moneta ry PolicyShock in Alter na tive Models (1965:11996:12)
M E A S U R I N G M O N E T A R Y P OL I C Y 89 3
-
8/7/2019 Bernanke-Measuring Monetary Policy
26/34
Qualitatively, the results from all five identifications are
r e a s on a b le , i n t h e s e n se of con for m i n g t o t h e p r e di ct i on s of
s t a n d a r d m od e ls a n d con v en t i on a l w is d om . I n e a ch ca s e , a n
expansionary monetary policy shock increases output relatively
rapidly (the peak effect is typically at twelveeighteen months),
an d ra ises th e price level more slowly (with relat ively litt le impactin the first year) but more persistently. However, quantitatively,
th e resu lts differ noticeably accordin g to the m eth od of ident ifying
policy shocks. F or example, un der th e NBR/TR identification t he
cumulative response of output to a policy shock of a given size is
estimated to be more than twice as large as found under the NBR
identification. Similarly, at four years the response of prices under
t h e N B R i d e n t ifi ca t i on i s f ou n d t o b e fou r t i m es g r ea t e r t h a n
u n d e r t h e F F R i de n t ifi ca t i on . F u r t h e r , t h e s e a l t er n a t i ve m e a -
sured responses often differ significantly (in both the economicand statistical senses) from those implied by the benchmark JI
model. Similar or greater divergences are found when the re-
sponses implied by the various models are compared for shorter
sample periods (omitted here to conserve space). Overall, these
differences are cert ain ly lar ge enough to ha ve importa nt effects on
the inferences one draws about the effects of monetary policy; and
thus, they underscore the need t o choose a model of the Feds
operat ing procedure tha t is a s nea rly correct a s possible.
H ow i s i t t h a t a l t er n a t i ve i n d ica t o r s of m on e t a r y p ol icy
innovations can ha ve such d ifferent implications for th e calcu-
lated impulse responses? Our analysis provides a simple frame-
work for understanding these differences in results. Let zFFR , z NBR ,
zN B R / T R , and zBR be th e policy shocks im plied by th e four overiden -
tified models, with signs chosen so th at a positive innovation
corresponds to an expansionary policy shock. Suppose that in fact
our just-identified model (with 0) is t rue. Then, u sing equa-
tion (11), we can write the putative measures of monetary policy
shocks implied by t he overident ified models in t erm s of the tr ue
structural shocks, as follows:
(13) zFF R u FF (1 d)vd v s (1 b )vb
(14) z NBR u NBR dvd vs bvb
(15) zN B R / T R u NBR du T R v
s bvb
(16) zBR uB R (u T R u NBR ) (1 d)vd vs bvb.
Note that the alternative policy-shock measures of (13)(16) are
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS89 4
-
8/7/2019 Bernanke-Measuring Monetary Policy
27/34
t h e s a m e ( or in t h e ca s e o f zFFR , proportional to) th e policy
innovations that would be constructed from more conventional
identifications based on the Choleski decomposition (see, e.g.,
Bernan ke an d Blinder [1992], Christiano, Eichenbau m, an d Evan s
[1994b], or Strongin [1995]).
From (13)(16) we see thatif the JI model of this paper iscorrecteach of the putative policy indicators z is contaminated,
i n t h e s en s e of p l a ci n g s om e w ei gh t on r e s er v es d e m a n d or
borrowings shocks. The degree of conta mination depends, of
cours e, on t he values of the pa ram eters. F or exam ple, if d is close
to one (as we have found to be the case, except dur ing the early
Volcker period), t hen th e nonborr owed reserves indicat or z NBRputs considerably more weight on reserves demand shocks rela-
tive to policy shocks than does the federal funds indicator zFF R
(compare equations (13) and (14)). To the extent that z NBR reflectsshocks to reserves demand rath er than to policy, th e associated
impu lse resp onses will not be reliable gu ides t o the effects of policy
innovations. Similarly, the Strongin measure zN B R / T R will be a
good mea sur e of monet ar y policy shocks only to the degr ee th at (or
dur ing periods in which) th e F eds r esponse to borrowing sh ocks,
b , is small (see equation (15)). Equations (13)(16) also imply
th at t he varian ces of th e structur al shocks play an importa nt r ole,
e.g., if the va rian ce of the borr owings sh ockvb is sufficiently small,
the Strongin measure may be a robust indicator of policy shockseven if b is not close to zero. Of course, conditional on our
estimated model, the least contaminated impulse responses are
th ose shown in the left pa nel of Figure II.
As another application of our approach to th e a nalysis of
empirical impulse responses, consider the recent debate on the
van ish ing liquidit y effect.A nu mb er of econom ist s, usin g non bor-
rowed r eserves a s an indicator of policy, h ave found tha t the
liquidity effectthe impact of a given increase in nonborrowed
reserves on the interest ratehas become much smaller or evendisappeared since 1982 [Pagan and Robertson 1995; Christiano
1995]. If correct, t his fin ding ha s importa nt pra ctical implicat ions
for policy-making. However, our approach suggests that this
result is largely due to the bias associated with using nonbor-
rowed reserves as the policy indicator.
To u n d e r s t an d t h is b ia s in m or e d et a il, n ot e fi r st t h a t ,
a ccor d in g t o t h e m od e l d e v el op e d i n t h e p r e se n t p a p er , t h e
ma gnitude of the liquidity effect is given by th e (3,2)-element in
t h e m a t r i x (I G)1A, which is 1/( ) (see equ at ion (11)).
M E A S U R I N G M O N E T A R Y P OL I C Y 89 5
-
8/7/2019 Bernanke-Measuring Monetary Policy
28/34
However, if nonborr owed reserves a re u sed as th e policy indicator,
and th e interest rat e (say, the federal funds ra te) follows immedi-
ately in t he ordering, th en t he liquidity effect will be measur ed as
the projection (regression coefficient) of the funds rate on nonbor-
rowed reserves. Calculatin g th is projection, we fin d t ha t
(17) Est imated liquidity effect
1
31 bb
2 dd2
(d)2d2 s
2 (b )2b24 ,
where the i2 ar e the varian ces of the st ru ctu ral shocks.
If d b 0 , s o t h a t t h e F e d i s t a r g e t i n g n o n b o r r o w e d
reserves, then the bias term (the second term in the brackets in
equat ion (17)) is zero. However, if d 0 a n d b 0 , a s o u r
estimates nearly always imply, the bias term is negative, i.e., them a g n it u d e of t h e l iq u id it y e ffe ct i s u n d e r s t a t ed b y u s in g t h e
nonborrowed reserves indicator. Indeed, using the param eter
estim at es for t he jus t-identified m odel for 19841996 (Table I), we
evaluate the bias term to be 1.035! Hence, if the true liquidity
effect for that period is of reasonable magnitude and the correct
sign, the estimated liquidity effect for the 19841996 period using
the nonborrowed reserves indicator will be small an d of th e wrong
sign. We note th at 19841996 is the only period we ha ve examined
for which t he calculated bias exceeds one in absolut e value (e.g.,the bias for the full sample is 0.777; for 19881996 it is 0.904).
Thus, although the magnitude of the l iquidity effect is always
seriously understated by the NBR model (assuming that the JI
model is true), only in the 19841996 period is its sign actually
revers ed. We conclude th at th e van ishing liquidity effect ma y
well be the result of using a biased indicator of monetary policy,
rat her t ha n of a chan ge in the economy.
The empirical a nalyses of this section are meant only to be
illustrative. N evertheless, we believe tha t they demonstra te thepotential of our method to clarify important debates about the
qua nt itat ive effects of cha nges in m oneta ry policy.
VI. A ME A SU R E O F T H E OVERALL STANCE OF MONETARY P OLICY
Our focus thu s far has been on modeling innovations to the
s t a n ce of m on e t a r y p ol icy, a s op p os e d t o t h e a n t i ci pa t e d or
end ogenous p ar t of policy (the policy ru le). As we h ave d iscuss ed,
the advantage of studying shocks to policy is that it allows us to
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS89 6
-
8/7/2019 Bernanke-Measuring Monetary Policy
29/34
gauge (at least roughly) th e effects of monetary policy on the
economy, with minimal identifying assumptions. In contra st,
empirical analysis of the effects of different monetary rules on the
economy is much more difficult; such an an alysis requires eith er
observations on a large number of monetary regimes, or else a
str uctur al model identified by str ong prior r estrictions.Nevertheless, it would be interesting to have an indicator of
monetary policy sta nce tha t includes the endogenous a s well as
th e exogenous (un forecast able) component of policy. Such an
indicator might be useful in characterizing the overall behavior of
th e Fede.g., the degree to which it a ccommodates various types
of shocksand in providing a general measure of current mone-
tary conditions. Indeed, central banks in a number of countries
curr ent ly use monet ar y condit ions indices, int ended to pr ovide
a s s es s m en t s of ov er a l l t i gh t n e s s or e a s e, i n t h e ir d a y-t o-d a ypolicy-makin g (see, e.g., Freedm an [1994]).
It is not difficult to devise a monetary conditions index, or
measure of overall policy stance, consistent with the framework of
this paper. For example, using the framework and notation of
Section II, consider the vector of variables A1(I G)P . T h i s
vector, which is observable given estimates of the structural VAR
system, is a full-ran k linear combinat ion of th e policy indicators
P , with the property that the orthogonalized VAR innovations of
i t s e le m en t s cor r e s pon d t o t h e s t r u ct u r a l d is t u r b a n ce s v . Inparticular, one element of this vector, call it p, h a s t h e p r o pe r t y
th at its VAR inn ovat ions correspond t o th e moneta ry policy shocks
derived by our appr oach.
I n a n a l og y t o t h e s ca l a r ca s e , i n w h ich t h e r e i s a s in g le
ob s er v a bl e v a r ia b le (e .g ., t h e fu n d s r a t e ) w h os e i n n ov a t ion s
correspond to t he policy shock, one might consider using p a s a n
overa ll measu re of policy. Indeed, u nder th e FF R models res tr ic-
tions, d 1 and b 1, p equals th e funds r ate; similar ly, und er
th e NBR m odels res tr ictions, p equals n onborrowed r eserves, etc.Bernanke and Mihov [1995] show t hat a measure constru cted in
this way correlates well ( in the full sample and in subsamples)
with other candidate indicators of policy, such as the Boschen-
Mills [1991] index discussed in the Introduction.
However, as a measure of overall policy stance, p has some
shortcomings: first , this indicator is not even approximately
con t i n u ou s ov er ch a n g e s i n r e gi m e ( e.g. , t h e fu n d s r a t e a n d
nonborrowed reserves growth are n ot in compara ble units, so that
a switch, say, from targeting the former to targeting the latter
M E A S U R I N G M O N E T A R Y P OL I C Y 89 7
-
8/7/2019 Bernanke-Measuring Monetary Policy
30/34
would show up as a discontinuity in the indicator). Second, this
m e a s u r e d o e s n o t p r o v i d e a n a t u r a l m e t r i c f o r t h i n k i n g a b o u t
whet her policy a t a given time is tight or easy (a similar
problem affects simple indicators like the level of the funds rate of
the growth rate of nominal nonborrowed reserves). However, we
have found that a simple transformation of this variable seems tocorrect both problems. Ana logous to th e n ormalization applied to
the r eserves aggregates in t he estima tion, to constru ct a final total
policy measure we normalize p at ea ch da te by subtra cting from it
a 36-month moving average of i ts own past values. This has the
effects of greatly moderating the incommensurable units problem,
as well as defining zero as t he benchmar k for normalmoneta ry
policy (normal, at least, in terms of recent experience).
Historical values for our suggested measure are shown in
Figure III.25
Also shown in th e figure, for compa rison, a re t he t wonarrative-based measures of monetary policy, the Romer-Romer
[1989] dates and the BoschenMills [1991] index (scaled to h ave
the same mean and variance as our proposed measure). Examina-
tion of Figure III suggests that our measure conforms well with
the Boschen-Mills index (the monthly correlation is 0.71), as well
as with other historical accounts of U. S. monetary policy. How-
ever, contra ctionary turn s in our indicator a ppear to lead ra ther
than to coincide with the Romer dates; by our measure, Romer
dates look more l ike points of maximum tightness in monetarypolicy, rat her t ha n points a t which policy chan ged from expan sion-
ary to contractionary.
Various exercises can be conducted using this indicator, in
conjunction with the basic VAR estimated in this paper. For
example, an impulse response analysis can be used to character-
ize t he dynam ic responses of monetar y policy to var ious types of
macroeconomic shocks, as identified in the nonpolicy block. We
l ea v e t h i s a n d ot h e r a n a l ys es u s in g ou r m e a s u r e o f m on e t a r y
conditions to futur e resear ch.
VII. CONCLUSION
We ha ve used a semi-str uctu ra l VAR app roach t o evaluat e
and develop measures of monetary policy based on reserve market
indicators. A principal conclusion is that no simple measure of
25. We use the par amet er estimat es from the switching-regime estimation of
Section IV, weight ed at ea ch dat e by the probability of each r egime, in const ru ctingthis series.
Q U A R T E R L Y J O U R N A L O F E C O N O M I CS89 8
-
8/7/2019 Bernanke-Measuring Monetary Policy
31/34
policy is appr opriate for the entire 19651996 period; cha nges in
op e r a t in g p r oce d u r e, s u ch a s t h os e t h a t occu r r e d d u r i n g t h e
19791982 Volcker experiment, imply changes in the preferred
indicator. For p ra ctitioners looking for a simple ind icat or of policy
stance, our results suggest that using the federal funds rate priorto 1979; nonborrowed reserves from 1979 to 1982; an d eith er t he
funds r ate or Strongins mea sur e in th e more recent period, will
give reasonable results. However, a more general and only slightly
more complicat ed altern ative is to base t he policy measu re on a n
estimated model of the market for bank reserves, along the lines
of our just-identified model.26 The latter approach ha s the advan-
26. A RATS procedure t hat est i mat es t he model , const ruct s t he resul t i ngpolicy indicators, an d calculates impulse r esponse functions (with st anda rd er rors)for a rbitrar y sets of nonpolicy variables is obtainable from the aut hors.
F IGURE II I
An Indicator of Overall Monetary Policy Stance
M E A S U R I N G M O N E T A R Y P OL I C Y 89 9
-
8/7/2019 Bernanke-Measuring Monetary Policy
32/34
tage of being able to incorporate the effects of possible changes in
reserve-mar ket str uctur e and in th e Feds operat ing procedures.
Unlike th e simpler indicators, our m ethod can a lso be generalized
to other count ries or p eriods. F inally, associated (an d consistent)
with our approach is a measure of overall monetary conditions,
which we believe could pr ove useful in both the ana lysis andconduct of monetary policy.
P RINCETON U NIVERSITY AND THE N ATIONAL BUREAU OF E CONOMIC RESEARCH
INSEAD AND CEPR
RE F E R E N C E S
Andrews, Donald W. K., Tests for Pa ram eter Insta bility an d Str uctura l Chan gewith Un known Cha nge Point,Econom etrica, LXI (1993), 821856.
Barra n, Fern ando, Virginie Coudert, and Benoit Mojon, The Transmission ofMoneta ry Policy in th e Eur opean Count ries,Cent re dEt udes Pr ospectives etDInforma tions Inter na tionales, Working Paper N o. 96-03, Febr ua ry 1996.
Bern an ke, Ben, Alterna tive Explan at ions of th e Money-Income Correla tion, inKarl Brunner and Allan Meltzer, eds., Real Business Cycles, Real Exchange
Rates, and Actual Policies, Carnegie-Rochester Conference Series on P ublicPolicy No. 25 (Amster dam : North-Holland, 1986)., On t he Predi ct i ve Power of Int erest Rat es and Int erest Rat e Spreads,Federa l Reserve Bank of Boston, New E ngland E conomic Review (NovemberDecember 1990), 5168.
Bernan ke, Ben, and Alan Blinder, The Federa l Funds Rate a nd th e Chan nels ofMoneta ry Tra nsm ission, American Economic Review, LXXXII (1