Professional Documents
Culture Documents
North-Holland
Bennett T. McCallum”
Carnegie Mellon Unic,ersity, Pittsburgh, PA 15213, USA
National Bureau of Economic Research, Cambridge, MA 02138, USA
This paper continues an ongoing investigation of the properties of a policy rule that specifies
settings of the monetary base that are designed to keep nominal GNP growing smoothly at a
noninflationary rate. Counterfactual historical simulations for 1923-1941 are conducted with the
rule and a small model of nominal GNP determination, estimated with U.S. quarterly data, with
residuals fed in as estimates of the behavioral shocks that occurred. The simulation results
indicate that nominal GNP would have been kept reasonably close to a steady three percent
path over 1923-1941 if the rule had been in effect.
1. Introduction
*The author is indebted to Michael Bordo, James Butkiewicz, Stephen Cecchetti, Louis Chan,
Allan Meltzer, Federic Mishkin, Eugene White, and an anonymous referee for helpful com-
ments.
‘The principal items are McCallum (1987, 1988a, 1990).
*A target scheme that weights real fluctuations more heavily than price level movements has
been discussed by Hall (1984), while a price level target has been advocated by Barro (1985),
Clark (1988), and Haraf (1986). A partial investigation of the feasibility of the latter is reported
in McCallum (1990).
31ndeed, this question was emphasized in the discussion following the original conference
presentation of McCallum (1988a), most explicitly in comments by Robert J. Hodrick and
Robert E. Lucas, Jr.
with quarterly data for- rhc years 1922-1941.4 Residuals from the estimated
relationships scrvc as cstimatcs of the behavioral shocks that actually oc-
curred and accordingly arc fed into the simulation process quarter by
quarter. The study’s strategy is thus similar to that employed in my previous
papers pertaining to the postwar era, except that robustness of the rule’s
performance with respect to alternative models is not explored nearly as
thoroughly.
The paper’s outline is as follows. In section 2, some facts concerning
1922-1941 are reviewed to indicate the nature of the problems faced by the
monetary base rule in attempting to cope with the nominal GNP collapse of
the early years of the Great Depression. In addition, one of the model’s key
relationships is estimated. Then section 3 develops specifications and reports
estimates for the remaining equations of the econometric model that will be
used together with the policy rule in simulations. The counterfactual policy
simulation results are described in section 4 for the basic version of the
model and some results for a few alternative specifications are summarized in
section 5. Brief concluding comments appear in section 6 and data series are
presented in an appendix.
2. Overview
It will be useful to begin by outlining the dimensions of the problem that
our policy rule will be required to overcome if it is to provide successful
stabilization in simulations over the period 1923-1941. Let X, denote the
logarithm of nominal GNP for quarter t, with the latter measured in billions
of dollars expressed at an annual rate. 5 Then the actual historical record is as
shown in fig. 1, where X, is plotted against time. Also shown is a smooth
target path (denoted xf) with a constant growth rate of three percent per
year - an increase of 0.00739 for the log of GNP in each quarter - drawn
through the actual x, value for the fourth quarter of 1922. (In this figure, and
those that follow, X, values are designated as LX and xp values as TAR.)
The sharp fall in X, between 1929.4 and 1933.1 reflects values of 4.61 =
log 100.92 and 3.91 = log 49.78 for those two dates, which imply an average
4The simulation period begins with 1923.1, rather than 1922.1, because initial conditions with
the latter would suggest a gradual introduction of the rule, as discussed in McCallum (1988a, pp.
195-198). The slightly longer period is used for estimation to provide more observations.
‘The quarterly values were developed by Robert .I. Gordon by interpolations of the Chow-Lin
(1971) type with industrial production as reported by Persons (1931) used as the interpolating
variable. Since the latter figures are seasonally adjusted [Persons (1931, p. 131)], so too are
Gordon’s GNP figures. They are presented in Balke and Gordon (1986). Quite recently, Romer
(1988) has constructed a revised series of annual GNP estimates for 1909-1928 that could be
used to develop new quarterly values for the early part of our sample. But since the main
discrepanies (in relation to the Commerce Department values) occur for years prior to 1922, that
exercise has not been conducted.
B. T. McCallum, A monetary base rule and the Great Depression
425-
4.00-
3.75
-
]-LX _--_-TAR]
Fig. 1. Time series plot of quarterly values for 1922.1-1941.4 of x, (log of nominal GNP) and XT
(target values), denoted LX and TAR, respectively.
drop of 5.3 percent per quarter over that span of three and one-fourth years.
Furthermore, despite its rapid climb during 1933-1936, the level of X,
remains far below the three percent target path continuously until late 1941.
The policy rule under consideration would have attempted to keep x, close
to the xp target path by means of quarterly adjustments in the growth rate of
the monetary b&e, a -policy instriment that- can be accurately controlled
by the Federal Reserve.6 Let b, denote the log of the base for quarter t.
Then the rule can be written as
“It might be argued that the Fed could not accurately control the base with the institutions of
the 1930s. That issue is beside the point; the controllability of the base under today’s conditions
is what is relevant to issue of the desirability of the proposed policy rule. In the present paper
the subject is whether that rule would have been effective, it if had been in force during the
1920s and 1930s. Such a regime is, of course, an alternative to the Gold Standard.
B. T. McCallum, A monetary base rule and the Great Depression 7
Here the impact of base growth is both sizable and prompt; the estimated
coefficients remain much the same if Ab, is used in place of Ab,_,. Also, the
inclusion of additional lagged values of Ax, and/or Ab, does not substan-
tially alter this relationship, nor does the inclusion of other variables such as
real GNP growth and/or the Treasury bill rate.
For the interwar years 1922-1941, by contrast, estimation of the same
specification as in (2) gives rise to the following results:#
‘These theories are the real business cycle theory of Kydland and Prescott (19821, the
monetary misperceptions theory of Lucas (1972) and Barre (1981), and a more Keynesian theory
with sluggish price adjustments and an expectations-augmented Phillips mechanism as embodied
the MPS model [recently described by Brayton and Mauskopf (1987)I. For more detail, the
reader is referred to McCallum (1988a).
‘Here the monetary base figures are quarterly averages of the high-powered money stock
values compiled by Friedman and Schwartz (1963, table B-3).
B. T. McCaNum, A monetary base rule and the Great Depression
I_COD ____.RODl
Fig. 2. Time series plot (1922.1-1941.4) of currency/deposit and reserve/deposit ratios, de-
noted COD and ROD, respectively.
in place of Ab,_l. Furthermore, if four lagged values of both Ax, and Ab,
are utilized, those for Ab, fail to provide incremental explanatory power, the
chi-square test statistic for the hypothesis of zero effect being 8.6 as com-
pared with a 0.05 critical value of 9.5.
These contrasting findings should come as no surprise to monetary
economists, since it is well known that dramatic shifts occurred during the
1930s in the currency/deposit and reserves/deposit ratios, here denoted
COD and ROD. The extent and sharpness of these shifts are indicated by the
time plots presented in fig. 2.9 Since a change in either of these ratios will
bring about a change in the money stock to base ratio, if the other does not
change in an offsetting manner, the GNP to base relationship will tend to
shift if the GNP to money stock relationship is unchanged.
The foregoing line of argument suggests that a critical issue is whether
there was a reasonably stable (i.e., unchanging) relationship between the
money stock and GNP during the interwar period. A small amount of
experimentation revealed that Ax, was quite strongly related to Am,_l,
where m, is the log of the Ml money stock. Three lagged Ax values were
“Here deposits are measured as Ml minus currency in circulation, with Ml statistics being
those calculated by Benjamin Friedman and reported by Balke and Gordon (1986), while
currency values are those of Friedman and Schwartz (1963, table A-l). Reserves are calculated
as base money minus currency in circulation.
B. T. McCallum, A monetary base rule and the Great Depression 9
“‘It should be emphasized that the d, dummy variable pertains only to the upturn in business,
not the preceding decline. Since no other dummies are introduced below, it is not the case that
they are being used to ‘explain’ the depression. They represent only the brief impact effect of the
early actions of the Roosevelt administration.
“The calculated value of the standard F statistic is only 1.1, as compared with a 0.05 critical
value of 2.2.
10 B. T. McCallum, A monetary base rule and the Great Depression
this specification will be retained as part of the basic model to be used in the
simulation studies of section 4.
Another element needed for our basic model is a relationship between the
monetary base, which is the instrument variable regulated by policy rule Cl),
and the money stock (again operationally represented by the Ml measure).
The first task of this section, accordingly, is to develop a model explaining
m, - b,, the log of the ratio of Ml to the base. Given this objective, one could
proceed either to model separately the behavior of the currency to deposit
ratio and the reserves to deposit ratio, using these two relationships and an
identity to determine the implied values of m, - b,, or to model movements
in m, - b, directly. In the present study the latter approach has been
adopted, largely for the sake of simplicity but also because preliminary
investigations suggested that there would be little benefit from following the
more detailed approach.
In considering the behavior of m, - b, over the interwar period, three
likely determinants spring to mind almost immediately. The first of these is
the public’s attitude toward the safety of its bank deposits, which was
crucially involved (as both cause and effect) in the banking panics of
1931-1933. Second is the level of reserve requirements, which was increased
sharply in 1936 and 1937. And third is the level of nominal interest rates,
which fell to exceedingly low values for the period 1933-1941 (and after).
Quantitative measures of the last two variables can be obtained in a straight-
forward manner,‘* but to represent the public’s attitude or confidence some
type of proxy must be devised. The strategy adopted in this study is to utilize
a measure reflecting the magnitude of bank failures. As a raw measure of the
latter, the ratio of deposits in suspended banks to deposits of all banks can
readily be calculated. i3 Let the log of this ratio for quarter t be denoted sod,.
Then it might be hypothesized that m, - b, would be related to sod, in some
distributed-lag fashion. Empirically it transpires that the strongest relation-
ship (among simple specifications) is between Am, - Ab, and sod,, which is
equivalent to a relationship between m, - b, and the cumulated sum of sod,
magnitudes. This relationship will be taken to prevail only through 1933.1,
“In the results reported below, the interest rate utilized is the 4-6 month prime commercial
paper rate as reported by Balke and Gordon (1986). For reserve requirements, the variable
utilized is the legally required ratio of reserves to demand deposits for reserve city banks. For
quarters during which the ratio changed, weighted averages of the initial and final figures were
utilized. The resulting series can be summarized as follows: 0.100 for 1922.1-1936.2, 0.125 for
1936.3, 0.150 for 1936.4, 0.1583 for 1937.1, 0.1917 for 1937.2, 0.200 for 1937.3-1938.1, 0.1792 for
1938.2, 0.175 for 1938.3-1941.3, and 0.1917 for 1941.4.
13The quarterly data on suspensions come from the September 1937 issue of the Federal
Reserve Bulletin.
B. T. &fcCallum, A monetary base rule and the Great Depression 11
Table 1
Alternative regressions with Am, - Ab, as dependent variable, 1922.1-1941.4.
d, 0.0730
(0.024)
Am,_, -Ab,_, 0.305 0.062 0.047 0.197
(0.106) (0.106) (0.089) (0.104)
Statistics
R2 0.230 0.270 0.373 0.439
SE 0.025 0.024 0.022 0.021
DW 2.29 1.92 1.78 1.95
-
“In cases 3 and 4, sod, equals zero for 1933.2-1941.4.
however, with zero effect assumed for later periods as a result of the altered
situation after the banking holiday and the creation (in 1934.1) of the Federal
Deposit Insurance Corporation.
Least squares estimates of several specifications, in each of which Am, -
Ab, is the dependent variable, are reported in table 1. There it will be seen
that the interest rate measure, AR,, does not enter significantly but that d,,
the previously-mentioned dummy variable for 1933.2, does. Also, the overall
explanatory power is greater when the sod, variable is included only for the
period up through 1933.1 as conjectured above. In sum, the preferred
relationship is as follows:
This equation, in which n; is the log of the required reserve ratio, will be
used as part of the basic model.
It could perhaps be argued that the bank-failures variable sod, should
simply be set at the value zero for the entire simulation period, the idea
12 B. T. McCallum, A monetary base rule and the Great Depression
being that bank failures would not have been an important determinant of
m, - b, if policy rule (1) had been in effect. That position will not be taken,
however, in this study. Instead, we will initially conduct our policy simula-
tions in a manner that permits bank failures to occur and affect m, - b, over
the period prior to 1933.2. Thus an additional relationship will be required,
one whose purpose is to explain quarter-to-quarter movements in the bank
failure variable, sod,. In specifying this equation, standard economic theory is
not very helpful, but it seems highly plausible that the unusually high sod,
values during 1931-1933 would be partially explicable by the poor business
conditions that were prevailing. Accordingly, the following relationship was
estimated and will be adopted for inclusion in our basic model:
Here the business-conditions variable x,_ , -x,*_ 1 has the anticipated nega-
tive impact: when x, is low, bank failures in t + 1 tend to be high.14 In (7) the
estimation period is 1922.1-1933.1, since one would not expect the same
relationship to be relevant after the institutional changes that took place in
the last three quarters of 1933.
A second way of proceeding, which should also be of interest, would
involve the counterfactual assumption that these institutional changes did not
take place. If the policy rule had prevented the depression, that is, it might
have also forestalled the creation of the FDIC. In this case, bank failures
would presumably have continued to occur as in the 1922-1932 period and to
have affected m, - b, as in that period. To represent this possibility, simula-
tions would be conducted with (7) remaining intact throughout 1923-1941
and with no change on the sod, coefficient in (6). More detail concerning this
second simulation strategy will be provided in section 4.
The foregoing sections have specified a system that includes the policy rule
(1) and a simple model of nominal GNP determination consisting of the
three equations (51, (61, and (7).15 In the policy simulations to be reported,
141twould perhaps be better to include a real measure of cyclical conditions instead of the
nominal GNP measure xI _ , -x;“- ,. That has not been done because our model is designed to
explain movements in nominal variables only.
“In terms of posited relationships and basic structure, this model is rather similar to a
five-equation system employed by Anderson and Butkiewicz (1980). Specificational details are
different, however, and the present model determines the currency to deposit ratio implicitly
rather than explicitly. One qualitative difference is that the present model does not rely upon
any measure (treated as exogenous) of ‘autonomous’ expenditures.
B. 7: McCallum, A monetary base rule and the Great Depression 13
3.75 5” I”_
? 3% 38 40
Fig. 3. Time series plot (1923.1-1941.4) of simulated values of x,, denoted LX, using basic
model and policy rule (1) with A = 0.25. Also shown are actual (LXA) and target (TAR) values
of x,.
these four equations are used to generate counterfactual time paths over the
period 1923.1-1941.4 for the four variables b,, m,, sodr, and x,. More
precisely, the simulations are conducted with actual historical vaIues of all
variables as of 1922.4 used as initial conditions and with residuals (for
1923.1-1941.4) from eqs. (51, (61, and (7) fed into the system each period as
estimates of the behavioral shocks that occurred. The reserve requirement
variable is held constant throughout the simulations. implying Arr, = 0, as a
natural complement to the poficy rule (11, while the dummy variable d, is of
course set equal to zero for all periods. Also, in the first set sod, is set equal
to zero for periods after 1933.1.”
The exercise just described has been conducted with five alternatives
values of the feedback coefficient h in rule (1). Consider first the results with
h - 0.25, which is the value emphasized in my first presentation of the rule
[McCallum (198711. The simulated time path is plotted (as LX) in fig. 3,
together with the target path X: and also the actual historical path of X,
(denoted LXA). It is immediately apparent from this figure that the simu-
lated behavior of nominal GNP is vastly superior to the actual outcome in
terms of its proximity to the ‘smooth and noninflationary’ target path. The
root-mean-square error (RMSE) vaIue reported in table 2 is 0.0922, as
compared with the actual historical value of 0.3269. Furthermore, the mean
16This is not to be interpreted as a claim that no bank failures would have occurred during
1933.2-1941.4 under our rule. but only as a way of eliminating the effect of sod, for eq. (5).
14 B. T. McCallum, A monetuy base rule and the Great Depression
Table 2
Properties of actual and simulated time paths of nominal GNP, 1923.1-1941.4.
Root mean
Mean Maximum square
x: -X, x, *--x , xp -x,
Actual historical 0.2095 0.765 0.3269
Simulated, rule (1)
A = 0.00 0.1811 0.7301 0.2880
A = 0.10 0.0263 0.2905 0.1342
A = 0.25 0.0052 0.1996 0.0922
A = 0.50 -0.0016 0.1378 0.0769
A = 1.00 0.0260 2.0622 0.6668
5.257
value of XT -x, is only 0.0052 for the simulated series, whereas the actual
mean was 0.2095. Indeed, it seems almost inconceivable that real output and
employment could have suffered declines anything like those actually experi-
enced if nominal GNP had followed the simulated path.
Simulated time paths for X, when the policy coefficient A is equal to 0.1
and 0.5 are shown in figs. 4 and 5, respectively. From there it is clear that
performance is better with stronger feedback (A = 0.5) and poorer with
weaker feedback (A = 0.11, a finding corroborated by the RMSE figures in
B. ?: McCallum, A monetary base rule and the Great Depression 15
425-
“The simulated volume of suspensions over deposits averages 0.0096 over 1930.1-1932.4 as
compared with the actual historical figure of 0.0151.
16 B. T. h4cCallum, A monetary base rule and the Great Depression
‘T‘40 ,3m
5.00
42:
2.5.
1.5
5
I_..-LH --__.LHA]
Fig, 8. Time series plot (1923.1-1941.4) of simulated and actual values of b, (log of monetary
base), denoted LH and LA4, respectively. Simulation is with basic model and A = 0.25.
4.50-
425-
4.00-
r_._LM .-_..LMA]
Fig. 9. Time series plot (1923.1-1941.4) of simulated and actual values of m, (log of Ml money
stock), denoted LM and LMA, respectively. Simulation is with basic model and A = 0.25.
18 B. T. McCallum, A monetary base rule and the Great Depression
Table 3
Properties of actual and simulated time paths of nominal GNP, 1923.1-1941.4 (alternative
institutional scenario).
Root mean
Mean Maximum square
XF -X, Xl* -x, xp -x,
almost double the $14.2 billion value of 1940X’s Such effects do not show up
clearly in the data, however: the 1941 values of est are positive but not
unusually large and neither Ag, nor Ag,_, (with g, = log of government
purchases) adds significant explanatory power to (5).
All of the foregoing results implicitly incorporate the assumption that the
FDIC was established and other banking regulations altered, with a major
effect on the generation of bank failures, at the end of 1933. As mentioned in
section 3, it should also be of interest to consider simulations in which the
implicit assumption is instead that no institutional change of this type takes
place. For this alternative scenario, simulations are conducted as before
except that eq. (7) remains intact through the entire simulation period and
the effective coefficient on sod, in (6) is not changed after 1933.1. In this case
residuals from (7) are unavailable for the period 1933.2-1941.4, so shock
estimates have to be generated in some other fashion. One possibility would
be to draw values randomly from a normal distribution with mean zero and
standard deviation 0.741. The procedure adopted, however, was to draw
shock estimates randomly from a finite distribution consisting of the 44
measured e7t residuals for 1922.1-1932.4.”
Numerical results for this alternative institutional scenario are reported in
table 3. There it will be seen that most of the error measures are larger than
in table 2, but to a minor extent. Basically, for intermediate A values
(0.1,0.25,0.5) the simulated paths are much like those obtained in the
previous set of experiments. This similarity is exemplified by fig. 10, pertain-
ing to the case with A = 0.25, which may usefully be compared with fig. 3.
IsThe difference was due primarily to a jump in federal defense expenditures from $2.3 to
$13.8 billion.
“Exclusion of the residual for 1933.1 seems warranted, given the uniqueness of the Banking
Holiday episode.
B. T. McCallum, A monetary base rule and the Great Depression
Fig. 10. Time series plot (1923.1-1941.4) of simulated values of x,, denoted LX, using alterna-
tive institutional scenario and A = 0.25. Also shown are actual (LZ4) and target (TAR) values
of x,.
5. Alternative specifications
In addition to the results obtained with the basic model of eqs. W-(71,
others have been developed with two alternative specifications. These alter-
natives are designed to explore the effect of (a> including more lagged terms
in eqs. W-(7) and (bl dropping the 1933.2 dummy variable d, from the
system. In both cases, the institutional scenario without creation of the FDIC
is adopted - that is, eqs. (6) and (71 are kept intact throughout the simulation
period.
It is of importance to experiment with additional lagged regressors in eqs.
W-(7) so as to ascertain whether the results reported above are an artifact of
the practice of including only those terms that have significant explanatory
power. That practice has the potential of shortening the model’s various
distributed-lag response times, thereby making feedback stabilization appear
unrealistically effective. To guard against that possibility, eqs. (51 and (6)
were re-estimated with four lagged terms included for all explanatory vari-
ables except the dummy d,. Thus the revised version of (5) includes Ax,_,
and Am,_, for j = 1,2,3,4 together with d, and d,_,, while the revised
version of (6) includes Arrt_j, sod,_,, and Am,_j - Ab,_j (for j = 1,2,3,4)
plus d,. Note that current values of Arr, and sod, are excluded in the latter,
a change that should tend to make stabilization more difficult. Also in this
20 B.T. McCallum, A monetary base rule and the Great Depression
5.25-
7
5.00-
4.75-
4.50-
4.25-
\
\ ;J
4.00- ’ I
‘bJ
3.75 4
w ‘30 m
Fig. 11. Time series plot (1923.1-1941.4) of simulated values of x,, denoted LX, using A = 0.25
and alternative model specification with dummies omitted. Also shown are actual (LXA) and
target (TAR) values of x,.
experiment eq. (7) was estimated with .~od,_~ and Ax,_~ (j = 1,2,3,4) as
regressors, Ax,_~ replacing x,_ I -x,*- 1 as another check against our basic
specification.
With the model revised in this manner, the simulation results with A = 0.25
in the policy rule (1) give rise to statistics for the target error xf -x, that are
quite close in value to those on the third line (for A = 0.25) of table 3. In
particular, the calculated mean value for XT -x, is 0.0148, the maximum is
0.2289, and the RMSE is 0.1005. The simulated time path, moreover, is too
similar to that of fig. 10 to warrant its inclusion.
The effect of dropping d, and d,_ 1 from (5) and d, from (6) - in addition
to the previous changes - is more substantial as the three x: -x, statistics
become 0.0061, 0.2931, and 0.1379, respectively. Thus, while the mean target
error is again close to zero, the maximum and RMSE values are somewhat
higher. From the plotted time path, shown in fig. 11, we can see that X, falls
farther below xf during early 1932 than in fig. 10. But the most undesirable
aspect of nominal GNP performance is the sharp ‘overshoot’ that occurs
during 1933 and early 1934, with x, values well in excess of x,*. This result
occurs, apparently, because the large positive residuals for 1933.2 and 1933.3
B. T. McCallum, A monetary base rule and the Great Depression 21
in the revised eq. (5) arrive after the upturn of late 1932 has already occurred
and while monetary growth is very rapid.
Consequently, it cannot be claimed that the role of the d, dummy is
innocuous. With the exclusion of d,, policy rule (1) keeps nominal GNP from
falling as far below the XT target as it did in actuality, but does not yield a
very attractive path. But while it is important to recognize that the influence
of d, is not trivial, it is also important not to leap to the false conclusion that
the existence of a substantial influence renders policy rule (1) unattractive.
For if the historical surge in GNP that took place in 1933.2-1933.3 was in
fact attributable to the unusual policy actions of the new Roosevelt adminis-
tration, as argued above, then the inclusion of d, in eqs. (5) and (6) is entirely
appropriate. And in this case the more satisfactory X, path of fig. 3 or fig. 10
indicates what rule (1) with A = 0.25 would have produced, according to our
estimates.
6. Concluding remarks
‘“This tendency can be illustrated in the context of the most well-developed of Lucas’s own
models, that of ‘Expectations and the Neutrality of Money’ (1972), for convenience log-lin-
earized as in McCallum (1984). The latter’s version with autoregressive money growth rates
features the money supply rule Am, = p Am,_, + E,, in which p is the basic policy parameter.
(Here mr is the log of the money stock and F, a white noise policy shock.) The results on p. 142
of McCallum (1984) enable one to determine that the log of per capita real output y, obeys the
following relation in this model: y, = h,, + [h&l - p)/(l + a,)][Am, - pAm,_,]. Thus the rela-
tionship between output and money growth rates depends sensitively on the policy parameter p.
By contrast, the log of nominal output - i.e., X, = y, + P, with p, the log of the price level - is in
this model given as x, =h,,-a,,+m,+Kl-PXb,-a,)/(l+a,)l [Am,-pAm,_,l. Again p
enters (unless h, = a,), but now in a comparatively unimportant way - the behavior of x, is
primarily dependent upon m, and only secondarily (if at all) on the term Am, - p Am,_,
“An important issue has been raised by the referee, who asks ‘whether a base rule directed at
maintaining the growth rate of Ml or M2 would have been more effective than one based on a
nominal GNP growth target’ since it would ‘operate with a much shorter lead time’. In view of
the role of Ml growth in eq. (5), it is apparent that a base rule analogous to (1) that kept m,
close to a smooth growth path could also induce nominal GNP to behave in a desirable manner.
But what target rate of Ml growth should be utilized in such a rule? As it happens, eq. (5)
implies that the average value of Am, needed to make Ax, equal 0.00739 on average is 0.01244
(which is about five percent per year). Conducting a simulation like that recorded in fig. 3, but
with an Ml target path that rises 0.01244 each quarter, yields a nominal GNP path that stays
somewhat less close to a smooth three percent growth path than in fig. 3; the RMSE is 0.1572.
But the main objection to which this rule is open is that policymakers in 1923 would not have
known what growth rate for Ml would be appropriate. A three percent rate, for example, would
have led to virtually zero growth in nominal GNP over 1923-1941 [according to eq. (5)].
Similarly, it is unclear in 1990 what rate of Ml growth would be appropriate over the next twenty
years. Some researchers believe that M2 velocity is more predictable, since it has been trendless
since 1954, but the growth rate of M2 velocity shifted sharply during the 1940s - see Friedman
and Schwartz (1963, p. 640). In any event, the purpose of this paper is not to argue that other
monetary rules would have been ineffective, but only that rule (l), which is highly operational,
would have been effective.
B.T.McCallum,
A monetary and the Great Depression
baserule 23
Table4
Table 4 (continued)
Appendix
The data series used in the present study are given in table 4 for the period
1918.1-1941.4. The series and their sources are as follows.
GNP: nominal gross national product, billions of dollars. Source: Balke
and Gordon (1986, table 2).
HPM: monetary base (high-powered money), billions of dollars. Source:
Friedman and Schwartz (1963, table B-31, average of monthly figures.
MI: money stock, Ml concept, billions of dollars. Source: Balke and
Gordon (1986, table 2). This series was compiled by Benjamin M. Friedman
by averaging the Friedman and Schwartz (1963) monthly figures and applying
an adjustment designed to make the values more compatible with current Ml
data (as of 1983).
CURRQ: currency in circulation (i.e., held by nonbank public), billions of
dollars. Source: Friedman and Schwartz (1963, table A-l), average of monthly
figures.
RRR: reserve requirement ratio for demand deposits in city reserve
banks. Source: Banking and Monetary Statistics, p. 400.
SUS: deposits of banks suspended, millions of dollars. Source: Federal
ReserLle Bulletin, September 1937, p. 909.
R: commercial paper rate, percentage per annum. Source: Balke and
Gordon (1986, table 2).
References
Anderson, Barry L. and James L. Butkiewicz, 1980, Money, spending, and the Great Depression,
Southern Economic Journal 47, 388403.
Balke, Nathan S. and Robert J. Gordon, 1986, Appendix B: Historical data, in: Robert J.
Gordon, ed., The American business cycle: Continuity and change (University of Chicago
Press for NBER, Chicago, IL).
Barro, Robert J., 1981, The equilibrium approach to business cycles, in: Money, expectations,
and business cycles (Academic Press, New York, NY).
Barro, Robert J., 1986, Recent developments in the theory of rules versus discretion, Economic
Journal 96, Suppl., 23-37.
Board of Governors of the Federal System, 1937, Federal Reserve Bulletin, 23, Sept.
Board of Governors of the Federal System, 1943, Banking and monetary statistics (Board of
Governors of the Federal Reserve System, Washington, DC).
Brayton, Flint and Eileen Mauskopf, 1987, Structure and uses of the MPS quarterly econometric
model of the United States, Federal Reserve Bulletin 73, 93-109.
Chow, Gregory C. and A. Lin, 1971, Best linear unbiased interpolation, distribution, and
extrapolation of time series by related series, Review of Economics and Statistics 53,
372-375.
Clark, Peter K., 1988, Comment, Journal of Money, Credit, and Banking 20, 476-478.
Friedman, Milton and Anna J. Schwartz, 1963, A monetary history of the United States,
1867-1960 (Princeton University Press, Princeton, NJ).
Hall, Robert E., 1984, Monetary strategy with an elastic price standard, in: Price stability and
public policy (Federal Reserve Bank of Kansas City, KS).
Haraf. William S., 1986, Monetary velocity and monetary rules, Cato Journal 6, 641-662.
Lucas. Robert E., Jr., 1972, Expectations and the neutrality of money, Journal of Economic
Theory 4, 1972, 103-124.
26 B. T. McCallum, A monetary base rule and the Great Depression
Lucas, Robert E., Jr., 1976, Econometric policy evaluation: A critique, Carnegie-Rochester
Conference Series on Public Policy 1, 19-46.
McCallum, Bennett T., 1984, A linearized version of Lucas’s neutrality model, Canadian Journal
of Economics 17, 138-145.
McCallum, Bennett T., 1987, The case for rules in the conduct of monetary policy: A concrete
example, Federal Reserve Bank of Richmond Economic Review 73, 10-18; also in:
Weltwirtschaftliches Archiv 123, 415-429.
McCallum, Bennett T., 1988a, Robustness properties of a rule for monetary policy,
Carnegie-Rochester Conference Series on Public Policy 29, 173-203.
McCallum, Bennett T., 1988b, Postwar developments in business cycle theory: A moderately
classical perspective, Journal of Money, Credit, and Banking 20, 459-471.
McCallum, Bennett T., 1990, Targets, indicators, and instruments of monetary policy, in:
William S. Haraf and Phillip Cagan, eds., Monetary policy for a changing financial environ-
ment (AEI Press, Washington, DC).
Persons, Warren M., 1931, Forecasting business cycles (Wiley, New York, NY).
Romer, Christina D., 1988, World War I and the postwar depression: A reinterpretation based
on alternative estimates of GNP, Journal of Monetary Economics 22, 91-115.