The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
June 22, 1984
SeasonalRevisions
wh ich seasonal patterns do not change from
year to year and the monetary aggregates
follow no trend or cycle. In this case, away
to estimate the seasonal component for, say,
January would be to calculate the difference
between the average of M1 for all Januarys
and the average of the series over all months.
Last February, the Federal Reserve
published revisions of the data for the 1983
monetary aggregates. The revisions
reflected new estimates of the seasonal and
benchmark adjustments routinely made for
M1 , M2 and M3. Such adjustments are
highly technical subjects that normally do
not arise in discussions of monetary policy,
but the revisions for 1983 were different.
They significantly raised the original estimates of M1 -growth during the M1 -monitori ng period establ ished for the second half of
1 983 by the Federal Open Market Committee (FOM C) -the Fed's chief monetary policymaking body. Monetary policy that had
been characterized as fairly "tight" under
the original M1 figures appeared "easier"
with the revised numbers (see the charts).
This Letter discusses the use of seasonal adjustments and the likelihood that "misleading" estimates ofM1 will re-occur.
The X-11 procedure extends this idea to allow for a trend/cycle component and changi ng seasonal patterns. The trend/cycle component is estimated with a centered moving
average of the series (where the data closest
to the month being adjusted receives the
most weight). The seasonal factors for January are calculated by taking the ratio of
each January in the sample to its respective
centered moving average. Once the seasonal factors are calculated in this way, they
are divided into the not-seasonally-adjusted
money series to obtain the seasonally adjusted monetary data used in policymaking.
Why and how?
Seasonal adjustments are designed to remove from the monetary statistics changes
that are due to seasonal variations in the
publ ic's need for money. For example, the
currency and checkable deposits in M1 tend
to build up prior to Christmas as they are
needed for shoppi ng, and then taper off after
the holiday season. Seasonal movements
therefore reflect temporary changes in the
public's demand to hold M1 and are
independent of the trends in macroeconomic variables such as interest rates and
GNP. Since seasonal movements in the public's money holdingshave no effect on the
future course of the economy, the Fed
attempts to accommodate seasonal
demands by formu lating its monetary targets
in terms of the seasonally ad justed
monetary data.
In employing centered moving averages,
X-11 makes equal use of past and future data.
However, when current data are being adjusted, the future data on the series are not
yet available. This is where the ARI M A part
of the estimation procedure comes in. An
ARI M A model "explains" the monetary aggregate on the basis of its past values. It is
used to project future values of the unadjusted series. The X-11 method then is applied using the actual past values and the
ARI M A-projected future values.
At the beginning of each year, usually in
February, the Fed replaces the projected
data that had been used in calculating the
origi nal seasonal factors for that year with
the actual data and revises the originally
estimated seasonal factors accordingly. By
their construction, seasonal revisions cannot alter growth in a monetary aggregate
over a year as a whole, but they may affect
the pattern of growthwithin a year.
The seasonal factors used in generating
these data are estimated with a statistical
method called X-11 ARI M A. To see how this
method works, consider the simple case in
1
.1'
JF
"'c\ ftci:) f{fl'O'\
'fi'J "5\
\&"11
'l,:r;\
\0
O c-:?:\
..\ n
IPD
L
\.\
Ii' IDl
fC».'
'Y.J
WI
11 \ ..,
"y,
(I.)
'-
i }'.·
...r.
C0 :'=i\ <(';;
'\;c=2)
,-",,\l.iL\
Opinions
in this nc'\'\/sletter do not
necessarilv reflect lhe views eli the nldl1 dgemeni
or the Federal
Bank of San Francisco,
or of the Board of Ciovernors of the h'deral
Reserve Svstem.
Growth in original Ml over the second-half
monitoring period was 5.5 percent, near the
bottom of the 5 to 9 percent monitoring
range. This seemingly slow growth in original Ml indicated that monetary policy was
fairly restrictive and suggested to some
analysts that there might be a recession in
1984. ReVisedMl gives a different picture. It
shows M 1 growi ng at a 7.2 -percent rate,
slightly above the midpoint of the monitoring range.
At the same time, the Fed makes benchmark
revisions in the monetary data. These revisions correct measurement error, whether
due to the later availability of data from depository institutions that report infrequently
or the discovery of reporting errors not detected earlier, in the seasonally unadjusted
series.
The 1983 adjustments
The monthly differences between original
and revised Ml in 1983 were quite large in
some cases. Expressedas annualized growth
rates, the largest monthly revision in 1983
was 7.6 percentage points in February. On
average in 1983, the absolute difference between revised and original Ml averaged 3.1
percentage points of annualized growth for
monthly data. Although large, revisions of
this size are not out of line with previous
experience-over the preceding three
years, revisions of the preceding year's
monthly data averaged 2.7 percentagepoints.
Why were seasonalrevisionsso large?
An important cause of the large 1983 seasonal revisions appears to be the ARIMA
forecasts of Ml, used as part of the calculation of the original 1983 seasonals,which
did not correctly predict the pattern of actual
M 1 growth over the year. In essence, the
ARI M A forecasts failed to anticipate the
rapid Ml growth in the first halfof1 983, and
the deceleration later in the year.
This is not surprising, since the ARI MA
model forecasts Ml on the basis of lagged
values of Ml only, and incorporates no information about the Fed's policy actions. The
rapid Ml growth in the latter half of 1982
and the fi rst half of 1983 appears to have
been significantly influenced by the sharp
decline in interest rates beginning in August
1982. Since the ARI M A model incorporated
no information about interest rate movements, it quite naturally underestimated Ml
growth in the period following the interest
rate decline.
Large monthly errors normally do not give a
misleading picture of the direction of monetary pol icy because the upward and downward adjustments tend to cancel out over
the span of several months. Thus, the rather
large monthly revisions noted above compare to the far smaller average quarterly and
semi-annual revisions of 0.8 and 0.2
percentage points in the period from
1 980 to 1982.
Last year was unusual in that there were
large revisions in the semi-annual data. A
long string of downward revisions in the
months of the fi rst half of 1983 were
matched by a long series of upward revisions in the second half, leading to an average (absolute) semi-annual revision of 1.3
percentage points. In the first halfofthe year,
benchmark revisions had no effect while
seasonal revisions lowered Ml growth by
0.9 percentage point. In the second half of
the year, benchmark changes caused an upward revision of 0.7 percent and seasonal
changes caused an upward revision of 1.0
percent, for a total of 1.7 percent.
When the actual data for 1983 replaced the
ARI M A forecast in the seasonal adjustment
program, the estimated seasonal factors for
the fi rst half of 1983 rose and forced a downward revision in estimated Ml growth. Since
seasonal effects must "wash-out" over the
year as a whole, the downward adjustment
of seasonally adjusted Ml growth in the first
half of the year required an upward adjustment of equal size in the second half.
These revisions reflect problems inherent in
seasonally adjusting a quantity that is signifi2
Chart 1
1983 Unrevised M1
Chart 2
1983 Revised M1
BIlIIon801
Dollars
540
BIUlonsol
Dollars
540
......
......
9 Percent Growth .........
520
......
9 PercentGrowt'Y ....
.......
.....
520
.... --5
-------Percent Growth
500
480 L-.l.-.l.-.L-.L-.l..-.l..-.l..-.l..-.l..-..L-I
J
F
M
A
M
J
J
A
S O N
0
effects were entirely concentrated in the last
half of the year. As discussed in the April
1984 Federal ReserveBulletin, a large portion of these revisions were related to
changes last year in reporting responsibilities, some associated with the introduction of new accounts, of some depositoiy
institutions, and are unlikely to happen
again now that these institutions are more
familiar with the new reporting system.
cantlyaffected by policy. As noted earlier,
the goal of seasonal adjustment is to remove
from the monetary aggregates movements in
the public's seasonal demandsfor money.
However, the seasonal adjustment procedures cannot distinguish between M1 movements caused by changes in demand and
those induced by changes in monetary policy. By its very nature, X-ll ARI M A wi II attribute part of such policy movements to seasonality, unless policy follows a regular cyclical pattern.
Second, although another large swing in M1
probably would cause another set of large
revisions in the seasonals, an M1 swing like
the one in 1 982-83 is not likely to re-occur.
Cumulative M1 movements of that size are
highly unusual in the post-war period.
None of this discussion is intended to suggest that information on monetary policy
should be used in calculating seasonal factors. Seasonal adjustment of monetary data
by the Federal Reserve is done under a strict
constraint to be objective. In other words,
monthly seasonal factors are calculated in
an objective (non-judgmental) way that can
be reproduced easily by the public. This
approach conforms to the recommendation
in 1981 of the Board's Committee of Experts
on Seasonal Adjustment Techniques, made
up of distinguished outside experts on this
subject. Attempting to cope with problems
of policy movements in money necessarily
would involve the kind of judgmental
adjustments that would violate the objectivity constraint under which the Federal
Reserve operates.
Finally, large seasonal and benchmark revisions in the second half of last year
happened to correspond to the second half
monitoring period established by the Federal Reserve for M1. Last year was the first
time that the FOM C established a range that
did not cover an entire one-year period.
Annual ranges have the advantage that
money growth from the end of one year to
the end of the next, by definition, cannot be
affected by seasonal revisions. One lesson to
be learned from 1983, then, is that studying
money growth over a semi-annual range involves a greater risk that data revisions will
alter any interpretation of tightness or ease in
monetary policy.
Will the problem re-occur? .
The revisions to M1 in 1983 raise an important policy issue. Are future seasonal revisions likely to change the picture of monetary pol icy as much as they did in 1983? The
answer is probably not. The problems in
1 983 were the result of the coincidence of
three events unlikely to occur simultaneously again. First, the benchmark revisions were unusually large and their net
In sum, it seems fair to conclude that unless
the F OM C establ ishes ranges shorter than a
year, and unless historically large cumulative movements in M1 are observed during
the year, the risk that data revisions will significantly distort one's picture of monetary
policy appears to be small.
John P.Judd
3
l.Sl::Il.:I
UOlllUI4Sl'. M • 4Pln • UOllCl10 • PPPAClN • 04 PPI
IIPMPH
•
PluloJIIP:)
PUOZIlV'
P>fSPIV
\ill\2?CS
'llIP::>'O:lSPUPJ:J
ues
Z!i':::'ON l1WH:Jd
G1Vd: J9VIS0d 'S'1l
llVW ssvnISHI:J
OUHOS:JHd
:iJ
B AN KI N G D ATA-TWE L F TH FEDERAL RESERVEDI STRI CT
(Dollar amounts in millions)
Selected Assetsand Liabilities
Large Commercial Banks
Loans, Leases and Investments 1 2
Loans and Leases1 6
Commercial and Industrial
Real estate
Loans to Individuals
Leases
U.S. Treasury and Agency Securities2
Other Securities 2
Total Deposits
Demand Deposits
Demand Deposits Adjusted 3
Other Transaction Balances4 .
Total Non-Transaction Balances6
Money Market Deposit
Accounts-Total
Time Deposits in Amounts of
$1 00,000 or more
Other Liabilities for Borrowed MoneyS
Weekly Averages
of Daily Figures
ReservePosition, All Reporting Banks
Excess Reserves (+ )/Deficiency (-)
Borrowings
Net free reserves (+ )/Net borrowed( -)
1
Amount
Outstanding
6/06/84
1 80,1 67
1 60,826
48,451
59,927
28,232
5,009
11,973
7,368
1 89,708
45,61 0
29,963
12,785
131,31 3
39,31 9
39,474
1 9,956
Weekended
6/4/84
-
Change
from
5/30/84
-
-
Change from 1 2/28/83
Percent
Dollar
Annualized
549
694
120
45
41
24
28
173
2,102
1,105
2,1 90
673
324
-
-
-
-
4,142
5,471
2,488
1,028
1,581
54
534
795
1,289
3,627
1,368
10
2,328
-
-
-
5.3
7.9
12.2
3.9
13.4
2.4
9.6
22.0
1.5
16.6
9.8
0.1
4.0
54
-
278
-
1.5
48
149
-
1,309
3,051
-
7.7
29.9
Weekended
5/21 /84
31
167
135
16
55
71
Includes loss reserves, unearned income, excludes interbank loans
2 Excludes trading account securities
Excludes U.S. government and depository institution deposits and cash items
ATS, N OW, Super N OW and savings accounts with telephone transfers
5 Includes borrowing via FRB, I T&L notes, Fed Funds, RPsand other sources
6 Includes items not shown separately
3
4
Editorial commentsmay be addressedto the editor (GregoryTong)or to the author .... Freecopiesof
Federal Reservepublications can be obtained from the Public Information Section, FederalReserve
Bank of San Francisco,P.O. Box 7702, SanFrancisco94120. Phone(415) 974-2246.
@FedFRASER