The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
IR
JkCD)
IT
JFifca\1hl
(CIt§(I : ; (G)
March 27, 1981
Foretelling the Future
In this time of constant change, private and
public policymakers are increasingly
demanding quantitative forecasts ofthe
economy. This demand has not gone unnoticed by the economics profession;
indeed, many individuals and organizations
now regularly make predictions of economic
conditions. The most widely followed
forecasts are based on large econometric
models, which attempt to measure the
structural relationships among various
economic variables.
The importance of these econometric models
cannot be overstated. All major private firms
and public agencies use models to indicate
how different policy actions might influence
the economy. Still, policymakers frequently
have difficulty in deciding which policy
actions to take because different models
forecast different results.
Structuralmodelling
The building of econometric models involves
economic theory as well as statistical
measurement. Economic theory is needed in
order to specify behavioral relationships in
the model. However, model bu iIders may
choose from a number oftheories concerning
the structure of the economy.
There are Keynesian models, monetarist
models, expectations models, supply-side
models, and models incorporating elements
of all these theories. When an economist
predicts that a change in a policy variable
(such as tax rates or the money supply) will
induce a change in another variable (such as
gross national product or the inflation rate),
the reader must realize that the prediction is
influenced by the assumptions about theoreticallinkages that the economist builds into
his econometric model.
No model has the power to foretell the futu re,
because of the many uncertainties affecting
the future. The predictions of such modern-
day oracles as Otto Eckstein (Data
Resources), Lawrence Klein (Wharton), or
John Rutledge (Claremont) are always subject
to question. Indeed, econometric modelling
is at a point where models can be built to
predict almost any set of numbers. This might
explain the difference of opinion concerning
the Administration's forecasts for 1 982,
which show the inflation rate slowing to 8.3
percent (measured by the consumer price
index) and real GN P growth rising to 4.2
percent. According to some critics, the
Administration's forecast is over-optimistic
because its underlying assumptions don't
coincide with generally accepted economic
theory or the past behavior of the economy.
Some critics widen their attack to model
building generally. According to this view,
the restrictions placed on econometric .
models are based on arbitrary choices among
reasonable alternatives. Economic theory
allows a great deal offlexibility in modelling
the economy, so that each model builder's
individual viewpoint determines which specification is most representative of the true
economic relationships. Some observers are
even more critical, such as Robert Lucas and
Thomas Sargent, the "rational expectations"
theorists. Lucas and Sargent assert that
"probabilistic microeconomic theory almost
never implies either the exclusion restrictions
suggested by Keynes or those imposed on
macroeconomic models."
The economics profession is aware of the
potential unreliability of structural macroeconomic modelling. Economists thus use
many other procedures in forecasting, both
independently and in conjunction with
structural macroeconomic models. They
frequently make judgmental adjustments in
their forecasts from macroeconomic models.
These adjustments, known technically as
"add factors", take into account factors not
considered in the models.
in
llewslettf'f
dc) not
[,efleet the vic\vs of the rnanagernent
01 the Federal Res{"rve Ban k of
Francisco
of the
nur (if lilt' BOdTel 01
I
Time-seriesalternative
they are capable of producing forecasts as
accurate as those produced by the large structural models. A comparison of such a timeseries forecast with the composite forecast of
the American Statistical Association-National
Bureau of Economic Research (ASA-NBER)
reveals that both are about equally accurate,
as measured either across variables or time
horizons (on the basis of mean absolute forecast error over the last five years). The ASANBER forecast can be viewed as a consensus
opinion, being the median forecast of some
40 to 50 economists who forecast on a regular basis.
An alternative procedure, time-series analysis, does not rely on detailed theoretical
relationships but instead attempts to capture
empirical regularities in the data. Economists
using this approach typically model the past
behavior of a variable, either independently
or in conjunction with other variables that are
felt to be leading indicators. Forecasters
choose a model primarily on the basis of
statistics that indicate how variables are
related, according to "reasonable" specifications of the data.
Time-series modelling is a method of estimating "reduced-form equations". Reducedform equations differ from structural models
in that they reflect the combined impact of
different influences. Every structural model
has a reduced-form representation, which is
simply a different representation of the
model. Reduced-form models have the
potential to forecast more accurately than
structural models, because they are not constrained by possibly spurious restrictions.
Additionally, time-series modelling can be
performed with vastly fewer resources, both
in time and money.
A number of studies have found time-series
forecasts to be roughly as accurate as those
based on large structural models with add
factors. However, forecasting accuracy is not
the only issue. First, time-series models have
undergone far less analysis than structural
models. It may be surprising that such models
have performed as well as they have. As more
research is directed towards time-series analysis, forecasting accuracy should be substantially improved. Moreover, the different
modelling approaches should be considered
complementary. By taking account of the
different information contained in the different forecasts, economists should be able to
gain greater insight into future economic
conditions.
A structural model builder would forecast
G N P by formulating equations which describe G N P components and their interactions with various sectors of the economy.
The equations would be specified on the
basis of assumed economic theory. A timeseries analyst, in contrast, might argue that
the theory used to identify these equations is
not valid. Instead, that analyst typically
would model GN P by using statistics which
show how G N P is related to its own past
values and to other variables that might indicate upcoming economic conditions, such as
the index of leading indicators.
Comparisonof forecasts
What does a typical time-series macroeconomic model actually predict for 1981? The
time-series model of the Federal Reserve
Bank of San Francisco is predicting an increase in real GN P of about 2.5 percent for
the year (4th quarter over 4th quarter}t an
inflation rate of 10.0 percent, and an average
unemployment rate of about 7.3 percent (see
table). This forecast can serve both as a prediction of the future and as a benchmark to
gauge other forecasts.
Small time-series models have been developed at the Federal Reserve Bank of San
Francisco and elsewhere to produce forecasts
of real GNP, the GN P deflator, and the unemployment rate. While these models are
-still in an embryonic state, tests indicate that
The time-series forecast shown here does not
indicate as sharp a slowdown as other forecasts do. The "Blue Chip" (Eggert) consensus
of 42 private economic forecasters, as of
2
March, calls for an increase in real G N P of
1.3 percent for the year, an inflation rate of
9.9 percent in the G N P deflator; and an
unemployment rate of 7.7 percent in the
fourth quarter of 1981. However, some widely quoted forecasts are in tune with the timeseries predictions.
deal of uncertainty is attached to any forecast.
Much depends on the assumptions in each
forecasting model, as well as the degree to
which new events in 1981 are similar, or
dissimilar, to those in the past. All forecasts
thus shouId be viewed with a degree of skepticism. Correctly foretelling the future course
of economic conditions requires a combination of technical skills and clairvoyance.
Skill is reflected in the various models, while
clairvoyance is needed to foresee 1981's
surprises.
Which of the many forecasts will be most
accurate in predicting the 1981 economy? A
variety of events, both controllable and uncontrollable, may take place to alter the performance of the economy. Indeed, a great
RobertJacobson
Time SeriesForecast
1981 1 -1 981 I V
Real G N P ($ billions)
Annual rate of change (%)
G N P Deflator (1972 = 100)
Annual rate of change (%)
Unemployment Rate (%)
19811
1981 II
1981 III
1981 I V
1494.4
2.1
188.2
9.9
7.4
1502.7
2.2
192.6
9.7
7.3
1512.5
2.6
197.3
10.1
7.2
1523.3
2.9
202.2
10.3
7.1
3
• 4eln • uo8aJO• epeAaN• o4epi
neMeH .. e!uJoJ!le:) .. euozp'v' .. e>lsel'v'
(G)
'Jlle':>'o:>spue'J:J
Ui!'S
ZSL'ON llWM:ld
OIVd
's'n
llVW S5Vl:>15"U1:J
BANKINGDATA-TWELFTHFEDERAL
RESERVE
DISTRICT
(Dollar amounts in millions)
Selected
Assets
andLiabilities
LargeCommercial
Banks
Loans(gross,adjusted)and investments*
Loans(gross,adjusted)- total#
Commercial and industrial
Realestate
Loansto individuals
Securitiesloans
U.5. Treasurysecurities*
Other securities*
Demand deposits - total#
Demand deposits - adjusted
Savingsdeposits total
Time deposits - total#
Individuals, part. & corp.
(Largenegotiable CD's)
WeeklyAverages
of OailyFigures
MemberBankReserve
Position
ExcessReserves(+)/Deficiency (- )
Borrowings
Net free reserves(+)/Net borrowed(-)
Amount
Outstanding
3/11/81
146,418
123,927
36,290
51,320
23,424
1,446
6,821
15,670
41,408
29,758
29,908
77,035
67,929
29,767
Weekended
3/11/81
n.a..
40
n.a.
Changefrom
year ago
Dollar
Percent
Change
from
3/4/81
-
7,807
7,465
2,022
6,308
- 1,042
466
85
257
- 2,586
- 2,257
2,297
16,663
16,173
8,353
470
585
504
84
- 104
57
131
- 16
-1,208
434
16
464
514
318
Weekended
3/4/81
n.a.
35
n.a.
5.6
6.4
5.9
14.0
- 4.3
47.6
1.3
1.7
- 5.9
- 7.0
8.3
27.6
31.2
39.0
Comparable
year-agoperiod
-
11
182
171
* Excludestrading account 5!=curities.
# Includes items not shown separately.
Editorialcomments
maybeaddressed
to theeditor(WilliamBurke)or to theauthor,, , , Freecopiesof this
andotherFederal
Reserve
publications
canbeobtainedbycallingor writingthePublicInformation
Section,
FederalReserve
Bankof SanFrancisco,
P,O.Box7702,Sanfrancisco94120.Phone(415)544-2184.
@FedFRASER