109

5
ECONOMIC FORECASTING

ECONOMIC FORECASTING

Forecasting We may begin our discussion of forecasting with a justification for

allocating resources to such efforts. It may be noted first that an
econometric model of economic phenomena is designed to enhance our
and Policy Analysis understanding of the inter play of forces behind the phenomena being
studied. It is safe to say that no amount of a priori theorizing will disclose
with Econometric Models the quantitative impact of one variable on another. Moreover, the very
preciseness of specification required in econometric model building should
necessitate very careful and logical thought on the part of the analyst. Such
thinking may often greatly enhance the analyst's theoretical view of the
phenomena being studied, even before any empirical fits are obtained.
Nonetheless, greater understanding of economic events should not be
In the previous chapters we have discussed in detail the estimation of the viewed necessarily as a goal unto itself. Rather, understanding should be
individual economic relationships that directly determine the balance of looked upon as a tool for optimal decision making designed to attain the
payments. The next step is the integration of these relationships into a com economic objectives of the society. The making of both public and private
prehensive quantitative model designed to provide understanding and, ulti economic decisions is measurably improved when there is a fundamental
mately, control of the economic environment. We shall see in this chapter understanding of the consequences of such decisions. Thus, for example,
that the problems of constructing a multi-equation econometric model are government policymakers will be interested in the impact of various
very much more difficult than the problems inherent in estimating monetary and fiscal policies on the domestic economy and balance of
individual relationships. In practice, these difficult problems have been payments, while private investors will want to know something about future
largely ignored and consequently only meager evidence exists concerning profit rates in the industries of immediate concern to them.
what should or should not be done when a model is constructed. For this This suggests that an econometric model and an associated forecast
reason this chapter should be taken to be tentative and suggestive. should be deemed successful insofar as decisions are made more
Historically, the quantitative analysis of the balance of payments began effectively. The discussion to follow will therefore be cast in a decision-
with the estimation of individual import and export demand equations. making context. Let us for the moment concern ourselves with some
Considerable effort was expended in the 1940's and early 1950's in the general considerations of model building.
meas urement of income and price elasticities. There was a preoccupation
during this time with the importance of the price variable, and it was General Concepts of Econometric Model Building An econometric
common for researchers to subject their price variables to tests of model is a set of equations, including statistically estimated relations, that
significance in an effort to credit or discredit the hypothesis that price seeks to explain in quantitative terms some observed economic phenomena.
changes had little or no effect on the flow of goods and services. Besides the econometric relations, the model may also include both identities
In the 1950's, quantitative inquiry concerning the determinants of the and noneconometric behavioral relations (e.g., a tax relationship specifying
balance of payments turned to multi-equation models. This shift in some legally determined percentage of income).
emphasis required, of course, the concurrent evolution of large-scale In order to focus on concepts and problems pertaining to econometric
electronic com puting facilities. It is now possible to build upon the many models in general, let us consider a model of the (allowing form
studies that have been made of particular relationships and to construct Y1 = /i(Y2, Y3, ... , Yn, X1, X2, ... , Xm)
relatively large and complex econometric models. Our concern in the
present chapter will be with the structure of such models, using as building (5.1)
blocks the relationships de veloped in the preceding chapters that dealt
with the flow of goods and
services, and of capital.

108
110 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
ECONOMIC FORECASTING 111

This model includes n equations, econometric or otherwise, each of which

describes a particular variable Yk as a function of the (n - I) remaining Y; tural model indicates the initial impact of one variable on another, while the
and m other variables, X;, (j = 1, ... , m). The Yi variables are called en reduced form indicates the final equilibrium.1
dogenous variables since these variables are explained by or, equivalently, are The reduced form system (5.2) will play the central role in both fore
endogenous to the model. Correspondingly, the X; variables are called ex casting and policy analysis. Forecasting will require the selection of a par
ogenous variables to indicate that no attempt is made to explain or forecast ticular set of values for the exogenous variables X;, which will be inserted
their values, or perhaps more accurately, that none of the endogenous vari into the reduced form equations to calculate forecast values for the en
ables influences the exogenous variables. dogenous variables Yi. Policy analysis will involve such a question as: "If
Before proceeding, we should perhaps caution the reader that this the value of a policy instrument Xi is altered, what will be the effect on
general representation of an econometric model does not imply that all the variable Yk?" The answer to such a question may be read directly from the
explanatory variables noted need be used in each equation. Rather, the ex kth reduced form equation.
planatory variables in each equation are potential choices, not necessary ones.
Consequently, we should take the functions Ji to be general enough to exclude Conditional and Unconditional Forecasts We have just described fore
any of the potential explanatory variables. Thus, for example, the case casting as requiring the selection of a particular set of values for the exogenous
when y1 depends on X3 and X7 alone is implicitly considered by the very variables, which together with the reduced form system will imply values of
general function /1. In addition, one or more of the equations may be the endogenous variables. This suggests that a forecaster may make either a
identities or behavioral relationships that are arrived at by other than conditional or an unconditional forecast. A conditional forecast will be of the
statistical form: "If the values of the exogenous variables turn out to be this, then the
methods. values of the endogenous variables will be that." In other words, the
Ordinarily the functions f; used in economic research will permit the forecast is conditioned on the future values of the exogenous variables. An
"solution" of the equation system (5.1). That is, we may express the en uncon ditional forecast may be represented as: "The values of the
dogenous variables as functions of the exogenous variables alone, as follows endogenous vari ables will be this." That is, the forecast is not conditioned
on any future events. The difference between a conditional and an
unconditional forecast should be clear. In the case of the conditional
Y1 = g1(X1, X2, ... , Xm) forecast, the forecaster in effect is admitting that he is unable or unwilling
Y2 = g2(X1, X2, ... , Xm) to project the values of the ex ogenous variables X;. With an unconditional
(5.2) forecast, however, the fore caster will in effect provide the projected values
of the exogenous variables.

1 We can illustrate these points with a simple demand and supply model
This of course requires that all of the relationships hold simultaneously,
which implies rapid adjustment to disequilibria. This assumption is not so Q = a+ {3P + -yD (i)
P=a+bQ+cS (ii)
strong when adjustment lags are included in individual relations.
Equation (i) is the demand equation indicating quantity demanded Q as .a linear function
When the model is expressed in the form of the first system (5.1), we of the price P and some demand factor D. Equation (ii) is the supply equation indicating
will call it a structural model. That is to say, System (5.1) represents the struc the supply price P as a linear function of the quantity sold Q and a supply factor S. This
ture of the economic phenomena. It is meant to indicate how any particular is a structural model with endogenous variables Q and P, and exogenous variables D and S.
The corresponding reduced form is
variable directly influences any endogenous variable. System (5.2), on the
other hand, is referred to as the reduced form. We may think of it as being Q = (1 - {3b)-1(a + -yD + {3a + (i')
an equilibrium system. To illustrate this point, we might consider the first {3cS) (ii')
equa P = (l - {3b)-1(ba + b-yD +a+ cS)
tion of the structural model (5.1). A change in X1 will induce a change in Yi, We see from Equation (i) that a one-unit increase in D will have an impact on the quantity
but this change in Y1 will induce changes in Y2, ... , Yn, which in turn alter demanded equal to 'Y units. This "shift" in the demand curve will tend to induce a rise in
y1 once more. Eventually, assuming stability, the system will settle down to the price due to the upward sloping supply function. Such a price increase will dampen
the impact of the change in D on the market-clearing quantity. The equilibrium result
an equilibrium again. This equilibrium influence of X1 on Y1 is expressed by can be read from the reduced form equation (i'), which indicates that the initial impact
the first equation of the reduced form system (5.2). Hence to repeat, the struc- of size 'Y will be damped by the factor (1 - {3b)-1.
 112 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
ECONOMIC FORECASTING
113
A pure conditional forecast is represented by the reduced form system
probability distributions, A and B, which represent competing informative
alone. This implicitly contains all of the conditional statements that can be
forecasts. Fo ecast A reflects relatively weak opinions about the event. Ify
1
made. A pure unconditional forecast will report only the forecast values of occurs, we will tend to favor forecast A, since y1 is more likely under that
endogenous variables. The reduced form would not be required here inas forecast; conversely, if Y2 occurs, forecast B would be preferred.
much as the values for the exogenous variables will have been specified.
Most econometric forecasts fall between the pure conditional and
unconditional cases. The reduced form or the structural model is therefore Probability Distri
ordinarily re ported. In addition, most likely values of the X; are used to bution B
calculate the forecast Y;. The user of the forecast is thus provided with both
the most likely forecast as the forecaster sees it, and a means to alter that
forecast in the event that the user disagrees with the forecaster's opinions Probability Dis
tribution A
about the future values of X;. Thus, this hybrid forecast is preferred, since it
provides the maximum of information.

Informative and Decision Forecasts A forecast will ultimately lead to

decisions on the part of the forecast user. The forecaster may play one of two
Y,
roles in this process. He may provide information about future events which
will be analyzed by some decision maker, or he may use the information to
make decisions himself. In the first instance, a forecast will provide proba FIGURE 5.1
bility statements about future events, while in the second the forecaster
Probability Forecasts
provides his view of what will occur without appending any probability
statements. In the latter event, the forecast user is implicitly denied those de
cision making powers that relate to the weighing of the likelihood of various This concludes our discussion of informative forecasts. Let us now turn
outcomes. to decision forecasts. We will view the process of decision making as deciding
A concrete example of the difference between these two types of fore upon an act a that depends upon the future value of an endogeneous variable
casts may be useful. If a forecaster predicts sales of one million units, pro -. Since Y is unknown, there is the possibility of making a nonoptimal de
duction will be geared to that level of sales. If, however, the forecast specifies c1s10n. The penalty for such a decision will be described by a loss function
the probability of other levels of sales, management may decide on a pro L(a, y), which describes the penalty for choosing act a, when y turns out to
duction level best suited to its goals. For example, there may be a reasonably be the true value of Y. A reasonable decision rule is to select a that
good chance of sales of two million units, and management may decide to minimizes the expected loss
have the excess capacity available in that event. E(L) = 2: L(a, y) P( Y = y) (5.3)
Let us consider first the informative forecast, which involves the fore y

caster as information provider but not as decision maker. An informative where P( Y = y) is the probability that the future value of Y will bey. This
forecast seeks to provide useful information about the future. This informa probability distribution is implicitly provided by the informative forecast.
tion takes on the form of a probability distribution, indicating the . _F?r any act a_t at is chosen, there will be a value of Y for which L(a, y)
forecaster's opinions regarding the likelihood of the possible values of the 1s mm1mal. A dec1s10n forecast will provide only this value of Y. This will
forecast vari able. A useful way of summarizing this distribution is in terms lead the forecast user necessarily to one particular act a. The forecaster will
of the most likely value and a 95 percent confidence interval. A point dec e on the value of the point estimate of Yin the same way that any other
forecast (without the confidence interval) cannot be an informative forecast dec1s10n maker would select an act given the probability distribution of y
since it does not provide the required probability distribution. P( Y), that is, on the basis of some loss function, explicit or otherwise. H
It is of considerable importance to be able to evaluate the quality of a fore therefore usurps the forecast user's decision-making powers.
cast after the event has occurred. A forecast may not be judged alone, but The evaluation of decision forecasts will necessarily be different from the
must be compared with [Link] competing forecasts. Figure 5.1 depicts two !
evaluation of informative forecasts. Since a decision forecast purports to
i
11
 114 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
ECONOMIC FORECASTING
115
minimize the loss we will incur from incorrect decision making, we will
quite naturally give the best grades to those forecasts that truly minimize must e used to arrive at particular values for the historically given variables.
the loss. This will necessitate a comparison of two or more alternative xpenence suggests that reasonably good accuracy can be obtained in such
circumstances.
forecasts. In the event that only one model is available, we may generate a
dummy fore cast for purposes of comparison. Suitable dummy forecasts Slowly changing variables make up a second set of exogenous influences
may, for ex ample, be the naive "no change" forecast or the projection of The appropriate alu_es to ass gn to these variables will presumably be easil;
past trends. calculated by proJectmg the historical paths. Examples would be population
and the labor force.
This discussion of the difference between informative and decision
fore casts is meant to emphasize that the form a forecast assumes will . T e third class of exogenous variables involves the set of government
depend on the forecaster's role in the decision process. In our judgment, the policy mstruments, such as tariff levels and taxes. From the point of view of
proper form of the forecast will most often be informative and unconditional, the government forecaster, the levels of these variables can be forecast with
with the reduced form system included to allow the forecast user some great acc_ura y. Although there may be considerable difficulty in projecting
adjustment of the forecast according to his own judgment about the future these policy mstruments when one is not privy to such information it is not
values of the exogenous variables. However, most forecasts have in fact possible to include them endogenously in an econometric model ince the
been of the de cision variety.2 government policyma_kers are unlikely to display a consistent pattern that cou!
Forecast Error There are three sources of forecast error. In the first d e well approximated by an econometric equation. Accordingly, the
place, there are natural disturbances to the true relationships. Secondly, we pr0Ject1on of go_v rnment policy will play a principal role in introducing error
mto an uncondit10nal forecast.
are using estimates to represent the true relationships. The discrepancy be
tween the estimates and the true relationships will result in forecast error. The three sets of variables just mentioned are exogenous both from the
Finally, the discrepancy between the true levels of the exogenous variables point of view of the particular model and from the point of view of the eco
and their estimated levels will induce additional forecast error. A nomic system generally. That is to say, they are inputs into the economic
conditional forecast will involve only the first two sources of error. An s st m and do not to any significant degree respond to events that occur
unconditional forecast will involve all three, although the last may swamp :-"ithm that systen:1. For this reason they are necessarily exogenous variables
m any econometnc model.
the other two. For this reason, practicing forecasters tend to neglect the first
two sources of error. . The fourth set of exogenous variables-is composed of variables that are
This concludes the essentially mechanistic aspects of forecasting. As yet, m f ct endogenous to the economic system in general, but that for some
the two most fundamental questions remain unanswered: What variables particular_ econometric :11odel are_ selected to be exogenous. For example,a
should be endogenous and what variables should be exogenous? Which ex one-eq at10n model of imports might express imports as a function of QNP
planatory variables should be used in each structural equation? Answers to a d pnces. I _such a model both GNP and prices are exogenous. To our
these questions are about one part mechanical and nine parts intuitive, as we mmd, the decis10n of what to include in this fourth set of exogenous variables
shall see. This is an area in which experience weighs very heavily. and what to include endogenously is the most troublesome problem that faces
an ec_onometric model builder. A model builder must ask himself if moving
Exogenous and Endogenous Variables Let us consider the first a vanabl from t e ourth class of exogenous variables to the set of endoge
question: which variables to have exogenous. Suits [22] provides in this ?ous var:ables will lillprove the forecast. There is unfortunately often little
regard a useful catalog of exogenous variables. The first type consists of mf rmat10n he can brmg to bear on this question. A related problem is the
those that are his torically given. These variables measure events that have o tm".al level of disaggregation. It is safe to say that a disaggregated model
occurred before the forecast period. A problem can arise, however, when a will yield more detail than an aggregated model, but it does not follow that
forecast for a par ticular period has to be made in the course of a period still the aggregates will be better forecast.
in progress and/or when the relevant data may not be fully collected. In . The development of econometric models quite naturally began with rel
such an event, some method of extrapolating the data already collected or atively small models with relatively few variables and equations. These
some projection scheme graduall evolved into larger and more complex models, predicated on the
assumption that the more variables endogenously determined, the better the
2 In some cases it may prove difficult to classify results as being informative or de
mod l. In recent years, however, the pendulum has tended to swing the other
cision forecasts. For instance, when the reduced form is reported, we will somewhat arbi waym response to arguments that the very large models are so complex that
trarily classify the forecast as informative since confidence intervals may be generated on
the basis of the forecast user's opinions about the future values of the exogenous variables.
116 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS

no one can fully grasp their fundamentals and that their relationship to the actual phenomena under study becomes more and more coincidental as they increase in size. According to this line of
argument, a small manageable model may thus be preferable.
The fifth and final set of exogenous influences in an econometric model
will consist of variables that represent rare events expected to influence the phenomena under study. For example, a commercial trade agreement soon to be ratified may be expected to
increase the flow of trade. Since the effect of such an agreement could not be estimated from historical data, we may want to adjust our historically estimated relationships to allow for
increased trade. This may be effected by putting exogenous variables that will reflect the likely magnitude of the event into one or more of the structural equations. The choice of the values
of these variables will rest entirely on good judgment.

Explanatory Variables Let us now turn to the other question raised earlier concerning the choice of explanatory variables for each structural equation. In the preceding chapters we
have discussed the estimation of in dividual equations and have argued that the appropriate explanatory vari ables should be suggested by theoretical considerations. This is not neces
sarily a preferred procedure for multi-equation systems, since when we com bine these individual relationships, we will want to avoid the possibility of relatively small estimation errors
being transmitted and amplified from equa tion to equation in such a way that the estimation error of the system as a whole far exceeds the sum of its parts. To put the problem precisely,
ordinary least squares applied individually to the structural equations of a large model is not the optimal estimating procedure when the goal is an accurate reduced form.
Although multi-equation estimation methods are available, they are
difficult to apply and have not been used to any great extent. In practice, most econometric model builders have used single-equation least squares methods of estimation with an experimental
approach designed to reduce the danger of an inaccurately estimated model. For instance, several regressions may be fitted and the "best" one selected. It is important here to emphasize that the
summary statistics that indicate good fits, for the individual structural equa tions, do not insure that the reduced form system is in itself very reliable. These summary statistics are calculated on
the assumption that the explana tory variables are given. In fact, many of the equations will include endoge nous variables whose values are not given to the forecaster, but rather are calculated
from the reduced form.
To express this differently, we observe that when the reduced form is calculated from the structural form, we will have to divide by the structural coefficients on any of the endogenous
variables that are also used as explana tory variables. If such a structural coefficient is small and unreliable, this di-
 POLICY ANALYSIS 117

vision will greatly amplify the inaccuracy of the reduced form. Accordingly, we should put a premium on large reliable structural coefficients for any en dogenous explanatory
variables. Whereas experimentation is to be avoided when inference on parameter values of structural equations is desired, such experimentation may be an absolute must in
order to arrive at a reasonably reliable reduced form.
The accuracy of the equation system as a whole may be assessed by using the historical values of the exogenous variables together with the reduced form to calculate
estimates of the endogenous variables that may be com pared with the actual historical values. Unfortunately, a discrepancy between the estimated and the historical values
may not lead directly to the particular equation that is the source of that error, since errors will be transmitted in a very complex fashion from equation to equation.

POLICY ANALYSIS

Some separate comments may be in order on the subject of policy analy sis. We have already indicated that policy analysis is no more than seeking an answer to a question of the
form: "If the government does this, what will be the result?" The answer to such a question can be read directly from the re duced form system (5.2) when the other exogenous
variables are given their fixed forecast levels. We should observe that policy analysis and conditional forecasts are essentially the same thing and may be distinguished only by the
emphasis placed on a particular set of the exogenous variables by policy analysis.
Another form of policy analysis is impact analysis, which deals with the question: "If the government alters this policy instrument by this amount, by how much will that
endogenous variable change?" Again the answer is con tained in the reduced form system, but this time may be expressed in terms of a policy multiplier
dY;
(5.4)
dXk
where g,- is the }th reduced form equation. Such a multiplier indicates the marginal response of the}th endogenous variable to a variation in the policy instrument Xk. A model
designed for impact analysis alone will be signifi cantly easier to construct than other models, since the choice of endogenous variables will be more or less straightforward. That is, an
impact model may have many exogenous variables and only a few endogenous ones. This will be undesirable in a forecasting model since in order to construct a forecast, the
118 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
ECONOMETRIC MODELS OF THE BALANCE OF PAYMENTS 119
levels of all the exogenous variables would have to be selected. The model it
self would play a minor role in the forecast and might be better discarded
altogether. A Model of the World Economy The basis of our discussion will be a
However, the impact analysis will be useful only when the policymakers ten-market model of the world economy. More complex models as well as
are able to react rapidly to economic events. Ordinarily the policy will be simpler ones could of course be constructed by disaggregating or aggregating
selected at some time before the actual event, in which case the policymaker the various markets we shall identify. Given the fact that most models that
must know both the policy multiplier and by how much the endogenous vari have been constructed to date are much simpler than the one we shall de scribe,
able will differ from its optimal level. A forecast would therefore be required. we shall lean in our discussion more towards simplicity than com plexity.
We will view the universe of economic events as the set of all economic
exchanges, each exchange involving one supplier and one demander. Ex
changes that may be considered to be essentially the same are classed to
ECONOMETRIC MODELS OF THE BALANCE OF PAYMENTS gether and called a market. The world economy is then a huge and complex
set of interrelated markets. The number of markets we might discuss is
limited only by the total number of exchanges, each exchange being at least
Having reviewed in general the main issues in the use of econometric models in some respect different from all others. What we wish to accomplish in
for forecasting and policy analysis, let us turn next to consider specifically terms of theory is to combine, condense, or drop most of these markets in
the question of the structure of econometric models that are designed to in order to simplify and to bring order to the seemingly chaotic events, and yet
vestigate the balance of payments and its components. That is, we will ex at the same time to maintain the essential features of the exchange phenome
amine in more detail the selection of endogenous and exogenous variables. non. Many of the issues that arise in the process of simplification are essen
As we have already implied, this question has not been and perhaps never tially empirical. It is therefore important that the model we use at the initial
will be satisfactorily answered. Indeed, the issues involved are at the very stages of empirical analysis be general enough to include many of the com
heart of economic science insofar as they relate to one's view of the world in peting views of the world economy. We may then allow the data to suggest
general and economic phenomena in particular and to the choice of simpli the appropriate simplifications that are empirically relevant.
fying assumptions that will help to order these phenomena in ways that will However, generality in specification combined with the usual data limi
improve our comprehension of and ultimately our control over them. tations will ordinarily leave the researcher with little results of any use.
In earlier chapters, we have seen that there. is considerable choice for Accordingly, he will be forced to impose his own theoretical views upon the
competing selections of explanatory variables in individual equations. When data. The point where theory should end and the data should take over in
we wish to combine equations into a comprehensive structural model, the the process is by its nature difficult to determine. The theoretical view we will
problems of choice become manifold. Ultimately, competing models will ha e present is a comparatively weak one, and we shall discuss at some length a
to be judged by performance. When a particular model performs poorly, 1t number of additional assumptions that will make the theoretical base
may be appropriately modified to improve performance. We nevertheless need stronger, but less general, as well.
some basis for departure in constructing a model. What should it be? What, The construction of our model of the world economy will require, first,
in other words, is the appropriate structure for the model? We cannot hope a classification of exchanges into a set of markets. Each market is meant to
to give a definitive and unambiguous answer to this question. But we can at include all of but only those exchanges that for our purposes may be con
least propose a framework that will help the researcher in making his own sidered to be essentially the same. Furthermore, every exchange may be
decisions and serve also as a means by which we can analyze certain specific easily classified into one of the markets. We shall abstract in particular, how
models that have in fact been used for forecasting and policy analysis. ever, from imported capital goods.
The process of model building should be attuned essentially to the Let us then divide the world into two hypothetical countries, domestic
tasks which the model is expected to perform. Our discussion here is meant and foreign, to be denoted by America and England. The balance of pay
to illustrate how a model may be constructed to perform a particular task, ments between these two countries will result from the complex interaction
namely forecasting the balance of payments. A more relevant model would of ten markets. These are the markets for: (1) American importables (English
include domestic effects as well; but the process of model construction is exportables); (2) American exportables (English importables); (3) American
essentially the same in both cases. securities; (4) English securities; (5) American home goods (not tradeable);
 120 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
ECONOMETRIC MODELS OF THE BALANCE OF PAYMENTS 121
(6) English home goods; (7) American capital goods; (8) English capital same in each country. Events in the securities market and the labor market
goods; (9) American labor; and (10) English labor. are assumed to have an insignificant impact on the goods markets and are thus
It is not difficult to see how events in each of these markets may influence neglected. Although this Keynesian model represents a much simpler view of
the balance of payments. The first four markets will determine directly the the world economy than our ten-market model, it may nevertheless capture
in ternational flows of goods and services and securities. The next four the essential features of the important phenomena, such as the international
markets for home goods and capital goods will compete with the transmission of the business cycle. As we shall see, most of the models which
exportables and im portables markets for the existing resources. For have in fact been constructed have this simple Keynesian structure.
example, price increases for American home goods or American capital Let us now explore in more detail for our broader model possible argu
goods will tend to lower the American supplies of importables and ments for its simplification that are. analogous to those arguments implicit in
exportables as real resources are shifted from those industries to the home or the Keynesian model above. We will want to consider each market's impact
capital goods industries in response to the price increase. Similarly, on the balance of payments to try to make some judgment as to whether that
American demand will shift from the home or capital good onto the market may be excluded from the model or aggregated with another market.
importables and exportables. All of these consequences of price increases in The decision to exclude a market may be based on one of two propositions.
the home or capital goods markets wiHtend to increase American imports We may feel that the impact of a market on the balance of payments is rela
and decrease American exports. tively slight. Alternatively we may observe the complexity of a market and
The capital goods markets will also play the highly important role of conclude that only very great research efforts could make any quantitative
determining the level of investment in each of the industries. Of course, events sense out of the observed events. Accordingly, an educated guess as to the
in the goods markets will have an important impact on this decision also. In future values of the variables controlled by this market will be as accurate as
addition, the securities markets will play a role in determining the interest any model forecast. In other words, the cost of an improved forecast in terms
rate and hence the demand for capital goods. Finally the labor markets will of research effort may not be worth the amount of improvement so afforded.
determine the wage rates, which will in turn influence supply functions in all In what follows we will argue only on the basis of the first proposition: re
the goods markets and the demand function in the capital goods market. moteness.
Considering our model from the United States' point of view, the first
This description is not meant to include all possible interactions, but rather is step of simplification may be to discard the foreign markets: English home
meant to indicate that each of the markets may play an important role in influencing goods, English capital goods, and English labor. The impact of these markets
the balance of payments. Nor is this model meant to be the only possible model of an on the U.S. balance of payments may be very remote. Nonetheless, they will
open economy. As we have mentioned, more complicated models as well as simpler influence the English supplies and demands of American importables and
ones could surely be constructed by varying the level of commodity or regional American exportables. We may want to decide how these functions will enter
disaggregation or aggregation. The present model is meant to be a sort of middle our model before we make a decision about discarding any of the three
ground from which to view all balance-of-payments models. It will provide a foreign markets. For instance, if we should decide that the price of imports
foundation for the construction of both simpler and more complex models as well as (English exports) is to be forecast exogenously, then the link between this
serve as a useful reference from which to evaluate other models. price and the events in the three English markets may be neglected, and these
three markets may be discarded.
For example, let us consider a simple two-country Keynesian model If the international flow of capital forms a small stable entry in the
balance of payments, the securities markets may also be neglected. However,
Yf = Ci + Ii + Xi - Mi i = 1, 2 (5.5)
if investment demand (capital goods demand) is responsive to interest rates
Ci= ai +biYi i = 1, 2 or other events in the securities markets, we may need to include the securi
(5.6) Mi= Xi= ei + mi Yi i j= 1, 2 ties markets endogenously in order to predict investment.
The first two markets-importables and exportables-are certainly re
(5.7) Yf = y-p i = 1, 2 (5.8) quired in any balance-of-payments model. Each of these markets includes
two demand functions and two supply functions. The American importables
where Yf, Yf, Ci, Mi, Xi, and Ii are aggregate demand, aggregate supply, market is made up of American demand and supply and British demand and
consumption, imports, exports, and (autonomous) investment. The model supply. If we like, we may think of the American demand for British exports
has only two markets-the markets for Country l's and Country 2's goods.
Home goods, capital goods, and exports are considered to be essentially the
122 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
EVALUATION OF ECONOMETRIC MODELS OF THE U.S. 123
as being the difference between the American demand for importables and
An econometric model was used to forecast the level of the components
the American supply, that is, an excess demand function. Nonetheless, we
of the U.S. current account in 1968. The capital account was projected
will have to realize that the American demand for British exports will be
exogenously on the basis of several assumptions about such things as growth,
influenced by American supply factors, in particular the capital stock em
profit, and depreciation rates. The econometric model of the current account
ployed by the American suppliers of import-competing goods. Of course, in
divided the world into three regions: the U.S., Western Europe, and the Rest
the short run the capital stock is fixed and we may ignore its impact on the
of the World. For our purposes, we need consider only the U.S. and Western
flow of goods. We have, therefore, justified the severing of one link between
European regions of the model. In this case the structure of the model may be
the consumable goods markets and the capital goods markets, at least in
represented by two equations: U.S. demand for Western European goods
short-run models. However, in models designed to forecast several years into
and Western Europe's demand for U.S. goods.
the future, it may be quite important to maintain this link between investment
and the supply capacity of the various industries. It should be clear that this model is extremely simple. If we use our
This discussion could in principle be pursued at much greater length. ten market model as a guide, we will be able to find only the markets for
We will nevertheless terminate it here since there is no end of variation in the inter national goods. In addition, these markets consist of a demand side
models that might be constructed to fit particular empirical circumstances. only. The real problem of forecasting is left to the selection of the
The construction of an econometric model is an exceedingly difficult task if exogenous variables, and the projection of the many balance-of-payment
done in a reasonable fashion. Since our discussion has been centered not on items that are excluded from the model. The bulk of the Brookings Report
"what to do" but rather on "how one might decide what to do," we will in fact discusses these problems and the econometric model is hidden in an
have succeeded in our goal if the reader has a flavor of the complexity of the appendix. One cannot help wondering if the model had a significant impact
issues and a feeling about how one might seek solutions. on the projection or whether it was merely window-dressing for a basically
noneconometric fore cast-

The Rhomberg and Boissonneault Model A model which provides an

EVALUATION OF ECONOMETRIC MODELS OF THE interesting contrast to the Brookings model has been presented by Rhomberg
and Boissonneault (R & B) [19]. The R & B model is essentially an expanded
U.S. BALANCE OF PAYMENTS
version of the Brookings model. The R & B forecast is more heavily depend
ent on the nature of their model. As in the Brookings work, the world is di
We have already stressed the point that the only norm with which to vided into three regions. The features of the model that interest us are best
judge a model is performance. Let us consider then f?r illustra ive_ purposes illustrated by examining only the equations for the U.S. and Western Europe.
The structure may then be reduced to six equations, three for each region:
some models that have been used for prediction or pohcy analysis with regard
a consumption function, an import-demand function, and an export-supply
to the United States. Using our ten-market model as a guide, we will first
price equation. This contrasts with the Brookings model, which contains
explore these models to deduce the implicit answers to the questions we have
only the import functions. Thus, the R & B model includes the supply side
posed. We will then ask how well the model performs. I the event that the
in the importables and exportables markets and also the demand side of the
model performs well, we will have a reason to favor the kmd of structure em
two home-goods markets (U.S. and Western Europe). Four of our ten
ployed. If the model performs poorly, we may wish to avoid such a structure
markets are included, but with only the demand side for two of them. There
in the future.
are no interest rates; no concept of capacity; no mechanism for generating
domestic prices; and so forth.
The Brookings Report The first model we will examine was constructed
by the staff and associates of the Brookings Institution!n 1962 under contr ct
As it has been repeatedly emphasized, the absence of certain relation
ships from a model does not constitute grounds for criticism. Performance
with the Council of Economic Advisors [21]. At that time there was growmg
is what counts. Since both the Brookings and R & B models were used to
concern over large balance-of-payments deficits and the resulting gold out flow
generate a forecast for 1968 under two sets of similar assumptions, we will
from the United States. Government policymakers accordingly sought answers
be able to assess their accuracy. The forecast changes are given in Table 5.1
to two important questions: was the situation likely to continue, and, if so,
to gether with the actual values for 1968. It is evident that the two models
what policy measures would be appropriate to remedy it?
sub stantiaHy underestimated the changes that occurred on both the export
and
 EVALUATION OF ECONOMETRIC MODELS OF THE U.S. 125
124 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS

The actual deterioration in the current account balance from 1961 to

import sides. The initial assumptions for the increases in :eal GN turned
1968 was $3.2 billion. This was in sharp contrast to the current account sur
out, interestingly enough, to be almost exact. While the mcrease m GNP
plus that had been projected on the two assumptions noted. It is of course
prices was in fact greater than that which had been projected for both regions,
too much to expect that all the events that occurred between these years
it is noteworthy that inflation proceeded substantially faster in Western
could have been foreseen. This is particularly true with regard to the impact
Europe, as had been hypothesized. The projected increase in U.S. export
of defense expenditures for Vietnam and the sharp rise in U.S. income and
prices was in fact too low, whereas the increase in Western Europe's export
prices in 1967-68 that resulted in large increases in imports. It is
prices was between the initial and alternative assumptions.
nevertheless the case that the models did not yield very accurate
projections. The moral is thus twofold: we need better models for forecasting
TABLE 5.1
purposes; and medium term forecasting is a risky business indeed.
Changes from 1961 to 1968 in U.S. Current Account
(In Billions of Current U.S. Dollars) t The Prachowny Model A substantially more disaggregated model of
the U.S. foreign sector than either of the foregoing has been constructed by
Initial Alternative
Assumptions b Prachowny [18] on a quarterly basis for the period 1953-64. As noted in
Assumptions "
Table 5.2, his foreign sector contained 23 equations, including identities.
Brookings R&B Brookings R&B Actual' Some flavor of the degree of complexity or lack thereof in the various foreign
Exports of goods and
services sector equations is evident from the variables that are listed. The domestic
+11.2 + 9.3 + 7.6 + 7.5 +13.5 sector of the model is relatively small and highly simplified, yet it permitted
Merchandise Prachowny to build in some of the domestic and foreign sector interrelation
Service + 3.0 + 3.8 + 3.0 + 3.3 + 8.3 ships.
TOTALd
+14.3 +13.1 +10.6 +10.8 +21.8 A very important point to recognize with regard to this model is the
Imports of goods and fact that although a great number of equations have been added, the
services structure of the goods markets remains the same as in the simple Keynesian
+ 8.9 + 8.1 + 7.9 + 7.1 model. That
is to say, the various demand components are added together into a single
Merchandise +18.5
Service + 1.6 + 3.1 + 1.5 + 2.7 + 6.5 demand term, "aggregate demand," and supply is assumed to be forthcoming
TOTALd +10.5 +11.2 + 9.4 + 9.7 +25.0 to meet any demand that arises. In other words, there is a single domestic
+ 1.0 - 3.2 goods market with the simple Keynesian supply response of providing all
Current account balance + 3.8 + 1.8 + 1.2 that is demanded. The problem of capacity limitations on production is ig
Allowance for E.E.C. nored, as is the related problem of price determination. Such a Keynesian
discrimination 0.6 0.6 model should thus be applied only to short-run periods with ample excess
Current account with capacity. When demand begins to impinge on productive capacity the Keynes
E.E.C. allowance + 4.4 + 1.8 + 1.8 + 1.0 ian supply response is quite unlikely to be an adequate description of
reality. Longer-run models will of course need to deal with both the determi
a Based on the following assumed percentage increases for the U.S. and Wes!ern nants of prices and the determinants of domestic capacity. These considera
Europe, respectively: real GNP, 43 and 33; GNP prices, 11 and 20; and export pnces, tions are dealt with to some extent in the other models of the U.S. economy
4 and 11.
b Based on the following assumed percentage increases for the U.S. and Wes_tern to be examined below.
Europe, respectively: real GNP, 36 and 29; GNP prices, 11 and 11; and export pnces, After estimating the various equations, 3 Prachowny was able to calculate
4 and 7. impact multipliers to analyze the effects of changes in various exogenous
, The actual percentage increases from 1961 to 19?8 for the U.S. an? Western Europe,
respectively, were: real GNP, 42.3 and 31.9 (partly estimated); GNP pnces, 16.9 and 27.2 variables on the balance of payments and GNP. In particular he considered
(partly estimated); and export prices, 9.8 and 9.8. the impact on the balance of payments of the imposition of the Interest
d May not be precise due to rounding.
t Adapted from R. R. Rhomberg and L. Boissonnea1:11t, "Effects of Income and
Price Changes on the U.S. Balance of Payments," Internat10nal Monetary Fund, S:aff 3
In 'general, his statistical fits were distinctly better for the current account than for
Papers, XI (March 1964), 82; U.S. Department of Commerce [29, p. 27]; and Orgamza the capital account equations.
tion for Economic Co-operation and Development [14].
 126 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
EVALUATION OF ECONOMETRIC MODELS OF THE U.S. 127

TABLE 5.2
TABLE 5.2 (Cont.)
Prachowny's Quarterly Model of the Foreign Sector
Equation Dependent Variable Major Explanatory Variables
of the U.S. Economy, 1953-64 t 16 Foreign purchases of U.S. Same interest differential as Equa-
private long-term securities ' tion 13; lagged purchases. -
Equation Dependent Variable Major Explanatory Variables
17 Repatriation of dividends and Sum of foreign direct investment and
Foreign Sector interest earned in the U.S. other private assets owned by for
1 Imports of consumer goods eigners times U.S. long-term interest
Real disposable personal income;
rate; differential growth rate of GNP
relative prices; lagged imports.
2 Imports of investment goods in other OECD countries and U.S.
Real expenditures on producers' 18 Repatriation of dividends Sarne interest differential as Equa
durable equipment; relative prices; earned abroad b tion 13; lagged investment times
lagged imports.
3 Imports of raw materials average of Canadian and U.K.
Manufacturing production; real
long-term interest rates.
change in nonfarm business inven 19 Repatriation of interest Same interest differential as Equa
tories; relative prices; lagged im earned abroad b tion 13; lagged private assets times
ports.
4 Merchandise exports average Canadian and U.K. long term
Real world exports (minus U.S. ex rates.
ports); relative prices;· U.S. direct 20 U.S. short-term capital Covered interest differential between
investment; trade credit; lagged
movements ' U.K. and U.S. Treasury bill rates;
exports.
5 U.S. payments for foreign exports; dummy variable for Volun
Current disposable personal income.
travel tary Restraint Program.
6 U.S. receipts for foreign 21 U.S. long-term claims U.S. Treasury bill rate; exports;
Canadian disposable personal in
travel from Canada against foreignersd dummy variables for Interest
come; Canada-U.S. exchange rate.
7 U.S. receipts for foreign Equalization Tax.
Sum of consumer expenditures in
travel from rest of world 22 Import identity
France, Germany,Italy, and the U.K.
8 Transportation, private remit- 23 Balance-of-payments identity
Imports; travel expenditures abroad.
tances, and other services
9 Transportation receipts Exports; travel receipts. Domestic Sector
10 Private remittances Lagged remittances. 24 Consumption expenditures Real disposable personal income;
11 Miscellaneous service pay- Lagged payments. lagged consumption.
ments 25 Nonresidential construction Based on Liu
12 Private miscellaneous service Canadian GNP; GNP in the Euro 26 Producers' durable equipment [11]. Based on
receipts pean OECD countries. 27 Residential construction Liu [11]. Based
13 U.S. direct investment Differential between U.S. long-term 28 Investment in nonfarrn on Liu [11].
abroad" interest rate on government bonds business inventories Real GNP; lagged stock.
and average of Canadian and U.K. 29 Disposable personal income
rates; lagged investment. 30 Industrial production Real GNP.
14 Foreign direct investment in Same interest differential as Equa 31 U.S. long-term interest rate Real GNP.
the U.S. tion 13. Average quarterly yield on U.S.
15 U.S. purchases of foreign Same interest differential as Equa 32 GNP identity Treasury bills; lagged rate.
long-term securities tion 13; dummy variable for Interest
Equalization Tax; lagged purchases. "Excludes second and third quarters of 1957.
b Beginning first quarter of 1959.
t Adapted from M. F. J. Prachowny, A Structural Model of the U.S. Balance of 'Beginning first quarter of 1959 through 1965.
Payments. Amsterdam: North-Holland Publishing Company, 1969. • Through 1965.
 128 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
EVALUATION OF ECONOMETRIC MODELS OF THE U.S. 129
Equalization Tax and the Voluntary Restraint Program during 1963-65. He
presented in addition the results of some simple simulation experiments on The Wharton quarterly model [7] of the U.S. economy consists of 47
the balance-of-payments and GNP impacts of a I percent increase in the equations with unknown parameters and 29 identities. It covers in its pub
Treasury bill rate coupled with some continuing changes in government ex lished version an estimation period from 1948 to 1964. Like the Michigan
penditures. model, it is Keynesian in nature with respect to the determination of aggre
While Prachowny's results are of considerable interest, they cannot of gate output and employment but includes equations for the determination of
course be taken literally in view of the comparative simplicity of the model. prices, wage rates, aggregate supply, and factor shares. It contains in addition
That is, while the model is explicit in its treatment of the demand side for in a small monetary subsector dealing with the determination of interest rates.
ternational transactions in goods, services, and financial instruments, it ab Imports are divided into three categories in the model: crude and processed
stracts almost completely as we have noted from supply considerations in food, crude materials and semimanufactured products, and all other imports
the relevant markets. Moreover, the absence of a mechanism generating do (including services). There is a single equation for exports.
mestic prices and interest rates in the model is an important limitation. If the The import equation for crude and processed food is estimated in per
model is to prove useful for purposes of forecasting and policy analysis, the capita form with real per capita personal disposable income and relative
measurement of the capital account relationships especially must be improved prices as the explanatory variables. Imports of crude materials and semi
and a linkage accomplished with a comprehensive model of the real and manufactured products are assumed to depend on sales originating in the
financial relationships of the domestic economy. Despite these reservations, manufacturing sector, the change in manufacturing inventories, and relative
Prachowny's work represents an important step in the construction of a fairly prices. All other imports are assumed to depend on real personal disposable
detailed model of the U.S. balance of payments. income, relative prices, and lagged imports. Exports are assumed to depend
upon an index of world trade (proxy for world income), relative prices, and
Other Models There are a number of models of the U.S. economy lagged exports. U.S. export prices are endogenously determined in the model
now in existence that deal almost exclusively with the domestic sector. This while the world trade and price variables are exogenous. In the use of the
reflects in part the orientation of the Keynesian system towards relationships model to generate ex ante forecasts and for purposes of policy simulation,
involved in a closed economy, an assumption which until recent years at values for imports and exports can be obtained as a component of aggregate
least has been plausible for the U.S. in view of the relatively small size of demand (7, pp. 50-69].
its foreign sector. There is some question now, however, about the The quarterly model of the U.S. economy developed by the Office of
appropriateness of these models in view of the increased importance of the Business Economics [29] consists of 36 equations and 13 identities covering
balance-of-payments con straint and the consequent increased sensitivity of components of GNP, prices and wage rates, labor force and employment
U.S. economic policy to in ternational economic influences. It may be related magnitudes, income components, monetary variables, and miscel
instructive to look briefly at how the foreign sector is handled in a number laneous variables introduced to close the model. The model contains two
of these models in order to obtain some impression of the work yet to be equations for imports. Imports other than crude materials and foodstuffs are
done. dependent on real disposable income and the ratio of nonwage to wage in
Let us consider first the Michigan econometric model of the U.S. come. Imports of crude materials and foodstuffs are dependent on the real
economy (30], which is an annual model in which equations are estimated value oflagged private GNP. Neither equation contains a relative price term,
for components of aggregate demand; productive capacity and it is interesting to note. Exports are treated exogenously in the model. None
employment; in come, labor costs, and prices; taxes and social insurance; of the financial items in the balance of payments is included. The model is
and the financial sector. The change in imports, which is included as a part thus able to make a forecast of net exports as a component of aggregate
of aggregate de mand, is held to be dependent mainly on a composite demand.5
relation between the change in nonfarm GNP and capacity utilization and
$3.4 billion for imports. The observed preliminary changes were $1.9 billion for exports
on the change in rela tive prices. Exports are taken to be exogenously and $2.1 billion for imports. The change in the balance of trade thus turned out to be
determined and to change at some specified rate. A forecast is thus made of - $0.2 billion rather than the $0.4 billion that had been forecast.
net exports as one of the components of aggregate demand.4 No forecast is A separate quarterly model is now being developed at Michigan and forecasts based
on it were first presented at the Annual Conference on the Economic Outlook in November
made of any of the financial items in the capital account. 1968. Net exports were treated in this version of the model as completely exogenous.
5
• For example, in November 1966 according to [30, p. 4], it was forecast that the The forecast for 1965 in billions of current dollars seasonally adjusted at annual
rates was as follows [29, pp. 26-27]:
1967 increase over 1966 (in 1958 prices) would amount to $3.8 billion for exports and
(Footnote continued on next page)
130 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS EVALUATION OF ECONOMETRIC MODELS OF THE U.S. 131
The Federal Reserve-MIT econometric model [2] is a quarterly model It consists of the following principal sectors: consumption; residential con
of the U.S. economy that focuses mainly on the financial sector and on the struction; inventories; orders; investment realizations; investment
links between this sector and those for goods and services. Its primary pur intentions; foreign sector; government revenues and expenditures;
pose is to quantify monetary policy and the effects this policy has on the production functions and factor income payments; wages and prices;
economy. The model consists of three principal blocks of equations: a agriculture; labor force; monetary sector; and the automotive industry. The
finan cial block; a fixed investment block; and a consumption-inventory purpose of having such a large model is to capture the workings of the
block, which includes as well income shares, imports, and federal personal economy as an interrelated sys tem and to be able to make forecasts and
taxes. There is a single equation for imports that are assumed to depend on analyze the effects of policy in great detail.
real GNP and a measure of capacity utilization. Dummy variables are Despite the size of the Brookings model, it contains a relatively simple
included to capture the effects of the 1959 steel strike and the 1965 dock foreign sector. There are two equations for imports and one for exports. The
strike. Rela tive prices were omitted because they were found to be imports of finished goods and services are assumed to depend on real dis
unimportant sta tistically. Exports are treated exogenously. None of the posable personal income, relative prices, and lagged imports. Imports of
financial items in the balance of payments are considered. Thus, we again crude materials, crude foodstuffs, and semimanufactures are assumed to de
can obtain a model fore cast of net exports as a part of aggregate demand.6 pend on the change in real nonfarm business inventories, real gross product
The model was also used to analyze by means of simulation the effects of a originating, relative prices, and lagged imports. Exports of goods and services
$1 billion increase in un borrowed reserves, a $5 billion increase in defense are assumed to depend on real world exports excluding U.S. exports,
spending, and a IO percent increase in the personal tax rate. relative prices, and lagged exports. Import prices are apparently taken to be
The Brookings econometric model of the U.S. economy [9] is a gigantic
exoge nous in the model, whereas export prices are generated in a rather
affair compared with the other models we have mentioned. It contains more
complex manner from the price deflators for five producing sectors. The
than 300 equations and has involved data collection for over 2000 variables.
model was estimated for 1948-60 and furnished the basis for some forecasts
for 1961-62 as well as a number of different simulation experiments
involving changes in government expenditures, government employment,
IQ 2Q 3Q 4Q Year personal income taxes with and without changes in monetary policy, and
p A p A p A p A p A changes in monetary policy. The detailed results of the forecast as well as
Exports (exogenous) 34.7 34.7 40.4 40.4 40.1 40.1 40.8 40.8 39.0 39.0 the policy simulations thus include estimated values of net exports in
Imports 28.2 28.6 30.3 32.4 30.7 32.7 32.0 33.9 30.3 31.9 constant dollars [9, pp. 20 and 41].
Net Exports 6.5 6.1 10.1 8.0 9.4 7.4 8.8 6.9 8.7 7.1 Although the equation specifications differ somewhat, it should be clear
from our discussion that the foreign sector is treated on a relatively very
Predicted (P) imports apparently fell short of actual (A) imports in each quarter. This simple basis in the most noteworthy of the econometric models of the U.S.
was due mainly to the underestimation of imports other than crude materials and food
stuffs. Predicted net exports were thus $1.6 billion below actual net exports. economy. This is in part a holdover from the period in which the foreign
6 In the published version of the model [2, p. 22] predictions were given only for sector played only a minor role in the economy. It also reflects the fact that
total imports (in billions of dollars at annual rates) in the context of the complete con the construction of a comprehensive model of the foreign sector that includes
sumption-inventory block:
international capital transactions is a very difficult task. It is obvious that
the foreign sector has to be treated comprehensively in countries that are
1965 1966
much more dependent on international trade and foreign capital markets
3Q 4Q IQ 2Q 3Q 4Q than may be the case for the U.S. 7 But given the increased importance of
Predicted imports 33.6 35.1 36.5 37.3 38.0 38.6 balance-of-payments policy considerations in the U.S. especially since the
Actual imports 32.9 34.4 36.0 37.1 39.0 39.7 early 1960's, much remains to be done to integrate the foreign and domestic
Difference 0.7 0.7 0.5 0.2 -1.0 -1.1 sectors in econometric models of the U.S. economy.

Allowing for data adjustments in the actual value of imports, predicted imports for the
7
See [10] for a quarterly econometric model of Canada, in which relationships
last two quarters of 1965 were closer to actual imports in the FED-MIT model than in describing foreign trade and international capital movements are of central importance.
the OBE model noted in the preceding footnote. Work is now in progress at the Bank of Canada to link the foregoing model with the
FED-MIT model of the U.S. economy.
 132 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
APPENDIX TO CHAPTER 5 133

CONCLUSION APPENDIX TO CHAPTER 5

"T\VO-GP. P" lvIODELS

Econometric model building is properly viewed as a tool for optimal de
cision making. A model builder may assume one of two roles in t e decision
making process. He may provide information about future events m the fori_n A separate species of model worth considering is the so-called two-gap model,
of probability statements, in which case a f?recast consume can make his which has been used to forecast the foreign aid requirements of developing
own choice of actions by weighing the likelihood of the vanous outcomes. countries. The genesis of two-gap models stems from the supposed rigidities
Alternatively, the forecaster may usurp these decision-making po":ers _and and lack of resources in less developed countries (LDC's) that may hamper
provide only a point estimate, in which case the forecast consumer :s dnven the effectiveness of traditional policy instruments in these countries in achiev
necessarily to a particular action. We have argued that the former (mforma ing their stated economic goals. In such circumstances, foreign resources in
tion-provider) role is the appropriate one. the form of grants and loans will help in the achievement of their goals.
The construction of an econometric model is much more an art than a The goals typically involve high levels of employment, balance in foreign
science. The choice of variables and model structure is exceedingly difficult payments, and rapid economic growth. We know from the theory of eco
and has often been made more by chance than by design. While the only n? nomic policy that at least the same number of policy instruments is required
rm with which to assess a model is performance, this unfortunately provides if the aforementioned goals are to be achieved. Thus, the policy instruments
little or no insight into the problem of model building. Bu what this norm in question will commonly involve fiscal, exchange-rate, and monetary
does suggest is that a model ought to evolve by annual adJustments, rath r poli cies. In effect then if these policies do not work well in the LDC's, by
than be created at a single point in time. However, any model must begm making foreign aid available, we are providing these countries with an
sometime, and its initial form may greatly influence its evolutionary path. additional instrument of policy. What we shall be concerned with in our
How, then, should the model begin? . . . . . discussion therefore is estimating the foreign resources required to meet a
We have tried to provide a thought-structure withm which the decis10ns reasonable set of goals established by an individual LDC subject to the help
involved in model building could be made. We first divided the universe of and coopera tion of a high-income, industrialized donor country.
economic exchanges into a set of mutually exclusive markets, in such a way The issues involved in forecasting foreign resource requirements are
that all exchanges within a market are essentially the sa e and that a y t"".o especially complex because it is impossible to know what is required from
exchanges from different markets are fundamenta y different. Keepmg m foreign sources over a period of five or ten years unless we have a fairly accu
mind just what our model is meant to do, we xa med each ?f the markets rate idea of what determines economic growth. Since our knowledge of the
to determine whether it could be discarded with little or no impact on the determinants of growth is unfortunately still rather limited, researchers in
performance of the model. Through this process an appropriate structure terested in constructing economic models for forecasting foreign needs have
may be selected. . had to make numerous simplifying assumptions. While the range of possible
Finally, we have considered some examples of eco ometnc m?dels that assumptions combined with the diversity of intellectual temperament of in
have actually been used in balance-of-payments forecastmg and policy analy dividual researchers might have resulted in a fairly wide variety of models,
sis in the case of the United States. Except for Prachowny'_s model, these the two-gap models to be discussed presently are all quite similar. There may
models have employed a fairly simple structure for the foreign sector. We accordingly be considerable room for alternative specifications of these
have been afforded therefore relatively little experience that would suggest models.
the most appropriate structure for balance-of-payments models. Hopefully, The models in question are highly aggregative and look upon economic
important advances in this area may be realized in the near future. growth as stemming exclusively from gross capital formation. Output ca pacity
is most typically given by the Harrod-Domar function, the product of the
constant output-capital ratio and the total capital stock

(5.A.l)
134 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
5
APPENDIX TO CHAPTER 135
where Q is the quantity of aggregate output, k is the capital-output ratio, and
In the ex ante sense, one gap will ordinarily be larger than the other. If
K is the capital stock. the investment plan is to be realized, foreign resources will be required to fill
In these models there are two limits to the amount of capital formation.
the larger of these two gaps. The smaller gap can then be widened to conform
The first is simply the lack of adequate resources. The economy in question with the ex post identity (5.A.2). For instance, if the export-import gap is the
may not have the capacity both to supply the consumption needs of its ex ante constraint and if foreign resources are available to fiil that gap, then
popu lation and to produce or trade for the capital goods required for either savings may be decreased or investment increased in order to bring
growth. To estimate the resource needs, it is necessary to calculate full about the ex post equality of the gaps.
employment out put and subtract from that the level of required real It should be noted that the foregoing considerations are applicable to
consumption. The figure obtained represents the supply of savings, or the developed as well as developing countries. The very important distinction,
real output available for capital formation, after consumption needs are however, is that the developed countries typically have greater mobility of
satisfied. In the event that the desired level of investment exceeds this resources, which will make their policies aimed at eliminating trade im
figure, foreign resources will be re quired to fulfill the investment objective. balances more effective. Many LDC's, in contrast, are forced to rely
This inadequacy of domestic productive capacity is usually referred to as heavily on imports especially of capital goods for investment purposes.
the savings-investment gap, reflecting the fact that the gap is the difference Once a level of investment is selected, the level of imports may be more or
between full employment real savings and the desired level of investment. less fixed. Given the level of exports, a trade imbalance may thus be a
The second constraint on capital formation is the foreign exchange or necessary con sequence of the investment program.
export-import gap. Accelerated growth is supposed typically to be It will be evident from our discussion that there are three fundamental
associated with rapidly expanding imports of goods and services. Exports, relationships used in two-gap models: the production function, savings func
on the other hand, will tend to grow as the product of the developed tion, and import function. Exports are typically thought to grow exogenously,
countries' growth rate and their income elasticity of demand for LDC and investment is calculated from the production function once a target
exports. Growth in exports may be relatively low due to the preponderance growth rate is selected. This is of course an exceedingly simple description.
of primary products in total LDC exports. The consequence of this disparity
in the rates of growth of imports and exports is that many LDC's will and
experience chronic balance of-payments deficits. Foreign resources will thus M == 10 + 0.3Q (b)
be required to :finance these deficits if the growth of GNP is to be where Q is the real value of GNP. If full employment Q [calculated from the capital stock
sustained. by relationship (5.A.1)] is at the level of 100, then savings and imports can be calculated
Although there is no ex ante relationship between the two gaps, they are as S = 10 and M = 40. Consider now the following cases with investment I and exports X
given in the parentheses.
observed to be the same due to the following ex post identity Case I: (I = 10, X = 40). The desired level of investment and the exogenous level
Q - (C + I) = S - I= X - M = IF (5.A.2) of exports are such that neither gap is operative. The equilibrium relationship (5.A.2) is
satisfied.
Thus, according to this identity, if the domestic purchases of goods (C + /) Case II: (I = 20, X = 40). Investment exceeds the available savings, but the export
import gap is not operative. An inflow of 10 from foreign sources is required to support
exceed aggregate output Q, this will be equivalent to the excess of invest this investment. The country uses the inflow to increase its imports from 40 to 50, thereby
ment I over savings S, which will in turn be equivalent to the excess of im ports creating a balance-of-payments deficit of 10. The ex post identity (5.A.2) is thus seen to
hold.
M over exports X, which finally is equal to foreign investment or the quantity
Case III: (I = 10, X = 30). Savings supply is sufficient, but there is a balance-of
of foreign resources made available for domestic use /p. payments deficit of 10. An inflow of 10 to finance the deficit can be used either to
If we consider these relationships in an ex ante or desired sense, the increase investment or to reduce savings (increase consumption). This change is such that
gaps need not be the same. In the event that the ex ante full employment the ex post relationship (5.A.2) holds.
In the event that the inflow from abroad is not forthcoming, the country's growth
levels do not conform to the equilibrium relationship (5.A.2), either objectives will not be fulfilled. But whatever occurs, the ex post relationship (5.A.2) must
government policy must adjust the ex ante values or the amount of hold. Thus, in Case II, the actual level of investment may be reduced to 10 or an additional
10 in real savings may be forced upon the economy. Case III will require the restriction of
income/expenditure will diverge from the full employment level to induce' a imports, perhaps through slower growth of GNP.
change in the actual values of expenditure to conform with the equilibrium The foregoing examples should indicate clearly the difference between the ex ante
equation (5.A.2).1 gaps and also the necessary equality of the ex post gaps. The examples also imply that
when both gaps are operative, foreign resources will be needed to fill the larger of the
1 This can be illuminated with the aid of an example. Suppose that savings and im two gaps.
ports are given by
Q - C = S = -10 + 0.2Q (a)
136 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
APPENDIX TO CHAPTER 5 137
It may be useful in any event to evaluate the model's relationships in more
detail. be an improvement in the capital-output ratio. That the capital-output ratio
varies substantially in fact among countries at different levels of development
The Production Function As indicated earlier, the production function attests to the significance of this point.
is typically of the Harrod-Domar type
The Savings Function Savings are typically related to output as
1
Q=-k K S =a+ {3Q (5.A.5)
This involves the assumption that the available labor is not a significant This is of course an extreme simplification. Furthermore, the use of
con straint on output, and that even if substitution between capital and labor historical data to determine the marginal savings rate is questionable.
is possible in production, no substitution in fact occurs. Furthermore, the Offhand, there would seem to be a presumption that a developing economy
con stant k rules out any possibility of shifting investment from less would experience rather wide shifts in the marginal savings rate over the
productive to more productive activities. time period for which a projection is being made. This would be particularly
The constant k may be estimated in a number of ways, the most true when the govern ment made a conscious effort to increase the savings
straight rate. In addition, the data observations which are available may not
forward of which is a regression of LiQ, on 11-1 accurately describe the savings function. When domestic investment
opportunities are absent, full employ ment will be sustainable only at low or
(5.A.3) zero savings. Government policy aimed at maintaining demand for full
employment would accordingly reduce savings below the theoretically
where 11-1 is the gross domestic capital formation in the previous period. A possible savings rate, which is to be estimated.3
similar approach, suggested by the United Nations group [27], specifies
The Import Function As before, by means of historical data, a simple
l t
Q, = a + -k I:, Ir (5.A.4) regression of imports on GNP is often used to explain imports
r=O
M = a +bQ (5.A.6)
where Ir refers to gross capital formation.2
These various estimates of the capital-output ratio are based on his The objections to such a procedure made with regard to the savings function
torical data, and it is taken for granted in this regard that the productivity of could be repeated here almost verbatim. An import function such as Equa
investment is immutably fixed. This is a rather restrictive assumption, for it tion (5.A.6) is excessively simple and is doubtfully stable. The use of historical
may well be that historical performance is oflimited relevance in the develop data ignores the very significant performance aspects of imports. That is, to
ment context. In fact, a most significant aspect of a development plan may the extent that relatively poor performance in the past is reflected in his
torically high levels of imports, should such performance be rewarded by
2 A slightly different formulation by Chenery and Eckstein [4] is based on the

propo sition that a part of the gross investment which occurs is allocated to replacement larger inflows of foreign assistance? Also, to what extent should we expect a
and social overhead capital. The amount thus allocated is a constant share z of current lowering of the import propensity due to import substitution? Are historically
produc tion. The capacity-creating investment is reduced to that extent low levels of imports the result of excessive government interference causing
various inefficiencies? If so, development objectives should allow for some
(c) what higher levels of imports to promote efficiency. These various considera
or tions suggest that the historical data on imports in the form of Equation
(5.A.6) are of doubtful relevance to a development projection.
(d)
3
On this point Chenery and Eckstein [4] argue that investment opportunities are
The constant term k in a regression of the form in Equation (d), which relates the ratio directly related to exports. They argue further that foreign capital inflow can substitute
of investment to the change in output to the inverse of the rate of growth of output, is for domestic savings, and that savings will be depressed by such capital flows. Their
seen to be the incremental capital-output ratio. arguments are supported by a regression of savings on GNP, inflow of foreign capital,
It may be noted further that Chenery and Bruno [3] in the case of Israel estimated k and the export/GNP ratio. As one would expect, they find the marginal propensity to
from input-output tables. save to be higher than that calculated by simple regressions of savings on output.
 APPENDIX TO CHAPTER 5 139
138 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS

A somewhat more reasonable description of imports is to disaggregate

into imports of consumption goods Mc and capital goods Mr
.,-, 0 .,-, 'SI" 'SI" 0
1.0C-:r--:10\0.,..;
w (5.A.7)
(5.A.8)

CO 00 \0 t-
indicating that imports for consumption are a function of gross output while
capital goods imports depend on investment alone. The use of historical
data to describe the import content of investment is more justifiable than its
c:: use in describing the required consumption-goods imports. In addition to
\-IN O tf') t
0 0 0 00 00 00 Tl these variables, Chenery and Eckstein [4] use reserves and export earnings
M ff) MN t'°'i N C:

: wa
. 9 "1"'-1 ..-I ....-,! ;:!
to re flect the scarcity of foreign exchange.
'-' 0
u Exports Exports X are typically assumed to grow exogenously through
time t
(5.A.9)
where A refers to the rate of growth. An alternative approach is to make ex
ports to developed countries a function of demand conditions there, usually
O O O 00 N <n
N N N N NN with output as the explanatory variable. Unfortunately, such a procedure
Ml \-IC'l000\00\C
oo'Sl''Sl"M'Sl"<nt cannot insure improved forecasting unless the value of output in the de veloped
N
economies can be reasonably well forecast.

Some Estimates of the Two Gaps Some indications of the orders of

magnitude of the two gaps are given in Table 5.A.l. Estimates of the
savings gap are about $6 billion to $7 billion annually for 1970, while
estimates of the foreign exchange gap, including both goods and services,
run on the order of $12 billion to $20 billion. These estimates are subject of
course to all the problems we have mentioned. That is, the projections
assume that there are fixed and stable relationships involving the capital-
"C)
V)
output ratio, the behavior of savings, and changes in foreign trade, and
(I)
I..
(I)
f2 12 f:: that despite their
;,- 0\ 0\ 0\ 0\ 0\ 0\
0 -I
simplicity these relationships capture the essence of economic growth in the
0
12 \-I T""'l "l"'-l \-1
1DJJc!,c!,c!ic!i
\-I T""'I

LDC's.
·u
'd0 :-;:;-.,.,.,.,\0\D\D\D
V) 0\ 0\ 0\ 0\ 0\ 0\
O\,-l,-1\-1,-l,-l"l"'-I

,-<
Conclusion The two-gap model is evidently subject to many
5 criticisms.4 Nonetheless, the model will prove or fail to prove itself only in
performance. Unfortunately, however, there is almost no way to assess its
performance. The implied forecasts are based on the assumption that the
"g' foreign aid re quired to fill the two gaps is in fact forthcoming. Inasmuch
(I)
C: as none of the
2 ii:°'
(I)
;:!
0
rJl
Ml
-1-R)
4
Many of the same considerations we have discussed are relevant also to
projections of debt servicing in which estimates are made, under various assumptions
0
s
U'.l
ci.
0. concerning the volume and terms of aid and the growth of current account receipts, of the
net financial flows from the industrialized countries to the LDC's and the proportions of
current account receipts that may be preempted by debt servicing.
 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS
REFERENCES 141
140
10. Helliwell, J. F., et al., "Econometric Analysis of Policy Choices for an Open
forecasts was actually accompanied by the "required" aid, the model cannot Economy," Review of Economics and Statistics, LI (November 1969), 383-99.
be said to have been tested. But even in the absence of such tests, all of the ob
jections we have outlined will properly make us skeptical of the results based 11. Liu, T. C., "An Exploratory Quarterly Econometric Model of Effective
Demand in the Postwar U.S. Economy," Econometrica, XXXI (July 1963),
upon two-gap models.5
301-48.
12. Maizels, A., Exports and Economic Growth of Developing Countries. Cambridge:
Cambridge University Press, 1968.
REFERENCES 13. Organization for Economic Co-operation and Development, M. K. Evans,
An Econometric Model of the French Economy. Paris: OECD, 1969.

1. Balassa, B., Trade Prospects for Developing Countries. Homewood: Richard D. 14. Organization for Economic Co-operation and Development, National Ac
Irwin for the Yale University Economic Growth Center, 1964. counts of OECD Countries, 1958-1967. Paris: OECD, 1969.

2. Board of Governors of the Federal Reserve System, Federal Reserve Bulletin, 15. Organization for Economic Co-operation and Development, Quantitative
Models as an Aid to Development Assistance Policy. Paris: OECD, 1967.
Vol. 54 (January 1968).
16. Organization for Economic Co-operation and Development, Techniques of
3. Chenery, H. B. and M. Bruno, "Development Alternatives in an Open
Economic Forecasting. Paris: OECD, 1965.
Economy," Economic Journal, LXXII (March 1962), 79-103.
17. Pincus, J., Trade, Aid and Development. New York: McGraw-Hill Book Com
4. Chenery, H.B. and P. Eckstein, "Development Alternatives for Latin America,"
pany for the Council on Foreign Relations, 1967.
Journal of Political Economy (forthcoming).
18. Prachowny, M. F. J., A Structural Model of the U.S. Balance of Payments.
5. Chenery, H.B. and A. M. Strout, "Foreign Assistance and Economic Develop
Amsterdam: North-Holland Publishing Company, 1969.
ment," American Economic Review, LVI (September 1966), 679-733.
6. Chenery, H. B. and A. M. Strout, "Foreign Assistance and Economic De 19. Rhomberg, R. R. and L. Boissonneault, "Effects of Income and Price Changes
velopment: Reply," American Economic Review, LVIII (September 1968), on the U.S. Balance of Payments," International Monetary Fund, Staff
Papers, XI (March 1964), 59-124.
897-911.
20. Rhomberg, R. R. and P. Fortucci, "Projection of U.S. Current Account
7. Evans, M. K. and L. R. Klein, The Wharton Econometric Forecasting Model.
Balance for 1964 from a World Trade Model," International Monetary Fund,
University of Pennsylvania, Department of Economics, Studies in Quantitative
Staff Papers, XI (November 1964), 414-33.
Economics, No. 2. Philadelphia, 1967.
21. Salant, W. S., et al., The United States Balance of Payments in 1968. Washing
8. Fei, J.C. H. and G. Ranis, "Foreign Assistance and Economic Development:
ton: The Brookings Institution, 1963.
Comment," American Economic Review, LVIII (September 1968), 897-911.
22. Suits, D. B., "Applied Econometric Forecasting and Policy Analysis " in
9. Fromm, G. and P. Taubman, Policy Simulations with an Econometric Model.
Forecasting on a Scientific Basis, Proceedings of an International Su mer
Washington: The Brookings Institution, 1968. Institute held in Curia, Portugal, September 1966. Lisbon: Centro de Eco
nomia e Financas, 1967.
, Indeed, some writers on foreign aid question the whole exercise of calculating the
two gaps. For example, Pincus [17, pp. 301 and 304] has stated: 23. Suits, D. B., "Forecasting and Analysis with an Econometric Model," American
The underlying issue is ethical, not technical. There is no way to estimate Economic Review, LII (March 1962), reprinted in R. A. Gordon and L. R.
the appropriate aid total except in light of agreed standards. But [Link] Klein, Readings in Business Cycles. Homewood: Richard D. Irwin Inc
about aid levels implies an agreement about goals and methods of reachmg 1965, pp. 597-625. ' .,
them••••
All econometric or impressionistic estimates of trade and savings gaps are in 24. Suits, D. B., The Theory and Application of Econometric Models. Center of
effect techniques of quantifying discontent according to a certain set of standards. Economic Research, Training Seminar Series. Athens, 1963.
They put the seal of rationality on [LDC] aspirations. Because it is a safe bet
that many LDC's could grow faster if they could get more aid, the inaccuracy 25. Theil, H., Economic Forecasts and Policy. Amsterdam: North-Holland Publish
of such measures does no great harm. In the eyes of those who support increases ing Company, 1961.
in foreign
aid, some gap is better than none.
142 FORECASTING AND POLICY ANALYSIS WITH ECONOMETRIC MODELS

26. United Nations, Macro-Economic Models for Planning and Policy-Making.

New York, 1967.
27. United Nations, Studies in Long Term Economic Projections for the World
Economy. New York, 1964.
28. U.S. Congress, Joint Economic Committee, The United States Balance of Pay
ments. Washington: U.S. Government Printing Office, 1963.
29. U.S. Department of Commerce, Office of Business Economics, Survey of Cur
rent Business, 46 (May 1966); 49 (June 1969).
30. University of Michigan, The Economic Outlook for 1968. Papers Presented to
the Fifteenth Annual Conference on the Economic Outlook. Ann Arbor, 1967.
31. Vanek, J., Estimating Foreign Resource Needs for Economic Development. New
York: McGraw-Hill Book Company, 1967.