In this lecture I propose to discuss the experience we have had with the method of model building as a contribution to economic science and the prospects for its further application. First of all I want to remind you of the essential features of models. In my opinion they are: (i) drawing up a list of the variables to be considered; (ii) drawing up a list of the equations or relations the variables have to obey and (iii) testing the validity of the equations, which implies the estimation of their coefficients, if any. As a consequence of especially (iii) we may have to revise (i) and (ii) so as to arrive at a satisfactory degree of realism of the theory embodied in the model. Then, the model may be used for various purposes, that is, for the solution of various problems. The advantages of models are, on one hand, that they force us to present a "complete" theory by which I mean a theory taking into account all relevant phenomena and relations and, on the other hand, the confrontation with observation, that is, reality. Of course these remarks are far from new.

While building models econometricians were often forced to supplement "literary" theories, since these often did not specify all relationships they were implicitly using.

Models have been built for a number of different purposes; first of all, for purposes of explaining actual developments and next, for finding ways of influencing actual development in some desired direction. Another aspect is whether short-term or long-term movements were the objective of either explanation or policies. There are large numbers of further alternatives of focussing. We will discuss some of them in this lecture.

First, I am going to discuss a number of experiences econometricians had with the activity of model building. Some of us were masters in hunting after high correlations, that is, good fits with observed values. In fact this was part of the art. Some of our critics thought this was easy enough and indicative of the futility of the activity. It wasn't always so easy, however. Some of the fits in our models never became very good, or, if finally they had been forced into a high correlation broke down a few years later. I am afraid that the first subject I tackled in my work for the League of Nations, namely to explain the fluctuations in investment activity never has become a great success. In the Netherlands Central Planning Bureau we found it safer, after some years, to ask industrialists for their investment programs rather than rely on an econometric explanation. Also government expenditures were among the variables difficult to explain. In both cases we may account for the lack of success by the fact that a small number of decision makers determine the picture and that hence random deviations will be important.

In a more general way many of us know that quite a few business cycle models were "forecasting" the turning points only after they had occurred. Ragnar Frisch (6) was quite right when, at an early stage of model building, he introduced random shocks as an essential element of the business cycle, leaving the cumulative process between turning points rather than the latter themselves as a thing that really could be explained by the models. Even so some turning points can be explained by the inner dynamics of economic systems.

In a number of cases models were hardly necessary to clarify some features of reality. I could not help thinking of this class of cases when recently I saw as a conclusion of one recent model that "Japan was a success in development". I thought we knew that already. Let me add, however, that the same model did explain something more.

During our hunting for good fits we did sometimes learn, as it should be. Thus, annual price fluctuations in beef prices could only be explained satisfactorily by the introduction, with a negative sign, of fodder prices some time before. High fodder prices force peasants to slaughter part of their livestock and hence depress beef prices. This was an aspect my collaborators and I had not been aware of and which text books on agricultural economics did not explicitly state. Another example was the explanation of the fluctuations in the general wage index in Britain before 1900 (20). No good fit could be obtained unless as one of the explanatory variables an index of mineral prices were included. After an intensive search I found that among the wages in various industries those for miners fluctuated by far the most and that for quite some time there prevailed a sliding-scale arrangement linking miners' wage rates directly with wholesale coal prices, something inconceivable to-day.

In several parts of our science, and I presume in other sciences as well, we must beware of following vogues too easily. Model building has become a vogue, just as, after that, linear programming or matrix algebra have become. Of course warnings against vogues are first of all coming from those who don't command the techniques implied. This is why I am myself inclined to hesitate to apply one of the two latter methods mentioned. But a critical examination of the structure of the problem before we try to solve it remains useful. And let me add immediately that linear programming does constitute a very useful technique in many cases indeed.

Returning to models, I am sometimes wondering whether, upon looking at some recent work by planners, I should not repeat the famous words by Goethe's Zauberlehrling "Die ich rief die Geister werd' ich nun nicht los" ("The ghosts I called I can't get rid of now"). Sometimes indeed some of our followers overdo model building.

In an attempt to evaluate what model building has contributed to the theory and practice of economic science I feel that at least we can say that models have had a didactic value. Often in our text books we bring simplified, not to say over-simplified, pictures of reality which nonetheless contribute to making understood some essential features of that reality. This is true already of some models inspired by Lord Keynes' fundamental work (10). It is true also of Leontief's input-output models (13). If I am allowed to quote a recent example I am guilty of myself, the same can be said of models introducing the difference between tradables and non-tradables (14). This model shows that if a country wants to eliminate a balance of payments deficit by living within its means, that is, by reducing its expenditure to its income, income itself is bound to fall and not so little.

What I called "didactic value" also stands for communication value. The ability of a planning expert to communicate with politicians and with citizens constitutes an important element in any type of democratic or semi-democratic planning and such communication can be enhanced by relatively simple models. In order not to misrepresent reality, however, there will be a need for a succession of models, as used in planning in stages or, as we now say, multi-level planning (12).

I do think, however, that the utility of models goes beyond their didactic value. They are a real and essential element in the preparation of well-coordinated policies. But they cannot do this job all by themselves. Models constitute a framework or a skeleton and the flesh and blood will have to be added by a lot of common sense and knowledge of details. Again, as a framework they can be of vital importance. Some of the simplest models were sufficient to show that the investment programmes recommended by the World Bank in its early days were not of the necessary order of magnitude. During the Great Depression already the same could be said of some of the programmes of an anti-cyclical character.

The framework I am referring to supplies the main ingredients for coordinating government policies at the level of a central government, that is coordinating the policies of the various ministries. Already many of the details concerning one ministry only would require the introduction of partial models, or could at least be left to them.

For short-term models sufficient time has elapsed already since their construction started, in order to test their forecasting performance. Several publications of the last ten years or so dealt with that subject, comparing, among other things, forecasts made in Scandinavian countries, Britain and the Netherlands. One score of success has been the number of turning points correctly predicted; this score is interesting since so-called primitive forecasts, that is extrapolations of past movements, are unable to produce turning points. Some models have been able to correctly forecast two-thirds of the turning points.

A need generally felt by model builders and their critics is the need for refinement, that is, for the introduction of many more variables. In a way this experience again was a lesson also to economists in general, since often their arguments run in terms not showing this degree of detail. One example we in Benelux experienced: real development showed that the grossly increased volume of trade between the three countries did not imply that whole two-digit industries were wiped out in one or the other country, but only much smaller subsectors. Here one has to introduce hundreds if not thousands of different products in order to do justice to reality. The same applies to the problem of the optimal division of labour (7) among all countries of the world, although the establishment of such an optimal division of labour may wipe out more important parts of two-digit industries. Here we tried to apply the Heckscher-Ohlin principle in a very concrete way. This also implies that for the choice of the best investment projects of some developing country much preciser information is needed than ordinary statistics can give us. It is a well-known experience that even so-called project data are far from sufficient to design the optimal development policy.

Two other examples of the need for refinement of models may illustrate the case. One has been often mentioned by Erik Lundberg (15) in his analyses of anti-cyclical policies, mainly financial policies. Partly we have a need here for much smaller time units and corresponding information, for these time units, on a number of relevant variables. Among the variables are a number of expectations not usually collected by statistical bureaus or even central bank statistical departments.

The other additional example of the need for more refined models and information can be taken from the experiences of the United Nations Research Institute for Social Development (UNRISD). The essential feature of this Institute's work is to include a number of so-called social variables. Leaving apart the question of how to define these variables - at present we have three different definitions competing - in two ways refinement is needed. On one hand, information about many more aspects of social phenomena is needed. Taking education as an example, the usual data available, such as enrolment, are too crude and should be specified, say, with regard to the type of education. On the other hand, data for smaller geographical units are needed, closer to what in other contexts is known as the "grass roots". The intuitive judgment of several sociologists that research and inquiries at this lowest level are by far more productive than the macrosocial research undertaken by UNRISD should be so interpreted (17). This implies that the question is not whether quantitative models are or are not productive. Precise knowledge about interrelationships can be obtained only by the technique of quantitative models; but the lack of homogeneity of crude information is the reason for the lack of success in the social area and hence refinement of the base material is the real need here.

All these refinements will make for ever more complicated models and therefore threat to make models unmanageable. This once again underlines the need for several stages of decision-making and hence of planning. As already said, the need for communication with the people and groups involved also points into the direction of this step-wise use of models. So also does the organizational aspect of decision making; a correspondence between the organizational setup of an optimum socio-economic order and the levels or stages of planning and the use of models for it is desirable. One of the future features of such a setup will also be the more precise location of the flows of information and, more particularly, the type of information needed.