Exploration Versus Prediction Models: Some Significant Aspects of the Systems Dynamics National Modeling Project

Published by F. Eberstadt & Co., Inc, Institutional Department
April 18, 1977

"You will reply that reality is under no obligation to be interesting. I shall reply that reality can get along without that obligation, but a hypothesis can’t."

—Jorge Luis Borges

Introduction

The Systems Dynamics Group of the Sloan School of Management at M.I.T. is in process of constructing a national economic model. This project draws on some twenty years of work, led by Professor Jay W. Forrester, on the modeling of economic decision-making through the methodology of systems dynamics analysis (see Jay W. Forrester, “A New View of Business Cycle Dynamics,” Journal of Portfolio Management, Fall 1976, and references incorporated therein).

Recently, we have been becoming progressively more acquainted with the work of the Systems Dynamics Group. In this regard, F. Eberstadt & Co., Inc. has sponsored seminars for institutional investors in New York and Boston at which Professor Nathaniel J. Mass, Director of the Systems Dynamics National Modeling Project, has presented the core analysis and discussed the implications of the Group’s ongoing efforts.

These notes are intended to identify certain aspects of the Systems Dynamics Group’s work which we consider to be particularly significant. We recognize that both the methodology and the conclusions of this work have raised controversial issues among professional economists as well as “common readers.” This has been particularly the case with respect to application of the systems dynamics approach to the very long-run analysis of world trends, as presented in Professor Forrester’s 1971 book, World Dynamics, and publicized most dramatically in The Limits of Growth. Here, we are exclusively concerned with the National Economic Model which the Systems Dynamics Group is now constructing.

In particular, we are concerned with the economic behavior and relations of the “production sector” of the national model, analysis of which we understand to be most mature. We intend to follow closely the development of other sectors of the model (e.g., the financial and government sectors) and the process by which all the sectors are assembled into an integrated model. The controversy which this model may arouse, both with regard to methodology and to implications for the functioning over time of an economic system, in our judgment would be largely reflective of the genuine analytical innovation which we deem it to express.

These notes discuss the following issues:

(1) The methodology of the National Modeling Project in contrast with the methodology of econometric analysis;

(2) The economic theory of the national model in contrast with “neo-classical” theory;

(3) Implications of the Model for the Present Economic Environment; and

(4) Implications of the Model for Government Policy and Future Economic Performance.

(The discussion of the implications of the systems dynamics model are, to a degree, independent of the analyses of the methodological and theoretical issues and may be read separately.)

We conclude by identifying certain questions on which, in our judgment, realization of the model’s potential will in part depend.

Methodological Issues

During 1974 and 1975, we published a series of Research Notes which discussed at some length certain critical theoretical and methodological failures of current economic analysis, particularly econometric techniques (see especially: The Great Uncertainty: The Problem of Political Economy in 1975, dated June 12, 1975). Against this background, we have been most interested in the approach embodied in the systems dynamics economic model.

Our critique of econometrics focused on four fundamental issues:

(1) In place of a theory of economic behavior and of functional economic relationships, econometrics abstracts statistical correlations from the available data;

(2) On the one hand, econometrics is perennially vulnerable to treating a statistical correlation as a causal relationship, when in principle statistical analysis cannot distinguish between cause and coincidence;

(3) On the other hand, econometrics’ putative identification of causal relationships is necessarily limited by the availability of statistical data and exposed to inaccuracies in the data that is available (i.e., econometrics analyzes a “statistical economy” whose correspondence to the actual, experienced environment in which lives are led and decisions are made is always uncertain and never complete); and, finally,

(4) As a forecasting technique, for which it has principally been developed, econometrics decisively depends upon projecting into the future correlations (mislabeled as behavioral relationships) based on past data (the “data base” problem, which arises in extreme form when current values of strategic variables, like interest rates or inflation rates, lie outside their previously observed ranges).

Econometrics, as noted, is a methodology oriented towards prediction: an econometric model stands or falls on its success in accurately predicting future values of economic statistics. The purpose of our critique, thus, was to assist in understanding how and why all econometric models had failed to predict the statistical data observed in 1974 (let alone the qualitative aspects of an inflationary, financial and political world crisis). We judge that the first prerequisite for understanding the Systems Dynamics Group’s economic model is to understand this: it is not a “prediction model”; rather it is an “explanation model.”

The purpose of a prediction model is to inform decisions made today with forecasts of the statistically observable future effects of an underlying, ongoing process of flux and change. The purpose of an explanation model is to provide understanding of that ongoing process itself. An explanation model, thus, can inform the formulation of policies which, in turn, will guide future specific decisions made in response to the inherently unpredictable, specific effects of ongoing processes of flux and change. Two conclusions result:

(1) The “fit” between an explanation model and any set of statistical data is neither a necessary nor a sufficient test of the adequacy of the model:

(a) Statistical fit is not a necessary test, because the available data themselves are not neutral and “objective”; they have been collected and categorized in line with pre-existing explanation models. Moreover, the data is always “contaminated” by the effects of events and processes (e.g., wars and droughts) which the model does not purport to explain; finally, data which corresponds to the categories of the explanation model may not, in fact, be available. (See below for further discussion of these points.)

(b) Statistical fit is not a sufficient test, because the correlation of observed data with the behavior of any model may in any case be coincidental, and to what extent it is or is not cannot in principle be determined.

(2) The Systems Dynamics Group’s work has been, in essence, to develop a relatively new technique (systems dynamics) which, in turn, embodies a radically new economic theory:

(a) The usefulness of the technique lies in its ability to accommodate and articulate the descriptive and contingent statements of the theory;

(b) The adequacy of the model depends upon the adequacy of the theory it embodies.

Once it is understood that the Systems Dynamics Group is engaged in the project of theoretically explaining real-world processes (rather than mechanically predicting future statistical manifestations of real-world processes), two separate issues can be distinguished. The first issue concerns the relationship between methodology and theory. The second issue concerns the relationship between theory and evidence.

The methodology of “systems dynamics” involves computer simulation. The usefulness of computer simulation for theoretical thinking arises from the following considerations. The world is very complicated: to explain seriously how any situation or process in the real world has come to be involves thinking through an extremely large number of potentially significant factors and relationships. In practical fact, it is very difficult to hold in one’s mind simultaneously all the factors and relationships that one “knows” to be in some sense important.

Translation of the theoretical enterprise from words into algebraic symbols is a temptation that may easily prove treacherous. Once a factor is translated into a symbol or a relationship is translated into an equation, two transformations take place. A description realistically enshrouded in the rich ambiguity of natural language is transformed into a “defined” term; a relationship whose complex and problematic nature is expressible through the grammar and syntax of prose is transformed into a bare algebraic equation. Especially in the relatively simplistic mathematics generally employed by social scientists, the manipulation of symbols and equations all too readily becomes an end in itself, a game unrelated to the greater project of comprehending those situations and processes to which the symbols and equations were originally intended to refer.

The computer would appear capable of modulating the theoretician’s risks in two respects. Even more than algebra it allows a very large number of factors and relationships to be held readily and, as it were, “consciously” available, so that the theoretician’s (not the computer’s) thinking is less likely to seize on too limited a selection of potentially relevant variables and run out their interactions to a necessarily inadequate conclusion. And more, through computer simulation, the interaction of that very large number of factors and relationships can be run out with facility. There is still, no doubt, a “translation” loss in going from the language of human discourse to the language of computer simulation. But, perhaps precisely because the result is a computer print-out, readily replaceable with a revised alternative, the temptation to read it as a primary record of lived experience, rather than as the programmed implications of a particular theoretical explanation of lived experience, may be materially reduced.

The computer, thus, is a tool for the theoretician. Like any tool, its capabilities condition both the purposes of its user and how the user thinks about his purposes. In this sense, it is no more “neutral” than a flint arrow to a cave man (without such a tool it is hard to imagine even approaching, let alone thriving on, carnivorous game). The conception of the project, however, the planning of the hunt, the stalking of the animal, the execution of what remains a technically demanding enterprise—all this is for the user of the tool. So we set about capturing, from a mass of undifferentiated data, significant meaning: stimulated perhaps by the technological opportunity to simulate reality.

But how to evaluate the significance of the meaning that is given to that mass of data? The conclusions set forth above, which contrast the systems dynamics project with econometric analysis, do not mean there is no relationship between this (or any other) explanation model and empirical evidence. Rather, they mean that the relationship is not a mechanical one in the sense that the embodied theory arises full-blown from analysis of supposedly objective “facts.” For, if it is indeed innovative, in order that it may be evaluated this (or any other) theory must itself generate its own “facts.”

That is precisely what occurred with respect to Keynes’ General Theory of the determination of aggregate employment and interest rates and of the role of money. Keynes defined and gave theoretical significance to a number of conceptual categories—in particular, aggregates of consumer income and expenditure and business investment—to which, quite literally, no empirical data then available corresponded. The collection of such data and its translation into usable form followed (though not exclusively) from the needs of Keynes’ Theory: the reversal of conventional empiricist theorizing about the relationship between theory and evidence.

Similarly, to evaluate the systems dynamics model in relation to empirical evidence will require the creation of relevant evidence—evidence defined by the model itself to be relevant—that does not now exist (e.g., evidence with respect to the factors that enter business decision-making and with respect to the relative development over historical time of the capital goods producing industries versus the consumer goods producing industries—see below). In fact, one may extend the argument a step further: one test for the “adequacy” of an innovative theory is precisely its fruitfulness in generating a need for new evidence. For this is a measure of the theory’s ability to ask new questions of reality, rather than merely to repeat the answers of received dogma. As we suggest in the conclusion of these notes, in this regard the Systems Dynamics Group’s theory appears usefully “adequate.”

The above discussion has approached certain core issues of the philosophy of science, particularly the relationship of empirical evidence and inductive logic to theoretical hypotheses (see, for example, T. Kuhn, The Structure of Scientific Revolutions and P. Feyerabend, Against Method). At this stage it is only important to re-iterate: The systems dynamics model does not compete with econometric models; it is fundamentally different in kind. Rather, the systems dynamics model does compete with the currently dominant economic theory, generally taught to college undergraduates as the “neo-classical synthesis.”

Theoretical Issues

Our interest in the work of the Systems Dynamics Group initially reflected dissatisfaction with econometric methodology both as a forecasting technique and, more basically, as a guide for understanding the functioning of an economic system. Our interest in that work was greatly heightened by recognition that, in fact, the systems dynamics model embodied an economic theory which, even at its present stage of development, differs from and is markedly superior to the dominant theory, the “neo-classical synthesis.” (It may be important to note here that the theoretical issues about to be discussed do not refer to the “Keynesian” or “Fiscalist” versus “Monetarist” controversy: Samuelson and other “Keynesians” and Friedman and other “Monetarists” adhere to the same basic principles of neo-classical economics; the content of the controversy between Fiscalists and Monetarists has more to do with political values than with theoretical insight and is all the more confused as each side attempts to “prove” its case by application of the irrelevant econometric methodology discussed above.)

The following discussion is necessarily limited to the theoretical issues raised by consideration of the systems dynamics model as it stands today and as we have been exposed to it. This means, specifically, that it is essentially limited to the model’s “production sector.” However, we judge the theoretical content of the model as it stands to be sufficiently rich and provocative as to justify critical analysis. The significant points at which we judge the systems dynamics model to differ from neo-classical theory are as follows:

1. The Theory of the Firm—The neo-classical theory of the firm is concerned with an abstract entity which, in the first instance, has no control over selling and buying prices and adjusts output in order to equate marginal revenue and costs; if it has power over prices (i.e., is a monopolist or monopsonist) it will set prices to obtain the same equalities at the margin; in any case, the neo-classical firm is a profit-maximizer operating with perfect information. By contrast, the systems dynamics model begins with a rich description of a “real-world” decision-making process in which information is incomplete but some ability exists to reduce risks; of particular importance, decisions about prices and output and employment levels are related to decisions about variables whose current values the firm itself knows and controls, especially levels of asset-holdings (inventories and capital goods); thus flows of money and goods are related to stocks of assets, which in turn realistically complicate the dynamic relationships between firms; each firm seeks to price outputs and inputs in the light of externally identifiable market conditions (as informed by order rates, inventory levels, etc.) and internal financial conditions.

Of even greater importance than the disparity in the initial descriptions is this: the neo-classical firm is described as it is because it is defined as a function of a general equilibrium model and its behavior is narrowly circumscribed by the conditions required for all the variables in the model to be in (or tending towards) equilibrium. In direct contrast, the systems dynamics model does not begin with a commitment to equilibrium and the behavior of the firms which constitute its production sector is not constrained by a priori conditions required for equilibrium to be realized. Thus, in turn, the description of the firm can be made as complex and realistic as direct observation of “real” firms suggests is appropriate (and as the technique of computer simulation allows). In particular, unlike the neo-classical firm operating in the stasis of a (mis-named) “competitive equilibrium,” the systems dynamics firm is allowed to compete and has the power to do so.

2. The Theory of Markets—As noted, neo-classical theory is a theory of general equilibrium, in which markets necessarily “clear” (all goods available for sale are sold and all effective demands are met) as prices move in response to available supply and effective demand. There is no similar requirement imposed on the markets in which the systems dynamics firms operate. In fact, on the contrary, at its core the systems dynamics theory is a theory of adaptation to perceived market disequilibrium. This represents a major theoretical distinction (perhaps the crucial distinction) between the two theories.

3. The Nature and Role of Capital Goods—The systems dynamics model distinguishes capital goods from consumer goods and, in turn, distinguishes the industries producing capital goods from those producing consumer goods. The reasons (well known to the “classical economists” of the nineteenth century from Ricardo through Marx) are: first, that capital goods are only used up (depreciated) slowly in the production process and, second, that the capital goods sector produces for its own consumption as well as for other (consumer good-producing) industries. Thus, as general economic expansion (contraction) takes place, the capital goods sector must expand (contract) disproportionately. This basic, structural characteristic of an economic system has, in effect, been “rediscovered” by the systems dynamics group (it has continued all along as an element of Marxian economic theory and as such is incorporated in the “new Cambridge” (England) school of economic theory). The mainstream “neo-classical” theory is virtually ignorant of it and of its consequences for the dynamic disequilibrium of economic systems.

4. Relationship between Micro-Theory and Macro-Theory—The neo-classical “synthesis” is, exactly, a synthesis of neo-classical micro-theory with “Keynesian” macro-theory. To generate a system of markets which as a whole would be in a general state of equilibrium, “Keynesian” macro-theory was grafted on to neo-classical theory in response to the massive departure from equilibrium of the economic system during the Great Depression. There is, thus, no integrated relationship between the neo-classical micro-theory of the firm and of markets and its macro-theory of aggregate income and expenditure (and of monetary stocks). The only relationship is that the equations expressing the aggregate relationships of the macro-theory can be appended to the system of equations expressing the market relationships of the micro-theory to produce a general system of equations which as a whole tends towards equilibrium. On the contrary, as we understand the systems dynamics model, the macro-behavior of the system—in all its modes of shifting disequilibrium—arises directly out of the behavior of its micro-elements and of the structural relationships between them.

The point of this discussion is not to characterize the systems dynamics theory, as it has been developed to date, either as “true” or as “truer” than neo-classical theory. Rather, we mean to indicate where we find the systems dynamics theory offers a more adequate explanation as to how an economic system functions than the neo-classical theory does.

As the theoretical implications of the complete national model are worked out, in fact, the systems dynamics theory may stand as a full conceptual alternative to the neo-classical synthesis. Thus, our interest in the theoretical status of the Systems Dynamics Group’s work is largely independent of the particular implications for present or future economic performance which we (or anyone else) draw out of it. (We also find interesting the correspondence between the systems dynamic theory and the economic theory developed by the “new Cambridge school”: one instance of this correspondence has been cited above; we understand that this is an example of “separate development” informed by a common concern for and knowledge of the history of economic thought.) Nonetheless, consideration of the theory’s implications is an appropriate aspect of evaluating its relationship to the “evidence.”