1Model Building in Economics is a nice and informative exercise in “small-m methodology.” Its intention is to provide a “Forest perspective” on model building, and not so much a discussion of individual “Trees.” If the Trees are meant to be individual models, such as game-theoretic models, DSGE models and simple econometrics models, the Forest and Trees metaphor is an apt description of the book, but if the Forest means methodology it is less so. Methodology is not presented from a Forest perspective but more as individual Trees: At first sight, there is not much cohesion between the many methodological observations one reads in this book. Only at second sight, at the very end of the book, an overarching theme that drives these many observations is revealed. In the concluding remarks of the Epilogue, Boland states explicitly (251):
Obviously by now you can see that I am primarily advocating that model builders should be more concerned with the realism of their assumptions, but, of course, some will say we should still allow for a sequential development of models – one that necessarily begins with highly unrealistic assumptions at its early stages. In other words, perhaps model building for model building’s sake should be allowed – but, I say, only if we do not just stop there and admire our elegant models. Specifically, my concern is that, before one starts making policy recommendations to governments, the realism of one’s model must be addressed and that widely used unrealistic assumptions must be rejected regardless of their mathematical elegance.
2Boland’s main concern is the “post-1980s view of models” that considers models as “instruments,” which should be considered as a (new) version of “instrumentalism.”
3Although I agree with Boland’s characterization of post-1980s methodology of model building as instrumentalism and in relation to this view seeing models as instruments, I do not share his criticism. This disagreement is based on our different views of four closely related facets of model building: the distinction between models and theories, the empirical assessment of models, the meaning of the term “unrealistic,” and the change of methodology since the 1980s.
4In the Prologue of his book, Boland describes what he considers as the main shift in model building in the 1980s. Until the 1980s, models were seen as particular representations of more general theories, that is, models were always models of a theory. Model building therefore consisted of three steps: (1) choose the basic behavioral theoretical assumptions that one is going to build a model of, (2) choose how each assumption is to be represented by a mathematical expression, and (3) choose how to test these assumptions empirically. In my view it is rather confusing that Boland labels this last step with the term “calibration,” a post-1980s term (introduced by Kydland and Prescott in 1982) which fits adequately the post-1980s view that models are better to be considered as instruments. Calibration refers to measuring instruments, and that is precisely the reason why Prescott and Kydland introduced this term to economics: their concept of models is that they have to be seen as measuring instruments.
5But more profoundly, this account of model building, which is according to Boland describing the actual pre-1980s practice, is a myth. It was preached as the official dogma, but never practiced. Even Jan Tinbergen who introduced the practice of empirical model building to economics did not build the first macroeconom(etr)ic models according to this stepwise schedule (see Boumans 2005).
6I agree with Boland that there is a difference between the pre- and post-1980 views, and that this difference is not about two different ideas of what constitutes a model in economics, but that there are two different views of what constitutes a theory (224). We however differ on the role of theories. According to Boland, since the 1980s “there is little if any distinction recognized between what is considered a theory and what is considered a model” (224). In my view, theories remained to be different from models, but they disappeared into the background. There is no difference between the functioning of models in economics before and after 1980, because before and after they function as “instruments of investigation” (Morgan and Morrison 1999). As a matter of fact, this denotation comes from Irving Fisher’s Mathematical Investigations in the Theory and Values of Prices (1892) to clarify how his hydraulic model should be employed.
7Because there is a difference between models and theories, their assessments will also have to be different. Theories as containers of general statements about the world have, indeed, to be tested whether they or their assumptions on which they are based are “true” or “false.” But instruments are built for a certain purpose, and therefore their “validation” is to assess how well they function to reach that purpose. That purpose can be anything: from better understanding a specific theory to measuring inflation rates in the Netherlands. This means that for each purpose the most appropriate kind of validation will be different.
8Boland traces the origins of instrumentalism in the early eighteenth century in Bishop Berkeley’s defence of Newton’s mechanics. The origins of models can be traced back to only the late nineteenth century in Heinrich Hertz’s (1899) discussion of the consequences of Maxwell’s new, that is, non-Newtonian, approach to physics. According to Hertz, a model is a representation (Bild), that should fulfil specific requirements. First and most fundamentally, the consequences of (that is, the inferences from) a representation of a mechanism must be the representation of the consequences of that mechanism. Hertz called this requirement “correctness.” Secondly, a representation should not contradict the rules of logic. Thirdly, the representation should be most “appropriate,” that is, it should contain as many as possible of the relevant essential relations and at the same time be as simple as possible. Appropriateness can only be tested in relation to the purpose of the representation, since one representation may be more suitable for one purpose and another for another. What the “essential relations” are is determined by the purpose of the model. It should be noted that none of these three requirements says that one should test these model relations directly. This is a relevant conclusion because a model is always a simplification of the system that is represented by it. Because it has to leave influences out, it will never be a true representation of that system. The classic example is the subway map of London: though it is a representation of the subway system, it is not a “true” representation: for example, the distances are “false.” But it is “appropriate” for the purpose it is designed: to find your way through London by using the subway. Theories should be tested on whether they are true or false, models should be tested on their correctness and appropriateness.
9Friedman did not use the terms appropriateness and correctness, but the term “realisticness” instead. Although he used this term sloppily, and hence causing a lot of confusion about what he actually meant to say, this term should not be confused with “truth,” as Samuelson (1963) did. Both Samuelson and Boland (244) are of course right to state that “You cannot work back from the observed truth of one’s deductions to the guaranteed truth status of the assumptions used to form those deductions.” But, this iron logic does not say that the empirical adequacy of the consequences does not inform us about the empirical adequacy of the assumptions from which the consequences were inferred. To the extent that we cannot test our assumptions in a ceteris paribus environment, model assumptions can only be tested with respect to their observable implications. Therefore models that are built for policy analysis can be rightfully tested for their predictive performance, because policy advice and predictions belong to the same kind of judgements: “what would happen if.” But testing predictive power is not the only way to validate the appropriateness of models used for the purpose of policy analysis. If a simulation shows the same dynamics as the real world, the model used in the simulation can be considered validated. The famous example is Irma and Frank Adelman’s (1959) simulation of a macro-econometric model of the USA.
10Although I am in fully accordance with David Hendry’s (1980) three golden methodological rules (and I guess Boland, too): “test, test and test,” these rules do not say what kind of tests should be employed. There is no general test that fits all models; the appropriate test is determined by the model’s purpose. Mathematical models built to gain an interpretation of a theory should be tested on their logical consistency, models for policy purposes should be tested on their predictive power, and models as measuring instruments should be calibrated.
11I am aware that the above discussion very much looks like a scholastic discussion among philosophers, and maybe it is. But if one puts this dispute aside, the book gives a very nice survey of the various practices of model building with the many small-m-methodological issues that go with them. For those who have attended one of the recent meetings of the History of Economics Society or of the International Network of Economic Method probably have seen Larry Boland taking pictures of the event, particular snapshots of conversations and presentations. This book is written with the same photographer’s attitude, it reads as a photo album: it surveys current issues and discussions vividly, extensively and quite exhaustively, almost without commentary. But as everyone knows, each photo is a comment.