Optimization as constraint: a comment on Mazzoleni and Nelson

“Couldn’t you just throw in a couple of Bellman equations?”

(an otherwise sympathetic Econometrica reviewer, rejecting a paper submitted by Winter, Dosi, and Kaniovski)

1. Introduction

Neoclassical economics continues to thrive, defying the predictions of a long line of diagnosticians who identified weaknesses that they thought to be fatal. When Richard Nelson and I joined that tradition in the 1970s and 1980s, we stood on the shoulders of many giants who had expressed similar skepticism about the neoclassical tradition. The significant feature of what we offered in 1982 was not an exceptionally novel critique, it was a relatively specific conception of a new path forward. We imagined that the promise of the new path might prevail where the old critique had not.

More recently, in the aftermath of the financial crisis and the Great Recession, there has been a new, and notably broad, wave of vigorous complaint against neoclassicism—though quite little from within the economics discipline itself. As in previous episodes, it is plain today that the convergence on the main elements of the critique is much stronger than any convergence on the nature of the path forward. While many “outside” critics can present a sound critique, they have had little to offer, and shown little prospect of effect, regarding the path forward for the discipline.

Mazzoleni and Nelson (2013) (hereafter MN) provide an informative and provocative account of the record of rebuffed challenges to the neoclassical paradigm. They parse the critiques among three broad headings: misrepresentation of the individual actors (flawed psychology), disregard of the role of social context (weak sociology), and explanatory priorities that neglect the phenomena of continuing economic change (indifference to the most consequential questions confronting the discipline). They document the case that these three sorts of objections have been made, and well made, in the past. They argue that the increasing mathematization of economic theorizing since World War II has amplified the negative consequences of the underlying flaws in the neoclassical account of economic life.

Their analysis is, in my view, basically correct (no surprise there). I do have to abstain from endorsing all of their specific points about the historical record, as several of these are actually quite new to me. Overall, their account offers a valuable, strong, and novel perspective on the specific obstacles that have long hindered the challenges to neoclassicism and impeded progress along alternative lines. A number of their themes certainly merit further discussion.

I will focus my comments here on the theoretical role of optimization, that is, on the neoclassical commitment to viewing individual economic actors as optimizers. That commitment lives in a rather subtle symbiosis with the second major commitment of neoclassicism, to equilibrium analysis.¹ First, “equilibrium” in the favored sense is conceptually intertwined with optimization in a way that is unfamiliar in the treatment of dynamical systems in most of the sciences—because a failure to optimize is itself regarded as a kind of disequilibrium. Thus, “equilibrium” connotes an element of optimization. But second, in a multi-actor system the optimization problem for an individual actor typically is not well-defined in the absence of some sort of equilibrium assumption about the system. Thus, optimization in a multi-actor context demands an equilibrium analysis. Due to these connections, efforts to separate the two ideas are generally found unappealing by neoclassical economists. Although my focus in this essay is on optimization, I consider it in the role it plays in neoclassical modeling, where it is typically linked to an equilibrium assumption. Other uses of optimization ideas and techniques—for example, in applied operations research—are not at issue in this essay.

It is clear that the challengers’ inability to overthrow the “optimization” or “rational choice” aspect of the neoclassical paradigm is a key rubric for an answer to the basic MN question about explanations for the failures of the successive challenges to neoclassicism. It is most obviously not only related to the “flawed psychology” theme that MN identify, but also intersects the other two.

I have been sporadically involved, over a long period, with the debate about whether actors optimize (see, e.g. Winter, 1964, 1975). I am thus well aware that many people consider the debate fruitless as well as tiresome. Still, a good deal of insight into the subject has gradually accumulated, at least some of which might be helpful in setting future research agendas. Further, it strikes me that some of the most important and most recent insights tend to get lost in the repetitive swirl of the old controversy about the status and adequacy of neoclassical economics. Also, it is striking how many different elements of the broader argument about neoclassicism appear specifically in the discussion of the merits of the commitment to optimizing actors. Is economics a science, or could it be? What is it really about? On what basis should theoretical premises be assessed? And so on. The subject is both complex and sprawling.

My principal thesis here is that the neoclassical commitment to optimizing actors has exacted a high opportunity cost because of its distorting effect on the discipline’s priorities. Some important questions cannot be addressed effectively with those tools, and even in the areas where the tools have yielded useful insight, they exact a further cost because of the barriers to complexity that they intrinsically involve. My view here is primarily forward-looking, that is, I am mainly concerned with whether continued reliance on optimization, in its currently dominant paradigmatic role, will be fruitful for the future—not on what it has yielded in the past.

To develop the case, I first seek to narrow and structure the question. Then I link the use of optimization assumptions to a broader methodological issue that extends well beyond economics, the practice of casting theoretical arguments in the form of abstract parables that have weak or indeterminate empirical reference. Section 4 highlights the particular role of the neoclassical enthusiasm for “modern” conceptions of rationality that are both strict and abstract, arguing that these conceptions are highly symbiotic with the “theory as parable” methodology. In the conclusion, I draw out some implications for the further development of the existing body of economic theory that shuns the neoclassical reliance on optimization.

2. Optimization: what is the question?

To narrow the issues, I begin with the key premises of my argument, recognizing that each such choice involves rejecting an alternative path—and on that path my subsequent points might be wrong or irrelevant. The most fundamental commitment concerns the nature of the economics discipline. I consider it to be concerned with pursuing an “intendedly scientific” understanding of the phenomena of economic reality.² I accept Marshall’s characterization of economics as “a study of mankind in the ordinary business of life.” (Marshall, 1920: 1). Alternative definitions often suggest that efficient resource allocation is the discipline’s defining concern, as in the Lionel Robbins statement quoted by Mazzoleni and Nelson (2013: 1415) that evokes “scarce means that have alternative uses.” Related definitions introduce a suggestion of optimization (“best use …”), or emphasize the analysis of “choice.”³ All of these alternatives are less expansive than Marshall’s definition, and each can be read as prejudging some important question about how the “ordinary business of life” is actually conducted.

In accepting Marshall’s definition, I do not need to argue that there is no field of study that corresponds roughly to the Robbins view, that is, is concerned primarily with resource allocation, or perhaps (nowadays) with exploring the logical implications of formal rationality. I only argue that an empirically grounded scientific understanding of the actual economy is a worthy objective, and that such a field of inquiry has a highly legitimate claim on the name “economics.” In that field, the relevance of optimization is not a given. The degree of emphasis it deserves might reasonably depend on whether real actors actually optimize (whatever that might mean), or if they do not, whether a paradigm based on optimization might nevertheless be broadly useful in understanding the real economy (for whatever reason). In contrast, if the discipline were actually about the normative analysis of allocative efficiency, it would be hard to deny optimization a central role.

Models featuring optimizing actors make a wide range of different contributions to understanding of the economy. The existence of this variety is one of the main sources of difficulty in the discussion about the merits of such models, because advocates of different assessments may have quite different impressions of what constitutes a contribution—or what would constitute a contribution in the future. The classic and still central example of such difference is the divide on the question of whether “optimization” is a process that ought to be observable in principle at the actor level, and if so what it looks like, versus the Friedman-derived “as if” interpretation (Friedman, 1953), which scorns the suggestion that indications of optimizing behavior should, if it really exists, be visible in the decision process itself.

The effect of Friedman’s interpretation is to shield what he called the “maximization of returns” hypothesis from evidence on decision processes. Of course, that evidence supports great skepticism about the empirical realism of optimization—and much more so today than in Friedman’s time. The legitimacy of this shielding move is an interesting methodological question, a question that is reasonably well defined and can be construed as being of a familiar type in the context of broader discussions in the philosophy of science. Much has been written about the question, and at least a portion of the commentary is informed by an awareness of the philosophical context.⁴ Especially in the early years, much of the commentary was critical of Friedman’s stance—but, as MN observe, Friedman’s view largely rules in the mainstream discipline today.⁵

Friedman, however, quite clearly understood economics as an empirical scientific discipline, guided by theoretical conceptions of the “hypotheses” that were worthy of test. His views on process evidence constrained the empirical relevance of received theory, but he certainly did not reject, in general, the idea that such relevance should be consequential for the acceptance of the theory. Whether that view is generally held among contemporary theorists seems doubtful, as the following section argues.

3. The epistemological status of parables

As MN explain, methodological discussion in economics has often featured comparisons of economics to other, more visibly successful disciplines, such as physics. Incidental to such comparisons, or independently, there have been attempts to place the economics case in the context of broader discussions in the philosophy of science. Frequently, these comparisons and discussions are conducted at a considerable distance from the details of the subject matter, that is, the actual pattern of scholarly practice, especially on the economics side. This is particularly true when the word “prediction” figures in the discussion, because prediction occupies a more central place in the larger methodological discussion than it does in the practice of most economists. It is true, of course, that prediction is focal some of the time for some economists: economists do a lot of different things. There is more concern with prediction in macroeconomics than in microeconomics, more in experimental and behavioral economics than in mainstream, neoclassical modeling. There remains, however, a large field of theoretical activity that does not aim at prediction.⁶ The usual accounts of scientific method in the philosophy of science do not do much to illuminate the significance of this activity. Although part of the answer is that scientific theories seek to explain as well as to predict, this only goes part way toward explaining the distinctive features of economic practice.

Some scholars have taken up this challenge.⁷ Although the terminology and the specifics differ across the writings of these scholars, there seems to be general convergence on at least two points. First, models are used as tools of inquiry in economics in a way that does indeed present some distinctive methodological issues. Second, an important part of the scientific value that economists derive from models comes from the multiple narratives that a good model can support, the answers to the question “what would happen if … ?” for a variety of cases. Thus, in an abstract sense, a model can be seen as akin to an experimental apparatus—although results are reached deductively and directly inform us only about the model itself, and not about the real world (Morgan, 2012: 242–246). The nature of the relationship (if any) to the real world is the nub of the remaining problem; it is quite a subtle one and seemingly not much clarified thus far.

Some years ago, David Teece and I took a stance on economic modeling that fits comfortably in the frame provided by more recent discussion:

“The dominant (neoclassical) mode combines unquestioning faith in the rational behavior paradigm as a framework, relative indifference to the delineation of the empirical phenomena that are thought to require theoretical explanation, and a delight in the construction of ‘parables of mechanism.’ Such parables provide a sharply defined view of an imaginary world in which the logic of a particular economic mechanism stands out with particular clarity. The insights generated by this method often seem valuable and compelling, but unfortunately there is often no attempt to bridge the vast gulf that separates the simple imaginary world with its isolated mechanism from the complex real world in which some analogous mechanism may, perhaps, operate.” Teece and Winter (1984: 118)

There are, however, some strong indicators for the degree to which a particular model is or is not parabolic.⁹

First, a parabolic model is offered without having any specific empirical reference in view, although it might be inspired by a familiar type of choice situation (as in the theory textbooks) or an anecdote about an alleged phenomenon. Second, the model involves simplifying assumptions that severely limit its potential for empirical reference. There are whole categories of assumptions that appear repeatedly in this role, including (i) restrictions to static or two-period analysis of what is clearly a fully intertemporal problem, (ii) assumptions of certainty, or of very simple structures for whatever elements of risk or uncertainty are admitted, and (iii) assumptions limiting the dimensionality of the problem—reflected particularly in the use of “representative agent” assumptions, and more generally in the fact that there are so often exactly two of whatever theoretical entities that there is not one of (consider the prominence of duopoly models in oligopoly theory. Of course, duopolies and near-duopolies do happen, though not often).

Needless to say, theorists do not adopt the extreme assumptions of parabolic models out of some impish impulse to stick a thumb in the eye of reality. They are introduced for the sake of getting on with the theoretical business. What is that business? It is to offer coherent ways of thinking and related insights about economic situations. That was always the business of theoretical economics, but as mathematical sophistication has risen, the business is increasingly focused on ways of thinking that meet demanding standards of logical coherence, expressed in logico-deductive modes of argument that claim to arrive at interesting destination B from origin A, with only mathematical manipulation and tight logic in between.

Even as intellectual styles have changed, theoretical arguments have influenced the thinking of the audience in much the same way. The theorist invites, “Think about it this way,” and then explains what “this way” is. In some cases, the theorist’s invitation is widely accepted. Readers enjoy a sense of intellectual satisfaction as the insight sinks in and as it becomes, in fact, a part of the reader’s own way of thinking about things. The resulting social impact can be remarkable, as Keynes testified in his famous statement about “madmen in authority.”

That such dramatic impacts could occur should not be mysterious, on two grounds. First and most important, human nature abhors an interpretive vacuum. We humans naturally thirst for understanding, and when that is hard to come by, we tend to grasp at straws—and particularly at straws that conform to our existing prejudices. Second, when theoretical models are found compelling, they are compelling as narratives: they appeal to the human propensity to appreciate a good story well told. Arguably, this aspect is at the heart of the persuasive power of even the most abstract and technical parables, though the persuasive effect may be limited to the small community of those who are sufficiently skilled in the art to appreciate the story.¹⁰

In my view, the great bulk of the theorizing that has advanced our microeconomic understanding over the years has been of this parabolic character, and it is a valuable intellectual heritage. To deny its value would be to deny my own subjective appreciation for much of that theoretical literature, not to speak of my own testimony in the form of selections for theory reading lists, or of my own hopes for the reactions to some of my own modest contributions that were offered in the spirit of mainstream parabolic theorizing.

But is it science? That is the question. It does not look much like (empirical) science, because the attention to empirical grounding is so casual, the treatment so impressionistic and so oriented to “let me tell you the moral of this story”—and suggestions for future empirical tests are so sparse. Typically, parables are offered as logically true stories about imaginary worlds, and that is the end of it.

Parables do, however, serve the cause of empirical science in various indirect ways. First, they are extremely important in pedagogy, and thus in transferring theoretical insight to new generations of researchers who might put it to empirical use. They can be valuable also to the community of experienced theorists, because a new parable can reveal unappreciated potentials, unrecognized difficulties or novel explanatory challenges lurking within the scope of an established paradigm. Finally, even if the original expression of some insight is isolated from the real world by bars of over-simplification, subsequent work may spring the doors of that cage—taking on, for example, the full intertemporal case, or the N country case—and thus facilitate more direct comparison with reality. Efforts of that generalizing kind obviously have had a large place on the agendas of theoretical research.

There are, however, some drawbacks of the parabolic style that derive specifically from the neoclassical commitment to optimization assumptions. On the one hand, optimization assumptions are often justified with the usual references to the usefulness of instrumental simplification for getting on with the explanatory work. Objections to the realism of such assumptions are waved off with the same dismissive gestures that would be directed to the objection that N = 2 (of whatever) is not a realistic case. Regarding optimization, however, it is clear that the contemporary modeling esthetic imposes high standards for the perfection of the optimization built into the model, and that this specific impulse is frequently at odds with the general drive toward simplification. Indeed, as the conceptual understanding of rationality has been elaborated and formalized since World War II, the level of complexity demanded in its representation has exploded (Dosi, 2013: xxiii).

Rational expectations theory, as put forward by Muth (1961) and vigorously promoted by Lucas (e.g. Lucas, 1972), is both a good representative of this complicating tendency and an important causal factor in the overall development. Muth proposed “… that expectations, since they are informed predictions of future events, are essentially the same as the relevant economic theory” (1961: 316). This imposed a challenging new discipline on theorists who wished to follow his lead, as many did. In principle, it required that expectations formation be confronted simultaneously with the specification of the rest of the model: when you specify your theory, you should also affirm that the model agents believe it to be the “relevant theory,” and take the consequences of that in the specification. For theorists who accept that principle, considerations of tractability generally require, in practice that the substantive core of the model, excluding the expectations part, must be even simpler than it would otherwise be. Given the “budget constraint” imposed by the bounded rationality of theorists, there is a substitution effect: higher standards in one respect imply lower standards in others, and the latter have given rise to narrow restrictions on statistical specifications as well as highly stylized treatment of the economic content of the models. Sometimes it even seems that sophisticated treatment of optimization is believed to have a talismanic effect: it is a charm that magically confers immunity against the consequences of absurd over-simplification in other parts of the model specification.¹¹

The neoclassical concern with formal rationality has had adverse consequences for economic science that go well beyond the significant effects just noted. In considering the above quotation from Muth, it is important not to overlook the words “the relevant economic theory” (emphasis supplied). With those words, and with the rest of his argument, Muth affirmed the primacy of an economic perspective in the explanatory domain of expectations formation. That was not, of course, a “hard sell” among his economist audience.

Consider the implications, however. One might think that understanding of expectation formation could be informed by reference to cognitive psychology, but in the neoclassical paradigm such reference is forbidden by an implicit appeal to disciplinary boundary conventions (“not economics”). Similarly, the institutional settings within which model actors are located have to be understandable in strictly economic terms, though realistic economic expectations in this domain might well be intertwined with political expectations. Similarly again, expectations regarding technology would likely be framed in the very narrow terms in which neoclassical economists generally see technology. The worthy ambition to endogenize expectations thus exerts subtle pressure toward describing the world as one in which only economic causation and random factors are at work—and this is “economic causation” as understood in neoclassical terms. Also, it is causation postulated in terms sufficiently simple to enable deductive analysis.¹²

Nelson and I complained, years ago, of the “autarky of economics” (Nelson and Winter, 1982: 405–407). The situation has only gotten worse in the mainstream core of neoclassicism, though on the periphery promising sectors of heterodox inquiry have become quasi-legitimate. The theoretical concern with formal rationality is a major contributor to the continuing autarky policy. As I argue below, the stance is intrinsically uncompromising and implies that neoclassicism cannot advance by absorbing alternative perspectives on the problems it claims to address.

4. A fork in the road

The account offered by MN takes note of a crucial fork in the historical road in these words:

“While the classical economists’ theoretical formulations were sufficiently broad and general to permit the assumption that economic behavior is rational to be interpreted simply as goal oriented and reasonably well informed and intelligent, the marginal utility theory (at least as formulated by Jevons) clearly associated rational with ‘maximizing’.” (Mazzoleni and Nelson, 2013: 1414)

Though the MN statement emphasizes the contributions of Jevons as marking the fork in the road of the rationality concept, it is clear that there was a protracted historical process that began earlier and extended much later. The “fork” actually consisted of a large number of specific turns that were taken as the new direction was set. Among the many scholars who placed the more recent markers on the optimization path I would nominate particularly Samuelson, Hicks, von Neumann and Morgenstern, Savage, and Arrow and Debreu. There are, of course, many others.

As MN also observe, it is not the case that the older and more flexible understanding of rationality has disappeared. Many economists today continue to employ that earlier understanding of rationality, a propensity that correlates strongly with primary reliance on verbal rather than mathematical argument. MN specifically mention Schelling, Olson and Coase in that connection. These are all credible nominations, but there are many others. A particularly important and distinctive case is Herbert Simon, who emphasized the importance of rationality while being an outspoken critic of the optimization version. My own list would also include Hirschman, Williamson, Ostrom and Sen, and also Akerlof, at least much of the time. One could also add a number of legal scholars who do a lot of economics, but little or no formal modeling, in their “law and economics” mix.

Another very instructive example is Edith Penrose. I cherish this one not only because of her many important insights into economic reality, but also because of one short statement that serves well to frame the distinction between the two kinds of rationality:

“There can be little doubt that the more complex an organization becomes, the more necessary it is to establish areas of quasi-automatic operation. The importance of routine as a means of taking care of some aspects of life in order that others may be given more attention has frequently been stressed.* The fact that many business decisions are not ‘genuine decisions,’ but are quasi-automatic and made routinely in response to accepted signals without a consideration of alternative choices has misled many into attacking the assumption that firms try to make as much money as they can—particularly where it can be shown that the rules governing the routine actions are not fully consistent with profit maximization.” (Penrose, 1952: 817)¹⁴

Particularly interesting to examine here is the relationship of the “fact” to what the “many” are “misled” about, and then the amplification provided by the final phrase, with its explicit reference to “profit maximization.” Although the statement might be puzzling on first encounter, it is fairly clear what is going on here. Penrose is standing up for the theoretical commitments to viewing business behavior as “goal oriented and reasonably well informed and intelligent” and to the centrality of “making money” among business goals. There are concessions to the reality of bounded rationality in the first sentence, and in the statement “without a consideration of alternative choices.” These concessions, however, do not significantly diminish the force of the claim that firms try to make as much money as they can. The emphasis is on the highly motivated, goal-directed character of the behavior—as in the earlier meaning of economic rationality. It is important to notice that the substance of that position actually differs only marginally from the position in Carnegie School behavioralism (Cyert and March, 1963), or from typical views among evolutionary economists today. Admittedly, there are some differences regarding the detailed account of the economic choices of both individuals and firms—for example, how closely do consumers calculate; how dominant is “making money” as a business goal for the large corporation? But these are relatively minor differences and the legitimacy of such issues is acknowledged in some branches of the mainstream.

Given a modern interpretation of rationality, however, we critics certainly are not misled in thinking that the behavioral facts affirmed by Penrose are at odds with profit maximization in a formal sense. There is a limited acknowledgment of this in the above quote (“… not fully consistent with profit maximization”), but the intellectual commitment Penrose strongly affirmed was to the motivational reality of the quest for profit, not to the theoretical rendering of the details. There is, however, no reason to think that quasi-automatic behavior and failure to consider alternatives produce only minor deviations from some hypothetical baseline of “true optimization.” It is quite plain that the consequences can be, in general, indeterminately large. Can such potential discrepancies be ignored, on the mere ground of the kinship at the motivational level between old-style rationality and modern optimization? If they can, what scientific value does the modern view add to the earlier one? What is gained by lavishing attention on fine points of optimization, if the reality judgments underpinning the modeling effort actually envisage no reliable distinction between optimization and habitual behavior? Consider the choice of a personal saving rate, for example, is there really no distinction between a forward-looking optimization of lifetime utility and adherence to an established consumption habit? If there is a difference, which idea offers the better first-approximation guide to typical behavior?

It is hard to imagine that there could be a viable understanding of the economics discipline that does not acknowledge the importance of economic motivation and does not admit that much economic behavior is “goal oriented and reasonably well informed and intelligent”—though on the latter point there is room for argument about the roles of deliberation and quasi-automatic habit (Winter, 2013). It is quite possible, however, to imagine an economics that does not rely significantly on optimization as a central behavioral assumption, though it might appear in other roles. Indeed, very little imagination is required to envisage this possibility, because the literature of evolutionary economics offers abundant illustration of just this approach.

The transubstantiation of rationality into optimizing seems, in retrospect, like an almost inevitable corollary of the broader move to formalize many of the arguments behind the received truths of economic theory, as they existed in the first part of the twentieth century. Indeed, as MN illuminate very well, the change in the understanding of rationality was entangled not only with mathematization and the quest for logical rigor, but also with even broader concerns. These included whether economics should be considered a science and if so whether it was akin to physics or something quite different. There were related concerns about the role of quantification and the interpretation of economic “laws,” if such exist.

To the rich discussion that MN provide, I would add only an observation about the chronology of change in the understanding of rationality. At the time when a number of forces were pushing toward a formal theory of goal-directed behavior, there were few, if any, alternatives to the idea of accomplishing that formalization by expressing economic motivation in the language of optimization. In the decades since theoretical economics took that optimization turn, the tool kit of models that are both formalizable and quasi-behavioral has expanded enormously. There are whole bodies of work associated, for example, with Simon’s “satisficing,” prospect theory, evolutionary games and the “rugged landscape” models in organization theory. A very wide variety of more specific models of learning and decision has been put forward. Many of these ideas are built into dynamic models, and sophistication regarding such models, whether analytical or computational also developed quite late. The lack of such sophistication further constrained the exploration of alternatives to (static) optimization. Historically, therefore, optimization was an early and unchallenged entrant in the developing sub-sector of the discipline where formalization of theory had become a concern. Though it had to displace established rivals in the broader sector of economics, it faced no competition that addressed concerns and boasted advantages analogous to its own. Thus, in the sub-sector, it quickly acquired the typical advantages of incumbency, as the MN analysis documents very well.

Although the quest for greater rigor was an important and explicit motive for the formalization move, esthetic considerations gradually became increasingly important in the judging of formal models. It is not surprising that rigor and esthetic appeal often advance hand in hand, as cleaning up an argument generally serves both purposes. There are innumerable illustrations of this pattern in the history of the discipline, especially since World War II. Establishing a good proof of a significant mathematical proposition is much like establishing efficient performance of a complex production task: not only is it rare for anybody to get it exactly right on the first try, but processes of incremental improvement can extend for decades. Such progress establishes an easier path for later generations of students to follow to achieve command of established knowledge—and on that ground at least, it is generally to be welcomed.

There is a downside, however. As MN observe, the neoclassical emphasis on formal modeling and maximizing choice is highly antagonistic to the acknowledgment of complexity, and has encouraged the suppression of complexity under each of the three headings featured in the long-term critiques. The more the assessment of models depends on esthetic considerations, the stronger this tendency becomes. In his analysis of how economists failed to predict the financial crisis, Paul Krugman proposed that they “mistook beauty, clad in impressive-looking mathematics, for truth” (Krugman, 2009b). He further commented that “… the temptation is always to keep on applying these extreme (neoclassical) simplifications, even where the evidence clearly shows that they are wrong. What economists have to do is learn to resist that temptation. But doing so will, inevitably, lead to a much messier, less pretty view” (Krugman, 2009a).

Krugman’s evocation of “a much messier, less pretty view” raises an important issue about the meaning of “complexity.” There is a body of theoretical technique that has developed under the rubric of “complexity theory,” which largely pursues the traditional scientific aspiration of explaining a lot by assuming a little. The assumptions may be unconventional, and unappealing to some schools of modeling, but they are not particularly “messy” and sometimes the results are even “pretty”—at least to the practiced eye. Although this sort of work is both appealing and valuable, it only reaches the threshold of the complexity problem. “Complexity” in the sense I mean here is better illustrated in areas that are traditionally of interest to economists like taxation, regulation, and science policy. It is dramatically illustrated in large-scale systems engineering, whether pursued for fundamental research (Mars exploration, the quest for the Higgs particle) or for more near-term objectives (the management of the artificial satellites supporting the GPS, weather forecasting, and other objectives). In these areas, simple parables may help sell copies of “airport books,” and they have a role in advocacy (propaganda), but they contribute very little toward getting the work done. Effectiveness generally depends on devising organizational arrangements that can tap the power of the division of labor among highly skilled specialists. This complex world could offer diverse and interesting opportunities to economists who are prepared to take it seriously, but such a prospect is not consistent with the culture of the discipline as it now exists. In particular, it is not consistent with the ideas about theory that dominate in that culture.

As noted previously, the problem with the neoclassical simplifications is not just the obvious one that attends the helpful ruling-out of messy complications, but that they erect a barrier against letting the complications back in. In his defense of instrumental simplification, Friedman offered the example of neglecting air resistance when predicting the behavior of a falling body (Friedman, 1953: 16–19). That this approximation can be adequate in some cases is clear. It is quite a different thing to argue that the theorists of the behavior of falling bodies should be forever barred from trying to incorporate the effects of the air resistance—on the ground that the theory without the air resistance is so much cleaner and esthetically appealing. That hypothetical injunction provides a close parallel to the effect of the “optimization constraint” in neoclassical economics today, where the return of complication is barred by the logical requirements of the framework of formal rationality. The role of the air resistance is played by the reality of computation costs and the neoclassical requirement that all of the choice alternatives must (in some sense) be visible to the actor for an optimizing choice to be possible.

These requirements do not similarly afflict the old-style understanding of rationality that MN describe. If that old understanding defined the ruling paradigm today, Simon’s concept of satisficing behavior (Simon, 1955) would be understood as just another way of thinking about rational behavior. As it is, satisficing is typically considered to be a heresy against the neoclassical orthodoxy, and the editors and reviewers of the discipline’s most prestigious journals certainly receive it as such.¹⁵ Another important heresy frequently practiced by builders of evolutionary models is treating goal-directed choice in intertemporal settings as driven by feedback rather than by foresight. Their idea of an instrumental simplification is to leave out the Bellman equations, at least in choice situations involving the indefinite future in a deeply uncertain world, where the behavioral implausibility of the optimization analysis is extreme (see epigraph). This again would not be heresy against old-style rationality; I wonder whether Edith Penrose ever heard of a Bellman equation.

5. Putting optimization in its place

The effect of the neoclassical commitment to optimization is to tilt research priorities in the direction of promoting the creation of parabolic models over empirical engagement with the complex reality and valuing beauty over truth. These costs should weigh in the overall assessment of neoclassicism, even if we were to waive the more familiar objections to optimization, for example, the objections featuring the discord with process evidence. (Of course, those objections would have been decisive long ago if the discipline had been seriously committed to putting its paradigmatic assumptions at risk, rather than focusing on verification opportunities.)

Meanwhile, progress has continued along the three paths that MN identify with the long-standing objections to neoclassicism. With some occasional exceptions, researchers in these areas have embraced some version of old-style rationality, but not the formalism of optimization favored in neoclassicism. The direct effects of that on the research progress are doubtless favorable, but the resulting ascribed status of the researchers as heretics certainly is not. It implies less promising career opportunities and diminished resources, which inevitably leads to some reduction of accomplishment. This effect is all the more consequential because serious engagement with complexity tends to be resource-intensive—as contemplation of large-scale science and technology projects clearly illustrates.

In the economics that might-have-been, the current neoclassicism would be occupying much less space under the big tent supported by two high-level commitments. The first is the indispensable commitment to the idea of economics as an intendedly scientific approach to understanding the economy, a pursuit that depends for its success on a vigorous, diverse, empiricism that is both guided and informed by theory. The second is the commitment to the rationality, old style, of economic actors. I do not say that it is impossible to do economics, as defined in the first commitment, without making the second commitment. I do say that very little of what has actually been in dispute—in particular, what has been in dispute between neoclassicism and the challengers—actually involves questions about the general plausibility of old-style rationality. It mainly involves the distinction between that and (notionally) true optimization.

In building the economics that might-be-yet, there are two kinds of relatively neglected projects that deserve attention by those who generally align themselves with the critics of neoclassicism. The first relates to defining the appropriate place of optimization in a reformed paradigm. For example, many evolutionary models posit situations in which two or three discrete alternatives confront an actor: a firm may choose at a particular time between sticking with the technique it has, going with a new technique it has invented, or attempting to copy the technique of a rival. The model stipulates simple ways of calculating the apparent merits of these alternatives, though not the long-run merits (which would require those Bellman equations and a model of the long-run environment to justify them). To assume that a firm makes the choice according to those apparent merits is to assume a kind of optimization. Should we disavow even that limited kind of optimization in our model building? If not, where do we draw the line? A second set of examples relates to “enacted rationality.” Some firms clearly make some choices, some of the time, by using actual, explicit, observable optimization calculations. In such cases, objections based on a clear clash with process evidence do not apply in the usual way; how should they be handled? Finally, and perhaps most important, there is the long-standing use of optimization models as a normative reference point for assessing real behaviors in real institutional contexts. That use could continue even if the postulated models of the real behavior did not involve reference to optimization at all. Should we abandon that normative use nevertheless, or treat it as a valuable part of the discipline’s heritage?

The second project involves improving the tool kit of simple behavioral models. An important strength of the optimization-and-equilibrium paradigm is that it offers a very flexible tool kit for making at least an initial approach to a wide range of different questions. True, the approach might well be misguided because of the limitations of the tools, but at least it is guided, and at least there is a rough consensus among neoclassicists about how it should be guided. In the heretical camps, in contrast, a rich assortment of possible approaches has accumulated over the years, but they are disorganized and unstructured. Modeling choices—when models are involved—tend to be ad hoc and only weakly justified in relation to nearby alternatives. Partly for this reason, the use of models is itself more limited than it might usefully be. Bringing some order into this picture would be an important step forward. My own view is that a fuller development of the satisficing approach could play the central organizing role, but I affirm the general importance of the project much more strongly than that particular suggestion.

In this comment, I have taken as a premise the conception of economics as an aspiring empirical science, and argued that its future success requires more serious engagement with complexity. The neoclassical commitment to optimization frequently stands in the way of such engagement, because the contemporary esthetics of theorizing push theorists away from both reality and complexity while promoting the construction of parabolic models. I note that the general position I have taken here actually corresponds rather closely to the general stance of the “institutionalists” in American economics of the interwar period—a point to which I was alerted by a belated reading of the stimulating analysis offered by Morgan and Rutherford (1998).

Reflecting on this insight, I recall that in my younger days I had a few senior colleagues of the institutionalist persuasion whom I faulted for resisting the idea that a little calculus might be helpful in expressing economic theory. I would still fault them for that, but this personal recollection underscores for me the complex historical relationships that MN have pointed to, between the mathematization wave and shifting views on the nature and direction of the discipline. In the end, I think those institutionalists were, largely and collectively, in the right—not about the value of calculus or mathematics in general, or the value of modeling, but about the nature of economics and the requirements for its progress as a scientific discipline.

Acknowledgements

Research support from the Mack Institute for Innovation Management is gratefully acknowledged. I am indebted to Giovanni Dosi and Richard Nelson for comments on an earlier draft. Special thanks go to Mary Morgan for her detailed feedback on the article. I hope that my revision succeeds in avoiding at least the worst of the pitfalls that were pointed out to me, but in any case I take full responsibility for the shortcomings of the final version.

References