The Two Blades of Occam’s Razor in Economics: Logical and Heuristic

Download full paper

Abstract

This paper is part of the general debate on the need to rethink economics as a human discipline using a heuristic to describe its object, on the need to explicitly reject the positivistic approach in neoclassical economics, and on the urgency to adopt a different methodology, grounded on a realistic set of initial assumptions able to cope with the complexity of the decision making process. The aim of this paper is to show the use of Occam’s razor in the economic debate around realism in economic modelling. Occam’s razor can be intended as a principle of logic that emphasizes simplicity, or as a heuristic tool that emphasizes parsimony. Neoclassical economists such as Samuelson have explicitly used Occam’s razor as a logical principle to highlight the strict logic and simplicity of neoclassical economics; neoclassical models, however, are based on unrealistic assumptions. Some approaches of heterodox economics, au contraire, have used Occam’s razor as a powerful heuristic tool to emphasize parsimony, building up models grounded on realistic initial assumptions and able to embed complexity into the general explanation of economic behavior, like in Simon’s bounded rationality (1955; 1957) and Hayek’s notion of human rationality (1948; 1952; 1974) and cultural and social evolution (1967; 1978).

Posted for comments on 6 Aug 2018, 9:25 am.

Comments (2)

  • Péter Cserne, Northumbria University, UK says:

    The express ambition of this thought-provoking paper is to flesh out a distinction between two senses of “Occam’s razor” in order to argue for more “realism” in economics.

    Arguments about the perceived excessive abstraction and lack of realism has a long history in economic theory, going back at least to the German Methodenstreit some 150 years ago. It has been a staple of methodological debates in economics ever since, with Friedman’s 1953 paper, the growth of experimental approaches and the behavioural turn in economics standing out as more recent milestones. (Notice how the milestone metaphor is suggestive of progress.) A related but less elaborate theme of the paper concerns the specificity of “human” or “social” science amongst the sciences. This theme has also been on the methodological agenda of economic theory since at least J.S. Mill’s engagement with Comte’s positivism.

    Against this background, Becchio’s attempt to exploit the analytical potential of “Occam’s razor” to reframe the discussion is both promising and risky. If one seeks to contribute to progress in these debates, reframing them in unexpected terms may be illuminating and rewarding. It would indeed be an important theoretical feat to demonstrate that “Occam’s razor” has the desired methodological implications for economics, justifying a move towards (whatever is understood as) “realism” and “parsimony”. The risk is, of course, that the attempt may turn out to be yet another highly abstract critical argument for “realism” in economics which rephrases old arguments in new terms but does not move the substantive discussion forward. As far as the paper is serving to economists some old (methodological) wine in new (Occamian) bottles, it depends on both the wine and the bottles how the experiment turns out.

    Let’s start with the “bottles”. The paper’s conceptual frame and terminological choices are surprising, especially to those who recall Albert Hirschman’s paper, entitled “Against Parsimony: Three Easy Ways of Complicating Some Categories of Economic Discourse.” i Defending heterodox economics as parsimonious, as Becchio does, suggests an innovative approach or at least a different rhetoric that may be just as persuasive in suggesting alternatives to neoclassical models in economics as the arch-heterodox Hirschman’s arguments for realism at the expense of parsimony.

    The distinguishing of simplicity and parsimony as a rhetorical framing seems to be inspired by a late paper by Herbert Simon (2001), an economics Nobel laureate and by general consensus a pioneer of what is now loosely called behavioural economics. Simon’s paper uses the dichotomy in a wide-ranging and highly formal discussion of information theory, and illustrates the more abstract points by a retrospective overview of some substantive arguments in his earlier work on empirical decision theory. For Simon, the distinction of simplicity and parsimony refers to how through the identification and representation of redundancies, science seeks to identify patterns in a set of data and then represent them in the most parsimonious way. In particular, while the title of Simon’s paper suggests an opposition of the two concepts, he in fact defines parsimony as a specific case of simplicity, itself being the reciprocal of the complexity of a data set.ii

    Becchio has, understandably, little to say on such matters of information theory; her focus is on substantive and methodological issues in economics. For her, the dichotomy of simplicity and parsimony serves as a suggestive rhetorical framing for the idea that seems to be driving the paper: we should be concerned at least as much with one aim of economic theorising: “realism” (one “blade” of Occam’s razor) than with the other: “simplicity” (the other “blade” of Occam’s razor). We have three concepts in two juxtapositions then. How the two criteria or aims of parsimony and realism relate to each other, is however not easy to grasp.

    In fact, to further complicate matters, the author suggests mapping the distinction of simplicity versus parsimony onto not one but two other dichotomies: epistemic (tool) versus methodological (principle)iii; and (pure) logical (tool) versus (a) heuristic.iv

    The second opposition between epistemic and methodological interpretation of Occam’s razor, as used in current general philosophy of science literature, suggests that simplicity and parsimony are concerned with two different problems: what is it there to be known and how we should pursue scientific inquiry. Yet this dichotomy is quickly abandoned. Becchio ultimately insists on the third opposition, between logic and heuristic. As I discuss below, the meaning of this opposition is somewhat opaque but it ultimately serves to illuminate the author’s primary concern: a certain “trade-off between simplicity and realism” which is specific to “the human sciences”.

    I will briefly return to this question of “bottles”, i.e. the conceptual and terminological choices of the paper at the end of this comment. Let’s now turn to the “wine”, i.e. the substantive points in the paper concerning the simplicity – realism trade-off and the argument for realism in economics.

    To start with, the reader is referred to some of the earlier explications of the idea of simplicity in the history of economics. The debate surrounding Milton Friedman’s 1953 paper on “The methodology of positive economics” is crucial in this respect. Yet, unfortunately, we are left without the author’s definite take on Friedman’s methodology. While she summarises some of the subsequent discussion (of special interest here is Ernst Nagel’s critique of Friedman’s methodology, based on the claim that Friedman failed to distinguish among three kinds of ‘unrealism’), the reader is left perplexed as to what the lesson of the debate is. It is also perplexing what is implied when the author puts Friedman and Samuelson in the same category of “positivists”, using the term with a clear disapproval. Could this mean that she considers the entire debate around Friedman’s methodological stance irrelevant or misguided? If so, why bother discussing with it? If valid and relevant points have been raised in the debate, what are these?

    As for simplicity, Becchio is right in referring to the difficulties in defining, let alone measuring it (p. 4 n 5). In a later turn of the argument, she even denies that under a reasonable definition of simplicity, neoclassical models are simple: their “simplicity becomes apparent and is misleading” (Becchio p. 8). Clearly, there is a case to argue that one kind of simplicity may lead to the lack of it in another sense, for instance when accounting for certain empirical phenomena requires ad hoc assumptions and cumbersome adjustments to an initially simple model. In fact, Becchio distinguishes four versions of simplicity. Even so, more conceptual clarity is needed here and a way of measuring simplicity in any of its versions, to have a meaningful assessment of various types of economic models in this term. Clearly, simplicity is not the only possible virtue of theories and models. More needs to be said about how simplicity competes with or complements other criteria of theoretical success.

    In a subsequent section, Becchio moves on to discuss “the reduction of complexity” (p. 12) achieved in rational choice theory, in particular through the homo oeconomicus model. Again, she seems to work under the assumption that by summarising certain claims raised in earlier debates, for instance, by quoting Rosenberg who called rational choice theory “formalized folk psychology”v, she has performed a satisfactory critical analysis. Her conclusion seems to be that “neoclassical economics […] does not explain the complexity of human actions in a realistic way.” (p. 13) What one should make of this claim in a serious contribution to the philosophy of economics depends crucially on what is meant by “realistic.” For the argument from realism to get off the ground, one needs an explicit conception of realism in theory-building. Although the references (to Nagel, Mäki and others) suggest that the author is aware of possible starting points, unfortunately this key concept is not explicated in a systematic way in the paper.

    In sum, section 2 offers, a brief and sketchy overview of some debates on “simplicity” and “realism” in “mainstream” or neoclassical economics. Up to a point, and in broad terms, one can agree with the author that “neoclassical economics rests on hypotheses and assumptions that are not just simple (in terms of Occam’s razor) but oversimplified and inconsistent with reality” (p. 8) Yet to make progress in the discussion, i.e. to provide a meaningful and reasonable criticism of neoclassical economics in terms of the simplicity-realism trade-off, requires further effort in explicating, operationalising and measuring both simplicity and “consistency with reality” as dimensions of theoretical success.

    In section 3, the author introduces what is supposed to be another interpretation (“blade”) of Occam’s razor which, in the hands of “heterodox” economists, leads to results that the author finds superior in terms of the simplicity realism trade-off. As we have seen, there is some ambivalence whether this second, previously neglected but preferable, interpretation of Occam’s razor is characterised as “methodological” or “heuristic” – the two terms are surely not coextensive. At any rate, the author seems to prefer the latter and operates with the term “heuristic” throughout.

    Yet this terminology generates some confusion right from the start. It is not clear whose use of heuristics is referred to. The term can refer to either (1) the methodological rule of thumb used by researchers while building theories or (2) to specific features of human behaviour under a certain theoretical description, namely when (economic) agents are modelled as using rules of thumb. This ambivalence is, as far as I can see, left unresolved and becomes especially pertinent in the discussion of Gigerenzer’s theory. One way to resolve it would be to naturalise the former use, i.e. (1’) re-describe theory-building as a human activity behavioristically, as a special case of human decision-making. Such a naturalised theory of science is perhaps broadly in line with Gigenrenzer’s or Herbert Simon’s approach to (economic) theorising.

    It seems, however, that the author tends to refer to heuristics in the original sense (1) and remain within the domain of analytical and normative philosophy of science, when contrasting one type of models or method (“neoclassical”, “static”, “mathematical-deductive”) with another, preferred, type of “alternative models, such as rhetoric, persuasion, metaphors, and heuristics.”

    In the second half of the paper, the author’s goal seems be both to explicate parsimony as one (or the?) adequate criterion of theoretical success, and to demonstrate how this criterion favours heterodox economic models over neoclassical ones. This is quite a task, so understandable there is hardly any room for more than assertions and hints. Thus, the author asserts that by using Occam’s razor as a “heuristic tool”, one could “build up a coherent and realistic theory of human behavior and economic dynamics in social terms.” It is mainly through hints at Herbert Simon’s ideas about satisficing behaviour and Gigerenzer’s evolutionary theory of heuristics that this claim is to be made plausible. Thus, the author suggests (p. 15):

    According to Simon, “human beings have reasons for what they do, but they seldom maximize utility, (…) given the complexities and uncertainties of the choice situations they face” (Simon 2001, 57). This is an explicit and direct application of the heuristic blade of Occam’s razor, focused especially on parsimony. If we intend satisficing as a humanly rational strategy, it appears simpler and more frugal than maximizing because in this scenario, individuals are following the rule of thumb in a dynamic context; while in the neoclassical scenario individuals are optimizers in a static and non-realistic context of perfect knowledge and complete information.

    Simon’s theory of satisficing is a clear, albeit sketchy example of how this heuristic works and is highly relevant in this context. Satisficing as an empirical generalisation is surely a possible starting point of a fruitful and powerful research programme.vi Yet it is worth keeping in mind that the model of human behaviour in terms of heuristics, just like any model, is subject to the simplicity/realism trade-off in the same manner as models of maximisation or optimisation are. At the very least, a meaningful comparison of neoclassical and satisficing or heuristic models requires the use of the same criteria of success. Thus, a fruitful question here is: under what criteria and in what contexts are these “alternative models” superior to the former ones.vii

    Let me return to the question of bottles, i.e. the terminological and conceptual choices of the paper and the possible theoretical motivations for these choices. Thus far, I have assumed that the main concern of the paper is a certain trade-off between simplicity and realism. As far as this is correct, there are various possible framings of the problem which can be compared in terms of transparency, rhetorical power and other criteria. As far as the explicit arguments in Becchio’s paper go, some bottles are better than others.

    It is doubtful whether the paper’s substantive argument for realism is made any stronger by linking it to either Simon’s 2001 paper or the mainstream philosophy of science distinction between epistemic and methodological interpretations of Occam’s razor. It is not clear how an invocation of Simon, even when he explicitly refers to Occam’s, provides adequate meta-theoretical tools for Becchio to characterise simplicity and realism as a trade-off: this latter terminology is not coherent with Simon’s specification of parsimony as a specific type of simplicity. Simon the information theorist uses an analytical distinction between simplicity and parsimony and does not talk about “realism.” Nor does he imply the choice of parsimony over simplicity as a superior criterion of inter-theoretical comparison. It seems that the invocation of parsimony or Occam’s razor is neither necessary or sufficient to support the comparison of various economic models in terms of realism and simplicity. In light of this, the author’s innovative rhetorical move to play out one interpretation of Occam’s razor against the other, and to juxtapose simplicity and parsimony seem valid metatheoretical steps but they are insufficiently motivated.

    To recall, Hirschman, the arch-heterodox economist followed another, rather conventional methodological route, viz. to argue in favour of realism and against parsimony. To put differently, Hirschman did not see realism and parsimony as necessarily pointing in the same direction. This suggests that there is a possible trade-off between the two. This conceptual framework allows for a rather simple, perhaps simplified, but transparent explanation of how standard and heterodox models are conventionally compared in economics: a gain in realism at the expense of simplicity.

    As a variant on this conventional strategy, consider Richard Thaler (1992, 198), another Nobel laureate and father of behavioural economics who once noted that devising accurately descriptive models of human behaviour is difficult because many theorists have a “strong allergic reaction to data.” Moreover, economic models based on the assumption of rationality are “elegant with precise predictions,” while behavioural research tends to be “messy, with much vaguer predictions.” He then asked a rhetorical question: “would you rather be elegant and precisely wrong, or messy and vaguely right?” (ibid.) This suggests that there are two important dimensions of inter-theoretical comparison; furthermore, one (truth, something similar to realism) is more important than the other (elegance, perhaps something akin to simplicity).

    Undoubtedly, the temptation to go beyond these two rather boring and unsophisticated metatheoretical frameworks is strong. This is possibly sufficient motivation for the rhetorical framing of Becchio’s paper: the dissociation of simplicity and parsimony and the association of parsimony and realism.

    However, if one starts to speculate about the ultimate underlying ambition of the paper, there is a certain ambiguity. Are the claims meant as conceptual or substantive ones. It may be possible to carve out an argument that the two criteria, realism and parsimony necessarily point in the same direction: this would require some conceptual heavy lifting. It is also possible that in a comparison of two theories or two models, the first is both more realistic and more parsimonious than the other. To make the substantive (empirical) argument that this is (or is not) the case for any two specific pairs (or n-tuples) of theories or models, one needs, apart from a minimal conceptual clarity, to operationalise and measure theories and models in these two dimensions. If Becchio wants us to take either of these less conventional routes further, she owes us a compelling argument.

    References

    Becchio, Giandomenica 2020. “The Two Blades of Occam’s Razor in Economics: Logical and Heuristic.” Economic Thought Open Discussion forum.
    Cserne, Péter 2019. “The Uneasy Case for Parsimony in (Law and) Economics: Conceptual, Empirical and Normative Arguments.” Global Jurist 18 (2019) 3, 20190001, DOI: https://doi.org/10.1515/gj-2019-0001
    Dietrich, Franz and Christian List. 2013. “A reason-based theory of rational choice.” Noûs 47: 104–134.
    Hirschman, Albert O. 1984. “Against Parsimony: Three Easy Ways of Complicating Some Categories of Economic Discourse.” American Economic Review 74: 89–96.
    Simon, Herbert A. 1997. “Science Seeks Parsimony, Not Simplicity. Searching For Patterns in Phenomena”, Carnegie Mellon University Complex Information Processing Working Paper #534, available at:
    http://digitalcollections.library.cmu.edu/awweb/awarchive?type=file&item=47027
    Simon, Herbert A. 2001. “Science Seeks Parsimony, Not Simplicity”, in Zellner A., Keuzenkamp H., McAleer M. (Eds.) Simplicity, Inference and Modelling: Keeping it Sophisticatedly Simple. Cambridge: Cambridge University Press, pp. 32-72.
    Thaler, Richard H. 1992. The Winner’s Curse: Paradoxes and Anomalies of Economic Life. Princeton: Princeton University Press.

    Endnotes

    i.Hirschman 1984. In Cserne 2019, I identify and analyse three arguments for parsimony in economics, using the term roughly in Hirschman’s sense.

    ii.“The primordial acts of science are to observe phenomena, to seek patterns (redundancy) in them, and to redescribe them in terms of the discovered patterns, thereby removing redundancy. The simplicity that is sought and found beautiful is the simplicity of parsimony, which rests, in turn, on the exploitation of redundancy. We do not seek the absolutely simplest law but the law that is simplest in relation to the range of phenomena it explains, that is most parsimonious.” Simon 1997 p. 6. I am quoting from what appears to be the working paper version of Simon’s 2001 chapter as I had difficulties accessing the 2001 version under the current circumstances.

    iii.“Occam’s razor in economics has been adopted as both an epistemic tool and a methodological principle.” Becchio p. 3.
    iv. “In using Occam’s razor as a pure logical tool, one that gives simplicity to a model, neoclassical economics commits a fault: it neglects the fact that in the human sciences a trade-off between simplicity and realism exists, and this trade-off cannot be confused with the degree of abstraction that characterizes the object of any natural science, like physics, because human behavior cannot be confused with the behavior of, say, atoms.” Becchio p.3.

    v.The subsequent quote from Buchanan (on p. 13) is not easy to interpret, nor is it clear what role it plays in Becchio’s argument. Is it an illustration of the criticised position or an endorsement of the author’s criticism? At the face of it, it does not seem to support either. It is difficult to discern what Buchanan means with the potentially confused spatial metaphor of “within the boundaries between…”. Does he refer to a theoretical space in between theories, i.e. a third one different from both? Or does he mean the intersection of the theories, thus recognising some common features of both?

    vi.Note that in spite of what Simon in the quote suggests, satisficing is different from explicitly modelling decision-making in terms of reasons, something that has been modelled in social choice theory recently, with no “heterodoxy” involved (Dietrich and List 2013).

    vii.At an earlier point in the argument, it seems as if the author wished to criticise neoclassical model-building as such. At least this is one’s impression when he characterised it as a “powerful rhetorical discourse” (p 8 n 11) To be sure, heterodox models are models nonetheless. Not to mention that a reference to rhetoric should not imply a criticism.

  • Steve Fleetwood says:

    I think the fundamental arguments in this paper are interesting and, I might add, alerted me to things I did not know about Occam’s razor (OR). But, at the moment, the paper reads like a set of notes for a paper rather than a paper. In short, it needs more work to bring out the fundamentals.

    If I understand Giandomenica correctly, she is arguing the following: OR has two blades: (i) simplicity and (ii) parsimony. Neoclassical economists have used the simplicity blade of OR as a logical principle to highlight the strict logic and simplicity of neoclassical models based upon unrealistic assumptions. Some heterodox economists have used OR’s parsimony blade, as a heuristic tool to build models on realistic initial assumptions.

    The first thing that is required is much better definitions of simplicity, complexity, parsimony, heuristics and realisticness – and a couple of others. Giandomenica may think she has done this, but, alas, she has not. These terms are exceptionally slippery and the meaning she gives to them needs to be make 100% clear.

    The same goes for the distinction between:

    an ontological razor (it is rational to believe the simpler theory), an epistemic razor (it is rational to adopt the simpler theory), and a methodological razor (it is rational to adopt the simpler theory because that is what the world is like

    I am not sure that ontological, epistemic and methodological are suitable labels for these three categories. For example, “this is what the world is like” (the way the world is, as Mäki puts it) is ontological not methodological. This needs clarifying.

    Incidentally, Simon’s bounded rationality does not dispense with Rational Economic Man (REM sorry for the sexist terminology). He keeps it, and simply reduces the information that REM can access. Genuinely dispensing with REM would mean embracing things like unconscious action – e.g. habit-guided action, or acting upon stereotypes. Personally, I think many heterodox economists could accept this, maybe even Hayek’s notion of tacit rule following. Simon does not quite break free of mainstream economics and is therefore not a good example of heterodoxy: Hayek is.

    The second thing that is needed is a clear structure for each of the two parts. Currently we have the following:

    1. OR, logical blade → simplicity → unrealistic assumptions. Unfortunately, in between we have a jumbled set of ideas and concepts – e.g. scientificity, axiomatics, deductivism, quantification, mathematization, not to mention points a, b, c and d on p 8. Personally, I see matters differently:1 tractability assumptions (i.e. necessary to make the maths work) → realistic assumptions. The reader needs to see a clear chain of causality from OR, to simplicity, to unrealisticness. Incidentally, the footnotes have another set of ideas and concepts. If they are part of t causal chain, then they need to be included.

    2. OR, heuristic blade → parsimony → realistic assumptions. Unfortunately, once again, in between we have a jumbled set of ideas and concepts – e.g. agency and institution/structure. Again personally, I would say: no need for mathematics → no need for tractability assumptions → no need for unrealistic assumptions. The reader needs to see a clear chain of causality from OR, to parsimony to unrealisticness.

    I see how and why Friedman fits into the argument (realisticness for him doesn’t matter) but I don’t see the point being made apropos Samualson.

    Perhaps the third thing that is needed is this. I think Giandomenica would be better advised to lose the Friedman & Samualson stuff (after all it is now very dated) and maybe pick on a case study, say, of a contemporary model that epitomises everything that is wrong with mainstream economics. This could, then, be compared to a contemporary non-mainstream `model´ that epitomises everything that is right with non-mainstream economics. Maybe Hayek would fit the bill here, or better still, a contemporary Austrian economist that completely rejects mainstream economics – I presume Giandomenica has a penchant for Austrians?

    Finally, Giandomenica has to deal, head-on, with the “all models are simplistic” claim. This is, after all, what made Friedman´s nonsense so powerful – as a rhetorical device. The fact is, all models are simple, in the (trivial) sense that they exclude stuff like Jupiter’s moons! I built a non-mainstream model of labour markets, and it could be said to be simple in this trivial sense. But, I did not exclude anything that was central to the fundamentals of labour markets, so in this sense it was not unrealistic. Giandomenica needs to say that simply excluding stuff (in order to simplify) does not necessarily make a model unrealistic – it all depends upon what is excluded. Excluding (say) institutions from a model of labour markets makes it unrealistic; excluding the hair colour of the HR manager’s eyes does not.

    Endnote

    1. I am not, of course, insisting that she agree with me, but I thought being given an alternative reading might help. Incidentally, this is Lawson´s argument.

Your comment

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Please note that your email address will not be published.