Should we Get Rid of the Preference Concept in Normative Economics?

Douglas Bernheim (with his co-author Antonio Rangel) is a major contributor of the recent developments within the so-called ‘behavioral welfare economics’ (BWE) research program. BWE can be viewed as an attempt to lay the theoretical foundations for the normative turn of behavioral economics that has been engaged in the late 1990s. Contrary to (too) many contributions within behavioral economics that merely take a casual approach to normative issues, the BWE pioneered by Bernheim and others builds on and extends the formal apparatus of welfare economics to cases of various behavioral inconsistencies. In this light, Bernheim’s recent paper “The Good, the Bad, and the Ugly: A Unified Approach to Behavioral Welfare Economics” published in the Journal of Benefit and Cost Analysis in which he attempts to develop a unified approach to BWE is worth reading for anyone interested in behavioral and normative economics.

Bernheim’s paper covers a lot of ground and makes many important points that I cannot discuss in detail here. I would retain however a very general idea that Bernheim never explicitly develops but which is nonetheless at the core of a key difficulty the normative turn of behavioral economics has given rise to, i.e. the problem related to the use of the preference concept in the normative analysis. Welfare economics has indeed historically made used of an account of welfare in terms of preference satisfaction. The pervasiveness of this account is due to several factors, at least two of them being especially prominent. First, the preference-satisfaction view of welfare is tightly related to a form of consumer sovereignty principle according to which economists should defer to individuals’ judgments regarding what is good or best for them. This principle may itself be justified on different grounds, for instance epistemic (individuals have a privileged access to the knowledge of what is good for them) or ethical (we must defer to individuals’ judgments because individuals are autonomous agents). Whatever the specific justification, the consumer sovereignty principle indicates that there are presumably strong reasons to view the satisfaction of preferences as being constitutive (or at least a good proxy) of welfare. Second, these reasons are strengthened by the formal isomorphism between the binary preference relation that is at the core of the notion of rationality in economics and the ‘better than’ relation that underlies any welfare judgment. As all microeconomic textbooks explain, rationality in economics correspond to the existence of well-ordered preferences over some domain of objects. In particular, this presupposes that the binary preference relation is transitive. On the other hand, there are good reasons to take the ‘better than’ relation as being intrinsically transitive – in the same way as the ‘taller than’ relation for instance. The fact that preferences and welfare judgments are well-ordered obviously makes easier (though obviously does not require) to assume that one can be directly mapped into the other. Problems however arise once it appears that individuals do not have well-ordered preferences. This is precisely the motivation behind the whole BWE approach.

As it is well-known, standard welfare economics essentially adopt a revealed preference stance and therefore associates welfare and choice. According to Bernheim, such an association is done on the basis of three premises:

Premise 1: Each agent is the best judge of our own well-being.

Premise 2: Our judgments are governed by coherent, stable preferences.

Premise 3: Our preferences guide our choices: when we seek to benefit ourselves.

Bernheim’s unified approach accepts (with several qualifications) premises 1 and 3 but rejects premise 2. This is important because, as I indicate just above, premise 2 has been essential in the economists’ endorsement of the preference-satisfaction view of welfare. In the normative writings of behavioral economists, premise 2 tends to be kept even in the face of experimental results by postulating that individuals have ‘true’ (or ‘latent or ‘inner’) preferences that failed to be revealed by choices. Crucially, these are these preferences rather than the actually revealed ones that are assumed to be well-ordered. Alternatively, it is sometimes assumed that individuals are endowed with several and mutually inconsistent preference orderings which, depending on the choice context, may determine one’s choice. Bernheim rejects, rightfully I believe, these views. He rather endorses the ‘constructed preference’ view according to which “I aggregate the many diverse aspects of my experience only when called upon to do so for a given purpose, such as making a choice or answering a question about my well-being”.

Bernheim’s endorsement of the ‘constructed preference’ view is important to understand why he continues to accept premises 1 and 3, even with qualifications. Regarding premise 1, he basically grounds it on the consumer sovereignty principle I discuss above. The qualification of premise 1 rests on the fact however that not all individuals’ judgments provide a correct assessment of their welfare. Berhneim argues for the distinction between ‘direct’ judgments, referring to ‘ultimate objectives’, and ‘indirect’ judgments, referring to the determination of means to achieve ultimate objectives. Only the latter but not the former are susceptible of mistakes that should be accounted for in the welfare analysis. Another way to state Bernheim’s claim is that behavioral economics cannot provide evidence for the fact that individuals are eating too much or insufficiently saving. But it can demonstrate that given their aims, individuals fail to make the best choices. Premise 3 then indicates that individuals’ aims, under appropriate conditions, indeed guide their choices. This makes choices often – but not always – a good indication of individuals’ welfare judgments. On this basis, Bernheim claims that choice-based welfare analysis should be regarded as preferable to approaches relying on self-reported well-being, including hedonistic measures favored by some behavioral economics. The unified approach to BWE Bernheim is suggesting proceeds through two steps: first, the behavioral welfare economists should determine the ‘welfare-relevant domain’ by identifying which choices merit deference (i.e. which choices are actually guided by welfare judgments); second he should construct a welfare criterion based on the properties of choices within that domain. Since there is no presumption that individuals’ preferences and judgments are well-ordered (rejection of premise 2), the resulting welfare-ordering will most of the time happened to be incomplete.

As I say above, the main lesson I retain from Bernheim’s paper is that we should probably completely abandon the preference concept in normative economics. To understand how I arrive at this conclusion, two points should be noted. First, Bernheim’s unified BWE approach does not rely on any assumption of the existence of true or inner preferences. There are choices on which the welfare analysis can build (the welfare-relevant domain) but this is completely unrelated to the existence of true preferences. The determination of the welfare-relevant domain will rather depend on the identification of mistakes or cases of lack of information and even once this is done, there is no guarantee that the choices will be coherent. Second, rather than preferences, Berhneim rather uses the term ‘judgments’ to refer to what is reflected by choices. Bernheim shows that ultimately the individual’s welfare judgments are reflected by a welfare relation that is isomorphic with a choice-based binary relation that has minimal formal properties (especially acyclicity). Interestingly, even if one rejects Bernheim’s choice-based approach, the alternative SRWB approach can also completely dispense with the preference concept since it does not need to stipulate that individual have well-ordered inner preferences.

As a conclusion, it may be worth remarking that Bernheim’s unified approach is somewhat ambiguous regarding the relationship between welfare judgments and choices. In a revealed-preference perspective, we may suppose that (possibly incoherent) choices reveal (possibly incoherent) welfare judgments. But at the same time, there is a sense in which Bernheim’s whole argument builds on the idea that choice-based welfare analysis is preferable because our choices are guided by our welfare judgments. In particular, the identification of mistakes is possible only because we can – at least in principle – identified either a gap between ultimate judgments of welfare and actual choices (what Bernheim calls ‘characterization mistakes’) or inconsistencies in judgments.  In either case, the relevant criterion for the welfare analysis is not that individuals are making inconsistent choices per se but rather how judgments and choices are articulated. Added to the fact that the unified approach naturally calls for the use of non-choice data, this indicates a real departure from the revealed-preference view that is still pervasive in welfare economics. All of this as interesting implications for another topic related to the normative turn of behavioral economics, i.e. libertarian paternalism, but I leave that for another post.

Wade Hands on the Representative Agent Assumption in Macroeconomics and Revealed Preference Theory

Note : This post has been originally published at the blog “Rationalité Limitée“

I am currently doing the ultimate refinements on my working paper “Sen’s Critique of Revealed Preference Theory and Its Neo-Samuelsonian Critique”. In the process, I have just finished to read several papers by Wade Hands on Paul Samuelson’s contribution to revealed preference theory and general equilibrium analysis. Hands’ papers are mainly historical but they also contain many interesting methodological and theoretical insights regarding the significance and the relevance of the representative agent assumption in modern economics.

Hands emphasizes several important points regarding how the representative agent assumption (RAA) came to be standard in macroeconomics but also – less obviously – in microeconomics. I start first with the macro side. In his paper “The Individual and the Market: Paul Samuelson on (Homothetic) Santa Claus Economics”, Hands convincingly argues that the RAA was not part of the standard Walrassian model of perfectly competitive markets of the mid-twentieth century. Indeed, this assumption was regarded as contrary to the fact that the Walrassian model is about decentralized economies with interacting and heterogeneous agents. As Arrow and Hahn put it on the very first page of their classical General Competitive Analysis:

“Whatever the source of the concept [equilibrium], the notion that a social system moved by independent actions in pursuit of different values is consistent with a final coherent state of balance, and one in which the outcomes may be quite different from those intended by the agents, is surely the most important intellectual contribution that economic thought has made to the general understanding of social process”.

In this book and in their subsequent writings, Arrow and Hahn (and others) do note that it is possible to assume the existence of some representative agent. They also note that this assumption would considerably simplify the theoretical work but at the cost of removing all the interesting content and findings from the analysis. This is in sharp contrast with the standard practice in contemporary macroeconomics through the so-called dynamic stochastic general equilibrium (DSGE) models where the economy is routinely modeled through the interaction between a representative household and a representative firm. Most of the time, regarding the demand side, the RAA is based on the key postulate of homogenous homothetic preferences (or utility functions). Homothetic preferences display an interesting feature: they do not exhibit income effects (i.e. the households’ preferences are invariant to income changes). When combined with the assumption that these homothetic preferences are identical across the population, this entails that the population of households behaves as a single rational agent maximizing utility under budget constraints. In other words, the market demand function exhibits the very same properties as individual demand functions and this function is “well-behaved”. The Walrassian model with the assumption of homogenous homothetic utility functions is an instance of what Samuelson used to call “Santa Claus economics”, i.e. economics based on models with very strong and unrealistic assumptions. As Hands show, though Samuelson himself has built and used such models several times, he was in general very skeptical regarding their relevance for general equilibrium analysis. The reason was not only empirical (preferences heterogeneity as well as income effects are clearly empirically significant features of modern economies) but also theoretical and methodological. For instance, Hands points out that assuming homogeneous homothetic preferences is the same than to apply directly the axioms of revealed preference theory to market (and not individual) demand functions. It also makes unnecessary all Samuelson’s contribution in the Foundations of Economic Analysis on the stability of competitive equilibrium since obviously if the representative agent’s preferences are well-behaved, we have a standard maximization problem with a unique and stable equilibrium.

This connects with the micro side of Hands’ papers. Hands shows that the RAA has also gained much popularity in microeconomics, though practitioners demonstrate less reflexivity about this than their macroeconomist colleagues. The rise of the RAA in microeconomics is particularly related to the development of a specific approach in demand analysis that Hands calls “contemporary revealed preference theory”. In a nutshell, contemporary revealed preference theory starts from a “restrictive domain” (i.e. a finite set of data regarding prices and quantities) and use some set of tools and axioms to infer a utility function “rationalizing” these data. This allows making prediction regarding demand in case of some parameters shift (e.g. changes in prices). Interestingly, the individuals almost completely disappear from the analysis: the axioms of revealed preference theory are directly applied to market demand functions without assuming that individual agents are “rational”. As Hands suggests, this is nothing but an instance of the RAA: it is simply assumed that market demand behaved as an individual agent and the axioms of revealed preference theory allow determining a utility function for this representative agent.

Samuelson is of course the pioneer of revealed preference theory. However, Samuelson’s aim was quite different. Essentially, it was to show the mathematical equivalence of ordinal utility theory and revealed preference theory and thus to demonstrate the possibility of conducting demand analysis without the use of psychologically-charged concepts such as utility. But, as for general equilibrium analysis, he was also reluctant to apply directly the revealed preference axioms to market demand function as it would be the same as to assume that there is some representative agent.

An interesting issue is of course how the RAA has gained such popularity in contemporary economics. Hands discusses several possibilities in his paper “The Contemporary Conundrum: Behavioral Economics, Rational Choice Theory, and the Representative Agents”. He suggests that social and technological features like the relative economic stability from the 1950s onwards on the one hand and the rise of computational power on the other have contributed to make issues of coordination and disequilibrium less salient and to facilitate empirical analysis of demand respectively. Another important factor (at least regarding the micro side) is the fast development of experimental economics and particularly the rise of the notion of “ecological rationality”. The latter has contributed to recognition that market rationality does not necessarily depend on individual rationality.

Interestingly, among the possible explanations, Hands rejects the possibility that the RAA is just the emanation of a deep trend in economics regarding the “profusion” (or the “proliferation”) of agency, i.e. the fact that the tools of choice theory are applied in a more and more abstract way to wider range of different types of economics agents: persons, firms, states, selves, … Hands argues that this explanation does not work in the case of the RAA because contrary to firms or states, the representative household does not actually exist in the world and does not make decisions. This is a fair point but it depends on an ontological commitment regarding what an economic agent actually is. A counterargument could be that an economic agent is simply but anything that can be represented as making consistent choice. This is indeed a nice way to rationalize (rather than to explain) at least part of the use of the RAA, as Don Ross demonstrates in several writings. A serious problem with this rationalization is that it almost inevitably leads to a divorce between positive and normative economics: if welfare is broadly defined as preference-satisfaction, then to assume that the representative agent is a “real” agent is of course very problematic. Incidentally, this was also one of the reasons why Samuelson rejected the RAA for general competitive analysis.

As a more personal conclusion, Hands’ history of the RAA highlights the fact post-WWII standard economics has struggled from its beginning with the difficult problem of reconciling issues of agency with issues of aggregation. The RAA is an obvious way of sidestepping problems related to aggregation as exemplified by the SMD theorem. But this led economics to a very abstract and formal view of agency (and rationality) which is made possible by the fact that the tools of choice theory (choice functions à la Sen or Arrow, decision theory) are themselves very abstract. This is not a problem per se; it depends on the problems economists are interested in. But the worry is precisely that in some cases this approach may not be adapted to the problem at stakes (the RAA in macroeconomics) while in others it may lead to a problematic redefinition of the scope of economics (such as to exclude whole part of normative economics from the field).