When Arrow and Debreu published their famous proof of the existence of competitive equilibrium in 1954 their work was met with extraordinary praise. In fact, approbation was so intense that there was hardly any substantive criticism of the paper. Not a good omen. But then again, who needs to cast doubts when you want to have faith?
The existence question is not only a technical question (i.e., finding out if a system of equations has a solution). In the grand narrative of market theory, the issue of existence of equilibrium is relevant because it concerns the reference point towards which disequilibrium prices (and allocations) are supposed to converge. The idea of market forces leading an economy to a point of equilibrium would be meaningless without certitude about the existence of the promised land.[i] In macroeconomic theory, this is so important that in some extreme cases (i.e., dynamic general equilibrium models), it is assumed that the economy is always in an equilibrium position.[ii]
The existence of a general competitive equilibrium haunted neoclassical economics since Walras. He knew that equality in the number of equations and unknowns was not enough to guarantee the existence of a solution.[iii] The problem was assumed away and the question of existence remained without an adequate answer for decades. [iv]
The Arrow-Debreu model changed the name of the game. As Koopmans (1957) noted, “in this model the problem is no longer conceived as that of proving that a certain set of equations has a solution.” The idea was now to show there was a specific state of the economic system where at some prices of goods the manifold individual price-taking maximization behaviors are mutually consistent.
The Arrow-Debreu (A-D) model was acclaimed as the first rigorous solution to this problem.[v] The essential idea of A-D is to build a model of an economy and to utilize tools developed in game theory by Nash (1950). The center of attention here is a fixed-point theorem, considered today as the most important tool in mathematical economics (Geanakoplos 1989).
Arrow and Debreu’s intuition was that because every equilibrium is a rest point, the fixed-point theorem lent itself beautifully for the proof of existence. The trick was to be able to interpret the fixed point as an (general) economic equilibrium.
In its simplest version, the procedure using this tool can be schematized as follows.[vi] Imagine you have a set P of price vectors each with the property Σpi = 1 (i.e., for each price vector in P the sum of its components is equal to 1). Now suppose you build a mapping f that transforms price vectors into new price vectors. This mapping transforms each price vector p in the set P to a point p’, that is, another price vector in P. The fixed point theorem tells us that under certain conditions, the mapping has a fixed point, i.e., there is a p* in P such that f(p*) = p*. The validity of the theorem depends on the properties of both the mapping and the set on which it is defined.[vii]
Suppose you build this mapping so that it represents the law of supply and demand, i.e., if demand is greater than supply, the mapping ensures that the price of that commodity increases. If it is less, the price will decrease. If the excess demand equals zero (i.e., supply equals demand) the price will remain unchanged. Now, under certain mathematical conditions, imagine you found that your mapping transformed one price vector into precisely the same vector. This is a fixed point at which the prices would remain unchanged. Suppose you also demonstrated that at this fixed point for every commodity with a strictly positive price, supply would be equal to demand and where a commodity’s price was zero the excess demand would be negative (i.e., positive excess supply). Your proof of existence would be complete.
But this is where one needs to look at the proof in more detail, especially at the structure of the mapping. In both Brouwer’s and Kakutani’s fixed point theorems, the mapping must be defined on a convex, compact set (i.e., a set that is closed and bounded). The unit price simplex P has this property. It is composed of all price vectors such that the addition of its components equals unity, Σpi = 1. This is strictly a mathematical necessity in order to be able to use the fixed-point theorem.
There are various mappings (transformational rules) that change price vectors into new price vectors and represent the law of supply and demand.[i] But we need to guarantee that the new price vectors are elements of P, the simplex, for one simple reason: the fixed-point theorem requires that the mapping be defined on a compact set. Thus, the mapping must also involve a normalization procedure in order to ensure that all of the new price vectors belong to the price simplex (i.e., they must have the desired property: Σpi = 1). And here is where things go awry: this normalization destroys the interpretation of the mapping as a representation of the law of supply and demand.
In our paper “The law of supply and demand in the proof of existence of general competitive equilibrium” (by Carlo Benetti, Alejandro Nadal and Carlos Salas) we demonstrate that none of the mappings used in the various proofs of existence of a general equilibrium involving a fixed point theorem is a good representation of the law of supply and demand.[ii] In some cases, commodities where demand is greater (less) than supply may have their price reduced (increased). The reason is straightforward: the normalization process needed to ensure that the new price vectors belong to the price simplex destroys the possibility of interpreting the mapping as consistent with the law of supply and demand. With the normalization procedure, price changes for one commodity not only depend on the sign of the excess demand of that specific commodity, but also on the sign of the excess demands of the other n – 1 commodities. This is not what the law of supply and demand indicates.[iii]
Our conclusion: the proof of existence of a general competitive equilibrium using a fixed-point theorem fails to possess an economic meaning.
In 1994 I had a brief meeting with professor Kenneth Arrow. I showed him the abstract of our paper. As he read it, he smiled and said “You and your colleagues are mistaken, you are confusing stability with existence and this is of course the source of your error”.
We had anticipated this type of reaction, so I replied: “No, we are clear on that. This has nothing to do with stability. We know the existence proof does not replicate a dynamic process. But you and professor Debreu claim that the mapping used in the proof of existence represents the law of supply and demand, and what we are saying is that the normalization procedure you have to use to ensure that the new price vector is in the unit simplex contradicts your claim”.
“Oh, it’s about the simplex”, exclaimed Professor Arrow, “Then you’re right”.
That simple comment confirmed our results: the mappings do not represent the law of supply and demand. As a consequence, the proof of existence of equilibrium using a fixed-point theorem is devoid of economic sense.[iv]
REFERENCES
Arrow, K. and F. Hahn (1971) General Competitive Analysis. San Francisco: Holden Day.
Geanakoplos, John (1989) “Arrow-Debreu Model of General Equilibrium”, General Equilibrium. The New Palgrave. (Eatwell,J., M. Milgate and P. Newman, eds.). New York: Norton.
Koopmans, T. (1957), Three Essays on the State of Economic Theory.
Nikaido, H. (1989) “Fixed-Point Theorems”, General Equilibrium. The New Palgrave. (Eatwell,J., M. Milgate and P. Newman, eds.). New York: Norton.
Smale, Steve (1982) “Global Analysis and Economics”, Handbook of Mathematical Economics, (Arrow, K. and M. Intrilligator, eds.). North-Holland.
Walras, L. (1952) Éléments d’Économie Politique Pure. Paris: Librairie Générale de Droit et de Jurisprudence.
Wickens, M. (2008) Macroeconomic Theory: A Dynamic General Equilibrium Approach. Princeton: Princeton University Press.
[i] See the exposition of Arrow and Hahn (1971: 25-7).
[ii] We analyzed four different mappings: Arrow-Debreu, Nikaido, Debreu, Arrow-Hahn. In Debreu’s version, commodities with a positive excess demand may see their price reduced to zero just because there are other commodities with greater positive excess demands!
[iii] In the mapping used by Debreu, the price of commodities with positive excess may be reduced all the way to zero just because there are other excess demands that are greater!
[iv] We leave out of this analysis the existence results of global analysis à la Smale (1982) because they rely on stronger assumptions. It is true that this line of analysis “is closer to the older traditions” as Smale claims but this is precisely what Arrow and Debreu wanted to leave behind by relying on a fixed point theorem.
[i] In Nikaido’s terms (Nikaido 1989: 139) existence of an equilibrium “is a primary premise of the theory, on which all its developments are built (…). Without this consistency, the theory is void.”
[ii] See Wickens (2008).
[iii] In his Éléments d’économie politique pure he showed that multiple equilibria were possible. He also considered the case where supply and demand curves lacked an intersection point. See Walras (1952: 66-68).
[iv] John von Neumann (1937) was the first to use a fixed point theorem in a proof of existence.
[v] Other authors presented similar proofs: McKenzie (1954), Gale (1955) and Nikaido (1956).
[vi] For a detailed description of the entire procedure, see chapter Nikaido (1968), Chapter 5.
[vii] Brouwer’s fixed point theorem for single valued functions requires that P be a non-empty, convex, closed and bounded set, and that f be continuous. Kakutani’s theorem requires that f be an upper-semicontinuous correspondence for which the image set f(p) is a non-empty, convex subset of P.
von Neumann used a Brower’s fixed point too, way before AD. However, he was working in a different tradition that, as noted by Kurz and Salvadori, should be seen in line with classical authors. In this respect, I think that to fully grasp the limitations of the supply and demand approach understanding of what Garegnani calls the change in the notion of equilibrium is essential.
HI Matías, thanks for bringing this up. On von Neumann, check footnote iv: the paper I quote was read in Princeton in 1932, (the original German version was published in 1938). Von Neumann used Brouwer’s fixed-point theorem to prove the existence of a saddle point for his famous minmax problem.
There are difficulties with the interpretation of Kurz-Salvadori because Arrow-Debreu models can also cover production of commodities by means of commodities (although other problems arise with long run equilibrium, which is related to Garegnani’s point).
The Classical tradition is better described as one in which two different classes of prices exist: market and natural (production) prices. They are determined by different economic forces. The former by the law of supply and demand, the latter by capital accumulation and demographics (Smith) or technology and distribution (Ricardo). Neoclassical theory only has one force behind all price determination. But both, Classical and Neoclassical theories have one thing in common: that they have not been able to demonstrate how the forces of competition lead to gravitation or convergence to equilibrium.
This is why the existence question has become so important for general equilibrium theory. In the absence of good results in stability, the proof of existence became the only positive output of general equilibrium theory. This is why the story in our paper is interesting, because we go to the heart of the proof of existence. Arrow, Debreu, Hahn, Nikaido, Gale and others claim the mappings they use in their proofs of existence (of general equilibrium) represent the law of supply and demand. We show those claims are empty: those mappings do not have that economic meaning. The conclusion is straightforward: the proof of existence of equilibrium is devoid of economic sense.
i actually have that paper—saw it awhile ago. i’m not totally convinced, except for the idea that arrow etc. doesn’t make economic sense. i think axiomatically or logically it is self-consistant but i’d have to look at it again. there’s a large semi-technical book on general equilibrium theory which was reviewed in american mathematicaly monthly which i read in parts—the idea was ‘existance is proven, but hardly ever will be reached’ (which is it seems the standard algorithmic and NP-completeness result —‘in the long run, we’re all dead’). most interesting in that book was results from the 60’s (possibly debreu) showing cycles rather than a unique equilibrium can exist—-quite similar to other results in biology (possibly isomorphic).
it would be nice to break the argument down. i guess i can always look at repec to see if anyone has cited the paper. one can interpret arrow’s response as simply saying ‘yes, get lost’.
There might be something to the surplus approach. I suggest the paper by Bellino and Serrano here http://sie.univpm.it/incontri/rsa51/Relazioni/Bellino-Serrano.pdf. I think it shows that market prices do gravitate around normal long run prices.
In fact, there is a lot going for the surplus approach. But I see some interesting difficulties in Bellino-Serrano. In their formalization of the gravitation process market prices appear as a modified system of prices of production (with sectoral profit rate disparities). This has two difficulties. First, it totally breaks away from Smith’s concept of prices as resolving into rents, wages and profits. In Smith there is no room for a component covering the cost of means of production. (This is Marx’s famous critique about the missing “4th component”). It is amazing that Ricardo thought about completing/correcting Smith without realizing he was grafting an entirely different concept of prices onto Smith’s theory. So gravitation in Smith does not involve this concept of prices. The second difficulty is that once accept this view of market prices (as a modified system of prices of production) it becomes impossible to construct a standard commodity that can be used in the study of a dynamic process. Price movements become unintelligible in the system of market prices. I’m afraid that Bellino and Serrano do not succeed in getting away from Steedman’s critique because of this difficulty. Thanks for bringing this up. We need more discussion around these points.
Very intriguing post. However, isn’t this “normalization procedure” a kind of “redenomination”, only more artificial and frequent? In real life, redenomination doesn’t break the law of supply and demand, so maybe this procedure won’t contradict the law of supply and demand either.
cf) http://en.wikipedia.org/wiki/Denomination_(currency)#List_of_currency_redenominations
Hi Himaginary: thanks for your comment! There is no money in a general equilibrium model. Everything is done in terms of relative prices. So there is no room for anything like a currency re-denomination. On the other hand, excess demand functions here are homogeneous of degree zero on these relative prices, so that multiplying the vector of prices by a scalar will not change anything.
But that’s not what’s going on in these proofs of existence. To use a fixed point theorem, the mappings must be such that every new price vector falls back into the price simplex, and this means that the price adjustment rule is totally changed. Because of this normalization procedure, the price adjustment rule says that the new price of a commodity must be a function of the excess demand of that commodity AND of the sign of all the other excess demands in the economy! That is not what the law of supply and demand says.
Thank you for reply.
Let me introduce Appendix A (in the last page) of the following paper:
http://www.geocities.jp/kiishimizu/pdf/uzawa-lpo.pdf
This proves existence of equilibrium for the function before normalization. (Normalized function appears only as middle-process.)
“Because of this normalization procedure, the price adjustment rule says that the new price of a commodity must be a function of the excess demand of that commodity AND of the sign of all the other excess demands in the economy!”
On this matter, I think the main culprit is Walras Law rather than normalization procedure.
Thanks Himaginary!
Tanaka just proves the point we are making in our paper. On the other hand, the role of Walras’ Law is something to consider. Again, in our paper we show that for the case of two commodities, some of the mappings appear to respect the law of supply and demand, but in fact, this is due more to Walras’ Law.
Thanks for your comments and the paper.
Quite the breakthrough on an established truism. Not sure I agree with the conclusions of all your work (have to consider it further), but I always appreciate when somone challenges the status quo.
Spot on with this write-up, I really believe this site needs far more consideration. I’ll possibly be again to read much more, thanks for that info.
air jordans
I’ve been absent for some time, but now I remember why I used to love this weblog. Thank you, I’ll try and check back more frequently. How frequently you update your website?
[…] [vii] Brouwer’s fixed point theorem for single valued functions requires that P be a non-empty, convex, closed and bounded set, and that f be continuous. Kakutani’s theorem requires that f be an upper-semicontinuous correspondence for which the image set f(p) is a non-empty, convex subset of P. – See more at: http://archives.dollarsandsense.org.user.s436.sureserver.com/newtcb/what-is-equilibrium-never-existed/#sthash.2f6MeNoM.dpuf […]