1 Economic markets as calculative and calculated collective devices Michel Callon and Fabian Muniesa* April 2002 (Very provisional draft, uncorrected English, references to be added) Paper to be presented at the New York Conference on Social Studies of Finance, Columbia University and the Social Science Research Council Pricing is a key element of the functioning of concrete markets and its description and explanation is one of the most central concerns of economic theory. Our paper is an attempt to give a new description of these mechanisms. We consider markets as collective organized devices which make it possible to reach compromises, not only on the nature of the goods to produce and distribute, but also on the value to be given to them1. In other words, we see markets as social mechanisms which, based on a multiplicity of antagonistic conceptions of welfare and goods or, in other words, contradictory and competing values and worth, produce acceptable compromises. In particular, one of the most significant achievements of economic markets is that they pragmatically are able to calculate, on scale one, shared values for tradable goods, when situations are fuzzy and uncertain (C. Smith). Many authors have underscored this conception of the market. We know that Babbage was inspired by it when he devised his calculating machine (ref. CampbellKelly and Aspray on Babbage’s observation of the Banker’s Clearinghouse in London). More recently Leontieff said: "The economic system can be viewed as a gigantic computing machine which tirelessly grinds out the solution of an unending stream of quantitative problems" (in: Essay in Economics). David Stark, drawing on Boltanski and Thévenot's work, adopted the same point of view in proposing the notion of heterarchy to define the organization of markets and to highlight their role in the definition of values given to goods2. What is important, in this conception, is the fact that collective machinery affords practical solutions to theoretically insoluble problems, by making commensurable that which was not. The aim of this paper is to advance the understanding of the functioning or, rather, of the different modalities of functioning, of this computandi instrumentum. At this point, we might feel compelled to provide a minimal definition of calculation. By putting calculation at the heart of our approach, we should indeed be Centre de Sociologie de l’Innovation. Ecole des mines de Paris. callon@csi.ensmp.fr ; muniesa@csi.ensmp.fr * “Entrepreneurship is the ability to keep multiple orders of worth in play and to exploit the resulting ambiguity … As an ability to exploit ambiguity, entrepreneurship is not the property of an individual … instead it has an organizational basis. That is organizational forms will differ in their capacity to sustain an ongoing rivalry among coexisting principles of evaluation. We use the term "heterarchy" to refer to the organizational form with this reflexive capacity … Restated, heterarchy is characterized by distributed intelligence and the organization of diversity” (ref: Stark). 2 2 careful to avoid to dangers. The first is to simply get back to an abstract and formal view of economic markets governed by impersonal laws (as the law of the demand) and constituted of disembodies economic agents reduced to their preferences and their (more or less extended) cognitive competencies. The second danger, more subtle and commonplace in sociology, would be to simply get rid of this cumbersome notion and to dissolve the problem of calculation into the anthropological, fine-grained account of a concrete market composition. For many anthropologists what was meant to be a calculative behavior turns to be a matter of mere judgment, mere choice, mere guess, or, when it can be observed, as something imported from elsewhere (institutions, culture, like in DiMaggio economic sociology). We learn from the ethnographic fieldwork that, either in the supermarket or in the trading room, people hardly ever calculate, they rarely perform any proper arithmetic combination: they interpret information, they decide what to do on the bases of heterogeneous and not always well-defined criteria (ref. Miller, Stark, KnorrCetina). In the end (and we are afraid that this is the normal consequence of taking calculation in its hard, arithmetic sense) nobody calculates. A quick glance at some etymology could allow us to temper this notion of calculation in order to reintroduce it in a realistic account of concrete market practices (and to usefully blur the boundary between mere judgment and mere calculation). In his analysis of the vocabulary for counting and estimating in Latin sources (ref. Vocabulaire indo-européen), Benveniste points out that there is an intimate link between computing and judging or estimating, in one hand, and that, in the other hand, computing refers to literal detachment (cutting) and re-attachment (leading to a result)3. Borrowing from him, we propose to give a very general (and provisional!) definition describing calculation as a three step process: • The entities taken into consideration are sorted out: they are detached and display in a single space (and here we have to make the effort of imagining all the possible forms of this single space: from a sheet of paper to a caddy at the supermarket, from a simple slate to the input band of a Turing machine). • Relations are drawn between them: more precisely, they are subjected to manipulation and transformation (movements to the left or to the right, from one place to another, scroll up and down, for real or just virtually, simulation at large). • A result is extracted: a new entity is produced (a ranking, a sum) that corresponds precisely to the relations (the manipulations) performed; this new entity creates a series of links, of attachments, between the entities originally taken into consideration. 3 First, in the two main words involved in this vocabulary, the roots duco and puto (and the verbs ducere and putare), there is a strong amalgamation of both the technical meaning related to calculation and a more abstract use as authoritative judgment, consideration or estimation. Rationem ducere means "to lead (to conduct) a count to its total (summa)." This meaning is transposed to a more general sense that we can translate as to judge or to believe: "to reach a conclusion after verifying all the elements of the problem, like when verifying a count, after successive elimination of all articles." Second, putare (and this is the etymological root for computing) was an original rural term for cutting or pruning plants (like vine), specially getting rid of the useless branches. The metaphorical use for counting (rationem putare) can be interpreted on the basis of the same literal sense: "to detach successively from the list (the count) the articles that have been verified" or "to verify all the articles, one by one, in order to reach an acceptable result". 3 Our argument will then proceed as follows. First we examine the question of the calculability of goods: in order to be calculated, goods must be calculable. In the second section we introduce the notion of calculative agencies to understand how these calculable goods are actually calculated. Thirdly we consider the rules and material devices that organize the encounter between individual supplies and demands, i. e. the specific organizations that allow for a calculated exchange and a market output. Those three elements define concrete markets as collective organized devices that calculate compromises on the values of goods. We shall meet different versions of our minimal definition of calculation in each of these three elements. In the fourth section we turn to the question of the design of this collective machine, insisting upon the role of modelling, simulation and R&D: the conception of economic markets is becoming an activity in its own right; consequently markets become explicitly perceived as the result of social engineering and could be at the heart of public debates that are no longer limited to the community of experts. The calculability of goods Conceiving of the (concrete) market as a device for calculating the value of goods means focusing the analysis on those goods and taking the singular market transaction itself as the starting point.4 In a market transaction a good changes hands. This good is given a price that constitutes its monetary value. The buyer, in exchange for the price she pays the seller, acquires a recognized and guaranteed right to use the good in a certain way for a specified period of time. Once the transaction has been concluded, the buyer and seller leave each other. That is why we normally refer to a process of alienation: the two agents engaged in the transaction become strangers again once it has been concluded (ref: Slater +Miller). At this stage several comments are called for. First, the good involved in the market transaction is not necessarily a physically delimited, tangible good such as a car or a fish. A good is a thing in the sense in which Durkheim says we must consider social facts as things, or in the sense in which Desrosières talks of statistics as things, or in which Thévenot speaks of the “investment in forms”. That is why a service (which has no physical reality) can be the object of a market transaction if it has first been transformed into a thing. This is what the most recent work on the service economy (Gadrey) shows us. The fact, for example, of having a car for certain use, under certain conditions and for a certain time (which in this case defines property rights, in the Anglo-American sense of the term), constitutes a good in the same way as the acquisition of a car whose buyer is the sole owner for an indefinite period5. These two things – the purchased car and the rented car – are equally stabilized, delimited and definable. They hold together (stand for themselves?), have boundaries delimiting them. In short, and this point is well established now, they have objective properties which allow the application of property rights and thus make them This is the consequence of choosing to consider the market as a set of interconnected transactions and not as an abstract market. 5 The conditions of use of a car that I own are framed (by traffic rules, the availability of infrastructure, the existence of after-sales service, guarantees, etc.) as much as are those attached to a rented car. 4 4 fit for a transfer of property6. Likewise, there is no difference of materiality between a fish sold on the Marseille fish market (see below) and a one-week holiday in the mountains bought by a Briton dreaming of sun and snow. Both cases concern things that "hold together" and can be appropriated because they have objectified properties. Materiality and physicality are not confused. This point was made by the anthropology of science a few years ago already when it analysed the conditions of individualization and circulation of scientific statements. A thing is a black box composed of a large number of heterogeneous elements which have been folded, arranged and connected lastingly to one another and which consequently objectify the thing that they constitute7. This thing that holds together, because it is held by the (human and nonhuman) elements comprising it, is a good if and only if its properties represent a value for the buyer that she can evaluate in relation to her own criteria. This valuation is expressed in a price or range of prices that the buyer is prepared to pay to appropriate the thing, that is, to become attached to it, to incorporate it into her world. Once she has acquired this good, whose properties she has evaluated before entering into the market transaction, the buyer becomes the owner of the good. The transformation is in fact twofold: not only is the good possessed by the owner, it also becomes her own good. As M. Strathern so rightly points out: by saying that when A is the owner of vehicle V, we are referring to a market transaction, while by saying that V is the good belonging to A, we emphasize the fact that it has been incorporated into the world of A, of which it has become an integral part. How can we describe the transformation of a thing which turns into a good that an economic agent values? How can we explain the integration into the buyer's world of a thing designed and produced outside of it? One way of addressing this difficult question, without being tempted by traditional analyses that resort to notions such as utilities and preferences (Cochoy, 2002), is to start with the illuminating notion of individualization (or singularization) of products proposed by Chamberlin. For Chamberlin, the product is an economic variable in itself, one that is strategic because, owing to its variability, it allows a singular adjustment between what the consumer (or a specific category of consumers) want(s), and what the This profound similarity between the object thing and the service thing is more difficult to perceive in Roman law than in Anglo-Saxon law. In property law, based on Roman law, things are considered as (or supposed to be) divisible between persons ("each entity is split among the persons who have a claim to it, and has as many parts as there are persons who have rights, like a sum of money divided between several claimants" (Ref: Marilyn Strathern). For property rights based on the common law notion of "bundles of rights", things are intrinsically compound. The common law notion of a bundle of rights naturally leads to the conception of abstract rights (a good example of which is copyright). To clearly illustrate the difference between the two types of right, Marilyn Strathern takes the example of money. In Roman law the distribution of property rights over a given sum of money is translated into a physical division of that sum of money. In the common law notion of property rights, it is as if that sum of money could abstractly be divided into a part used for purchases of food, another for purchases of accommodation, another for leisure, etc. From an anthropological point of view these two perspectives are complementary: in one case persons are defined by ownership of things (which proliferate by division), while in the second case the proliferation of ownership which defines the objectivity of things allows the proliferation of persons. The notion of property is at the heart of this reversal. 7 A thing is a network in a black box, which is ponctualized. 6 5 supplier offers. This individualization process consists in the shaping of a product whose properties are progressively defined in such a way that it can enter the consumer's world, mix with it and become attached to it. We previously suggested calling this progressive adjustment a product qualification process. The market transaction concludes this successful adjustment. Throughout the process, the thing, which is a product undergoing qualification, is transformed into a good with adjusted properties and successful individualization. The transfer can then take place. The good leaves the world of supply, breaks away from it (which is possible since it is an objectified thing, that holds together), and slots into another world, that of the buyer, which has been configured to receive it. It becomes entangled in the networks of socio-technical relations constituting the buyer's world. This work of adjustment is the substance of any market transaction; only the terms of its implementation vary. But the essential point made by Chamberlin is that, in all cases, there can be no transaction without the individualization of the product, that is, without the qualification and adjustment implied by such individualization. This position leads him to the extreme argument that advertising, the seller's location and the salesperson's smile are not simply cosmetic contrivances used superficially and artificially at a late stage to personalize an otherwise impersonal good. The famous black T Ford, for which the customer can choose the colour as long as it is black, is no exception to the rule. In general a black Ford does not exist, there are only model T Fords, which may be black but which are individualized as soon as a buyers decide to turn them into constituent and often central elements of their own worlds8. The question that then arises concerns the obviously multiple and varied conditions and modalities of this process of individualization and singularization of products through which they are transformed into goods adjusted to the buyer's world. To understand this diversity one has to bear in mind the twofold constraint weighing on a product if it is to become a good: that of objectification (it has to be a thing) and that of singularization (it has to be a thing whose properties have been adjusted to the buyer's world, if necessary by transforming that world to produce the adjustment and attachment it triggers). Objectification and singularization are produced simultaneously; objectified properties are those that allow the individualization of the good. A description of the variety of the modalities of objectificationindividualization of goods does not fall within the scope of this article. Many studies on the subject are now available. We simply want to make a few points which help to clarify the conditions of the calculability of goods. First, the properties which define the good as an individual good and constitute its profile or identity are neither intrinsic nor extrinsic. Two symmetrical mistakes have to be avoided. The first would be to say that its properties are constitutive elements of the good. The second would be to say the opposite, that is, that the good is simply a blank screen on which social representations are projected, This point of view is now easy to agree with. The omnipresence of innovation and the increasing role of services have familiarized us with the idea of a singularization and individualization of products. But when Chamberlin was writing, deep in the era of Fordism, it took a lot of courage and imagination to put forward such claims. Chamberlin himself, in his discussion of the essential difference between monopolistic competition and imperfect competition, highlighted the fact that readers did not understand the argument (Chapter 4). 8 6 since its properties, reduced to mere meanings, are given to it by consumers or society. In the first case the good is considered to be objectively describable, for example in an unquestionable list of characteristics. In the second, all possibility of objectification of the good is denied, and the multiplicity of points of view and representations is emphasized. On the one hand Lancaster; on the other, Baudrillard . As soon as one agrees that there can be no market transaction without a process of objectification and singularization, the opposition dissolves. The purchase is not the result of a subject-object encounter, external to both, but of a process of attachment which, from qualification to re-qualification of the product, leads to the individualization of its properties. This does not mean that all re-qualifications are possible or that all strategies of attachment are equally probable. Properties are coproduced, which does not prevent them from being either real or singular. Second, this co-production of objectified and singular properties, which allows the appropriation of a good and its entry into the buyer's world, necessitates a sequence of work starting with the product design. This work involves also a large number of intermediaries who can be called professionals of qualification (including in the downstream part of the process, the packagers, merchandizers, marketers, advertisers, but also those who arrange goods on shelves, decide the localization of agencies, and so on: ref: Cochoy) and whose role is starting to be studied in earnest. Brand names, like personal relations with salespersons, are clearly part of this process of qualification and singularization. Third, the adjustment process increasingly involves a systematic and sometimes lengthy exploration of the network of entanglement constituting the (potential) buyer's world, especially in the service sector. One of the skills of designers, sellers and their colleagues is to be able to investigate, with the buyer, her attachments in order to be able to propose new ones and to deduce the most appropriate profile. This work is never more clearly visible than in market transactions on the Internet.9 Yet, as Miller shows, in a traditional market like the car market (i.e. sales and not rental), this exploratory work that allows the definition, objectification and individualization of the good to which the customer decides to become attached, also plays a large part. There is therefore no irreducible opposition between entanglement and market transaction (Miller does not see it, but Slater, commenting on Miller, rightly points it out). The paradox is that, to achieve the alienation of the good through which the buyers and the sellers will become strangers, increasingly heavy investments in the investigation of entanglements and in their reconfiguration are necessary. In short, to expand the market it is necessary to produce more and more entanglements. The proliferation of merchandize goes hand in hand with the proliferation of non-market attachments (ref Licoppe)10. FIGURE 1 ABOUT HERE Amazon is a good example of these strategies. To every buyer is associated the list of her previous purchases. Amazon, on the basis of this record, is able to make individualized suggestions about what could (or should!) be the next purchases. 10 The seller might be described as Serres parasites. He investigates and calculates relations without being caught up in the world he is trying to explore and to reconfigurate. Ref: Tiago. Morena. 9 7 We have said enough to introduce the notion of calculability of goods. By being endowed with properties, which both objectify and individualize it, a good becomes calculable by the consumer who has been enrolled in this process of objectification-individualization. By qualifying a good, that is, objectifying properties, which by individualizing it produce its adjustment to a singular consumer, its evaluation by its future buyer is made possible. In other words, for the buyer, evaluating a good is evaluating her attachment to it. In the evaluation of the good, and consequently in the fixing of its monetary value, we see the key role played by qualification, that is, by the objectification of singular properties that individualize the product. No calculation of the value of a good is possible if it has not been individualized, for the value of a good is nothing but the value of the buyer's attachment to that good, an attachment that implies the incorporation of the good into the buyer's world. The buyer can embark on that evaluation only if the good has been endowed with properties which make distinctions enabling her both to define the good and to compare it to other goods (see Cochoy and Buridan). In the vocabulary of professionals of qualification, this work has a name: positioning. As marketing manuals say, positioning defines consumer-targets (singularization) while defining the field of competition. Individualizing a good means endowing it with properties that distinguish it from other goods which then become comparable11. Economic life is a long series of such re-positionings or, in our vocabulary, re-qualifications12. Comparability might be achieved by a large variety of means, like the establishment of metrological networks to objectify and measure certain properties, the creation of quality labels and, more generally, the enforcement of quality standards (FED, Thévenot). Being individualized the good counts for the buyer who can consequently make calculation on it. Being comparable to other, at least substitutable, goods, it provides the buyer with the possibility to measure just how much she is attached to that particular good rather than to any other)13. The value of the good is calculable because its properties make it calculable A good is therefore not calculable "in itself" (as assumed by so-called hedonistic methods, for example), which does not mean that it has no importance in the calculation of its value by and for the consumer (as in the reduction of goods to their symbolic dimension). It is made calculable through a process of qualification which necessarily involves individualization, that is, some form of participation by the buyer (or one of her representatives), and in the same time some sort of similarities between differentiated goods. 14 This is the first step of what we defined as the calculating process (another way of phrasing that is to say that calculating means calculating positions: cf. The Laws of the Markets). 12 Slater notes, for example, how low-calorie bread, through a series of requalifications (packaging, distribution network, advertising, etc.) was repositioned to spread from the narrow market of women wanting to diet, to the broader health product market. This repositioning produced a new good with new properties, combined with older properties that were maintained but whose significance was thus modified. 13 This situation is clearly described by economic theory in the concepts of cross elasticity or, more generally, opportunity costs. 14 On the centrality of operations of classification, Bowker + Star, Sorting things out.. Conventions of quality as analyzed by Eymard-Duvernay and Thévenot play a crucial role in creating diverse possibilities of equivalencies and similarities between different individualized goods. Ref Descola. 11 8 As suggested above, this process constitutes any market transaction. We can nevertheless maintain that as such it assumes an increasingly central role. Investments made in order to achieve such individualization are continuously growing. One reasons is probably increasing competition which naturally makes attachments more and more difficult since they are constantly threatened by the strategies of detachment devised and deployed by competitors (ref: Callon et alii, The economy of qualities). Another reason is the spread of NICT which facilitate this singularization and the adjustments it involves. The service sector clearly illustrates this evolution, but it is probably in financial markets and, more particularly, derivatives markets, that the trend is most clearly visible. There, the product, elaborated by a bank for an industrial client who wants to be protected against financial risks, is often reduced to a mathematical formula, usually very complicated, which can be used at any point in time to calculate the financial yield of this product in relation to variations in the values of other financial entities. The product, a mathematical formula, is both objectified and singularized: objectified because the properties which qualify it by defining it have progressively been stabilized and set; and singularized because these properties have been determined in such a way as to adjust it as well as possible to suit the client's needs. This mathematical formula generates a number that can be compared to other numbers. It is singular and comparable, and consequently calculable, but in a way that is immediate (Lépinay). More generally, but probably to a less extreme degree, goods whose qualification leads to quantified properties – as in the case of industrial convention of quality (B&T) – are profiled to allow numeric calculation. In all instances, calculability involves the dual process of objectification of properties which, by being individualized, allow the attachment of goods and at the same time the evaluation of that attachment whose monetary equivalent provides a numeric and relative measurement. Calculative agencies Calculable goods make calculation possible. But for calculation to take place, there have to be calcutating agents15. Rather than talking of calculating agents, I prefer the idea of calculative agencies (Ref Callon: Laws of the m). First, this notion highlights the idea that they are not individuals, people, but most often hybrid collectives, what B. Latour has called centers of calculation. Second, it emphasizes the fact, related to the first, that To calculate does not mean necessarily doing mathematical or even numerical calculations. Calculation starts by establishing distinctions between things, states of the world, by ranking them and by imagining and evaluating courses of actions associated to those things or states as well as their consequences. By giving such an extended, but usual, meaning to the notion of calculation we try to avoid the conventional but too sharp distinction between judgement and calculation. Judgement, as far as market transactions are concerned, is better to be considered as a rudimentary form of calculation based, like in the Pierre Bourdieu sociology of consumption, on a very simple operation of social distinction. In order to avoid such an opposition that leads to the classical divide between what is categorized as social and what is categorized as economic, we prefer to consider a whole spectrum of types of calculation from the most simple and elementary ones to the more sophisticated and complicated ones. 15 9 these agencies are equipped with instruments: the calculation does not take place only in human minds; it is distributed amongst humans and non-humans (see Mac Kenzie). The notion of distribution is of course crucial. It is not enough to say that human flesh-and-blood agents, faced with difficult calculations, use tools, without which they would never be able to manage. The famous thesis of M. Weber on the role of double entry bookkeeping (DEBK) in the rise of capitalism is a good illustration of this possible misinterpretation. Yamey, criticizing Weber's thesis, or rather its reformulation by Sombart, affirmed, for example, that without entrepreneuship a simple bookkeeping tool could not have led to the rise of capitalism. It is not double entry bookkeeping that calculates, he says, but the human agent, in this case the entrepreneur who decided to use it (ref). In front of another audience I should have spent a good deal of time disproving this very commonly held thesis, especially in economics. For Chandler and then Williamson, but also for Simon, it is because they are faced with complicated accounting tasks that agents, to relieve their brains and enhance their performance, devise efficient tools and create rules and routines or set up organizational devices to calculate for them. The notion of a distributed calculative agency is more demanding and makes it possible to avoid this objection. The entrepreneur, who is supposed to calculate the return on investment, does not use the DEBK to effect a more precise, rapid and exact calculation, the idea of which existed before the tool itself. It is the couple constituted of the entrepreneur + the double entry bookkeeping that conceives of this calculation and performs it. We could go so far as to say that DEBK, simply by being there, available, proposes this calculation to the entrepreneur who accepts the invitation and in return asks DBEK to do the calculation16. That is why, in order to be clearly understood by economists and management scientists, Miller is probably right to use the notion of calculative practices (see also Vollmer) to avoid turning once again to an instrumental view of calculative tools. But, because the notion of practice is too general and too restrictive (it is for example contradictory with the very idea of agency), we prefer that of calculative agency which enables one to maintain the possibility, under certain circumstances, of strategic action while authorizing a great diversity of forms of agencies. The market transaction, in which the good switches hands from the seller to the buyer, takes place between two calculative agencies, each calculating the good in its own way. These calculations (sometimes) result in an agreement on the price (which may be set beforehand or negotiated during the transaction). The question is now: how can calculative agencies be characterized? To answer this question we can rely upon a set of studies which are not only very rich but also cover a very complete range of calculative practices developed in economic activities, on both the supply and the demand side (Stark, Miller, Vollmer, Powell, Akrich and Law, Berry, Lave, Conein, Cochoy, Preda). These studies show in particular the existence of a series of asymmetries which could be summarized as follows. • There is strong asymmetry between calculative agencies. Put simply, agencies' calculative powers and the complexity of the operations they can perform See the notion of affordance interestingly translated into French by promission (promise + permission). 16 10 are very unequally distributed. For example, the face-to-face between supply and demand in a supermarket involves (at least) two calculative agencies, although one of them (that of the supply) is infinitely better equipped than the other. To be sure, the consumer is never alone, isolated. He is also distributed and relies in particular on references, marks, and all sort of preformatted and precalculated information provided by the supermarket and its arrangement. Packaging, the organization of shelves, relations of proximity between products, brand names, labels, and testing by tasting17, all constitute a system of distributed cognition which participates actively in the qualification process. The consumer can also perform tests at home, with friends and family, and discuss the results. He reads magazines and books that guide his choices. He is engaged in relations of prescription that multiply by division his reflection and action. His companion, who has discussed the matter with colleagues at work, draws up a list, a veritable cognitive prosthesis, of products to buy. Consumer unions organize comparative tests that conclude with multicriteria evaluations (Di Maggio, Conein, Cochoy, Méadel, Akrich & Rabeharisoa, Dubuisson, Mallart). But no matter how strong it is, the calculative agency that evaluates the attachment of goods to the consumer's world remains weak compared to the calculative power of the supply which is usually heavily equipped at least in the case of a supermarket. From design right down to the positioning of products on shelves, a series of professionals is involved, all working in a coordinated way, exploring the distributed consumer's world to better integrate the product into it by playing skillfully on product qualification. The supermarket material device, which, as we have just pointed out, is mobilized by consumers in the calculation of goods, is itself largely the result of a calculation performed by the supply. By inspecting the shelves, reading labels and reading guides or instruction manuals, the consumer continues a calculation planned by professionals of qualification. There are even life-size supermarkets set up to investigate and observe consumer behaviours. Thus, we clearly have two calculative agencies that meet, and one of the two is better equipped than the other. From this gap we should not conclude that asymmetries always develop in the same direction or that they are set for once and for all. In some cases the buyer has calculative power superior to that of the seller (as in certain sub-contracting markets). In other cases one of the agencies, initially in an inferior position, gradually acquires tools allowing it to change the balance of power and to become more active in the process of qualification and individualization: the struggle for more autonomy or more recognition might explain why some agencies make efforts to enhance their calculative equipment. The increasing role of consumer organization or the obligation for enterprises to take environmental criteria into consideration in their own calculations are two examples of such a reversal. In other words, and this point is now well documented, the difference between the calculative capacity of a child hesitating between two packets of sweets or between two pokemon, and that of the manager of a supermarket does not stem Involvement of consumers in testing activities become more and more frequents. Recently a French bank has proposed to its potential clients to test its services during a six-month period without any engagement. 17 11 (only) from their own calculative competencies. They are essentially the consequence of the asymmetry of equipment. For example, as J. Lave clearly showed, the fact that a consumer does not use arithmetic when he shops does not mean that he does not calculate. He does calculate, but uses other instruments and procedures 18The difference is that facing him is a multitude of professionals armed with computers, comparing code bars, adding up prices to the last cent, monitoring variations in stock, measuring returns on investment, evaluating the impact of an advertising campaign, establishing econometric models intended to give a stylized representation of worlds of consumption, etc.19 • The second asymmetry concerns the type of calculation. The customer calculates an attachment; the seller performs a series of calculations in order to evaluate stocks, for example, or to measure profits or markets shares. It is clearly in this respect that we can say that values are radically different. A particular consumer may be attached to a good which he finds aesthetically attractive, and agree to pay a price for that attachment, while the seller will simply record a rate of return on the investment. But this asymmetry (what elements are entered into the calculation?) is not always definitive. For some market transactions the buyer may acquire specific tools, sometimes even provided by the seller, as it might be the case in the property market with the calculated negotiation of the granting of loans, for example. In this case the buyer does not only evaluate his attachment to a good that has been adjusted to his world; he becomes in addition a calculative agent capable of budgeting her own expenses and income. (cf. Bourdieu). On the other hand a seller, first configured as a classical capitalist calculative agency, can suddenly drop his usual calculating tools and re-evaluate the good he was about to sell by measuring his own attachment to it. • The third asymmetry is more difficult to define but nevertheless very important. It describes a given calculative agency's ability to integrate the results of other calculative agencies and to capitalize on them. That is the case for example of a firm that decides to create decentralized profit centers or business units, and delegates calculations to this large number of distributed calculative agencies. It then simply collects and summarize the numbers produced by those agencies (see Miller). On the supply side, the extent of this double movement of proliferation and integration of diversified calculative agencies, taking part in the process of good qualification, is obviously variable. But it is generally more developed than on the demand side. The consumer usually does not have this capacity to mobilize (and control) a large number of autonomous calculative agencies. Imagine what the market transaction would be if the client in a supermarket, standing in front of a shelf, had direct access to the calculative capacities of a consumer union or to the results of laboratory tests performed at his request. This asymmetry can act inversely, depending on the nature of the market transaction. It can also change. Aren’t you amazed by the number of buyers in cf. Cochoy and the notion of "qualcul" ("qualculation") to denote this form of calculation that is not necessarily based on arithmetic. 19 The current situation in Europe is very interesting from this point of view. We witness sellers and consumers armed with calculators converting local currencies into Euros – calculators distributed freely, which redistribute the use of digital calculation. 18 12 supermarkets who use their mobile phones to consult, in real time, distant interlocutors whose opinion they are obviously requesting. The Internet and ecommerce give the buyer some ability to mobilize diverse calculative agencies accessible on line, especially in the exemplary case of financial markets.20 The qualification of products, that is, the objectification-singularization of their properties, affords the conditions for calculation. This calculation is performed by calculative agencies who are all differently equipped and organized. Convergence on an agreement is the result of possible stabilization of the good whose calculable and calculated properties result in a single evaluation depending, in particular, on the calculative tools and devices used. But this agreement also depends on the mechanisms and modalities through which the different calculative agencies connect and encounter one another. It is this calculated encounter that constitutes the third point of entry into an analysis of the market as a calculative device. Calculated exchanges Calculable products which are calculated by calculative agencies involved in a transaction: the description of the market as a calculative device is starting to become clearer. To complete it, we still have to consider the mechanisms through which supply and demand meet each other. These encounters, also calculated (see later), help to make the market a calculative device capable of making (incommensurable) values commensurable. The very notion of a market is highly problematical, for it often implies the existence of an abstract space in which aggregated demands and supplies encounter and cross one another, and through successive adjustments end up defining what is commonly called the market price. In this conception, the abstract market is a structure that frames and formats every individual commercial transaction. It is totally in line with the famous definition proposed by Cournot: "Economists understand by the term market not any particular marketplace in which things are bought and sold, but the whole region in which buyers and sellers are in such free intercourse with one another that the prices of the same goods tend to equality easily and quickly". (ref) If, as many authors have noted, concrete markets are under-analysed, it is probably due to the key position that this abstract conception of the market has long occupied in economic theory. This conception has not only helped to make concrete markets invisible, it has also led to insurmountable logical and theoretical difficulties. The issue over which market theory has stumbled is obviously mechanisms of aggregation of individual and separate supplies and demands. The case of Walrasian equilibrium is a perfect example. The problem is not only that an exceptional set of conditions have to be met in order to produce a Walrasian-like type of environment (see Guerrien). Also, in referring to the famous tâtonnement process according to For a good illustration of this point see the study by Lépinay and Rousseau on amateur online traders. 20 13 which actors discover an equilibrium price, Léon Walras himself will not provide a proper description of the mode of functioning of this mechanism.21 When we talk about abstract conceptions of the market, we mean precisely this: attempts to find a coherent explanatory principle that are based on a shortcut, an ellipsis. All situations in where the market is invoked as a single, abstract principle misses, at some point or at another, the explanatory key that will describe the aggregative effect. Of course, concrete markets can be devised so to meet some of the characteristics of abstract markets: we think about anonymity procedures, clearinghouse procedures, or auction protocols. Garcia (ref.) has provided us with a very interesting example about the construction, under the direct influence of neoclassical theory, of a blind auction protocol for strawberry trading in a French rural market. In financial markets, automated quotation can offer us other interesting examples in this vein (ref. Muniesa). But, besides the fact that such examples are still quite rare, those markets will not be abstract markets in the sense of markets in where the aggregative phenomena is to be “abstracted”.22 Our approach enables us to avoid this dead end. We have chosen as a starting point the market transaction itself, not the macrostructure of a hypothetical abstract market, but what could be called its microstructure (a concept borrowed from financial market specialists).23 With this focus on singular market transactions we do not put aside the phenomena of aggregation and production of possible macroregularities; on the contrary, it facilitates their understanding and analysis, for two reasons at least. First because (see above) the process of qualification of products that transforms them into calculable goods leads to the temporary establishment of differences and similarities. These, by making them substitutable, to varying degrees, for other (objectified and singularized) goods, produce a series of connections and interdependencies between market transactions. The calculation of an individualized good involves necessarily other goods for which it can be substituted. As Chamberlin says: "the whole is a […] network of markets linked together on the basis of one market per seller".24 Competition between calculative agencies, which is a matter of consumers' attachment and detachment, is inscribed into the calculation of products during their process of individualization-differentiation.25 See Teira Serrano for an exploration of this fundamental elusiveness. Walras used the example of Paris Stock Exchange as an intuitive illustration of his tâtonnement process (Ingrao and Israel). But the mechanism that allow actors to find (to calculate) the accurate price, dynamically and without trading at “false prices”, is missing from his theoretical construction. 22 Those markets can be referred to as abstract markets if we understand this expression not in the sense of “abstracted” but in the sense of “mediated”. Simmel has developed a theory of monetary abstraction that fits these second use of the notion (ref. Philosophy of money). 23 The term is widely used in regards to financial markets (ref. R. A. Schwartz, M. O’Hara, A. Madhavan, R. Lee). The role played by technological frames has recently been emphasized through the term market technostructure (see R. A. Schwartz). Microstructure is a convenient notion to designate a set of limited transactions and agents within a particular exchange architecture. It must not be understood as the opposite of macrostructures. 24 H. White used this statement when considering that a firm's consumers were the mirror in which it could observe the behavior of rival firms. 25 Here again we must underline the role played by standards, legal classifications, etc., in the construction of spaces of similarities. 21 14 This organization of competition – and it is on this point that we want to concentrate now – also implies a set of rules and devices whose existence, hidden by the concept of the anonymous abstract market, has progressively been discovered and recognized. This discovery is primarily owing to the practical problems raised by the organization of financial markets and electronic markets, but also to the upsurge of the experimental economics (P. Mirowski). The expression of singular supplies and demands, as well as their encounters – in other words the establishment of a set of connected market transactions –, cannot be reduced to the unproblematical intersection of two aggregated curves. These intersections, or encounters, which concern a large number of separate calculative agencies, must themselves be organized. It is the modalities of this organization, generally not well known, on which we would now like to focus. The idea that the organization of the linking up of supply and demand is an issue in its own right is not, as such, anything new. The classical distinction between auction markets, markets in which prices are the result of pairwise negotiations, and markets in which prices are posted, shows the diversity of forms of organization. This diversity is even greater if we take into consideration the existence of different types of auction and the possibility of hybrid forms (see Klemperer for an overview of auction theory). This diversity obviously raises several questions. Two are relevant here. The first relates to the description and characterization of this diversity; the second to relations between the forms of organization of market transactions and the characteristics of the aggregated market (can one, for example, link certain rules to certain performance criteria such as efficiency, liquidity, transparency, fluidity, etc.). The following two examples, presented in a stylized form, will enable us to clarify these questions a little and to suggest possible answers. First, using an example borrowed from Kirman – that of the Marseille fish market – we show: a) that these microstructures exist (and are complex) and b) that they can produce global and unpredictable regularities. In the case of the Marseille fish market, the aggregated market is competitive whereas the behaviours of individual agents are non-competitive. In short, there is something that can be described as an aggregated market, but its structure, if we stick to this word, is the outcome of the organized composition of a great number of singular market transactions; it is not the frame that defines the rules and format of these transactions. We then provide a few examples of the problem posed by the organization of the calculated encounter between supply and demand, based on studies of financial markets and their microstructures, and particularly the interpretation proposed by F. Muniesa. • As Kirman notes, fish markets are a very old institution that can be found as far back as ancient Rome. They are also a traditional subject of study in economics: "Firstly fish is a perishable good and the fact that stocks cannot be carried over makes the formal analysis of the market simpler. Secondly the organisation of such markets varies from location to location with little obvious reasons. In Iceland for example there are 32 auctions, 18 of these are English, i.e. rising price and 14 are Dutch, i.e. descending price. At Lorient in France fish is sold through a combination of pairwise trading and auction, whilst at Sete it is sold by Dutch auction. The fish market in 15 Sidney is conducted as two simultaneous Dutch auctions" (ref Kirman, p. 4). He adds: "The comparison of different outcomes under different forms of organisation is an obvious research topic but one which has not received much attention to date". (ref ibid, 4).The Marseille fish market is very interesting for several reasons : - First because very detailed data on transactions exist. The data set consists of the details of every individual transaction made over a period of three years. The data is systematically collected and recorded by the Marseille Chamber of Commerce which manages the market. The following information is provided for each transaction: the name of the buyer; the name of the seller; the type of fish; the weight of the batch; the price per kilo at which it was sold; the order of the transaction in the daily sales of the seller. The data run from 1 January 1988 to 30 June 1991. The total number of transactions for which we have data is 237,162. - Next, because its organization is original. On the wholesale fish market for Marseille, situated on the coast at Saumaty on the northern outskirts of the city, and open every day of the year from 2 a.m. to 6 a.m.26, over 500 buyers27 and 45 sellers come together, although they are not all present every day. Over 130 types of fish are traded. Prices are not posted and total stocks are not common knowledge at the beginning of the day. All transactions are pairwise. There is little negotiation and prices can reasonably be regarded as take it or leave it prices given by the seller. The analysis of these data shows that a high proportion of buyers is loyal to their sellers. Moreover, in the same day a seller agrees to different prices for different customers, that sometimes vary widely: "successive prices charged for the same type of fish to different buyers may differ as much as 30%". This distribution is stable from one day to the next and, contrary to what may be expected, during the same day we witness no drop in price. Paradoxically, it seems, it is the most loyal customers who are systematically charged the highest prices. - Finally, because the gap between the real market and the abstract market is maximal. On the fish market, individual behaviors obviously do not correspond to what one expects in a competitive market. Yet the economist's calculation of the aggregated market shows that it follows perfectly the law of demand: when the demand increases prices increase and vice-versa. This example shows two things. First, it shows the complexity of rules which organize the expression of supply and demand and the encounter between the two, and which specify the conditions of the calculation of goods that precedes the transaction. We clearly see from this case that it is not enough to say that products are calculable (and so they are for, as shown by the buyers' loyalty to their sellers, the singularization of goods, despite their apparent standardization, is very high: we are really in the configuration described by Chamberlin). Nor is it enough to say that agents are equipped with calculative tools (and they clearly are because buyers are in turn sellers who have to balance their accounts). To account for the realization of transactions, it is necessary to add – and this is obviously Kirman's contribution – that transactions take place in a very well defined spatial and temporal frame, that they are pairwise, with an undisclosed and un-negotiated set price, a high degree of variability of prices for the same seller and the same fish, strong loyalties, and so on. 26 27 The fish market is now open during the day rather than at night. In fact some 1,400 buyers appear in the records but many of these were hardly present at all. 16 The fact that the transaction takes place at a certain price is the result of all these rules that along with calculable products and calculative agents contribute to constitute the fish market as a powerful calculating socio-technical device. To demonstrate the existence and functioning of this machine, Kirman very cautiously develops a model into which he enters certain rules governing the encounter between supply and demand, certain calculative competencies of agents, and certain simplified forms of qualifications of products (a sole sold by seller X is radically different to one sold by seller Y). This model enables him to obtain some of the valuations produced by the real market (distribution of stable prices over time; higher prices for the most loyal buyers, etc.). This simulation shows that, to explain the prices on which transactions are concluded, it is necessary to consider that the rules organizing encounters between supply and demand participate directly in the calculation of values: in this sense they are an essential component of the calculative device. The second result highlighted by simulation (and in which Kirman was keenly interested) is that the composition of these non-competitive micro-transactions results in the constitution of an aggregated competitive market. This result probably explains why, when the market was to be reorganized, neither buyers nor sellers were in favour of the change. The market organization evidently satisfies the agents, taken individually, since they engage in loyal behaviors over long periods; it also shows a collective efficiency that could hardly be improved on. The important point in this demonstration is that there is no one best way: the efficiency of the abstract market does not imply a single form of competitive organization. Note, also, that the aggregated abstract market is a reality calculated by the economist and not a frame that formats transactions. Transactions are framed and formatted by a set of rules that have nothing to do with the abstract market and serve no purpose in its description. In fact there are more than rules, because a large number of material devices are also required for the organization of these encounters and the effectiveness of the rules. Can we still talk of an aggregated market? Yes, but provided that we do not exclude the mechanisms of aggregation and mise en relation, for they alone explain the possible production of regularities28. The famous law of the market (when demand increases, prices increase) is one of the possible results. And even when it is observable, it does not act like a structural constraint. In particular, we should beware of the dialectic temptation to consider that this regularity, once produced, frames market transactions. What this example shows so remarkably well is that between macro-regularities and particular transactions lie the qualification of products, the calculating agencies, and the rules and material devices organizing the encounter between supply and demand. Any analysis that considers the market without describing all these elements has missed the point. • The second example will help us to advance in our understanding of these mechanisms. What is surprising in the case of such a classical and old market as the fish market, is far less so in the case of more modern ones such as financial markets (For the exemplarity of financial markets see Mac Kenzie). In fact it was for the description and analysis of the functioning of the latter that the tremendously useful notion of “microstructure” was developed. This concept enables us to go from the 28 Sophisticated and complicated behaviors may lead to simple aggregate properties. 17 organization of individual market transactions to possible global regularities or, more exactly, to ask open questions about mechanisms of aggregation that until now were a neglected or underestimated problem.29 How are supply and demand expressed? By what procedures are they connected? What type of dynamic equilibrium is achieved? These questions, by no means new, lead to the well-known theoretical invention of the Walrasian auctioneer, a fictional character responsible for the thankless task of adjusting demand and supply around a market price. This virtual, moral and efficient being has been severely tested by NICT. It has been necessary to explain very precisely the different operations to be performed in order to produce the adjustment of supply and demand and pricing. 30 To denote these rules which are now partly inscribed in computer programs, P. Mirowski has suggested using the notion of an algorithm that clarifies the steps to follow. These algorithms presuppose technical devices, calculative competencies, and identified and stabilized goods. In electronic financial markets, more than in other markets, they must be made visible and formalized. The computerization of real financial markets shows the diversity of possible algorithms and the socio-technical devices they mobilize. It also reveals the difficulty of evaluating the overall effects produced by these rules. We will simply give a few examples here to illustrate some of the multiple choices with which the designers of algorithms are faced (ref: Muniesa). In the early automation of Paris Stock Exchange (automated quotation was introduced in 1986), the actors at work were confronted to a multiplicity of strategies and to wide opened controversies in where regulatory issues, computer concerns and political decisions were consistently blurred. An important choice, for instance, was to opt for either an order-driven market or a quote-driven market (those options were both matters of institutional organization and technology framework). In the former, price determination and allocation between matching counterparts are delegated to the machine, while in the latter it is a flesh-and-blood agent, the market maker, who publishes his own bid and ask prices. In an electronic order-driven market, orders transmitted to an electronic order book are conveniently queued and matched by a double-auction algorithm according to a complex set of priority rules. For opening prices, for instance, the logic program calculates aggregated functions of supply and demand: more precisely, it searches the price that will maximize the overall volume exchanged and minimize the amount of unexecuted orders. For the quote-driven alternative, it is the market makers who, by trial and error (and in competition between them), try to link up supply and demand. The order-driven solution was, in the end, the alternative favored at Paris Stock Exchange. Toronto Stock Exchange’s technology CATS (Computer Assisted Trading System) was adopted, and carefully adapted to the local environment in order to abolish open outcry. Was this technology performing a strict order-driven architecture, a fully blind auction protocol? In fact, and depending on an amount of heterogeneous circumstances, we often deal with hybrid market forms, combining order-driven, quote-driven and OTC (Over the Counter) elements. This was the case Provided that microstructure is not used to designate a separate level of analysis. “Walras himself could not have been so far as us in the details” (our emphasis), said one of the leading actors of exchange automation at Paris Stock Exchange, referring to the complexity of the call auction algorithm (Muniesa 2000, p. 151). 29 30 18 at Paris Stock Exchange with the implementation of a parallel OTC system for block orders, or with the decision of disclosing the identity of potential counterparts in the trading screens (allowing traders to one-to-one negotiation prior to order matching). Other choices illustrate this possible and effective diversity of algorithms. For example in 1994 the abolition of round lots quotation at Paris Stock Exchange, a remnant of the open outcry system, made it possible to absorb individual shareholding that was difficult to take into consideration when trading round lots on the market. The determination of opening and closing prices is also a good example of the variety of algorithmic solutions31. The switch from continuous time quotation to a single-price call auction is a choice that has significant consequences on the setting of closing prices32. The list of options for order formats, order matching, price determination and shares allocation could be extended. This list shows both the existence and the crucial role of these algorithmic solutions. They not only frame the expression of supply and demand but also determine the actual way in which what economists call price discovery operates. This discovery is a complicated business because it consists in taking into account – but how? – a large number of supplies and demands that have to be connected – but how? – to each other. Instead of two aggregated curves crossing each other, we have a wide variety of algorithmic configurations relying on material, technical and organizational devices, and on embodied competencies, all of which produce widely diverse arrangements. Finally these experiences of automation and computerization of financial markets show that there is no straightforward relation between the choice of an algorithm and the resulting performances for the aggregated market.33 Once again, the one best way is invalidated – which is hardly surprising for the sociologist – and along with it the claim that overall performance can be deduced analytically from options concerning the microstructures of transactions. From the fish market and the financial markets examples at least three tentative conclusions can be drawn: • There exists a wide variety of rules and material devices to organize the encounter and aggregation of multiple supplies and demands. These combinations of rules and material devices act, as suggested by P. Mirowski, as algorithms which determine how transaction are connected one to each other and how they can act one upon each other. This is why it is possible to say that this encounter is calculated. The existence of these procedures and their diversity have long been hidden or even denied by the belief in the existence of an abstract market, a structure framing singular market transactions. The recognition of the existence and importance of “An opening algorithm is a communication technology for framing allowable messages between traders, together with a set of rules or algorithms that transforms traders’ signals into a set of opening price and quantity allocations amongst market participants” (ref: Domowitz). 32 This is one solution for avoiding manipulation of prices at closure. But this solution have to confront other alternatives: Paris Stock Exchange was one of the first marketplaces to implement a call auction at close, while other markets opted for taking the average price of last trades to determine closing prices (see Muniesa 2000. 33 To take just one example among many, Ian Domowitz, a specialist in financial markets recognizes that: "there is a tremendous variety of possible opening mechanisms. There is no single opening method that is 'best' for all markets". (ref) 31 19 these rules and devices, that we can agree to call algorithmic configurations,34 prompts us to reject the inaccurate and simplistic view of pricing based on the intersection of two aggregated curves. To be sure, there is intersection, but the calculation of these intersections is obtained with a complex algorithm that organizes the exploration of supplies and demands and governs their encounters and interconnection. The fish market and financial markets illustrate at once the complexity, diversity and irreducible nature of these algorithmic configurations. We can not talk, abstractly, of the Marseille fish market or the Paris Stock Exchange without providing the precise description of these configurations. "The" market does not exist independently of those rules that, in a certain sense, define what the market is. We can see why the analyses of real and of abstract markets have been dissociated from each other for so long; the mechanisms of aggregation and composition were simply ignored or simplified to the extreme. The reader will notice that we use the notion of algorithmic configuration to refer to a wide variety of arrangements that does not always imply computerization. We also apply it to organized market space or, we might say, market architecture. This includes regulatory issues at large, even if such issues are not explicitly translated, at some point or another, into machines (but still always translated into socio-technical devices). Should we temper this use of the notion of algorithm and underline its metaphorical character? We prefer not. First, and this is a lesson learned from the history and the sociology of computers, a computer is not far from behind precisely a matter of organized social space (see Campbell-Kelly and Aspray on Babbage and on Von Neumann, and Collins on machine-like action). But second, and more important, we are interested in emphasizing the logical properties of an algorithm when referring to market organization. An algorithm is a logical program: the emphasis is precisely on coupling the notion of logic (often used to refer to a single operative principle, like when we hear about the logic of the market or the logic of capitalism) with the notion of program (a concrete set of instructions that are contingent to a specific situation and/or task, and that implies a variety of possible solutions). If there is a notion explicitly devised to define the complexity and multiplicity of logical steps needed the achievement of a concrete operation in a concrete situation, this is the notion of algorithm: “The distinguishing feature of an algorithm is that all vagueness must be eliminated; the rules must describe operations that are so simple and so well defined that they can be executed by a machine.” (Knuth 1996: 59). • These algorithmic configurations produce overall effects, more or less stable regularities. But these regularities can never be directly deduced from the observation of the conditions in which singular transactions take place. The effects of composition preclude such deduction. Conversely there is no straightforward way from individual transactions to overall regularities. Put in a nutshell, there is really something that can be described as an aggregated market but this aggregated market does not constitute a constraining frame; it is the outcome of the organized encounter of a large number of individual supplies and demands. In addition, and this point is extremely important, the total explicitation of these rules is probably an impossible task, especially since their functioning and interpretation imply material devices and 34 A more correct term would be socio-technical algorithm. 20 tacit know-how that can not be formalized (see Callon 1998 for a similar argument about framing and overflowing). • From the above we conclude that algorithmic configurations always have a situated character which makes their identical transposition to other places and other markets improbable. Experiments, simulations, modelling Let us recap. A real market can be described (at least partially) as a device for the evaluation of goods which are the objects of transactions. This calculation is possible only if the goods are calculable, by the calculating agencies, and if an (more or less) stabilized and formalized algorithmic configuration organizing the encounter between supply and demand exists. This simple statement shows the diversity of possible forms of organization of markets taken as devices for calculating prices. A good can be made calculable, that is, individualized and objectified, in a thousand different ways. Calculative agencies are as numerous and varied as the tools they use and the hybrid collectives to which those tools belong. Algorithmic configurations are also similarly multiple and diversified. Moreover, relations between these three components make the picture even much more complicated.35 The diversity of forms of organization of markets raises a series of questions. One of them concerns their origins. For example, how can the organization of the Marseille fish market be explained? Why this calculative device and not another? Likewise, how can the organization of the calculation of Paris Stock Exchange, the NASDAQ, etc. be explained? This question is very general and could be applied to any concrete market, for instance the early nineteenth century cement market in the US or car rental in France. This type of exploration is a major undertaking. We will limit ourselves here to a few thoughts intended to highlight a recent trend that complicates usual interpretations in terms of self-organization but makes realist, in the same time, the idea of the malleability of economic worlds. For a growing number of markets there are actors who explicitly raise the question of the practical organization of the calculation of the value of goods. This calculating machine, the concrete market, is the object of more and more research, experimentation and development. Studies such as those of Kirman obviously pave the way for simulations and experiments that could raise the question of other forms of organization for the auction fish market. This possibility, outlined very roughly for the Marseille fish market, is already a reality for other markets. Francesco Guala (ref) provides a very striking illustration of this design and development work, that could be called the social engineering of markets. He relates the history of the auctioning of Hertzian frequencies by the US federal government in the early 1990s. The stated aim of the government (in reality the Federal Communication Commission) was multiple. The auction was to be organized in such For sake of clarity we have neatly distinguished the three components, but there are obvious interrelations. For example the classification of goods is both linked to their calculability and to the organization of the encounter between supplies and demands. But from an analytical point view they can be nevertheless distinguished. 35 21 a way as to allocate frequencies to those companies that would exploit them most efficiently with optimum valorization. They also had to avoid monopoly situations and facilitate the promotion of small companies, rural telephone companies and firms owned by representatives of minority groups or by women. The terms of reference therefore combined considerations and values that were political, economic and social. The main question was: what form of organization should be chosen so that the market reaches a compromise between these different conceptions of the good? The answer proved to be very difficult to find. 36 But the most interesting aspect, for us, is that it was formulated after a real R&D process, in the sense given to the term in Frascatti's manual. To launch reflection on the matter, the federal agency proposed an auction system in two stages (combinatorial auction)37. The firms concerned did not know what to think about that. Two of them, Pacific Bell and Air Touch Communication, decided to recruit three of the best economists specialized in game theory (including Milgrom and Wilson) and asked them to evaluate the proposal and possibly to make counter-proposals. The economists soon gave their verdict: no theoretical evaluation was possible; the problem was too complicated. They simply – although it can be considered as a lot - provided examples of the functioning of the algorithm that gave inefficient allocations. Then, based on various theoretical intuitions, they proposed a different organization, that of simultaneous ascending bid auction. But there again, it turned out that the theoretical demonstration of the effects produced was proved impossible. To compare the different systems, it was decided to launch laboratory tests. Designing the experiments raised a whole series of questions that the economists had not considered in their theoretical models. For example, how could the non-circulation of certain information (concerning the number of players, decisions to withdraw, etc.) be guaranteed? What rate of incrementation should be chosen for the auctions? How many rounds should be organized to allow the revision of strategies without favouring some participants? What time lapse should be observed between rounds? What conditions of eligibility should be set for participation? What terms should govern withdrawal? How should rules be formulated and transmitted? These options, among others, concerned rules as such but also implied the establishment of a set of practical arrangements and forms of material organization. From abstract theoretical reflection (game theory) the economists moved on to real experimental laboratory research in which theory and practice, deduction, induction and bricolage were combined. Like in any R&D process, once certain solutions and devices had been tested, the problem of scaling up and development arose. The lab research was followed by a first scale one test, monitored by observers. After further adjustments, the real auction was launched. Its organization was thus the outcome of a genuine R&D process. Mainly because it concerned a market of interdependent goods. Wavelengths cover territories and have to be juxtaposed to allow long-distance communications. 37 In which goods are initially auctioned in packages, with a sealed-bid mechanism, and then on an individual basis. 36 22 In the final analysis, we can say that "the" wavelength allocation market was designed in a laboratory and developed through a series of experiments performed on different scales. In these experiments the different components and performances of this particular calculative device were tested and progressively evaluated. Economics and economy have probably never been so directly related to each other. The term market-maker would be apt for this social engineering which, like any elaboration of machinery, is the result of pluri-disciplinary cooperation. This automatically brings to mind the birth of thermodynamics and everything it owes to the observation and mastery of steam engines. The economy of markets is (at last!) entering its scientific, because experimental, phase. It is focusing on the functioning and organization of real markets and becoming aware of the fact that no simple analytical treatment can get through its complexity. It may eventually be possible to design and to operate purified calculative devices, close to abstract models. However, there is still a long way to go before reaching the point at which abstract and concrete markets are adjusted to each other, like machines and their theory, thermodynamics38. In our view this evolution, which is taking economics from formal theory of abstract markets to the experimental science of real markets conceived as complex devices for calculating the value of goods, and which simultaneously causes the organization of these markets to depend on laboratory work undertaken under the leadership of economists and their colleagues, informaticians or psychosociologists, is concerning a growing number of markets. As we have seen, this is the case of financial markets in which the conception of microstructures is already generating a huge volume of experimental research. It is also the case of e-markets and may perhaps be the case one day of the fish markets for which Kirman proposed models that could lead to real research programmes to further our understanding of their functioning All these experiments should contribute to a real economic engineering. After all, the now famous strawberry market described by Garcia is the result of a real social engineering process, performed under the responsibility of an economist (ref). Naturally, in these R&D studies it is the algorithms that link up supply and demand on which most of the effort is concentrated, probably because that is where the biggest lack of knowledge is to be found. In this lab research and the consequent developments, hypotheses are made on agents' calculative competencies and on the calculability of goods. But they do not yet generate the same theoretical and experimental investments even if interesting assumptions are made on the distribution of competencies between agents and rules, especially by evolutionary economics: as, for example, calculative competencies are reduced to a minimum, the idea being to find the most efficient algorithmic configurations that could function even with "stupid" agents). But such simplifications on the competencies of calculative agents are often possible only because the products are themselves highly stabilized and qualified in advance. If we were to simulate a market for which the calculability of products was of prime importance, such simplifications would For an interesting discussion of the epistemological questions raised by this new form of economics see Guala. 38 23 probably be counter-productive39. We could, moreover, imagine that in such cases ongoing observation of the functioning of scale 1 markets might constitute the best research strategy, especially since these non-stabilized markets, with changing calculative agencies, make intensive use of communication technologies, thus facilitating observations and calculations on those observations. In short, we need to plan experiments performed directly on scale 1. This opens a vast field for research. It may well be that other social sciences like sociology and anthropology are called on to carry through such experiments. Conclusion We started this paper with the now widely accepted hypothesis that markets are organized, that their forms of organization vary, and that none of them can be considered to correspond to the one best way. This hypothesis seems particularly productive when we conceive of markets as devices for calculating the value of goods, and for finding out, at the same time, acceptable compromises between separate worlds and values that are uncertain, fuzzy and contradictory. To sum up: economic markets are not frames or (macro)structures. The abstract model of markets and its spatial metaphor, as epitomized by Cournot definition, are both inadequate to describe how goods are evaluated. Instead we have showed how fruitful could be the concrete organized markets approach. We have thus suggested that this markets could be described as consisting of three interrelated elements: for it to be possible to calculate the value of goods and to set prices which lead to an agreement, the products have to be calculable, calculative agencies have to calculate them, and algorithmic configurations have to organize the expression and encounter of supply and demand. Apart from highlighting these three elements, which enable us to shed some light on the diversity of organizational forms of real markets, we have emphasized the importance of research and development in the creation and testing of these calculating devices. Actors can thus conceive of a better adjustment between real markets and the goals they intend to give to them. We can imagine that not only the social sciences and computer sciences but also other stakeholders could become involved in these R&D and design activities. Consequently, everything that is said on technical democracy, on public debate and on the involvement of concerned groups in technical issues, applies to economics as soon as it becomes a real experimental science based on laboratory research (Ref Callon, Lascoumes and Barthe). A final objection must be at least briefly discussed. Behind the variety of forms of calculation, which is the main result of our analysis, is there not the same logics, that of calculation? Economic markets could be said hegemonic because they impose more and more widely and forcefully the same law: you shall calculate! In a certain sense this is correct. But one of the advantage of our approach is: first to clarify the fact that there are many ways for calculating values and for reaching compromises; Here we naturally think of the lab work that would be required to clarify the effects produced by the granting of a particular type of property right on genetic sequences (EST), software or cultural products. It would be the only possibility for getting out of the current deadlock in which specialists cannot agree because they confine themselves to abstract reasoning. We would then talk not of applied economics but of economics as an experimental science. 39 24 and second, and more important, to suggest that open discussions and even public debates are possible on how to make and organize these calculations and even on how to refuse that certain values be calculated. More generally speaking, calculation, as cogently demonstrated by Porter and Barry (ref: ), do not mean depolitizing. One way for making this point, in the case of economic markets, is to consider what economic theory calls market failures. One can imagine failures related to the impossibility of performing calculations or to the unsatisfactory nature of the calculations. One can, for example, try to understand why this calculating machine sometimes breaks down, so that the calculation cannot be done. Algorithms can also be criticized for not taking into account certain calculative agencies that would like to participate in the calculation. The conditions of calculability of certain products (and hence their qualification) could be criticized and redefined, for instance by demanding and imposing that products take into account environmental effects or children's rights. Finally certain goods could be made uncalculable, by making them public (as said in economic theory), that is, common and non-individualized. At the end, instead of abstract (global) markets that impose their (in)human logic which is possibly counterbalanced by local relations and identities, we have markets that can be organized in different ways, especially even when the question is how to find out a compromises between different conceptions (of the value) of goods. The qualification of goods Seller G Buyer sociotechnical world of the buyer attachment individualisation entanglement PROCESS OF QUALIFICATION 25 FIGURE 1