sign in sign up
<<
Home , Industrial organization

INDUSTRIAL ORGANIZATION: THE ROLE AND CONTRIBUTION OF * By HENRY GRABOWSKI, Yale University and DENNIS MUELLER, Cornell University

Industrial organization traditionally has been growth and efficiency grounds and the like. one of the more empirically-oriented branches of While the field therefore has an empirical focus, . Many of the hypotheses peculiar to it research on these questions also requires a strong have been inductively derived (e.g., the Berle and underlying theoretical structure if results applica- Means study [1]). Its literature contains a heavy ble for policy are to be meaningfully evaluated. dosage of case study and descriptive statistical In its limiting extremes, traditional microeconomic material. Questions of the proper measurement of theory offers a basis for formulating hypotheses various phenomena, the quality of data, and in- on such issues. Certainly, one of the most power- dex number problems also occupy a prominent ful input- generators of normative results place in the literature. in all economic inquiry is the microeconomist's While the field has been somewhat slow to take model of perfect . With this model at up the use of econometric techniques in its em- the core and a good deal of pirical work, recent years have witnessed the ap- analysis available for other structures, the pearance of a large number of studies on concen- empiricist can draw on a considerable body of re- tration, innovation, advertising, mergers, and sults in formulating his speculative hypotheses. other industrial organization subjects that have Unfortunately, as one begins to pull at the employed mutliple regression analysis and similar various assumptions underlying the perfectly econometric tools. This development is encourag- competitive model, one finds that it is possible to ing and suggests that the IO may some rationalize almost any kind of empirical relation. day be able to derive more general hypotheses and Consider the range of solutions that are normally theories than have been possible up to this point. possible under an oligopolistic . Unfortunately trends in mergers and firm diversi- Suppose, for example, one is interested in investi- fication [2] [5] during the post-World War II gating how affects a firm's allocations period have created (or exposed) deficiencies both for activities such as advertising and R and D. It in the stock of theories that can be tested and in is possible on a priori theoretical grounds to have the data available for testing them. Indeed, these a whole range of contrasting solutions including trends have been so broad and strong that they the aggressive behavior of an extreme rivalry situ- threaten to vitiate future efforts to apply econ- ation, the passive behavior of the Cournot solu- ometric techniques to firm and industry data and tion, and the collusive behavior of a cooperative force the IO empiricist back to a case-to-case arrangement. Moreover, the empirical conditions approach-with the unit of investigation being under which one might expect to find any of these the large, diversified firm. Hence, we stand in the solutions are very nebulously formulated and danger of seeing the period of infancy in the ap- tend to resist quantification. They consist of fac- plication of econometrices to industrial organiza- tors such as the amount of the pres- tion coincide with its zenith, unless we are able to ent, the goals of the firms, their attitude toward develop better theories and/or come up with risk, and the historical pattern of the market's better data than are presently available. The re- development. It is, therefore, difficult to predict mainder of this paper is devoted to a discussion of in advance what might happen here solely on the these two problems. basis of economic theory. These problems have been compounded by the I. Problems Associated with a Lack of Theory rise in importance of the large diversified firm. Industrial organization is intimately concerned Microeconomic theory is basically a theory of the with matters of , ultimately at- one-product firm. It has little to offer us by way tempting to provide insights into questions such of testable hypotheses of how multi-industry as the optimal forms of market structure on firms behave. Nor does it provide us with a well- accepted framework for analyzing the behavior of * The authors are indebted to the National Science Foundation for financial support under grants GS-2677 divisions of diversified companies. Yet the only and GS-2678 which greatly facilitated completion of sources of data we have come in these forms. this paper. Either we employ the census data for industries, 100 INDUSTRIAL ORGANIZATION 101 which are aggregates of the figures for single Yet anyone who has attempted to deal empirically product firms, the major divisions of diversified with dynamic behavior under uncertainty will firms based in the industry, and minor divisions find some common bonds with this approach. For of firms based outside of the industry, or we resort example, the stock reaction model prevalent in in- to data on firms and accept the aggregation over vestment studies is at base an error learning model the many diverse products the typical corporation with close ties to the satisficer's rule of thumb de- now produces. Indeed, there is a basic dilemma in- cision making. Indeed, because traditional theory herent in tailoring our current theories to accom- is basically static in nature, mainly for reasons of modate these existing data sources; that is, any analytical tractability, investigation of dynamic simplifying assumption we might choose to make behavior and lag structures must proceed very to allow us to use one body of data, are likely to much on an inductive trial and error basis. destroy the usefulness of the other. For a variety of reasons, therefore, one seldom For example, suppose we decide to assume that has much beyond a few speculative presumptions the diversified firm acts as a monolithic when relating existing theory to the policy- maximizing institution. This would imply that oriented issues of industrial organization. What each of its subdivisions would act in a manner con- are the consequences of this lack of a general sistent with the maximization of profit for the theoretical structure in our hypothesis formula- whole enterprise. This might lead it to buy from tions? There would seem to be at least two basic other divisions of the parent firm, or from outside concerns. First, the lack of theory makes it very suppliers as part of a reciprocity strategy, at difficult to discriminate and select among the which do not maximize the division's profit. various hypotheses relevant to a particular case. Similarly it might undertake R and D and adver- A theory serves the useful function of implying tising which are only profitable because of their certain events and ruling out others as inconsis- strong complementarity with the activities of tent with its basic assumptions. Testing the con- other divisions. Whatever the merits of this as- sistency or inconsistency of a theory over a variety sumption with regard to the diversified firm, its of empirical relations, then, allows one to make use is obviously inconsistent with the assumption informed judgments on its general acceptability. that the units of observation maximize profit If one observes a particular set of relations with- when data by industries as gathered by the Cen- out a well-defined theoretical framework, how- sus of are employed. ever, it is difficult to discern subtle contradictions The sheer growth in the size of corporations either internally or with past results and it thus has produced bureaucratic structures within the becomes difficult to reject hypotheses which are firm that make traditional models of entrepreneur- not consistent with the data. Undoubtedly, this ially led or stockholder controlled firms seem is one reason why opposing hypotheses tend to frightfully obsolete. Economics and some of its linger on with such vigor in the industrial organi- sister disciplines have responded admirably to this zation literature. challenge, generating a large number of - A second problem, related to the one discussed ing behavioral hypotheses based on managerial above, concerns the stability over time and across motives, organizational theory, etc. However, various cross-sections of any relationships ob- this proliferation of competing hypotheses only served in this manner. If, for example, one ob- makes the empirically-oriented economist's job serves a particular relation between, say, a market that much more difficult by expanding the range structure variable and a measure of industrial of plausible alternatives from which he must efficiency which ultimately rests on a more com- choose. plex set of unknown and unobserved relations, Most of these "managerial" models have re- then the observed relation is subject to unexpected tained the central postulate of economic rational- changes even if the basic structure remains in- ity embodied in the maximization assumption and variant. Unless one has a comprehensive theoret- have merely attempted to recast it to new situ- ical framework for analyzing a particular struc- ations. Others have suggested, however, that a ture, the generalization of results beyond the "satisficing" and rule of thumb decision-making limiting confines of one's sample is not warranted. approach is more fruitful for explaining manager- Nevertheless, as Professor Mason reminds us, ial behavior in an uncertain environment. The "inoone who is other than eclectic, methodologi- implications of this motivational postulate for cally speaking, has any business in the field of in- the empiricist are to fractionalize his world view dustrial organization" [4]. Decisions of public completely and make his research efforts wholly policy resting on economic considerations cannot inductive in nature. For this reason, most econ- be postponed while our theories are being refined. omists have chosen not to adopt this approach. In many policy areas, research therefore must 102 AMERICAN ECONOMIC ASSOCIATION necessarily go forth as best it can on current met. The results that emerge from all of this are analytical structures. Research of this nature rightly regarded as only tentative and at best in- must frequently be inductive in spirit and con- dicative of the potential for further, more inten- sider as many alternatives as possible over what- sive research on the hypothesis. Unfortunately, ever data samples are available. In reporting re- such research is seldom undertaken, since data of sults, the researcher in these areas should also set higher quality than were used in the original study forth his basic research design and in particular rarely appear. indicate how other approaches compare with the The industrial organization economist probably main findings. Too often one sees only the "tip of has been no more guilty of committing these sins an iceberg" of this type of emprical analysis, pre- than his colleagues in other branches of economics. ceded by an elegant but artificial attempt to make However, high levels of merger activity in recent the study appear deductive in nature. However, years and the trend toward conglomerate organi- in such circumstances, it is much more instructive zational entities have rapidly increased the to know, at least in a general way, what is the magnitude of data problems in this area. explanatory power of other variables and func- Mergers can effectively destroy data in both tional specifications. Comparison of results with cross-section and time series studies. In time more "naive" models is also highly desirable and series work a merger may change the characteris- frequently ignored in these circumstances. istics of the firm sufficiently to preclude entirely At the same time, empirical research in IO will the combining of the firm's data for before and most certainly benefit in the future as it has in the after the merger. At a minimum it forces the past from continued research efforts on a number analyst to give up one or more degrees of freedom of analytical frontiers. Given the significant cur- by deleting an observation or adding dummy rent trends toward large size and diversification variables to the model. Mergers can create similar in our , a high priority certainly exists for problems when cross-section data are used. If theoretical studies investigating the firm in a lagged variables are employed, the merging firm more disaggregate fashion and ones delving into must be dropped from the sample in the year of how the pieces of such organizations fit together the merger. Since the number of firms in the sam- to form rational decision-making entities. It ple and their identities will thus vary over time, would seem that only by gaining a greater knowl- the researcher's ability to draw inferences by com- edge of what is happening at the very "micro" paring the results for homogeneous cross-sections level can we begin to make any real progress in is reduced or eliminated. Even when no lagged understanding relations at the market structure variables are employed, the potential for com- level. While this kind of disaggregate approach is paring cross-sections over time may be impaired likely to generate theoretical models involving by the changing composition of the sample as a very complex sets of relations, simultaneous result of mergers. equation techniques are becoming increasingly The increasing extent of firm diversification available to help facilitate the use of these models across product and industry lines also creates in econometric applications. On the other hand, serious data problems. The implicit assumption obtaining data at the intrafirm level to the degree in cross-section studies is that all of the firms in required by such models is a serious problem as the sample follow similar behavioral patterns. data sources now stand and one which will be in- This assumption seems most plausible if all of the tensively considered below. firms produce a single well-defined product and least appealing if they each manufacture a differ- II. Problems Associated with the Lack of Data ent array of products only a few of which may be Data problems are both universally acknowl- produced by the other firms with which they are edged and ignored by . The individual grouped. scholar is at the mercy of the private firm or The empiricist who employs Census of Manu- government bureau for most of his data, for he facturing data on industries faces problems dif- possesses neither the financial nor in some cases ferent in kind but similar in nature to those cited legal power to collect them himself. Even if he is above. For example, in addition to the usual ag- astute enough to choose the best data available gregation problems he now must contend with for his particular model, they are frequently in- the fact that his data are aggregations of whole adequate for the task. Rather than abandon his firms, divisions of firms, and parts of divisions of model to the limbo of untested hypotheses, the firms. If cross-section regressions are estimated, economist typically "adjusts" the data and trun- he must contend with the fact that parts of a sin- cates the model to the point where an uncom- gle firm may be included in a number of industries fortable compromise between theory and data is and hence that the observations are not strictly INDUSTRIAL ORGANIZATION 103 independent. And so it goes. These are problems Indeed, for reasons of keeping the data confiden- with which we are all familiar, and yet all ignore. tial, he usually can identify the characteristics of Many researchers have tried to avoid some of his sample only by phrases like "ten large chemi- these problems by simply dropping the trouble- cal firms" or the like. some observations (firms or industries) from their The aggregate character of data, even at the sample. While this may have been a reasonable corporate level, will provide obstacles in such a procedure as long as these phenomena occurred study. For example, the size variable normally with some rarity, this no longer is the case. At a available will be the firm sales or assets across all minimum this procedure is now very costly in the industries in which it produces rather than the terms of lost degrees of freedom. More impor- corresponding measure in its major industrial tantly, the remaining samples of nonmerging and activity. Hence, unless diversification by firms uindiversifiedfirms or industries with only single- into more or less research intensive industries ac- product firms can no longer be assumed to be in tivities is negligible, or if not, at least independent any sense randomly generated. Indeed, they are of size, this hybrid measure will influence the rather peculiar groups of enterprises that seem in- Schumpeterian relations estimated for each in- tent upon swimming against current economic dustry grouping. Where diversification is sub- trends. stantial, the resulting estimate may bear little relation to the true structural relation pertaining III. Some Examples to a particular kind of industrial activity. The extent to which these deficiencies in data To help illustrate these points let us take a combine to frustrate the attempts to test hy- closer look at the firms in the chemical industry. potheses is probably best illustrated by relating Moody's for 1969 lists 183 firms which produce them to an area in which the authors have par- chemicals. Of these 183 firms, 101 can be classified ticipated: the recent proliferation of studies test- as based outside of the industry, or in the case of ing the Schumpeterian hypothesis. The central the conglomerates, as having no real base industry question being investigated here is whether the at all. These range in scope from companies whose level of innovative activity tends to increase bases are in related chemical processing industries more or less proportionately than size in a partic- such as petroleum refining, rubber products, and ular industrv or line of industrial activity. Micro- drugs to companies whose base industry involves economic theory provides conflicting strands of a distinctly nonchemical technology such as elec- thought concerning the expected relationship tronics, alarm systems, shoe machinery, and here and, therefore, an empirical inquiry into the uranium mining. Obviously with over half of the structural relations across various industry sam- firms producing chemicals being based outside of ples and time periods exemplifies the inductive the industry, any assumptions regarding be- approach suggested for such situations above. havioral homogeneity for all chemical producing The most desirable measure of a firm's inputs firms are quite hazardous. Even among the 82 to innovational activity for such a study is data firms that are based in the chemical industry there regarding the level and character of R and D ex- is a great range in the products and constellations penditures, standardized by some uniform defini- of products they manufacture. Park Chemical, tional criteria. The National Science Foundation for example, is a fairly specialized manufacturer has been collecting such data since 1957 for a very of metallurgical and automotive chemicals. Union large sample of firms but they are able to provide Carbide also produces metallurgical and automo- information on it only aggregated over broad in- tive chemicals. In addition, however, it produces dustrial classifications, roughly comparable to a whole line of chemicals that includes agricultural the SIC 2-digit level categories. These reported pesticides, alcoholates, fluorocarbon propellants data are therefore not very useful in applications and refrigerants, fumigants, latexes for paints, at the microeconomic level indicated above. Many lubricants, plasticizers, silicones, oils, resins, ure- firms publicly report their annual R and D ex- thane foams, etc. Moreover, the 1966 Fortune penditures, but usually say little concerning the Plant and Product Directory indicates that it pro- nature of such expenses or how their definitions duces products in no less than 22 different 3-digit relate to the NSF criteria. The researcher in this industries outside of chemicals, and it has divi- area must therefore work through an array of sions manufacturing carbon products, consumer questionable figures often surrounded by a strong products, electronics, foods, gases and equipment, sense of firm security and secrecy. Even if he is metals and minerals, nuclear products, plastics, successful in obtaining R and D data privately and textile materials. Despite the chemical base from firms, he will have problems getting a com- they share in common there is probably a greater plete or representative sample in any industry. degree of homogeneity in the product mix of 104 AMERICAN ECONOMIC ASSOCIATION

Union Carbide and some of the large diversified particular importance here are data on firm op- petroleum and acknowledged conglomerate com- erations on a fairly disaggregated basis. In seeking panies than there is in the product lines of Union these data we have a potentially powerful ally in Carbide and Park Chemical. the stockholder. Both the stockholder and the Nor are things much better if one limits one's economist have exhibited a heightened interest sample to the dozen or so largest firms. While each of late in finding out precisely what is taking place large chemical company produces any single prod- within the sprawling diversified corporation. uct in common with a number of the other large For this reason we would like to join in support- chemical firms, the array and rankings of firms ing the suggestion that a large data bank be estab- differs from product to product. Hence, it is diffi- lished on a national level. In an age where great cult to speak of a single leader for the chemical technological advances have been made in the industry although DuPont is by far the largest. data communication and processing field, one Nor is it possible to predict which firms are most could certainly devise a system whereby such likely to introduce innovations in a specific area data could be made available for analysis, while or what and from whom reactions can be ex- at the same time respecting their confidential pected. nature at the individual firm level. A computer Given these facts, it would appear very hazard- system hookup from a national data bank to ous to set forth an estimated structural relation various academic centers which would involve on the Schumpeterian question based on a regres- central processing and the communication of sion over aggregate corporate figures for the larg- results without any sight of certain confidential est dozen or so chemically-based firms. As shown data inputs is an obvious development which is above, such a procedure ignores the significant now feasible and highly desirable if the rational contributions of nonchemically based firms and development of public policies is to be encouraged. includes the nonchemical production of those Barring the creation of a supply of data of this based in the industry.' Indeed, the pointed dis- proportion, econometric work in industrial organi- cussions and debates over the validity of observed zation can be expected to move forward rather results which have turned on questions of regres- slowly. Indeed, a return in emphasis to the case sion specifications and proper deflating procedures study approach might even be in order-with the and the like (for a survey here, see [3 ]) may very subject of each case study being a single large well be academic, given the weaknesses of the diversified firm. These organizations each possess data bases common to all these studies. a sufficient amount of economic importance to warrant such special treatment. And with the IV. Some Recommendations analytical tools and information currently avail- By focusing attention on current trends in con- able to us only a very thorough analysis of each centration, the publication of Berle and Means's giant firm can be expected to yield convincing classic study in 1932 [1] helped launch systematic conclusions. Certainly, the days when one can efforts to gather data on industry and aggregate toss the total sales, profit, capital , concentration. In this we are very fortunate. The etc., figures for G.M., 3M, Sperry Rand, and the important questions concerning the absolute other diversified giants into the electronic hopper levels and trends in industrial concentration are and view the results with any confidence are among the best documented issues in the field. numbered. If similar progress is to be made on the many REFFERENCES other hypotheses that warrant testing, then data A. BERLEAND G. C. MEANS,The Modern Corpora- them must be collected 1. A. appropriate for testing tion and (New York, 1932). and made available to researchers in the field. Of 2. H. F. HOUGHTON,statement in Economic Concentra- tion, Hearings before the Subcommittee on Antitrust 1 inter- Some indication of the extent of intra- and and of the Committee on the Judiciary in a of large industry coverage to be expected sample (Washington, 1964), pp. 156-78. firms is provided from Bureau of chemically-based MARKHAM,"Market Structure, Business figures for 1963 [6]. For ex- 3. J. W. Census 1965, ample, the percent of the of shipments accounted Conduct and Innovation," A.E.R., May, for by the eight largest firms whose primary enterprise pp. 323-32. category code is 28A (basic chemicals, plastics, synthet- 4. E. S. MASON,Economic Concentration and the Mo- ics) is only 36 percent in organic chemicals, 25 percent nopoly Problem (Cambridge, Mass., 1957). in inorganic chemicals, and 30 percent in plastics [6, 5. W. F. MUELLER,statement in Economic Concentra- that p. 289]. At the same time, this source indicates tion, op. cit., pp. 502-18. in 28A ap- for the 16 largest firms enterprise category, 6. U. S. Bureau of the Census, Concentration Ratios in proximately 30 percent of value added and 40 percent Industry, 1963, Part II (Washington, of total is in a different enterprise cate- Manufacturing gory [6, p. 310]. 1966).