Navigation – Plan du site

AccueilNuméros13-3The Computerization of Economics....Computer Operators and Software E...

The Computerization of Economics. Computers, Programming, and the Internet in the History of Economics

Computer Operators and Software Engineers at Data Resources Inc.

Opérateurs informatiques et ingénieurs logiciel chez Data Resources Inc.
Pedro G. Duarte et Francesco Sergi
p. 723-768

Résumés

Cet article présente le rôle central des ordinateurs et des logiciels pour le succès de Data Resources Inc. (DRI), une des principales entreprises d’économétrie aux Etats Unis pendant les années 1970 et 1980. DRI a été fondée en 1968 et a été dirigée jusqu’en 1984 par Otto E. Eckstein, un économiste d’Harvard. En s’appuyant sur ses réseaux académiques et au sein des administrations publiques, Eckstein réussit à construire pour DRI un modèle macroéconométrique à large échelle, qui produisait des prévisions économiques vendues aux entreprises et aux administrations publiques. Dans cet article, nous mettons en évidence la manière dont DRI avait organisé ses structures informatiques, en support à la modélisation macroéconométrique aussi bien qu’à d’autres de ses nombreuses activités commerciales en relation avec l’analyse économique et l’économétrie. Nous défendons l’idée que DRI constitue un cas d’étude essentiel dans l’étude de l’histoire de la modélisation macroéconométrique à large échelle, précisément en raison de son utilisation singulière des outils informatiques, en particulier de la technologie du time-sharing, et de sa capacité à développer des nouveaux logiciels pour les économistes. Ces aspects des pratiques de DRI liées à l’informatique sont documentés grâce à des entretiens semi-directifs avec Peter White (un ancien opérateur informatique et ingénieur logiciel de DRI), ainsi que d’économistes qui ont été impliqués directement ou indirectement dans les activités de DRI.

Haut de page

Texte intégral

  • 1 The literature on the history of macroeconometric models in policymaking institutions includes: for (...)
  • 2 For a broader perspective on the relationships between economics and business until the interwar pe (...)

1Macroeconometric models went for profit in the 1960s. After having been developed since the pioneering efforts of Jan Tinbergen (1936; 1937; 1939), in the 1960s such models connected academia, governmental agencies in different countries (mainly central banks), and now also consulting firms.1 In Massachusetts, Data Resources Inc. (DRI hereafter) was established in 1968 by Harvard macroeconomist Otto E. Eckstein and the financier Donald B. Marron, and it started operating in 1969. But DRI was not alone in the hype of what Smith (1994) called the “market for macroeconometric models.” In Philadelphia, the Economics Research Unit was created within the economics department of the University of Pennsylvania in 1961 and started distributing the Wharton Quarterly Model. But it was in 1969 that the Wharton Econometric Forecasting Associates, Inc. (WEFA) was incorporated by the Trustees of the University of Pennsylvania as a “not for profit” organization, and was led by Lawrence Klein (PhD MIT 1944). In the same year, Kleins colleague at the Wharton School and member of WEFAs Board of Directors, Michael K. Evans (PhD Brown 1964), broke up with WEFA and struck a deal with Chase Manhattan Bank to create Chase Econometrics Forecasting Associates. These companies had corporate and government clients and became the three main actors in a thriving market in the 1970s.2

2This article draws attention of historians of economics to the role played by DRI in this industry and in broadly disseminating macroeconomic expertise and, most particularly, expertise based on large-scale macroeconometric modeling. Moreover, macroeconometric modeling within corporations such as DRI was far from being simply an “application” of macroeconometrics as developed in academia or in policymaking institutions; it constituted, rather, a frontier for technical innovation in handling macroeconometric models and in spreading their use to businesses and government institutions.

3Recent contributions to the history of macroeconomics have shifted the focus of investigation from theoretical debates within academia (as depicted notably in De Vroey, 2016) to the practice of macroeconomic modeling, particularly in policymaking institutions (see e.g. Boumans and Duarte, 2019). This shift in focus has called attention to the “practice” of macroeconomics, to the interaction between economists and policymakers, and to the importance of policy routines (forecasting, economic scenarios, etc.) in determining modeling choices. In this literature, the materiality of practices (funding, availability of data, policy demands, etc.) is central: this also points to, among other factors, the importance of computers and software for macroeconomics. Indeed, computer tools have been often highlighted, though rarely studied in detail, by historians of macroeconomics as creating both constraints and opportunities for the discipline (computational capacities, costs, possibility of simulation, handling more/different data, complexity/size of models, tractability issues, and so on). This also results from the lack of sources on computer-related practices, since macroeconomists often were, in their publications, quite elusive about computer-related details.

4In this article, we investigate such computer-related aspects of the practice of large-scale macroeconometric modeling. This choice is motivated by the fact that computer-related specificities constitute one of the most distinctive aspects of the DRI business model. We document how DRI’s success relied indeed on the development and on the commercialization of a broad economic database, of computing capacities with time-sharing technology, and of software facilitating econometric work for DRI customers. The case of DRI illustrates how, within the industry of private consulting, the use of computers and software played a distinctive, driving role for the development and the dissemination, outside of academia, of macroeconomic knowledge and modeling practices.

5Historicizing the modeling practices of private companies such as DRI, as well as the computer-related aspects of such practices, constitutes an important challenge. Published sources on such practices are scarce or non-existent—though the model structure and technical discussions on solving and estimating them were published in academic journals (or as PhD theses) by those economists developing them. Moreover, the records of private companies, especially those who ceased to exist (such as DRI), are rarely accessible, when they survive at all. Even first-hand experience of those who worked for private companies such as DRI are difficult to collect, since these protagonists are mostly “hidden figures” of history—that is, they leave little trace in terms of traditional sources for historians (such as publications and archives), and are therefore difficult to locate and to contact. The only possibility is to document those practices and computational details while the actors and the companies were still around. This is what Kenneth Kraemer, Sigfried Dickhoven, Susan Fallows Tierney, and John Leslie King did in their 1987 book, Datawars: The Politics of Modeling in Federal Policymaking (Kraemer et al., 1987). They were able to interview several DRI people, consult documents, and interview also customers to offer a rich analysis of how DRI and its services went into the US federal government and helped institutionalizing modeling as a policymaking tool.

6Unexpectedly, in April 2021, when Œconomia – History/Philosophy/Methodology circulated its call for papers “The Computerization of Economics,” the editors of the special issue received a message from Peter White, who introduced himself as follows:

I worked at Data Resources Inc. in Lexington, MA, USA from 1972 to 1979. I was a software engineer, not an economist, though. And in 1972, I was a very junior one. By 1979 I was responsible for most of the infrastructure code in the EPS (Econometric Programming System), the modeling Program that DRI produced, used, and licensed. (Peter White, personal communication, 15/04/22)

  • 3 See the interviews list at the end of this article. Unfortunately, it was not possible to conduct a (...)

7Peter Whites experience at DRI proved of great interest, insofar as it helped us document, based on his unique first-hand account, the role of computers and software at DRI during the 1970s. We complemented our correspondence and interview with White with other semi-structured interviews with other people involved, directly or indirectly, with DRI’s activities.3 We combined these oral sources with published materials from DRI, a few relevant archives, newspaper sources, and a few written recollections. But it was the use of oral history that allowed us to flesh out the concrete practice of using computers in macroeconomic modeling at DRI during the 1970s. Thanks to oral history, it became possible to “take a walk” through DRI’s computer room and to observe the functioning of the computer infrastructure, as well as the division of labor and the interactions between macroeconomists and other DRI professionals. At the end of this “tour,” we shall better understand what kind of constraints and opportunities were created by computers and software for macroeconomics during this period.

8We start our analysis with a presentation of the origins of DRI and contextualize its activities within the development of macroeconomics in the 1970s and the emergence of the industry of macroeconometric modeling (Section 1). In the second section, we detail the functioning of computer infrastructure at DRI and how this supported economists (within and outside DRI) in their analytical work. The third section presents the development of the Econometric Programming System (EPS) by DRI and locates the design and functions of EPS with respect to other software of the time. In the fourth section, we briefly present the difficulties encountered by DRI in the early 1980s, after it was bought by McGraw-Hill. Finally, we will present, in our concluding remarks, some general lessons for the history of macroeconomics that can be drawn from the history of DRI.

1. DRI and the Thriving Industry of Macroeconometrics

1.1 Wall Street, the Council of Economic Advisors, and Harvard: The Origins of DRI

  • 4 Kraemer et al. (1987, 70-72); Davidson (1969); “Preliminary Prospectus, Data Resources Inc. Common (...)
  • 5 During his life, Marron would have prominent executive roles well-beyond DRI, as for instance the c (...)

9DRI was incorporated in Delaware in December 1968 and was open to customers in the Fall of 1969. It was an outgrowth of a consulting program that was initiated by Donald Marron’s New York institutional brokerage house, Mitchell, Hutchins & Co.4 Marron was a financier, who had built a career in Wall Street since the early 1950s.5 In the mid-1960s, he had the idea of creating a “consulting program” that would provide portfolio managers with useful insights into macroeconomic conditions: “I therefore put the following question to a number of my colleagues on Wall Street: ‘Who is the best young economist you can think of to fill this role?’ The most frequent reply was ‘Otto Eckstein’” (Marron, 1984, 537).

10Eckstein enjoyed such a well-established public reputation because of his role as a member of the Council of Economic Advisors in the years 1964-1966 (Arenson, 1984). Eckstein had also a prestigious academic career. A Princeton undergraduate in economics (1951), he completed his PhD in economics at Harvard (1955), and he became a Harvard Professor of economics. In 1957 he had a short stay at the RAND Corporation, when he got interested in simulation. Back to Harvard in 1958, he became involved in a macroeconomic modeling project together with his colleague James Duesenberry (PhD University of Michigan 1948) and with Brookings’ Gary Fromm (Kraemer et al. 1987, 70). They were invited to participate at the 1959 “Conference on Economic Instability” organized by the Social Science Research Council (SSRC) and they became part of the project of building what later became the SSRC/CES-Brookings model, with the ambition of building the most highly disaggregated model of the US economy (Acosta and Pinzon-Fuchs, 2019, 538; Bodkin et al., 1991, 95-110). Eckstein went to Washington, DC, to be a staff director for a research project at the Joint Economic Committee of Congress.

  • 6 “Johnson conversation with Otto Eckstein on May 04, 1964,” Lyndon B. Johnson Presidential Recording (...)

11Thus Eckstein built an “outstanding reputation as an economist,” which made of him a “natural” candidate for entering the CEA (in replacement of John B. Lewis)—as President Johnson explained to Eckstein, on the phone, when offering him the job.6 His previous theoretical and empirical research on the determination of wages and prices (Eckstein and Wilson, 1962; Eckstein, 1964) led him, during his time at the CEA, to assume the responsibility for “wage-price guidepost policies” and for “developing some of the programs of the Great Society,” which was a set of domestic programs launched by President Johnson in 1964-1965 that included, among other things, the Civil Rights Act of 1964, the War on Poverty, and Medicare and Medicaid (Arenson, 1984).

12After leaving the CEA in 1966, Eckstein spent a year at a behavioral science research center at Stanford University where he developed “the idea of creating an information service to provide economic data for analysis, a notion that had come up earlier in discussion with Fromm and Duesenberry” during their earlier joint work (Kraemer et al. 1987, 70-71). In that same year Eckstein accepted Marron’s invitation to start a new “consulting program,” which took them across the US, meeting with various corporation executives to discuss current economic conditions and prospected development of the US economy. However:

traveling was a tedious and time-consuming way to impart information to our clients, and sometime in 1967 Otto suggested to me that perhaps we could use a computer instead. (Marron, 1984, 537)

13The original idea by Eckstein was for customers to be able to access information through time-sharing, a brand new technology that allowed multiple users to operate simultaneously on a mainframe computer via remote terminals connected over phone lines. The idea then blossomed in establishing a company, whose “product concept developed by Otto had four basic elements: the [macroeconometric] model itself, a large data base, a computer, and the software” (Marron, 1984, 537).

14Eckstein had been, more generally, interested in making the most recent econometric techniques available to business and the government for conducting their activities. His own experience had convinced him that economic analysis outside academia was not benefitting from the most recent advances in econometrics, especially when applied to macroeconomics. In a 1972-1973 correspondence with Harvard’s colleague Kenneth Galbraith, spurred by a controversy about the blurred lines between the Harvard Economic Departments and DRI (cf. infra), Eckstein presented a retrospective account of his motivations to establish DRI:

After my most recent period of full-time government service in 1966, my views on the economy were sought by business and financial organizations. I quickly discovered that they made little use of macro economics or econometrics. … Econometrics, which always looked to me to be a very practical way to establish quantitative relationships, received little use and remained an academic plaything. I had already discovered in the government that even macro-decisions were made on the basis of very crude quantitative work, without the benefit of the thirty years of methodological development of econometrics. (Eckstein to Galbraith, December 20th, 1972; in Collier, 2022)

15Marron (1984, 537) emphasizes that Brookings’ experience with collecting and handling large dataset was paramount to Eckstein. This is no wonder: the procedures for collecting, homogenizing, storing, and exploiting a large dataset to be used for macroeconometric modeling (and beyond) was one key aspect of the SSRC/CES project in the early 1960s, which was then passed into the Brookings Institution (Acosta and Pinzon-Fuchs, 2019). Two Brookings economists even joined DRI at its foundation, taking on the responsibility of the data part (Marron, 1984, 537): John Ahlstrom, one of the programmers of the Brookings model, and James Craig, who had been the databank manager of the Brookings model (Greenberger et al.,1976, 217). Eckstein envisioned that data provided by DRI, and made easily accessible through time-sharing, would change the pace of the dissemination of econometric modeling.

  • 7 For a comprehensive lists of contributors to DRI models through the years see Eckstein et al. (1974 (...)

16For the purposes of its new consulting firm, Eckstein envisioned building his own macroeconometric model, instead of relying on an existing one. Thanks both to his early involvement with the macroeconometric modeling community and to his Harvard network, Eckstein was able to ask other economists to contribute to the building of his model for DRI. Marron (1984, 537) cited as early contributors to the DRI models Fromm, Martin Feldstein (D.Phil. Oxford 1967), Lester Thurow (PhD Harvard 1964), Dale Jorgenson (PhD Harvard 1959), and Edward W. Green.7 All the economists consulted by Eckstein were offered shares of DRI as a compensation for their help:

  • 8 In the 1976 booklet for the first public offering of DRI shares, there is a list of the principal s (...)

I therefore proposed that we offer stock in the company to nearly all of the country's leading econometricians in exchange for commercial rights to their academic work. Otto questioned the wisdom of distributing stock to individuals who might contribute little to our company, but agreed that this was a small price to pay for technical assistance that might prove vital to the project. … Therefore, we signed up most of the leading lights of econometrics, including Lawrence Klein, Martin Feldstein, Lester Thurow, Marc Nerlove, Dale Jorgenson, and Robert Hall. (Marron, 1984, 538)8

  • 9 Marron (1984) mentioned one million dollars as the initial capital of DRI, whilst Eckstein spoke of (...)
  • 10 On Harvard Business Barometers, see Morgan (1990, chap. 2), Armatte (1992), Friedman (2009), and Le (...)

17DRI relied mostly on venture capital, channeled through Marron’s Mitchell Hutchins & Co., but mainly provided by the First Security Corp. of New York (a branch of First National Bank), which happened to be chaired by Robert Dennison, a graduate from Harvard Business School (Eckstein to Galbraith, 08/01/1973, cited in Collier, 2022).9 Eckstein told Galbraith that other possibilities of funding were explored, including a foundation financed project at Harvard, which was “ruled out immediately because of the poor experience with the Harvard Economic Barometers of the late 1920’s, an episode with which I was familiar from reading the archives of The Review of Economics and Statistics.” (Eckstein to Galbraith, 08/01/1973 cited in Collier, 2022)10

18DRI started its operations in the Summer of 1969, with its headquarter in Lexington, which was just ten miles (a thirty-minute drive) from Harvard. Marron (1984) claims that this location had been instrumental in the success of DRI, as it could both profit from the proximity of MIT/Harvard, as well as make possible for Eckstein to drive from his home to DRI offices at any time (he spent even weekends working at DRI according to Marron, 1984, 538). At the launching of DRI, The Boston Globe described the goals of DRI by its founders:

We want to become the basic national information utility on economics.’ That’s the aim of Otto Eckstein and a group of expert colleagues who have set up Data Resources, Inc., in quest of that interesting goal (Davidson, 1969).

19Otto Eckstein and Donald Marron served as de facto co-CEOs of DRI at its creation in 1968, but Eckstein took full responsibility of running DRI, and remained the CEO throughout the 1970s. However, this double role of Eckstein (Harvard professor and CEO of DRI) and, more generally, the relationship between Harvard University and DRI had been under scrutiny. In 1972-1973, John K. Galbraith, another Harvard faculty member, raised the issue of a potential conflict of interest, stirring an internal debate with Eckstein, Feldstein, Jorgenson, and Duesenberry (Collier, 2022). Later in 1973, a special commission within Harvard Economics Department ruled that “business connections between Economics professors and outside corporations do not interfere with hiring decisions and teaching practices.” (Harvard Crimson, 20/03/1873, quoted in Collier, 2022)

1.2 DRI’s Main Products and Structure

  • 11 Davidson (1969) reported that “DRI’s client list [included] Morgan Guaranty Trust, Girard Trust (Ph (...)
  • 12 The Congressional Budget and Impoundment Act of 1974 established the Congressional Budget Office an (...)

20From its beginning DRI developed a business model that sold products and services to the government, universities and private companies. The governmental network was instrumental to DRI’s success. After creating the macroeconometric model and making it available to clients in 1969, “Eckstein enlisted the help of Charles Warden, a retired military officer who had become acquainted with Eckstein in Washington in 1950-60,” and who went to Harvard for obtaining a PhD, following Eckstein’s advice (Kraemer et al., 1987, 72). After that Warden returned to Washington where, with Eckstein’s patronage, he secured the position of special assistant of the chairman of the Council of Economic Advisers, a post he held from 1966 to 1969. Warden’s job brought him into extensive contact with major actors in the economic policymaking process, and his managerial skills were broadened through dealing with the complicated technical and political aspects of national economic policymaking” (Kraemer et al., 1987, 72-73). After securing some revenue from the private sector, DRI opened an office in Washington, DC, to market its services to the government.11 Assisted by Warden, Gary Fromm became the first director of DRI’s Washington office (Kraemer et al., 1987, 75). Taking advantage of the few competitive bidding restriction on contracting private companies by the federal government in the early 1970s, DRI secured as its first federal users “the Library of Congress, which then was the primary research and analysis arm of Congress, and the Treasury Department, which had primary responsibility for analyses used by the executive branch in the creation of the annual budget” (Kraemer et al., 1987, 76). This list was soon enlarged to include major economic policy agencies of the executive branch: the Office of Management and Budget, the Treasury Department, and the Council of Economic Advisers; Congress’ Congressional Budget Office; and the Federal Reserve Board (Kraemer et al., 1987, 28).12

  • 13 McManus recalled that an “international macroeconomics group” would also work on collecting data fo (...)
  • 14 We could not get a precise organizational chart of the company. But the 1976 prospectus for the pub (...)

21DRI was structured around different “groups” of economic analysis, from the macroeconometric model group to those doing both sectoral and industry-level analysis. Kevin McManus, an economist working at DRI from 1976 to 1983, recalls that most groups would typically involve around 10 people, mainly economists, with 1 or 2 PhD graduates (McManus, interview). Each group would oversee data collection and produce economic analysis for a specific sector of the US economy or a specific industry (McManus, interview).13 McManus, for instance, was part of the “steel sector group” that would collect and compile data on steel and iron markets (received from iron and steel corporations), then use it to produce data plots, and to produce a forecasting of the demand for iron and steel, which feed a quarterly DRI publication sent to customers (McManus, interview). Forecasting was produced relying on an econometric model, taking for instance automotive industry’ planned production as an indicator for future demand for iron and steel, or data on past iron and steel shipments as leading indicator for future demand; the econometric model was further disaggregating demand by end-industry (McManus, interview). The “steel group” itself would be organized around different sub-sectors of analysis (raw steel production, steel-shipments by end-markets, …). Other DRI groups would conduct similar activities for the pulp and paper industry, for the transportation industry, for the energy sector, for the agricultural sector, for banking and financial activities, for the housing sector, and so forth.14

  • 15 For instance, “car sales” forecasts produced by the large-scale model would be used by the “steel g (...)

22The analyses of each group would receive inputs from other groups, but most importantly from the “national forecasting group” (or “macroeconomics group”), a group of around 30 people in charge of DRI’s large-scale macroeconometric model.15 This was “the most prestigious” group within the company, since the model was conceived as the cornerstone of DRI’s activities and reputation (McManus, interview). For instance, it was this group who was mostly in charge of DRI’s main outlet, the monthly Data Resources Review, which featured macroeconomic forecasts among other sectoral analyses.

23The “hype” of DRI forecasts was recounted in these terms by the New York Times:

Young people flocked to work with Eckstein and D.R.I. economic forecasts were required reading from the White House to corporate board rooms. … Throughout the 1970’s, D.R.I. reigned supreme in economics, extracting hefty fees from corporations that felt the need to know what D.R.I. thought and were afraid to be caught without it. (Wayne, 1984)

  • 16 McManus also recalls that the macroeconomics group was the DRI group that had more systematic inter (...)

24Macroeconomists in academia were also following up the DRI forecasts.16 Robert Solow, for instance, recalls:

I was a middle-aged DRI addict; I always looked forward to the monthly Review. Whenever I felt impelled to think about what is likely to happen in the near future, or what the differential effect of this or that policy change would be, I know my natural inclination was to start with the DRI forecast or the DRI estimate. (Solow, 1985, 81)

  • 17 For instance, in the 1970s the US House Budget Committee had consulted DRI for building their own m (...)

25DRI had also federal and local government contracts.17 However, corporate customers made up the most considerable part of DRI’s business. The New York Times estimated that two thirds of the DRI revenues in the mid-1970s came from corporate customers:

Under time-sharing, a corporation paid D.R.I. to plug into its big computer, which housed the D.R.I. macroeconomic model and massive amounts of raw data. Corporate economists would manipulate the data in the computer, throw in some of their own assumptions and make economic forecasts. Demand for time-sharing was so strong that D.R.I. grew quickly from its 1969 start—to $3 million in sales in 1972, and to $72 million by 1982. (Wayne, 1984)

  • 18 These numbers come from DRI’s Annual Reports of 1975 and 1976, available in box 27, folders “Data R (...)
  • 19 “Data Resources Inc., 1976 Annual Report,” p. 19, folder “Data Resources Inc., 1976-78,” box 27, Ma (...)

26Although it is hard to check the sales data presented, it is the case that DRI’s overall finance registered impressive gains in a short time span. Comparing some indicators in 1969 and in 1976 we see that: its revenues increased from $27,100 to $17,176,000, its operating expenses grew from $629,600 to $14,403,300, resulting in a change of net income from -$602,500 to $1,501,700.18 By 1976 DRI had regional offices in San Francisco, Chicago, Houston, New York, Pittsburgh, besides Washington, and international offices in Brussels and Toronto.19 More generally, these were rosy times for the private consultancies. In 1982 the big three econometric firms (DRI, Chase, and Wharton) collected more than $120 million from about 200 clients in business and government.

  • 20 Mead Corp. was, at the time, a Fortune 500 company, with headquarters in Dayton, Ohio, and dozens o (...)
  • 21 Luke recounted that he had a previous experience, in a similar role, within NRC Corp., a large comp (...)
  • 22 Moreover, one had the possibility also to dig more precisely in how the forecast was built (look at (...)

27To provide an illustrative example about the use of DRI’s services by corporate clients, we have the recollections of Jim Luke, an economist who worked as Head of planning at Mead Corp. (1980-1984), a major paper company (Luke, interview).20 Luke’s role at Mead Corp. was to help planning production, sales, and financing of the firm, typically through an annual 5-year rolling financial plan. This required forecasting the sales of different branches of the firm, at regional level, and by type of product. The planning process had proven difficult in the years preceding Luke’s arrival: he recalls that he was brought to Mead to implement a more “scientific” approach to planning, based on econometrics.21 DRI services were helpful in this task for at least two reasons. First, DRI provided his customers with access to the macroeconomic forecast of a large set of variables. Luke recalls that, in the econometric models of paper sales he had to build for Mead, the most relevant variables would be, for instance, the level of employment (driving demand for paper), or the interest rate (driving paper mills’ financing cost). It would have been impossible for Mead to forecast these variables on its own, since these would require building their own macroeconometric model.22 Second, Luke had access to all DRI data. This access to data, he emphasized, was amazing because of the quantity of variables, the length of historical series, and also the access to forecasts for most individual variables (even outside those of the main model). As Bodkin et al. (1991, 122) described:

  • 23 White also recalls: “For DRI, an awful lot of the attraction for the customers was access to the da (...)

The DRI Model is regarded as the centre-piece of the DRI information system, but the information system is much broader than the basic model. (For instance, fewer than 1000 time series are required to maintain the basic model, while the information system contains over 20 000 time series.)23

28This large amount of data was highly useful to the work of Luke and other economists working, like him, into private corporations, because they could use these data to perform their own regressions.

29Yet another product was sold by DRI to its clients. DRI offered its customers consulting services: DRI’s specialist provided analysis or report on specific issues of interest to the customer (McManus, interview). This was done by a specific consultant group instead of the economists from sectoral groups. However, McManus confirms that most of the direct interaction between DRI economists and corporate clients would happen with corporation strategic planners (like Luke) or with corporation market analysts.

30For different subsets of these services, DRI customers will pay a yearly subscription worth, in 1972, between $1,500 (basic subscription, giving access only to a limited number of services) and $12,000 (for access to full services), that is, in today’s terms, between approximatively $10,650 and $85,000 (Greenberger et al., 1976, 219). On top of this subscription, each client using DRI’s computers through time-sharing would be billed according to CPU-time used (cf. infra, Section 2.2). A few years later, according to Lodewijks (1989, 36), “The average client for Data Resources Inc. [paid] $75,000 a year for a variety of its forecasting services,” that is, approximatively, an overall bill for DRI’s services of $185,000 a year in today’s term.

1.3 The DRI Macroeconometric Model

  • 24 See Bodkin (1999) for a historical panorama on computation developments of macroeconometric models, (...)

31The DRI quarterly model of the U.S. economy, developed throughout the 1970s, was representative of the class of large-scale macroeconometric models earlier advocated by Klein and Goldberger (1955) and then developed further by the joint effort of academics and policymaking institutions in the 1960s. Eckstein (1983, 3) explicitly claims this legacy for his own model, while he located DRI’s model into a “second generation” of macroeconometric models, which included the models developed by SSRC/CES-Brookings, by MIT-Fed, and by Wharton (Eckstein 1983, 3-4).24

  • 25 According to Eckstein (1983, xi), the model was rebuilt “after the oil shock,” since “the behavior (...)

32The original DRI model (1969-1972; presented in Eckstein et al., 1974) went through substantial refurbishment after 1974.25 The version of the model available at the end of the 1970s consisted of roughly 700 equations, of which 375 stochastic behavioral equations (Eckstein, 1983). The economy represented by the model was “disaggregated” into various blocks, each further disaggregated in further details, following what Hoover (2012) has identified as the “Klein” microfoundational program. Economists in charge of the model designed specific blocks for various sectors and industries (whilst the models developed within other groups were rarely feeding into the large-scale macroeconometric model; McManus, interview): the “demand” block (212 equations), the “production” block (208 equations), and the “financial” block (193 equations; see details in Eckstein, 1983, Table 16; see also Bodkin et al., 1991, 123). The disaggregation within blocks was conducted by differentiating industries, economic agents, or institutions.

  • 26 These simulation techniques were so important that Klein (1975, 21) claimed that:

33Not only the macroeconometric models of the 1960s were developed by teams of researchers connecting academia (typically as an inter-university effort) and governmental bodies, but also each one built on previous efforts and offered some novelties in comparison to the competitors. The SSRC/CES-Brookings model, which, as we argued, constituted an important experience for Eckstein, brought several innovations that were further developed in other models, including DRI’s main model. We focus on three of them. First, the large model (with non-linearities) was estimated with quarterly US data from 1949 to 1960 (a sample of approximately 60 observations) thanks to the splitting of the system of equations “into several interacting blocks” that were “ordered in a recursive chain” (Bodkin et al., 1991, 101). Using the recursive estimation techniques just developed by Herman Wold (1954; 1960), Franklin Fisher (PhD Harvard 1960) showed that due to the block recursiveness of the Brookings model, each block could be consistently estimated by a system of single equation least square regressions (Fisher, 1965; see also Fisher, 1970; and the exposition in Bodkin et al., 1991, 101-103). Second, the application of the Brookings model to policy simulations was tested by “ex post forecasts for the period from the first quarter of 1961 to the fourth quarter of 1964” (Bodkin et al., 1991, 103). Third, the Brookings model advanced a computer simulation method that was used later on in several other models.26

  • 27 See Acosta and Rubin (2019) for an analysis of the banking sector in several macroeconometric model (...)

34The other key macroeconometric model developed in the 1960s under the leadership of Franco Modigliani and Albert Ando was the MIT-Penn-Social Science Research Council (MPS) model of the Board of Governors of the Federal Reserve. One of the key characteristics of this model was a “detailed financial sector which became a nuclear part” of the model, and which “grew out of the Brookings project” (Bodkin et al., 1991, 108). This sector featured a detailed description of the monetary markets (with several different interest rates: on corporate bonds, commercial loans, municipal bonds, and mortgages), detailed links of these markets with the real sector, and various exogenous policy instruments (see Cooper, 1972).27 Finally, the MPS model also relied on the technique of block simulations and was heavily used in studies of stabilization policies in the US.

35The Wharton model developed by Klein since 1963 would eventually build on Brookings’ and spliced the macroeconometric model with an inter-industry input-output model, yet initially being a much smaller model than Brookings’, but quickly growing in size (the third version of the model, of 1972, had approximately 200 equations and 100 exogenous variables; Bodkin et al., 1991, 126).

36The DRI model of the US economy was a quarterly model (as the Brookings, MPS and Wharton models) re-estimated each year. It was larger than the MPS and Wharton models, and smaller than Brookings’. With a detailed financial sector inspired in the MPS model, it was also organized in blocks of equations like the Brookings model and included input-output linkages throughout industries like in the Wharton model.

37The DRI model’s main characteristic, shared with the Wharton model, was the focus on ex ante forecasts that were released quarterly to DRI clients, and constituted an objective factor in establishing a “track record” by which credibility was established:

Because of the importance of the paying client to the development of the model, there has been an important emphasis on (ex ante) forecasting, but other aspects (such as the development of satellite or peripheral models by users of the basic model) have also received some attention. (Bodkin et al., 1991, 122)

38According to Bodkin et al. (1991, 119) the macroeconometric models in the 1960s became more complex and sophisticated due to five interrelated trends that continued unabated in vigor in the 1970s: “improvements in computer technology, the implementation of a team approach to econometric modeling, the development of an institutional framework largely independent of the originator(s) of the models, detailed and large-scale modeling, and increasing theoretical sophistication.” So, the 1970s witnessed “the rise of the econometrics consulting house as a commercial entity.” (ibid.) And as the authors put it: “Perhaps the most successful econometric model ever built, from the point of view of the market test of the product, was the DRI Model of the US economy” (ibid., 122).

  • 28 As mentioned by Goutsmedt et al. (2019), Eckstein was precisely one of those involved with pushing (...)

39The recent history of macroeconomics has provided strong evidence that large-scale macroeconometric modeling was still in an ascending phase of development all along the 1970s (Bodkin et al., 1991; Boumans and Duarte, 2019). Contrary to what is sometime argued in contemporary “potted histories” of macroeconomics, neither monetarism nor new classical macroeconomics had halted the development and use of these models, especially within policymaking institutions (see e.g. Goutsmedt et al., 2019 on the resistances against the Lucas Critique).28 The commercial success of DRI during the 1970s and early 1980s confirms once more the popularity of large-scale macroeconomic modeling during this period. Naylor (1981) estimated that probably less than 100 US companies used econometric models in 1970. This number went to something between 750 and 1,000 in 1980, estimated from the number of clients of Chase Econometrics, DRI, and Wharton. He surveyed 234 corporate members of the National Association for Business Economists (out of 1691 questionaries sent), and 90% of them subscribed to an econometric service bureau. The survey indicated that DRI’s services were used by 68.3% of the firms subscribing to a econometric service bureau, followed by Chase Econometrics (32.4%) and Wharton (17.6%).

2. A Walk through DRI Computer Room

40As we saw, Marron (1984, 537) argued that Eckstein’s original insight for DRI’s business identity was to develop a model, a large dataset, a computer with time-sharing, and a software, with these four elements characterizing DRI’s information system. In this section, we describe the working of the computer infrastructure at DRI, mostly relying on Peter White’s recollections.

2.1 The Burroughs Computer

41When establishing DRI, Marron and Eckstein had been thinking about choosing the “right hardware” (Marron, 1984, 538). They narrowed down the choice to two companies: University Computing and Burroughs. Marron argued that they finally favored Burroughs for three reasons. First, with Burroughs, DRI would have their own (or leased) hardware, while University Computing was only proposing time-sharing. Second, Burroughs was also located in Lexington, where Eckstein lived and where DRI would be located, so they actually “shared the same building” (for the first “two-three years,” i.e., 1969-1971/1972; Marron, 1984, 538). Finally, Burroughs was itself interested in working with DRI as this would open a new market for them and “establish their technical credibility” outside military and governmental circles (White, interview; Gray and Smith, 2003; 2009). Indeed, the B5500 computer, although commercially successful, was not a widespread type of machine at the time. Burroughs “was using DRI as a showcase to prove to the computing world that their equipment was as good or better than IBM’s” (White, interview). This was also emphasized by Marron:

At the time [1968] Burroughs’ primary customers were financial institutions. If a B-5500 machine were used to deliver Otto’s product to the large industrial companies that were expected to be DRI’s main customers, we argued, Burroughs could establish its technical credibility with those companies and persuade them to use other types of Burroughs equipment as well. (Marron, 1984, 539)

42White (interview) also adds that “as a showcase, both Burroughs and DRI bigwigs would occasionally bring people through the computer room to show off the equipment and prove how well it was working.

  • 29 Overall, the computer business at the turn of the 1950s was segmented with respect to these two dif (...)
  • 30 The operating system of the B5000, the “Master Control Programm” (MCP), was written in ESPOL (a sup (...)
  • 31 It is usually mentioned in the history of programming that the consensus around ALGOL experienced a (...)

43Although Burroughs’ origins are rooted in the nineteenth century and in the business of mechanical calculators for commercial use, the company had taken the turn of building large computer systems in the 1950s, starting with the D825, produced for the US military (Cortada, 1993; see also Yost, 2013). The first commercial machine produced by Burroughs was the B5000, which was designed to continue Burroughs’ service to both its scientific and business clients (especially banks, which had been using previous Burroughs’ equipment for bookkeeping procedures; Waychoff et al., 1986, 10-13).29 B5000 was designed based on advances made by Burrough’s in multiprogramming (the ability of a computer to run several programs simultaneously) and programming based on recent high-level (i.e., machine-independent) languages, ALGOL and COBOL (Waychoff et al., 1986, 15-18; 21-22; Gray and Smith 2009).30 ALGOL, developed from the 1958, was expected to become a new international standard for scientific programming—a bet that was taken up by Burroughs when developing the B5000, although it turned out to be wrong (Waychoff et al., 1986, 8).31 COBOL (“Common Business Oriented Language”) was designed from 1959 by Grace Hopper within project UNIVAC of Remington-Rand Corp: COBOL had “English-like” syntax and was hence aimed at better meeting the needs of business customers.

44The first B5000 was delivered to the United Technologies Center of United Aircraft (an aerospace conglomerate) in 1963. The second went to NASA a month later, and then 20 machines went to a host of clients including the First Pennsylvania Banking Company (Gray and Smith, 2009, 45). In 1964, Burroughs started commercializing the B5500 months after IBM released the so-called third-generation computer, System/360 (one in which transistors were replaced by integrated circuits; Gray and Smith, 2009). B5500 was three times faster than B5000 and used disks instead of drums, all making it a great commercial success that “expanded the B5000’s small customer base in the large-scale computer arena” (Gray and Smith, 2009, 44). Burroughs perceptively made its product more competitive by also offering a FORTRAN compiler for the B5500, letting them to even persuade IBM clients to switch to the B5500 (Gray and Smith, 2009, 46).

45The B5500 production lasted until the late 1970s, lingering on at various universities (Gray and Smith, 2009, 46). When DRI acquired two B5500, in 1969 (and eventually acquiring a third B5500 in the mid-1970s), the machine had been thus on the market for several years already. Later, in 1977, DRI’s B5500 were replaced by the new B7700 computer (also from Burroughs), which had a faster CPU and more RAM.

46Figure 1 below illustrates the physical layout for one single B5500 computer. The actual computer is constituted of the big cabinets in the background of the picture. Data were recorded by several magnetic tape drives (right-hand side of the picture).

Figure 1. Burroughs 5500 Computer

Figure 1. Burroughs 5500 Computer

From Burroughs’ brochure. Computer History Musem’s collection (https://www.computerhistory.org/​revolution/​mainframe-computers/​7/​166/​674, retrieved 03/11/23)

  • 32 “In addition to that, there was a guy, named Joe Neal, who was in charge of the networks for DRI. H (...)

47The operation of the two (then three) B5500 for DRIs activities required the intervention of three professional figures: software engineers, field engineers, and computer operators.32 With the cooperation of these three professionals, economists at DRI were able to build and use datasets and models to conduct their analyses, and customers of DRI were able to access DRI database and use DRI computers to perform their own analyses.

48DRI software engineers were in charge of creating the tools that would be used both by DRI economists and customers (cf. Section 3). Field engineers were Burroughs’ employees (and not DRIs). From one to three field engineers were permanently working at the DRI computer room to ensure that the B5500 was functioning at the best of their capabilities. If necessary, they provided fixing and maintenance “on the spot”: White recalls that field engineers

implement[ed] field corrections to the hardware, whenever necessary, that is, literally, opening computer boxes and lay them down on the floor, adjusting or cabling wires with soldering irons, while the computers were still running and performing all the work requested by DRI economists and customers (White, interview).

2.2 Computer Operators and Data Inputting and Backup

  • 33 We could not retrieve information about Druin’s background: he was an engineer, and likely trained (...)

49DRI’s own computer operators were mostly in charge of two separate aspects: the daily backup of data (cf. infra) and running diagnostic tools (which had been designed by Burroughs). The latter checked the optimal use of computer capacities (that the machine was “running smoothly”). There were usually two computer operators working in the computer room, under the supervision of Roger Druin, the chief computer operator during the early 1970s.33

50Computer operator could be a career-entry job. For instance, White had started his career at DRI in 1971 as a computer operator during a break in his studies: at the time, he was a student at MIT, preparing a BSEE (Bachelor of Science in Electrical Engineering) in Computer Science. After graduating, White moved to a software engineering position (White, interview).

51The computer room at DRI, where computer operators would spend their working day, had a “T-shaped” layout, as White recalls:

The physical layout of the computer room was T-shaped. What normally is the long side of the “T” (going up and down) was very short. That short area was where computer operators spent most of their time. There is where the consoles were. Note that, at this time, none of the consoles were CRT [cathode ray tube]: it was all paper. So there were consoles there, one for each of the main computers, and then a couple of extras. We also had the tape drives there. There was also a large line printer, the kind that used 8 ½ by 14 [inches] paper. The horizontal part of the T was where the [B5500] computers actually were, and also the network equipment … In addition to the actual computers, this [the horizontal part of the T] was where the disk drives were. (Peter White, Interview)

52During the day, one or two field engineers and two computer operators would be walking up and down the room to perform their duties:

Mostly we stood. The computer terminals were on stands that were approximately [1.40/1.50m high], and we worked on those, because we were always walking back and forth from the tape drives. Even when we were not doing backups at night, economists would send us data on tape, and we had to load them. So we would walk around, also picking up printouts, and stacking them … [printouts] were all labelled (whose they were) and we will stack them in the appropriate spot for people to pick up. There was one spot in the computer room where you could sit down at a terminal, but there was not a desk as such. (White, interview)

53Software engineers and economists would not stay in the computer room. While economists would come in mainly to pick up the printouts and walk out, software engineers would stand around and chat with computer operators, since “they had more in common.”

  • 34 Working days at DRI were 16-hour days, but the night shift (3pm to 11pm at night) was a distinctive (...)

54Peter White recalls that “one of his main duties” as a computer operator was “to create backups of the econometric data on the computers on a daily basis.” This was a well-organized routine, consisting of “incremental backups” from Mondays to Thursdays, and a “full backup” on Fridays. Backups would take place during the “night shift” and they will take between from 3 to 6 hours.34 A backup corresponded to recording on magnetic tapes “of all the data on the disks on the computers (econometric data, software source code, etc.).” These tapes consisted of a set of 6-9 large reels, 1/2 (inches) wide (= 12mm), and 2400 (feet) long (=731.52 meters); Monday-Thursday backups would take 5-10 tapes less than the Fridays backup. The tapes reel ran on 6-9 spinning tape drives” (cf. the right-hand side of Figure 1).

  • 35 “Tapes were reused for a while. So they will eventually wear out. The actual magnetic material on t (...)

55The duty of the computer operator consisted in “mounting, dismounting, and labelling tapes that entire [backup] shift.” The resulting backup tapes were stored offsite, then reused “every month or two,” until they had been used “too many times and had to be replaced.” This operation, although constituting a routine, would be a delicate one, subject to unforeseen glitches or malfunctions: the role of computer operators was precisely to oversee the process and to intervene in case anything went wrong.35 Backups were kept for ensuring that economic models as well as data (DRIs most valuable business assets) would not be jeopardized by any hardware failures: “Sometimes the [computer] disk drives would die … the heads would drop from the disk and scrap the magnetic media of the disk. That happened” (White, interview).

56Although in charge of backing up economic data, computer operators were not directly involved with the collection of economic data. Within DRI, each economic group was in charge of collecting some data, related to the specific industry or economic sector placed under their responsibility. McManus (interview) recalls that “each group had a ‘data banker’; she (they were always ‘shes’) would receive statistical reports and she would type in the numbers.” While inputting data was done at DRI almost always by women, DRI economists were roughly 50% men and 50% women (McManus, interview). Women also played an important role as programmers in DRI software engineering team (cf. infra, Section 3).

57Concerning data inputting, White also recalls that:

there was a group in New York [of maybe five to ten people], working for DRI, whose sole job was to input the data. They would pull out the Wall Street Journal and other business publications and literally read it of the paper and type [the data] in. Thats what they did all day long. (White, interview)

58While this specific group would directly “type in” the economic data into DRIs computer database, some other economists would need help to turn economic data they needed from paper format to computer-readable format. White recalls that Roger Druin, the head computer operator at DRI, had as a secondary job the conversion of paper data into cards:

  • 36 IBM cards with data were regrouped in these “boxes,” containing around 2000 cards each; each card “ (...)

The economists often were looking for people to extract numeric data from printed reports published by the various national governments as well as the big corporations. … My supervisor at the time moonlighted by keypunching the numeric data onto IBM data cards using an IBM 029 Keypunch he had in the front hall of his home. He would get assignments from the economists and show up at work with 10-20 boxes of these cards. (White, interview)36

59Time series constituted the core of DRI’s database. Most series were monthly, with monthly data also occupying the most space on DRI’s computers; but annual, semi-annual, quarterly, and even weekly data were available to DRI economists and customers (White, interview). This rich database came however with some burden in handling heterogenous frequencies in data. Specific programs were to be designed to handle this, as White recalls:

I remember being surprised at the time that some years had 53 weeks, depending on how you defined the end of the week. If you define the end of the week as Thursday, you count how many Thursdays in a year, and some of them had 52, some of them had 53. It was kind of amusing when that happened. But the software had to handle it, and so did the models. Also, if they were working with Detroit and the car industry, the data were on a 10-day basis. Because that was the way Detroit worked at the time. Back then it was every ten days they got a new data point. ... There were a number of programs that DRI wrote to consolidate data, for example to convert quarterly data on annual data, or to distribute data, for example to convert it into weekly data. Again, this was designed to match what economists were trying to do whit their models. (White, interview)

60The practices we described in this section highlight two general aspects on the role of computers for macroeconomics in this era. First, we can observe how the actual operation of computers was not at all within the domain of economists’ skills or duties. In the 1970s, economists’ use of computing capacities for both data gathering and econometric or modeling activities had to rely on other professional figures (mostly electronic engineers). These professional figures played a pivotal role in the functioning of DRI’s core business, although fewer in numbers—in 1976, around 17% of people working at DRI had computer-related duties, against around 32% involved with forecast and modeling (cf. supra, fn. 14). This division of labor manifested itself not only in differences in training, in daily routines, and tasks, but also in a physical separation between the two groups. Second, the computer practices needed for inputting and backing up data were a significant effort to make data (and particularly, time series) ready to be used by economists both within DRI and outside (the economists working for DRI’s customer firms). We have seen, in particular, how the storage and preservation of data constituted a central aspect of computer practices at DRI, a structural element for computer operators’ working routine.

2.3 The Time-Sharing Technology

  • 37 Christopher Strachey and John McCarthy are today credited with pioneering contributions, in the end (...)

61The time-sharing technology emerged in the early 1960s as the solution to the problem that computers were too large, too expensive, and too costly to maintain, limiting the access to computational capacities.37 The first commercial time-sharing was offered by General Electric in 1965 (using the “BASIC” software) and in 1968 the company was providing service “to 100,000 users in the United States and abroad” with a rate of “$10 an hour.” Time-sharing brought computers to schools and businesses, and in 1965 it created the hope “that there would be an outlet in every kitchen for housewives to figure their shopping lists” (Jensen 1968).

62The basic idea of time-sharing was precisely to make possible for several users to access a computer independently and simultaneously, and being able to input requests, to access existing programs (from a “library” of programs) or running their own programs. Connection to the computer was made through phone lines, while the user would be able to operate the computer through typewriter-kind of “terminals,” plugged into their phone. Outcomes of the works inputted through terminals would either appear back on the terminal (as one-line output) or, in the case of larger amounts of output data, they would be mailed to the user by the time-sharing company.

63When, in the end of the 1960s, Marron and Eckstein were preparing to set up DRI, the time-sharing technology had been at the center of an investment fever:

In the late 1960s new computing companies were springing up almost weekly; Business Week caught the spirit of the times when it quoted an entrepreneur as boasting, “All you have to do is get a few good guys, go down to lower Broadway, and yell ‘timesharing’ and they bury you up to your neck in money. Then you rent a computer and you’re in business. (Marron, 1984, 538)

64Indeed, several companies in the US had made time-sharing their main business. A 1972 Guide to Time Sharing (Auerbach Publisher Inc., 1972) listed more than one hundred companies providing time-sharing services across the US. The most famous were ComShare, based in Ann Arbor, Michigan and led by Richard Crandall (an electrical engineer and mathematician) and Tymshare, based in Cupertino, California (see Crandall, 2002). Large computer companies like IBM also provided their own time-sharing services to their customers (Auerbach Publisher Inc., 1972).

  • 38 Obviously, economists working at DRI would also access the computer through the same time-sharing s (...)

65The business of time-sharing companies relied on billing customers based on the time use of CPU capacities and disk usage they had. DRI adopted this business model, although, like some other companies (like ComShare; Randall, 2002), they also provided clients with a library of programs that the customers could access (White, interview). Moreover, every user had her own personal storage space for files.38 Costs of such services were sufficiently high to restrict the use of time-sharing to large companies. For instance, ComShare charged $10 to $20 per hour (depending on the speed of connection; see Auerbach Publisher Inc., 1972, 156), which would be equivalent approximately to $700 to $1,400 today; plus, a fixed monthly subscription rate of $400 (that is, around $2,800 today).

66DRI’s time-sharing interface was provided to DRI by Burroughs. Each B5500 could support 10 to 15 simultaneous users, that is, DRI’s computer infrastructure with three B5500 could host 30 to 45 simultaneous users through time-sharing (White, interview). Later, with the new B7700, this number became much higher. The Burroughs interface for time-sharing was considered particularly user-friendly:

  • 39 However, when looking at the list of computer systems supported by the largest time-sharing compani (...)

Burroughs had actually quite a nice time-sharing user interface. I have had some experience with IBM’s user interface, through school [at MIT]. Burroughs was from my point of view much nicer. It was easier to understand. You did not have to worry about producing job cards to get your program to run. Basically, you had a program that someone had produced for you, that is, some of the programs from DRI. You just gave the name of the program and whatever parameters the program was looking for, and it will just run it. (White, interview)39

67In summary, the time-sharing business model of DRI relied more particularly on the programs that were tailored by DRI. We now turn to their development.

3. The Development and Use of Econometric Programming System (EPS)

  • 40 Unfortunately, we could not retrieve any documentation or archival evidence about the functioning o (...)
  • 41 The 1976 booklet for the first public offering of DRI shares listed two other software developed by (...)

68Three distinct programs were originally at the core of DRI’s activities: AID, EPL, and MODSIM (McManus, interview)—and Kraemer et al. (1987, 74) added another program to this list, MODEL. AID was a data banking program: it would be used by DRI employees (mostly women, as mentioned above) to input economic data into DRI database. The second program was EPL (Econometric Programming Language), which would fetch data out of the database, and perform statistical analysis, econometric ordinary-least-square regressions, and data plots. Finally, the program MODSIM (Model Simulation) complemented EPL and handled modeling with simultaneous equations, and, in particular the large-scale macroeconometric model (McManus, interview).40 Finally, MODEL was developed by a young programmer hired by DRI, Robert Lacey, and allowed users to explore econometric modeling (such as changing from quarterly to annual data, creating and running forecasts under different assumptions; Kraemer et al., 1987, 74).41

  • 42 Although apparently not having direct connection with software development at DRI, one major outcom (...)

69The development of EPL represents another case for the cooperation between academic economists and DRI, which was facilitated by Eckstein’s position in the field of macroeconomics. In the mid-1960s, a group of macroeconometricians at MIT were trying to develop new software supporting econometric analysis for academic purposes, including macroeconometric modeling.42 Robert Hall, a PhD student at MIT (from 1965 to 1967), recalls:

Existing software was totally primitive, with fixed data formats and really clumsy ways to manipulate data. I needed something better for my own research and I saw an opening for helping other people. I knew that computers could be taught more flexible ways because the FORTRAN compiler knew how to decode data flexibility and how to use algebraic notation to describe data manipulation. Also, I understood recursion which made clean programming of flexible tools pretty straightforward. (Hall, personal communication)

  • 43 TSP was further developed in the 1960s, at MIT by J. Phillip Cooper, Robert Gordon, Charles Bischof (...)

70The result of Halls effort, with support of the National Science Foundation (Kraemer et al., 1987, 74), was “TSP” (Time-Series Processor), a program for econometrics which was originally designed in FORTRAN and run on an IBM 1600, then on the central IBM-360 at MIT.43

  • 44 Lacey will remain head of software engineers at DRI until 1978, when he left to start his own compa (...)
  • 45 Renfro (2004, 407) also mentions the “important design contribution” to EPS made by Edward Green in (...)

71Hall knew Eckstein but he was also acquainted with Lacey, who became DRI’s head software engineer (Hall, personal communication). Lacey, although not an economist, was a Harvard student in computer science, with possibly a minor in economics (although he apparently dropped-out before completing his degree). He was described by all the interviewees as a particularly “brilliant” programmer and the mind behind all DRI software” (McManus, interview; White, interview).44 DRI bought TSP and Lacey commissioned Hall to adapt it to the DRI computer system, resulting in EPL (Hall, personal correspondence; Kraemer et al., 1987, 74).45 This was a time with no copyright protection for computer programs in machine-readable form, which came only in 1976 (Samuelson, 1984).

  • 46 “Data Resources Inc., 1976 Annual Report,” p. 3, folder “Data Resources Inc., 1976-78,” box 27, Mar (...)
  • 47 We were unable to ascertain the background and respective role of all DRI programmers. However, rel (...)

72The original programming environment at DRI (AID, EPL, MODSIM, MODEL) was however deemed unsatisfying. In 1975-1977, Lacey headed the design of a new software that would combine the original programs, providing a single interface to users (Kraemer et al., 1987, 74). The new project was named Econometric Programming System” (EPS) and it was launched to customers in January 1977.46 Peter White, who had become a DRI software engineer in 1975, joined Laceys team of software engineers, which included also Greg George, Paul Lecain, John Powley, June Rokoff, Nancy Rosenfield, Steve Rothman, Marsha Shift, and John Schmitz. They were later joined by Kevin McManus, who moved from the steel group to programming.47

73Routine interactions between DRI economists and software engineers were more frequent than between economists and computer operators. Insofar as all economists at DRI were using DRI software to get their economic work done, they all needed guidance or help from software engineers (McManus, interview). Economists would also raise specific requests for changes, commands, or specific operations. This experience was at the core of the EPS development, as White recalls: EPS was aimed at providing a generalized platform for doing econometric modeling, [making it] easier for the economists to do their own programming; … basically, it was a language for non-programmers to write models in” (White, interview).

  • 48 White also recalls that ALGOL entailed “some specific efficiency issues” in the computations relate (...)

74While Lacey did the basic design,” White implemented functions and operators in the EPS programming language:” Robert [Lacey] and 2-3 others, including myself, were designing a way to convert the giant set of equations that formed the economistsmodel into ALGOL” (White, interview).48 According to White, EPS attained its goal:

At the time I was a computer operator, it was not very easy for an outsider (a non-DRI employee) to do modeling by himself. You had to be a real expert to do so. By the time I have left DRI [1979], the software we produced made it much easier for people to do that on their own. They can just sit there, write econometric equations, and then push them to the software and get the results … without having to involve a DRI economist. (White, interview)

  • 49 Luke compares DRI services accessible to EPS to modern-day FRED (Federal Reserve Economic Data, the (...)

75This positive assessment is not only echoed by Charles Renfro (2004, 407) but also confirmed by users, like Jim Luke. He recalls how easy and intuitive he found the EPS interface, how efficient the design of commands was, and how accessible the documentation was, at least for someone familiar with econometrics (Luke, interview).49 Comparing to another widespread software, SPSS, to which he got acquainted to during his studies, EPS was more relevantly designed for econometric work.

76Since its launching in 1977 EPS was continuously developed throughout the 1970s and the beginning of the 1980s following also the demands of DRI economists:

The economists would come to us and say: “Peter, my model would do so much better if there was a function that calculated this particular thing” … I would sit down with the economist, get the details of what he was looking for, then I would sit down with Bob [Lacey] and Paul [Lecain] and we would work out from the programming side of it. And then it would go and be implemented. … There were definitely some big releases. Where it was planned, and there would be some amount of fanfare about the accomplishment. The biggest one I can think of was producing the [main DRI] model as machine code. But if someone would come in and say to me, that would be done in a day, or in a week, we just added it. (White, interview)

77EPS grew larger and larger, to the point that the EPS group stopped printing out on paper the full extent of EPS code, preferring storage on microfilm (White, interview). The synthetic EPS documentation that we could consult (courtesy of Peter White; DRI, 1978a; 1978b) give however a general insight into the vast number of functions and operations supported by EPS. From their workspace, users could create their own personalized dataset, selecting relevant data from DRI and applying various transformation to data (changing frequencies, making seasonal adjustments, and so far). From the selected data, the user could easily construct descriptive statistics, plots and other graphical representations, and, of course, conduct econometric or modeling work. Figure 2 below provides a first list of the possibilities for estimating single equations models, as well as carrying out the usual batteries of tests.

Figure 2. Econometric Commands and Functions in EPS

Figure 2. Econometric Commands and Functions in EPS

Source: DRI (1978b, 4).

78One could notice the syntax of EPS, which is, as pointed out earlier, very easy to understand and to use for econometricians, as it relies on names for commands and functions that are much informative and clear.

4. DRI After Eckstein: The Bumpy 1980s and 1990s

  • 50 The due payment of an original loan of $1,500,000 was then $2,805,000 (“Preliminary Prospectus, Dat (...)
  • 51 McGraw-Hill was already one of the “big three” American educational publishers, featuring other div (...)
  • 52 Actually, the acquisition by McGraw-Hill benefited all workers of DRI, as well as those macroeconom (...)

79At the height of its success, DRI became a NYSE listed company in 1976, with the proceeds used to prepay the loan (the principal plus interest) used to acquire new computer equipment.50 Three years later, on July 1979, DRI was acquired by McGraw-Hill Inc. and Joseph E. Kasputys (PhD Harvard Business School) became its new president.51 According to Eckstein, this acquisition would help DRI to “do more with [its] technology,” while preserving DRI’s independence (“they won’t tell DRI what to forecast about the economy”; Bendheim, 1979; see also Rowe, 1979). The purchase was of a considerable size ($103 million dollars; Kraemer et al., 1987, 80): but DRI was then a company making $3.1 million profit per year (Rowe, 1979). Thanks to the acquisition of DRI by McGraw-Hill, Eckstein became multimillionaire.52

80Since the late 1970s, with the growing of DRI and the more intense competition in its market, a rift among top officials developed about what should be “the main components of DRI’s service mix” (Kraemer et al., 1987, 79). A dispute on how to organize the company’s market then intensified and some key senior executives departed. Charles Warden left in 1977, and in June 1978 “Stephen Browne, vice president in charge of industrial economics, left DRI with six of DRI’s top consultants to start their own company called Planning Economics Group” (Kraemer et al., 1987, 79). Senior vice-president Dennis O’Brien indicated also that he wanted to leave DRI, and he soon did. “Within two weeks, Marron, who for two years had been holding of offers by large investors to buy the company, arranged the sale of DRI to McGraw-Hill” (Kraemer et al., 1987, 80). The press suggested that these departures had to do with financial issues in the remuneration of personnel or with “a fast growth of the company, [which] had impaired communication among executives” (Rowe, 1979). In addition to those departures, the company itself seemed to have understood in the late 1970s the big challenges it had ahead: “Oddly, D.R.I. predicted the demise of its key business as far back as 1979, when it saw that smaller computers would eventually replace big mainframes, like its Burroughs. … And, to make matters worse for D.R.I., the decline of time-sharing came just as the recession of 1981-82 forced corporations to reduce outside expenses” (Wayne, 1984).

81Eckstein died of cancer in March 1984, at age 56 (Marron, 1984, 542; Wayne, 1984). He had relinquished his duties as CEO of DRI since 1981, but until the summer of 1982 he continued to supervise the DRI macroeconomic forecasts published in the Data Resources Review (Arenson, 1984), as well as the publishing of a complete presentation of the DRI macroeconometric model (Eckstein, 1983). The passing of Eckstein left notably a reputational gap that the new directors of DRI tried to fill by hiring Martin Feldstein as a “consultant” (Wayne, 1984). Nonetheless, DRI continued to experience a serious fall in profits (Wayne, 1984).

82Around 1985, McGraw-Hill appointed J. Phillip Cooper to lead a large business sector of its Standard and Poors subsidiary, to which DRI had been attached to.53 Cooper (interview) recalls that business and personal relations between DRI employees and its owner company, McGraw-Hill, were very tense. Cooper’s lead of DRI resulted in the company being profitable again in 1987. Besides managerial changes, Cooper pushed for a renewal of DRI’s computer infrastructure, introducing two new Unisys mainframes to replace the five to six Burrough’s mainframes still operating at DRI, as well as the new Varians mini-computers introduced in the early 1980s.54 Thanks to these changes, DRI lived through the 1990s. But in 2001, DRI formally ceased to exist, as it was merged with WEFA and Primark Decision Economics to form a new company, named Global Insights.55

  • 56 Cooper (interview) also remembers that, when he arrived at DRI around 1985, DRI’s macroeconometric (...)
  • 57 In 1981 Burroughs even lost “two lawsuits by B700 and B800 systems users who said that the computer (...)

83The loss of ‘hype’ of DRI in the mid-1980s had to do with the main elements of its original project: a large-scale macroeconometric model, a large set of economic data, a computer infrastructure and its computational capacities (accessible throughout time-sharing), and a software (EPS) making it easy for users to perform their own econometric analysis. On the one hand, as already well-documented by the literature on the history of macroeconomics, the large-scale macroeconometric modeling did not go immediately out of fashion in the early 1980s (this eventually happened in the mid-1990s, with the raise of a new type of macroeconomic models).56 On the other hand, the interest for economic data, especially for businesses, never ceased. But two crucial elements of DRI’s original business were clearly declining from the beginning of the 1980s: mainframe computers and time-sharing would soon be replaced by personal computers and, later on, by the internet; Burroughs’ computers, as well as ALGOL, would disappear accordingly. DRI’s new management seemed aware of these changes: “We are in the midst of a major technological change where the technology is moving from 100 percent reliance on time-sharing to a situation where clients can have powerful personal computers on their desks,” DRI’s president Kasputys declared to the New York Times (Wayne, 1984). Such problems that DRI was going through coincided with difficulties that Burroughs was having in producing market-competitive large-scale computers since the 1960s. As Gray and Smith (2009, 44) told: “By the end of the 1970s, faced with IBM’s dominant market position, Burroughs was struggling to maintain its position in the computer industry. … and Burroughs found itself in serious difficulties at the beginning of the 1980s.”57 DRI adapted its software EPS through the 1980s, but a PC version of EPS was released only in 1990 (Renfro, 2004, 407).

84In short, this indicates how computer tools were pivotal in the commercial success of DRI in the 1970s: once the technological advance of DRI’s information system had declined, DRI’s business started to decline as well.

5. Some Lessons for the History of Macroeconomics

85This article tried to uncover a neglected aspect of the history of US macroeconomics of the 1960s and 1970s—that is, the wide dissemination, through private companies of macroeconomic expertise. This was the ambition of Eckstein when founding DRI, since he thought that business and financial organizations “made little use of macroeconomics or econometrics”; an ambition in many ways met by success, judging from the “hype” of DRI’s analysis among the private companies and the government, and academic economists alike. The type of expertise popularized by DRI and similar companies relied in particular on large-scale macroeconometric used for forecasting. Historians of economics had already emphasized how, during this period, the use of large-scale macroeconometric models became entangled with macroeconomists’ practice of public expertise, especially within (or at the service of) policymaking institutions (in the US and abroad). Our contribution, by studying the functioning of DRI, the services it sold to its customers, draws attention to the simultaneous emergence of a large and far-reaching market for such macroeconomic expertise, relying on a “package” of services that included macroeconometric models, computer power (with time-sharing), large dataset, and user-friendly software. In short, large firms like DRI, selling their service both to public institutions and to private businesses, seem thus to crucially contribute to the larger dissemination of macroeconomic knowledge (and newly crafted econometric and modeling techniques).

86Moreover, we argue that the private business of macroeconomics was not a “by-product” or a “side” aspect in the development of the discipline. Indeed, this business was carried by key contributors to the large-scale macroeconometric approach (such as Eckstein at DRI, and Klein at Wharton, just to mention two main figures) and many of their graduate students, who operated at the boundaries between academia and private business. As a result, macroeconomics firms such as DRI were at the theoretical and empirical frontier of large-scale macroeconometric modeling, producing models that were at par with models crafted by policymaking institutions and academia. Thus, we argue, it would be relevant for the history of macroeconomics to investigate more thoroughly the private business of macroeconometric modeling, since this is one active area for contributions to the development of macroeconomics during this era.

87In addition, our contribution to the study of macroeconomics as a practice is to focus on how the macroeconometric modeling of the 1960s and 1970s has been entangled with the use of computers. The case of DRI shows how the computer infrastructure contributed to the success of macroeconomics during this era. However, somehow contrasting with usual narratives, the main driver of this success was neither the increase of computational capacities of computers per se, nor the generalization of mainframes (which remained overall, a costly and relatively scarce infrastructure throughout the 1970s). We particularly emphasized two computer-related innovations that were driving the expansion of business like DRI: time-sharing and software development. The adoption, in particular, of the time-sharing technology contributed crucially to make possible to a larger and larger number of economists, including those working for private businesses to access computing capacities, data and models required to perform (macro)econometric analysis. The second aspect driving the successful dissemination of macroeconomic expertise has been software development, tailored to allow (macro)economists to use computers rapidly and intuitively (that is, with relatively little training) for applying a standard toolbox of econometric and modeling techniques.

88We highlighted as well that computer-related transformations were not deux ex machina, and that their contribution to the dissemination of macroeconomics has been conditional to the crafting of well-defined and sophisticated social and technical routines.

  • 58 To which, we should add, the key role of business executives and financiers, who, like Marron (and, (...)

89First, we have highlighted how the intensive use of computers was associated with a division of labor, involving economists as well as several professional figures (software engineers, field engineers, and computer operators).58 In such a division of labor, macroeconomists were not very much “near the computer:” their role was mainly confined to one of users, away from the key aspects of computer practice such as data collection/inputting, data transformation and distribution, data storage and safe-keeping through backups, the maintenance and operation of the computer facilities. However, a few economists played a pivotal role in designing the computer infrastructure; most particularly, a few economists (over several dozen employed by DRI, for instance) contributed actively to software design. This entailed several key choices, that we documented at length in the case of DRI: the choice of “the right hardware,” related computer languages, and software design.

90But what could the peak of DRI’s popularity, until the mid-1980s, and its later misfortunes in the late 1980s and 1990s, add to the later history of large-scale macroeconometric models? As already emphasized by historians, criticisms raised in academia against these models did not bring to their demise in the early 1980s. These models were continuously developed, during this decade, by policymaking institutions—and, we add, by private business such as DRI (which had policymaking institutions as customers), still enjoying a great deal of popularity and commercial success. The case of DRI helps documenting at least two reasons for this phenomenon.

91First, theoretical innovations, like rational expectations, or theoretical criticisms (like Lucas, 1976) fell short in questioning the core of DRI’s forecasting business. Established “track-record” (as emphasized by Eckstein, 1983) in forecasting was likely to outweigh, for DRI’s customers and the general public, potential flaws in the theoretical mechanisms providing the forecast. The ability of large-scale macroeconometric models to provide “unconditional” forecasting was not questioned, even by their harsher critics (Lucas, 1976, 20).

92Second, one could easily draw a parallel between the reasons, outlined above, for successful dissemination of large-scale macroeconometric modeling, and the reasons for the failure of their criticisms, in the 1970s and early 1980s, to provide a competitive alternative. As emphasized by Cherrier et al. (this issue), the dissemination of alternative modeling practices relying on rational expectations was limited by their ability to solve tractability and portability issues. They were solved by developing a very different type of computer-infrastructure (based on personal computers, packages like Dynare, and an open-software culture). To say it otherwise, the computer-infrastructure and business model of the large-scale macroeconometric modeling industry of the 1970s persisted in the 1980s since an alternative was not yet available. All this shows how the embeddedness of macroeconomic analysis, and particularly of large-scale macroeconometric modeling, into computer infrastructure has been crucial in the dissemination and development of macroeconomics from the 1970s.

We are most grateful to Peter White for sharing his memories and devoting so much time to discussing with us his experience at DRI: without him, this article would simply not have seen the light. We also thank all other interviewees who kindly accepted to answer our questions and to comment on an earlier version of this article. We are the only responsible of any remaining inaccuracies. The article also benefited greatly from two anonymous referees’ comments and from Marcel Boumans, Cléo Chassonnery-Zaïgouche, and Pierrick Dechaux’s suggestions as the editors in charge of our article. Pedro Duarte wishes to thank the National Council for Scientific and Technological Development - CNPq (306408/2022-8) for financial support.

Haut de page

Bibliographie

Interviews and Correspondence

Peter White, personal communication, 04/15/2021; 04/19/2021; 11/15/2021; interview, 11/15/2021.

J. Phillip Cooper, interview, 03/23/2022.

Jim Luke, interview, 03/08/2022.

Kevin McManus, interview, 03/17/2022.

Robert Hall, personal communication, 03/15/2022; 03/18/2022; 03/22/2022.

Bibliography

Acosta, Juan and Beatrice Cherrier. 2021. The Transformation of Economic Analysis at the Board of Governors of the Federal Reserve System During the 1960s. Journal of the History of Economic Thought, 43(3): 323-349.

Acosta, Juan and Erich Pinzón-Fuchs. 2019. Peddling Macroeconometric Modeling and Quantitative Policy Analysis: The Early Years of the SSRCs Committee on Economic Stability, 1959-1963. Œconomia. History, Methodology, Philosophy, 9(3): 537-558.

Acosta, Juan and Goulven Rubin. 2019. Bank Behavior in Large-Scale Macroeconometric Models of the 1960s. History of Political Economy, 51(3): 471-491.

Arenson, Karen W. 1984. Otto Eckstein, Educator Who Led in Economic Forecasting. The New York Times, March 23. https://www.nytimes.com/1984/03/23/obituaries/otto-ecktein-educator-who-led-in-economic-forecasting.html [retrieved 03/11/21].

Armatte, Michel. 1992. Conjonctions, conjoncture et conjecture : les baromètres économiques (1885-1930). Histoire et mesure, 7(1-2): 99-149.

Auerbach Publishers Inc. 1972 Auerbach Guide to Time Sharing. Philadelphia: Auerbach Publishers Inc.

Backhouse, Roger E. and Beatrice Cherrier. 2019. The Ordinary Business of Macroeconometric Modeling: Working on the Fed-MIT-Penn Model, 1964-74. History of Political Economy, 51(3): 425-447.

Bashin, Kim. 2011. How McGraw-Hill Went from Publishing Railroad Journals To Downgrading The US Debt. Business Insider, August 8. https://www.businessinsider.com/mcgraw-hill-textbooks-standard-poors-2011-8?r=US&IR=T [retrieved 10/08/2022].

Bendhein, Kim. 1979. McGraw Hill Inc. Plans to Buy Company Run by Otto Eckstein; Data Resources Stock Jumps. The Crimson, July 7, https://www.thecrimson.com/article/1979/7/17/mcgraw-hill-inc-plans-to-buy [retrieved 02/11/21].

Bernstein, Peter L. 1993. Capital Ideas: The Improbable Origins of Modern Wall Street. New York: Simon and Schuster.

Binotti, Annetta Maria and Enrico Ghiani. 2008. At the Origins of the NAIRU: Short-Term Economic Policy and the Development of the First Italian Macroeconometric Models. History of Economic Ideas, 16(1/2): 103-132.

Bodkin, Ronald G. 1999. Computation in Macroeconometric Model-Building: Some Historical Aspects. In Daniel J. Slottje (ed.), Advances in Econometrics, Income Distribution and Scientific Methodology. Essays in Honor of Camilo Dagum. Heidelberg: Physica-Verlag, 41-60.

Bodkin, Ronald G., Lawrence R. Klein, and Kanta Marwah. 1991. A History of Macroeconometric Model-Building. Aldershot: Edward Elgar.

Boumans, Marcel and Pedro G. Duarte. 2019. The History of Macroeconometric Modeling: An Introduction. History of Political Economy, 51(3): 391-400.

Brayton, Flint, Andrew Levin, Ralph Lyon, and John C. Williams. 1997. The Evolution of Macro Models at the Federal Reserve Board. Carnegie-Rochester Conference Series on Public Policy, 47: 43-81.

Collier, Irwin. 2022. Harvard. The Data Resources Inc. Connection. Galbraith Asks Eckstein, Feldstein, Jorgensen. 1972. Economics in the Rear-View Mirror, February 22. https://www.irwincollier.com/harvard-the-data-resources-inc-connection-galbraith-asks-eckstein-feldstein-jorgensen-1972/ [retrieved 26/06/2022].

Committee of the Budget. 1975. Hearings, Reports and Prints of the House Committee on the Budget. Ninety-Fourth Congress, First Session. Washington DC: US Government Printing Office.

Congressional Budget Office. 1976. Five-Year Budget Projections: Fiscal Years 1978-1982. Washington DC: Congressional Budget Office. https://www.cbo.gov/sites/default/files/94th-congress-1975-1976/reports/1976_12_fiscal.pdf [retrieved 19/07/2022].

Cooper, J. Phillip. 1972. Development of the Monetary Sector: Prediction and Policy Analysis in the FRB-MIT-Penn Model. Lexington: Lexington Books.

Cortada, James W. 1993. Before the Computer: IBM, NCR, Burroughs, and Remington Rand and the Industry They Created, 1865-1956. Princeton: Princeton University Press.

Crandall, Richard L. 2002. Oral History Interview with Richard Crandall. Charles Babbage Institute. University of Minnesota Digital Conservancy. https://hdl.handle.net/11299/107231 [retrieved 17/07/2022].

Data Resources Inc. 1978a. EPS. Overview. Lexington: Data Resources Inc.

Data Resources Inc. 1978b. EPS. Summary. Concepts, Commands, Functions, and Operators. Lexington: Data Resources Inc.

Davidson, Bruce. 1969. DRI. Utility of Econometrics. The Boston Globe, July 12, 102.

den Butter, Frank A.G. and Mary S. Morgan (eds). 2000. Empirical Models and Policy-Making: Interactions and Institutions. London: Routledge.

De Vroey, Michel. 2016. A History of Macroeconomics. From Keynes to Lucas and Beyond. Cambridge: Cambridge University Press.

Eckstein, Otto E. 1964. A Theory of the Wage-Price Process in Modern Industry. The Review of Economic Studies, 31(4): 267-286.

Eckstein, Otto E. 1978. The Great Recession. With a Postscript on Stagflation. Amsterdam: North-Holland Publishing.

Eckstein, Otto E. 1983. The DRI Model of the U.S. Economy. New York: McGraw Hill.

Eckstein, Otto E., Edward W. Green, and Allen Sinai. 1974. The Data Resources Model: Uses, Structure, and Analysis of the U.S. Economy. International Economic Review, 15(3): 595-615.

Eckstein, Otto E. and Thomas A. Wilson. 1962. The Determination of Money Wages in American Industry. The Quarterly Journal of Economics, 76(3): 379-414.

Fisher, Franklin. 1965. Dynamic Structure and Estimation in Economy-Wide Econometric Models. In James S. Duesenberry, Gary S. Froom, Lawrence R. Klein, and Edwin Kuh (eds), The Brookings Quarterly Econometric Model of the United States. Chicago: Rand McNally & Co, Chapter 15.

Fisher, Franklin. 1970. A Correspondence Principle for Simultaneous Equation Models. Econometrica, 38(1): 73-92.

Friedman, Walter A. 2009. The Harvard Economic Service and the Problems of Forecasting. History of Political Economy, 41(1): 57-88.

Goutsmedt, Aurélien, Erich Pinzón-Fuchs, Matthieu Renault, and Francesco Sergi. 2019. Reacting to the Lucas Critique: The Keynesians’ Replies. History of Political Economy, 51(3): 535-556.

Gray, George T. and Ronald Q. Smith. 2003. Before the B5000: Burroughs Computers, 1951-1963. IEEE Annals of the History of Computing, 25(2): 50-61.

Gray, George T. and Ronald Q. Smith. 2009. After the B5000: Burroughs Third-Generation Computers 1964-1980. IEEE Annals of the History of Computing, 31(2): 44-55.

Greenberger, Martin, Matthew A. Crenson, and Brian L. Crissey. 1976. Models in the Policy Process. Public Decision Making in the Computer Era. New York: Russel Sage Foundation.

Haigh, Thomas. 2017. The History of UNIX in the History of Software. Cahiers d’histoire du Cnam, 7(7-8): 77-90.

Halsmayer, Verena. 2017. A Model to “Make Decisions and Take Actions”: Leif Johansen’s Multisector Growth Model, Computerized Macroeconomic Planning, and Resilient Infrastructures for Policymaking. History of Political Economy, 49(supp.): 158-186.

Hoover, Kevin D. 2012. Microfoundational Programs. In Pedro Garcia Duarte and Gilberto Tadeu Lima (eds.), Microfoundations Reconsidered: The Relationship of Micro and Macroeconomics in Historical Perspective. Cheltenham: Edward Elgar, 19-61.

Klein, Lawrence R. 1975. Research Contributions of the SSRC-Brookins Econometric Model Project. A Decade in Review. In Gary Fromm and Lawrence Klein (eds), The Brookings Model: Perspective and Recent Developments. Amsterdam: North-Holland, 11-29.

Klott, Gary. 1985. Chase Econometrics. The New York Times, October 5. https://www.nytimes.com/1985/05/10/business/chase-econometrics.html [retrieved 10/08/2022].

Kraemer, Kenneth L., Sigfried Dickhoven, Susan Fallows Tierney, and John Leslie King. 1987. Datawars: The Politics of Modeling in Federal Policymaking. New York: Columbia University Press.

Lenel, Laetitia. 2021. Searching for a Tide Table for Business: Interwar Conceptions of Statistical Inference in Business Forecasting. History of Political Economy, 53(Suppl.): 139-174.

Lodewijks, John. 1989. Macroeconometric Models and the Methodology of Macroeconomics. Journal of the History of Economic Thought [HES Bulletin], 11(1), 33-58.

Lucas, Robert E. 1976. Econometric Policy Evaluation: A Critique. Carnegie-Rochester Conference Series on Public Policy, 1:19-46.

Marron, Donald B. 1984. Otto Eckstein and the Founding of Data Resources, Inc. The Review of Economics and Statistics, 66(4): 537-542.

Morgan, Mary S. 1990. The History of Econometric Ideas. Cambridge: Cambridge University Press.

Naylor, Thomas H. 1981. Experience With Corporate Econometric Models: A Survey. Business Economics, 16(1): 79-83.

Padilla, Mariela. 2019. Donald B. Marron, Financier, Art Collector and Philanthropist, Dies at 85. The New York Times, August 12, https://www.nytimes.com/2019/12/08/us/donald-marron-dead.html [retrieved 03/05/2022].

Pyfer, James. 2018a. Project MAC. Encyclopedia Britannica. https://www.britannica.com/topic/Project-Mac [retrieved 30/05/2023].

Pyfer, James. 2018b. “The Virtues of Sharing”. Computer History Museum. https://www.computerhistory.org/revolution/mainframe-computers/7/178 [retrieved 30/04/2023].

Rancan, Antonella. 2019. Empirical Macroeconomics in a Policy Context: The Fed-MIT-Penn Model versus the St. Louis Model, 1965-75. History of Political Economy, 51(3): 449-470.

Rancan, Antonella. 2020. On the Rationalization of Italian Policy Making: At the Origin of the Bank of Italy First Macroeconometric Model (1964-1970). History of Economic Ideas, 28(2): 1-27.

Rancan, Antonella. 2022. The Place of the Phillips Curve in Macroeconometric Models: The Case of the Federal Reserve Board’s Model (1966-1980s). Journal of the History of Economic Thought, 44(2): 161-181.

Renfro, Charles. G. 2004. A Compendium of Existing Econometric Software Packages. Journal of Economic and Social Measurement, 29(1-3): 359-409.

Rowe, James. 1979. McGraw-Hill Inc. Will Purchase Data Resources for $103 Million. The Washington Post, July 16. https://www.washingtonpost.com/archive/business/1979/07/16/mcgraw-hill-inc-will-purchase-data-resources-for-103-million/26bc39ea-4f2d-4742-90fe-b7ef7313e342/ [retrieved 02/11/21].

Samuelson, Pamela. 1984. CONTU Revisited: The Case against Copyright Protection for Computer Programs in Machine-Readable Form. Duke Law Journal, 1984(4): 663-769.

Shenk, Timothy. 2016. Inventing the American Economy. PhD dissertation, Columbia University. https://doi.org/10.7916/D8NZ87N1.

Smith, Ron. 1994. The Macromodeling Industry: Structure, Conduct and Performance. In Stephen Hall (ed.), Applied Economic Forecasting Techniques. London: Harvester Wheatsheaf, 68-88.

Solow, Robert. 1985. Reflections on Macroeconomic Modeling; Confessions of a DRI Addict. Eastern Economic Journal, 11(1): 79-83.

Tinbergen, Jan. 1936. Kan hier te Lande, al dan Niet na Overheidsingrijpen, een Verbetering van de Binnenlandse Conjunctuur Intreden, ook zonder Verbetering van onze Exportpositie? In Prae-Adviezen voor de Vereeniging voor de Staatshuishoudkunde en de Statistiek. Den Haag: Nijhoff, 62-108. Translated into English and published in in Leo H. Klaassert, Leendert M. Koyck, and H. Johannes Witteveen (eds), Jan Tinbergen Selected Papers. North Holland: Amsterdam, 1959, 37-84.

Tinbergen, Jan. 1937. An Econometric Approach to Business Cycle Problems. Paris: Hermann.

Tinbergen, Jan. 1939. Statistical Testing of Business-Cycle Theories. Geneva: League of Nations Economic Intelligence Service.

Van Horn, Robert and Edward Nik-Khah (eds). 2015. Symposium on the Contributions of Business to Economics. History of Political Economy, 49(2): 164-314.

Waychoff, Richard, Lloyd Turner, Robert F. Rosin, Ralph W. Pearson, G. Clark Oliphint, F. Brad MacKenzie, Ray W. MacDonald, Duncan N. MacDonald, William D. Lonergan, Norman L. Kreuder, Paul D. King, Joseph T. Hootman, Erwin A. Hauck, John E. Hale, Bernard A. Galler, James Ford, Ray R. Eppert, Benjamin A. Dent, David M. Dahm, Bobby A. Creech, George A. Collins, Henri Berce, Robert S. Barton. 1986. Burroughs B 5000 Conference. Charles Babbage Institute and University of Minnesota Digital Conservancy. https://hdl.handle.net/11299/107105 [retrieved 17/07/2022].

Wayne, Leslie. 1984. Dismantling the Innovative D.R.I. New York Times, December 16, 4.

Wold, Herman. 1954. Causality and Econometrics. Econometrica, 22(2): 162-177.

Wold, Herman. 1960. A Generalization of Causal Chain Models. Econometrica, 28(2): 443-463.

Yost, Jeffrey R. 2013. Appropriation and Independence: BTM, Burroughs, and IBM at the Advent of the Computer Industry. IEEE Annals of the History of Computing, 35(4): 5-17.

Haut de page

Notes

1 The literature on the history of macroeconometric models in policymaking institutions includes: for the Federal Reserve Board of Governors, see Brayton et al. (1997), Backhouse and Cherrier (2019), Rancan (2019; 2022), and Acosta and Cherrier (2021); for the Social Science Research Council, see Lodewijks (1989) and Acosta and Pinzón-Fuchs (2019); for the Norwegian planning, Halsmayer (2017); for the Bank of Italy, Binotti and Ghiani (2008) and Rancan (2020); and the several country studies in den Butter and Morgan (2003) and in Bodkin et al., (1991, chap. 6-12). This latter book also recounts the development of several macroeconometric models in the US (the Brookings Institution model, the Bureau of Economic Analysis/US Department of Commerce model, the Wharton model, and others). See also Shenk (2016, chap. 5).

2 For a broader perspective on the relationships between economics and business until the interwar period, see Van Horn and Nik-Khah (2015). Postwar relations between economists and business have focused on few domains, with special emphasis on financial economics and financial activities (see e.g. Bernstein, 1993). Smith (1994) and den Butter and Morgan (1998) discuss some important aspects of the macroeconometric model-building industry in general.

3 See the interviews list at the end of this article. Unfortunately, it was not possible to conduct a more systematic round of interviews with more and diverse profiles. Since the company records are non-existent, this makes it difficult to make a comprehensive list of potentially relevant interviewees. Moreover, given that time has passed, our interviewees had lost contact with most of their colleague. Finally, several key figures (including Eckstein) are deceased.

4 Kraemer et al. (1987, 70-72); Davidson (1969); “Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 11, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University.

5 During his life, Marron would have prominent executive roles well-beyond DRI, as for instance the chair of UBS Americas. After his death in 2019, the New York Times summarized his main achievements: “[Marron] led some of Wall Street’s most powerful firms, played a shaping role in the expansion of the Museum of Modern Art in New York and delved into policy issues as a member of the Council on Foreign Relations and other organizations” (Padilla, 2019). In a 1976 booklet for the first public offering of DRI shares, Marron is described as also being “a director of the New York Stock Exchange, a governor of the Securities Industry Association and a member of the National Market Advisory Board” (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 18, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University).

6 “Johnson conversation with Otto Eckstein on May 04, 1964,” Lyndon B. Johnson Presidential Recordings, WH6405.02/3337, Miller Center, University of Virginia. Johnson was of course advised to appoint Eckstein by Walter Heller, then Chairman of the CEA (see “Johnson conversation with Walter Heller, on May 04, 1964,” Lyndon B. Johnson Presidential Recordings, WH6405.02/3337, Miller Center, University of Virginia).

7 For a comprehensive lists of contributors to DRI models through the years see Eckstein et al. (1974, 595), and Eckstein (1983, xiii). Feldstein and Thurow were Harvard professors at this time, and Jorgenson had just moved from Berkeley to Harvard in 1969. Green was a graduate student at the University of Pennsylvania in 1965 and became a Research Fellow in the late 1960s. Soon after his PhD, Thurow became a staff economist at Johnson’s CEA, brought by his advisor, Eckstein. In 1968 Thurow became an assistant professor of economics at MIT.

8 In the 1976 booklet for the first public offering of DRI shares, there is a list of the principal shareholders of DRI. Besides the economists just mentioned and also Fromm, Green, and Donald McLagan (DRI vice-president), there were also the economists F. Gerard Adams (an econometrician at University of Pennsylvania, who worked with Klein), Frederick Bamber (an economist with an MBA from University of Pennsylvania working at DRI), James A. Craig (formerly at Brookings, working at DRI), Anthony J. Finizza (PhD Chicago 1969, working at DRI in 1969-1977), Thomas L. Hall, and Philip K. Verleger (PhD MIT 1971, a Senior Staff Economist on President Ford’s Council of Economic Advisers and Director of the Office of Energy Policy at the US Treasury in President Carter’s administration)—this list also included the journalist and political commentator Bill Moyers, who was special assistant to President Johnson from 1963 to 1967, during which he was also the White House Press Secretary (1965-1967), and director of the Council on Foreign Relations (1967-1974) (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 23, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University).

9 Marron (1984) mentioned one million dollars as the initial capital of DRI, whilst Eckstein spoke of “a 7-digit figure” (Eckstein to Galbraith, 08/01/1973, cited in Collier, 2022).

10 On Harvard Business Barometers, see Morgan (1990, chap. 2), Armatte (1992), Friedman (2009), and Lenel (2021).

11 Davidson (1969) reported that “DRI’s client list [included] Morgan Guaranty Trust, Girard Trust (Philadelphia), Sears, Roebuck; the Channing funds, Aetna Life and Casualty, A.T.&T., Bank of New York, Federal National Mortgage Association., Connecticut General Life, Baden Securities, and Mitchell, Hutchins.”

12 The Congressional Budget and Impoundment Act of 1974 established the Congressional Budget Office and made the budget analysis its responsibility (no longer to be done by the Library of Congress).

13 McManus recalled that an “international macroeconomics group” would also work on collecting data for other countries and producing an international forecast. DRI had also activities and clients outside the US, notably in Europe (with an European office set in West Germany). Due to a lack of sources on this aspect of DRI’s activities, we restrain our analysis to DRI US-based activities.

14 We could not get a precise organizational chart of the company. But the 1976 prospectus for the public offering registered this: “As of June 30, 1976, DRI had 280 employees”: “90 in model development and forecasting [32%], 18 in marketing [7%], 79 in service consulting [28%], 47 in data processing and other related internal computer activities [17%] and 46 in general and administrative work [16%].” Among these employees 30 had a PhD degree, 40 MBA, 47 Master’s, and 84 Bachelor’s (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 17, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers).

15 For instance, “car sales” forecasts produced by the large-scale model would be used by the “steel group” to forecast steel demand (McManus, interview).

16 McManus also recalls that the macroeconomics group was the DRI group that had more systematic interaction with academics, receiving academics for seminars or having members co-authoring articles with academics.

17 For instance, in the 1970s the US House Budget Committee had consulted DRI for building their own macroeconometric model, which was then hosted at DRI computers (Committee on the Budget, 1975, 29). Other congressional activities relied on DRI data and forecasts (see e.g. Congressional Budget Office, 1976, 49). Kraemer et al. (1987, 84) argued that “[the] U.S. government became a major user of DRI services. An evaluation of the extent of use of macroeconomic modeling by federal agencies conducted by the General Accounting Office in 1978 indicated that the DRI model was being used in seventy-two activities where macroeconomic analysis is required.”

18 These numbers come from DRI’s Annual Reports of 1975 and 1976, available in box 27, folders “Data Resources Inc., 1972” and “Data Resources Inc., 1976-78,” Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University. The number of subscribers increased 225% from 1971 to 1975, with the average revenue per subscriber increasing from $15,625 in 1971 to $32,427 in 1975 (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 12, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers).

19 “Data Resources Inc., 1976 Annual Report,” p. 19, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University.

20 Mead Corp. was, at the time, a Fortune 500 company, with headquarters in Dayton, Ohio, and dozens of regional branches and offices. Mead sold different qualities and types of paper, cardboard, but also printers and other services (such as an innovative “online” law library).

21 Luke recounted that he had a previous experience, in a similar role, within NRC Corp., a large computer company. At Mead Corp., executives were mostly “paper people,” knowledgeable about paper business but often without any other specific qualification (building their way up from salesmen to executive roles). Nonetheless, these people were sensitive and aware about the reputational stakes of having a more “scientific” approach to planning. On the other hand, since Mead was a big corporation, headquarter management roles were filled with graduates of Ivy League business schools, most of which would have been aware of recent developments in econometrics and aware of the potential of these new methods for running businesses. Finally, there was also an important role of external consultants (Boston Consulting Group and later McKenzie), who were constantly around nudging things in this direction (Luke, interview).

22 Moreover, one had the possibility also to dig more precisely in how the forecast was built (look at the model, at the assumptions), instead of just having “the number” of the forecast. This allowed clients to make judgement and adjustments eventually when entering “the number” of the forecast in his own regressions (Luke, interview).

23 White also recalls: “For DRI, an awful lot of the attraction for the customers was access to the data. I mean, access to the model was important, but most of the customers wanted to do their own modeling using the data” (White, interview).

24 See Bodkin (1999) for a historical panorama on computation developments of macroeconometric models, including those of the 1970s.

25 According to Eckstein (1983, xi), the model was rebuilt “after the oil shock,” since “the behavior of the economy changed considerably in the mid-1970s and various aspects of supply and finance needed to be represented more explicitly.”

26 These simulation techniques were so important that Klein (1975, 21) claimed that:

The procedures that are now routine in computer dynamic simulation at virtually every econometric center of the world stem from the research done between 1964 and 1969 at the Brookings model project. This is not to claim that simulation techniques would not have been developed, but they might not have developed in their present form, and it is virtually certain that they would have been seriously delayed but for the technological breakthroughs of the project.

27 See Acosta and Rubin (2019) for an analysis of the banking sector in several macroeconometric models of the 1960s, including the MPS.

28 As mentioned by Goutsmedt et al. (2019), Eckstein was precisely one of those involved with pushing back against the Lucas Critique, on the ground of its weak empirical relevance (Eckstein, 1978). Eckstein, still in 1983, argued: “The basic rationale for continuing to use structural models in the face of the rational expectations criticism is this: changes in policy regime seem to have been among the minor sources of structural change of the economy and of forecasting error in the actual historical record” (Eckstein, 1983, xi-xii; 40-46).

29 Overall, the computer business at the turn of the 1950s was segmented with respect to these two different types of clients, business and scientific. Business clients would need computers to handle commercial data and operations (like inventories, sale numbers, accounting, producing paychecks, and so on), while scientific clients would be rather interested in performing computations for scientific or engineering purposes (Cortada, 1993). Government procurements, including for military and aerospace activities, were of course a third important market. Moreover, note that the market for computer was also geographically segmented: for instance, several European countries would have local computer firms, disputing the market to a few transnational firms like IBM.

30 The operating system of the B5000, the “Master Control Programm” (MCP), was written in ESPOL (a superset of ALGOL). For details on ALGOL and COBOL, see infra (Section 3).

31 It is usually mentioned in the history of programming that the consensus around ALGOL experienced a sudden stop with the 1968 NATO Conference on Software Engineering, when a part of the community turned their effort into developing an alternative language (UNIX; see e.g. Haigh, 2017).

32 “In addition to that, there was a guy, named Joe Neal, who was in charge of the networks for DRI. He would wander in and out [of the computer room]: he had an office for himself somewhere else in the building, but he would come in to adjust the equipment and make sure that networks were available for the customers” (White, interview). According to Greenberger et al. (1976, 219) and Kraemer et al. (1987, 73), DRI’s phone network for time-sharing was provided and maintained by another time-sharing company, Tymshare (cf. infra, 2.2).

33 We could not retrieve information about Druin’s background: he was an engineer, and likely trained in computer science in the Army (White, interview).

34 Working days at DRI were 16-hour days, but the night shift (3pm to 11pm at night) was a distinctive duty of field engineers and computer operators, although software engineers and a few economists could be working also during these late hours. Note that the computers were in operation 24/7, unless a physical problem required to shut the machines down (White, interview).

35 “Tapes were reused for a while. So they will eventually wear out. The actual magnetic material on the tape will basically fall off … Sometimes the tape drive would chew the tape rather than flow it freely. It was always exciting when that happened, particularly if you were in the middle of a 2400-foot reel: you had to sit there and put the crinkled and ruined tape out. And then the field engineers had to go there and figure out why it got chewed up. … We might run out of tape, and I will have to go find some. … I remember once or twice we basically went 95% of the way through the backup and then there were no more tapes.” (White, interview)

36 IBM cards with data were regrouped in these “boxes,” containing around 2000 cards each; each card “would have 5-8 columns of data,” to form a time series.

37 Christopher Strachey and John McCarthy are today credited with pioneering contributions, in the end of the 1950s, converging toward the idea of “time-sharing”: the former worked in the UK, at the National Research and Development Corporation (https://history.computer.org/pioneers/strachey.html, retrieved 17/07/2022); the latter worked at the MIT (http://jmc.stanford.edu/computing-science/timesharing.html, retrieved 17/07/2022). The first working “Compatible Time-Sharing System” (CTSS) software was demonstrated in 1961 at MIT, which was then expanded and developed by MIT’s Project on Mathematics and Computation (MAC) founded in 1963 with the support of the Defense Department and the National Science Foundation (and operating until 1973; Pyfer, 2018a). In that same year the Dartmouth College unveiled the Dartmouth Time-Sharing System (DTSS), using the BASIC programming to make it easy for beginners to use (Pyfer, 2018b). This inaugurated a new era: “the first common multi-user systems, with dozens of people online at the same time. Typical 1960s users are a mix of businesspeople, bank employees, students and researchers, and military personnel” (https://www.computerhistory.org/timeline/, retrieved 19/02/2022).

38 Obviously, economists working at DRI would also access the computer through the same time-sharing system and have access to similar capabilities. Every economist at DRI had a paper terminal in her office, connected to the computers (McManus, interview). The time-sharing system was set up in such a way that DRI economists’ would always have priority access to computational capacities (White, interview).

39 However, when looking at the list of computer systems supported by the largest time-sharing companies (Auerbach Publisher Inc., 1972), Burroughs appears as relatively small with respect to its competitors.

40 Unfortunately, we could not retrieve any documentation or archival evidence about the functioning or origins of these three programs, except for EPL.

41 The 1976 booklet for the first public offering of DRI shares listed two other software developed by DRI: PATSY and SURVEY, but we were unable to find more information about them (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 15, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University).

42 Although apparently not having direct connection with software development at DRI, one major outcome of software development at MIT was TROLL, developed initially in 1966 under the lead of Edwin Kuh, with the contribution of J. Phillip Cooper and Mark Eisner (Renfro, 2004, 388; Cooper, interview). Hall also contributed to TROLL in 1967, integrating the features for time-series econometrics and analytical derivatives into TROLL (Hall, personal communication). From the 1970s, TROLL rapidly became one of the major software for macroeconometric modeling (distributed by INTEX Inc.), still representing today the major commercial software for large-scale modeling, especially in policymaking institutions.

43 TSP was further developed in the 1960s, at MIT by J. Phillip Cooper, Robert Gordon, Charles Bischoff, and Richard Sutch, and, at Princeton, by Ray Fair. In the 1970s, the development of TSP was taken on by two different groups. On the one hand, Bronwyn Hall (Robert Hall’s wife), in Berkeley, where TSP was distributed commercially and became one of the most widespread econometric software (Renfro, 2004, 390), then giving also birth to a PC version (Micro-TSP) and, later on, to EViews (developed by David Lilien; Hall, personal communication; Renfro, 2004, 367). On the other hand, Cooper developed TSP into his Econometric Software Package (ESP), which “incorporated autocorrelation correction, 3-stage least squares estimation, simultaneous, non-linear model simulation, Box-Jenkins ARIMA time-series modeling, and other capabilities” (Cooper, interview); ESP was also adapted in 1982 to work on IBM personal computers.

44 Lacey will remain head of software engineers at DRI until 1978, when he left to start his own company. He was replaced by Craig Jackson (McManus, interview). Unfortunately, we were unable to ascertain Lacey’s exact background, training, and career after DRI.

45 Renfro (2004, 407) also mentions the “important design contribution” to EPS made by Edward Green in close cooperation with Lacey—which was also confirmed by White (interview); however, we were unable to assess its exact content.

46 “Data Resources Inc., 1976 Annual Report,” p. 3, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University.

47 We were unable to ascertain the background and respective role of all DRI programmers. However, relying on White’s and McManus’s memories, one could say with certainty that, conversely to computer operators, software engineering roles were accessible also to those with an initial training in economics but with an elective training in programming or an interest in programming. This was the case of McManus, of Greg George, and possibly of Marsha Schiff (who was in charge of the relations between the programming group and the economist groups at DRI).

48 White also recalls that ALGOL entailed “some specific efficiency issues” in the computations related to this type of model, some hinting to the fact that it might not have been the best-suited language for macroeconometric modeling. Indeed, we are unable to find any examples of any large-scale macroeconometric model of the time, besides DRI’s, that was written in ALGOL. FORTRAN seems to have been by far the dominant language for macroeconomists. But when we look at the corporate users of econometric models, Naylor’s (1981) survey brings FORTRAN as the most used language (35%), but followed closely by EPS (33.3%).

49 Luke compares DRI services accessible to EPS to modern-day FRED (Federal Reserve Economic Data, the online database maintained by the Federal Reserve Bank of Sant Louis).

50 The due payment of an original loan of $1,500,000 was then $2,805,000 (“Preliminary Prospectus, Data Resources Inc. Common Stock, William Blair & Company,” Aug. 31 1976, p. 4, folder “Data Resources Inc., 1976-78,” box 27, Marc L. Nerlove papers, David M. Rubenstein Rare Book & Manuscript Library, Duke University).

51 McGraw-Hill was already one of the “big three” American educational publishers, featuring other divisions, including legal and medical publishing, films, financial services (notably with the acquisition of Standard & Poor’s in 1966), and computer software distribution (Bashin, 2011). McGraw-Hill also acquired, in 1985, Chase Econometrics, henceforth controlling two of the three major econometric consulting firms of the 1970s (Klott, 1985).

52 Actually, the acquisition by McGraw-Hill benefited all workers of DRI, as well as those macroeconomists who had been receiving DRI’s shares as a compensation for their technical assistance (cf. supra). However, it is probably safe to say that Eckstein was the first economist who became rich thanks to macroeconometric modeling!

53 Cooper had been a PhD student of Franco Modigliani at MIT, and he had been working on monetary macroeconomics and contributing to the MPS model. He completed his PhD in 1972 (Cooper, 1972). Cooper worked at Chicago Business School (1969-1972) before starting a career as an Executive, first within a startup software company, originally utilizing TROLL. Cooper was involved in the early developments of econometric software at MIT, especially by contributing to the development of TROLL’s econometric capabilities; he also developed his own software (called ‘ESP’), starting from Hall’s TSP code, which Cooper commercialized to universities and companies since the early 1970s (Cooper, interview).

54 Unisys was established in 1986 from the merge of Burroughs and Sperry Corporation—a merge initiated by Burroughs as an hostile takeover. Sperry was a large industrial conglomerate, whose computer branch has been producing, since the 1950s, the UNIVAC mainframes series. Unisys was thus, at the time, the second largest US computer corporation.

55 https://ihsmarkit.com/btp/global-insight-economics-country-risk.html [retrieved 10/08/2022].

56 Cooper (interview) also remembers that, when he arrived at DRI around 1985, DRI’s macroeconometric model was not considered as a major problem to the company.

57 In 1981 Burroughs even lost “two lawsuits by B700 and B800 systems users who said that the computers’ capabilities had been mis-represented” (Gray and Smith, 2009, 54).

58 To which, we should add, the key role of business executives and financiers, who, like Marron (and, later, Cooper) in the case of DRI, provided both the financial support and the management skills necessary for get this business going, expanding, and eventually restructured to a new reality.

Haut de page

Table des illustrations

Titre Figure 1. Burroughs 5500 Computer
Crédits From Burroughs’ brochure. Computer History Musem’s collection (https://www.computerhistory.org/​revolution/​mainframe-computers/​7/​166/​674, retrieved 03/11/23)
URL http://journals.openedition.org/oeconomia/docannexe/image/15635/img-1.jpg
Fichier image/jpeg, 28k
Titre Figure 2. Econometric Commands and Functions in EPS
Légende Source: DRI (1978b, 4).
URL http://journals.openedition.org/oeconomia/docannexe/image/15635/img-2.jpg
Fichier image/jpeg, 171k
Haut de page

Pour citer cet article

Référence papier

Pedro G. Duarte et Francesco Sergi, « Computer Operators and Software Engineers at Data Resources Inc. »Œconomia, 13-3 | 2023, 723-768.

Référence électronique

Pedro G. Duarte et Francesco Sergi, « Computer Operators and Software Engineers at Data Resources Inc. »Œconomia [En ligne], 13-3 | 2023, mis en ligne le 01 septembre 2023, consulté le 07 février 2026. URL : http://journals.openedition.org/oeconomia/15635 ; DOI : https://doi.org/10.4000/oeconomia.15635

Haut de page

Auteurs

Pedro G. Duarte

Insper Institute of Education and Research, Brazil. pedro.duarte@insper.edu.br

Francesco Sergi

Université Paris Est Créteil, France. francesco.sergi@u-pec.fr

Articles du même auteur

Haut de page

Droits d’auteur

CC-BY-NC-ND-4.0

Le texte seul est utilisable sous licence CC BY-NC-ND 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont susceptibles d’être soumis à des autorisations d’usage spécifiques.

Haut de page
Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search