I. INTRODUCTION
Academic literature on postwar macroeconomics refers to Milton Friedman’s and Edmund Phelps’s expectational Phillips curve as a turning point. The central role played by expectations in explaining wage change and the acceleration of inflation, along with its implications for aggregate demand policies, are told as a challenge to which Keynesian economists surrendered in the 1970s (Gordon Reference Gordon2011, p. 20; Hall Reference Hall1976; King Reference King2008, pp. 331–332; Okun Reference Okun1980, p. 818; Mankiw and Reis Reference Mankiw and Reis2018, p. 82; Santomero and Seater Reference Santomero and Seater1978, p. 515; and Snowdon and Vane Reference Snowdon and Vane2005, p. 140, among others).
James Forder (Reference Forder2014) investigated the history of the Phillips curve, arguing that this standard narrative is a myth—a myth he demolished by carefully documenting the very different types of empirical, theoretical, and policy-oriented works associated with the label “Phillips curve.” Forder shows that Friedman’s 1967 Presidential Address was not path-breaking and that the post-1968 literature was not devoted to responding to his challenge. In particular, empirical studies after Friedman (Reference Friedman1968a) did not merely concentrate on establishing whether the long-run Phillips curve was vertical and did not have the expectational argument as its principal objective.
Forder examines an extensive academic and non-academic literature but devotes little attention to macroeconometric models, how they embedded the wage and price sector within a larger system of equations, and whether the macroeconometric literature was affected by Friedman’s and Phelps’s accelerationist hypothesis. According to Forder, macroeconometric models were short-run forecasting models, thus little involved in the long-run Phillips curve debate. Moreover, because they were conceived as “works in progress” that continuously developed, it would be hard to grasp how the Phillips curve relationships were approached (Forder Reference Forder2014, p. 99).Footnote 1
To investigate whether and how macroeconometric model builders confronted the academic debate on the Phillips curve, on wage change, expectations, and the policy trade-off is the object of this article. I concentrate on the first Federal Reserve Board’s model, the so-called Federal Reserve-MIT-University of Pennsylvania (henceforth FMP-MPS) model, resulting from a collaboration between economists from the Division of Research and Statistics at the Board, under Frank de Leeuw’s supervision, and academic economists (especially PhD students), under the directorship of Franco Modigliani and Albert Ando.Footnote 2 The collaboration formally lasted from 1966 to 1970 with the delivery of the model to the Federal Reserve Board of Governors. Since then, model builders from the Board and from the academia went on to update and modify the model separately.Footnote 3
The close interactions of the two teams were likely to open the model construction to academic studies and debates, trying to combine theoretical achievements with econometric practice, policy simulations, and forecasts. Indeed, its first version closely followed the macroeconomics mainstreams, i.e., Keynesian economics and structural econometric modeling.Footnote 4 Nevertheless, from the late 1970s there was no longer such a close parallelism between the evolution of the discipline and the macroeconomics practiced in public institutions.Footnote 5 By looking at the model updating and respecification, it becomes apparent that major changes in the academic research lines did not affect the model-building agenda at the Division of Research and Statistics. Flint Brayton et al. (Reference Brayton, Levin, Tryon and Williams1997), from the Federal Reserve Board, explain the model modifications over the 1970s and 1980s as the result of economic events (oil shocks), institutional reforms (the freeing of exchange rates in 1973), and, to a lesser extent, academic developments. However, they also acknowledge that the model specification responded to “shifts in the consensus view,” arguing, “As a new consensus developed in the late 1980s and early 1990s, the Federal Reserve models were redesigned to incorporate, at least partially, the evolving new paradigm” (Brayton et al. Reference Brayton, Levin, Tryon and Williams1997, pp. 3, 5). In other words, model builders at the Board were aware of the evolution of the discipline, but their research agenda had different timing and paths.Footnote 6 The FMP-MPS model was maintained in operation until the mid-1990s, and the wage and price sector, considered the “culprit” of the rational expectations revolution, was little modified and regarded as one of the more stable parts of the model (Brayton et al. Reference Brayton, Levin, Tryon and Williams1997, p. 3; see Ando and Brayton Reference Ando, Brayton and Cross1995). Under this perspective, the paper is a case study of the changing traction of academic disputes in non-academic economic spheres by looking at the way an ambiguous, yet much-debated, object called the Phillips curve was used by the FMP-MPS modelers, some being more academically oriented and others more policy oriented. Initially they were all involved in academic disputes, but from the 1970s onwards discussions on what to model at the Federal Reserve Board appear to follow a different, more distant, and independent dynamics on their own.Footnote 7
II. WAGE AND UNEMPLOYMENT RELATIONSHIPS: THE STRUGGLE WITH THE PHILLIPS CURVE
The FMP-MPS model was designed to qualitatively and quantitatively analyze the channels through which monetary and fiscal actions influenced aggregate demand. Different research groups formed by academic economists and from the Federal Reserve Board staff worked on separate blocks of the macroeconometric model. While the latter especially concentrated on policy simulations and forecasting, academic economists were also interested in the model theoretical underpinning and consistency.Footnote 8 In his interviews with Robert Hetzel, James Pierce recalls that the model forecasts were needed to complete the Greenbook, which represented the basis for policy discussions at the Federal Open Market Committee (FOMC) meetings. However, he also acknowledged that at the Board they were all aware of the model’s shortcomings in forecasting and that a great deal of judgment and model adjustments were always required (Pierce Reference Pierce1995, pp. 4–6).Footnote 9
As for the academic team, Ando and Modigliani insisted on the connections between their macroeconomic analysis and the model structural relationships, arguing that the FMP-MPS model was not conceived as only a short-run forecasting model. It was also concerned with an understanding of the long-run equilibrium properties with the steady state that should act as a reference point for the system’s dynamics.Footnote 10 Discussing an early version at the 1968 meeting of the Econometric Society, Modigliani explained the reasons for the building of another macroeconometric model, stressing its distinctive purposes: “The focus of our research … differs in several respects from any of the existing models. In the first place our main concern was not with forecasting but rather with analyzing the mechanisms through which policy tools, and particularly the tools of monetary policy, affected economic activity, and the nature and magnitude of the lags involved in the process” (Reference Modigliani1968, Box RW 15, p. 1). Therefore, Modigliani added, the “specification of the model should be dictated as closely as possible by the existing body of economic theory, blended with relevant institutional considerations, rather than by the goal of obtaining the closest possible fit” (Reference Modigliani1968, Box RW 15, p. 2, MP).Footnote 11 Ando also suggested that the model could have been an important teaching device for graduate courses to reduce the gap between economic theory and practice (undated, Box RW 15, p. 1, MP).Footnote 12 He explained that the model performance should not be based on its forecasts but on its ability to catch the system behavior “over a fairly long period” (undated, Box RW 15, p. 11, MP).
The model was structured along Keynesian lines with aggregate demand and money supply that influenced real output by way of short- and long-run interest rates determined by two term structure equations. Underlying the short-run dynamics there was a modified version of Robert Solow’s growth model (see Modigliani Reference Modigliani1975b; Szenberg and Ramrattan Reference Szenberg and Ramrattan2008, ch. 6; Rancan Reference Rancan2020). As for the construction of the wage and price sector, different groups of modelers from the Board and the MIT worked on it in different phases of the project, following similar lines of reasoning. It was finally based on a “standard” Phillips curve relating the rate of wage change to past inflation and unemployment. Wage changes were transmitted on prices through a markup over costs in the oligopolistic commodities markets.
The wage and price block of equations was introduced endogenously only in the last version of the model after several attempts. Difficulties in assessing the Phillips curve’s theoretical basis and its consistency with the model as a whole, along with empirical fit and data problems, were behind these attempts. Most of the attention concentrated on a proper specification of the relationship between wage change and unemployment, which should be able to account for the structural modifications of the labor market, particularly its segmentation and an appropriate measure of the unemployment variable.
Alfred P. Tella and Nelson J. Tinsley, both from the research staff of the Board of Governors, worked on a first specification of the wage and price sector, a version of which they discussed at the 1967 annual meeting of the American Statistical Association. Their focal points were the measure of unemployment adjusted for hidden unemployment, and of potential output, with a distinction between full capacity output and full employment output (Tella and Tinsley Reference Tella and Tinsley1968, pp. 669–670).Footnote 13 However, their specification was probably unsatisfactory since in the first published report on the FMP-MPS model, tests and simulations were carried out with the wage and price sector still exogenous (see de Leeuw and Gramlich Reference de Leeuw and Gramlich1968). In a 1968 memorandum it was claimed that its proper specification was the main gap in the last version of the macroeconometric model. After a description of the sector’s main features, and particularly of the recursive property from prices to wages, it was concluded that despite all the econometric attention, the wage equation was still the weakest: “We are still not clear as to whose behavior is being represented and what the basis is for some of the variables in the equation” (Anonymous, April 18, 1968, p. 6, MP).
Modigliani, as well, presenting a preliminary version of the FMP-MPS model, only mentioned the wage equation. He focused to a large extent on the model’s theoretical structure and long-run steady state properties. He analyzed the price equation for a given wage and introduced only verbally and very briefly the wage equation, stating, without further details, that the basic model of wage determination has “elements in common with the so-called Phillips curve, in that the rate of change of money wages is related, in part, to unemployment. However, it differs from it in a number of significant respects” (Reference Modigliani1968, Box RW 15, p. 45, MP). The model had an equation measuring the “speed of adjustment” of prices to wages that, Modigliani explained, might depend on the extent of pressure in the labor market: “when unemployment is very low, employers are led to bid up wages in the effort to secure scarce labor, then prices may adjust faster to the change of wages” (Reference Modigliani1968, Box RW 15, p. 45, MP). He concluded that partly for data problems, the wage and price sector had not yet reached a “satisfactory state,” and, therefore, to close the model and for policy simulations, the wage rate was taken as exogenous (Reference Modigliani1968, Box RW 15, pp. 44–45, MP).
Soon after Tella and Tinsley, George de Menil, a PhD student at MIT, and Jared Enzler (from the Board) began to work on this block of equations. It was their model, to which I return below, that was embedded in Ando and Modigliani’s version of the FMP-MPS model discussed at the American Economic Association meeting of December 1968. They were, however, still cautious about it, arguing that it was in progress although it tracked the data “surprisingly well” (Ando and Modigliani Reference Ando and Modigliani1969, p. 311). Indeed, a few months earlier, in a September 1968 report, Ando claimed that they had not yet been able to put together a wage equation that was acceptable:
[W]e are taking the level of wages as exogenous for the time being. But it is absolutely essential that the determination of the wage level be made endogenous to the system, in order that the system be considered an adequate description of the U.S. economy. Furthermore, it is difficult to justify the procedure in which the time pattern of wages is assumed to be the same under different sets of monetary and fiscal policies. Franco has worked on the estimation of the wage equation rather hard this summer, and he expected to continue to do so until we have resolved this vexing problem, with the assistance of George de Menil of Boston College. (Reference Modigliani1968, Box CO1, p. 5, MP)
At the Board of Governors meeting of May 1969, de Menil discussed a paper on the determinants of the price level in which the wage and price equations were not simulated together (de Menil Reference de Menil1969, Box RW 14, MP). Only in the October 1969 report could Ando finally write that “Franco and de Menil have worked on the specification and estimation of a wage equation … [which explains] the data reasonably well.… Their efforts appear to be beginning to pay off, and hopefully by early 1970, we will have a fairly satisfactory wage equation that can be incorporated into the model” (Ando Reference Ando1969, Box CO1, p. 2, MP).
Ando and Modigliani (Reference Ando and Modigliani1969) explained that the wage equation was built in the tradition of Phillips (Reference Phillips1958), George Perry (Reference Perry1966), and Norman Simler and Alfred Tella (Reference Simler and Tella1968), whose focus was the structure of the labor market. They especially concentrated on the composition of the labor force and its modifications. Simler and Tella showed that to be a good proxy of the excess demand, the reported unemployment must be adjusted by considering hidden unemployment—that is, workers discouraged from participating in the labor force—whose size varied directly with the level of employment (Reference Simler and Tella1968, pp. 3–4; see also Santomero and Seater Reference Santomero and Seater1978, p. 507). Simler, Tella, and Perry were concerned with an understanding of whether the economy was becoming more inflation-prone, looking at the labor market behavior.
Even if model builders were working on the model specification when Phelps (Reference Phelps1967) and Friedman (Reference Friedman1968a) put forward the accelerationist hypothesis, and the wage equation had since the beginning lagged inflation among the explanatory variables, the role played by change in price expectations was not addressed. The rate of change of prices was apparently read as the influence of changes in the cost of living on the wage determination process. It is not clear whether it was also understood in terms of expectations of future price change. In any case, it was not the focal point of the model specification.Footnote 14 To have a proper modeling of the labor market behavior was seen as being as important as modeling price expectations for an understanding of the wage determination process and, since the mid-1960s, to explain the worsening of the trade-off between inflation and unemployment.
III. THE PHILLIPS CURVE TRADE-OFF
In 1970 the Federal Reserve Board and the Social Science Research Council (SSRC) sponsored a conference on “The Econometrics of Price Determination,” devoted to understanding the inflation process through a survey of the most well-known macroeconometric models of that time: the OBE, the Wharton model, the Brookings model, the FMP-MPS model, and the St. Louis model. It was, therefore, representative of the state of art on the modeling of the wage and price sector of macroeconometric models. Of all of them, only the St. Louis model embedded a long-run vertical Phillips curve, arguing that its shape was not merely an “academic question” because of its relevant policy implications (Andersen and Carlson Reference Andersen, Carlson and Eckstein1972, p. 167), and Robert Lucas applied rational expectations to test Friedman’s natural rate of unemployment (Lucas Reference Lucas and Eckstein1972). In his overview of the conference results, James Tobin referred only briefly to Lucas’s now famous paper and emphasized instead the “widely accepted” explanation of inflation that emerged from the conference, according to which inflation was the result of interactions between a labor market approaching full employment and oligopolistic price setting. Tobin concluded that even if a “pragmatist” might agree in principle with the natural rate hypothesis, he also believed that “a trade-off does exist for policy-makers as well as statisticians” (Reference Lucas and Eckstein1972, p. 13).
Discussing the FMP-MPS wage and price equations, de Menil and Enzler (Reference de Menil and Enzler1972) first pointed out the importance of interactions among real variables, prices, and wages in macroeconometric models, which imply a trade-off between unemployment and inflation that is valid in both the short and long runs. They faced the verticality issue on empirical ground on the basis of the price coefficient estimates. In a footnote, they also criticized Lucas’s dismissal of adaptive expectations and the conclusion that inflation increases output only when it is unexpected.
Their wage equation was quite “standard” with money wages that responded slowly to a disequilibrium in the labor market, and to the lagged rate of change of prices:
$$ \frac{PL_t-{PL}_{t-1}}{PL_{t-1}}={d}_0+{d}_1\;{\left[\frac{LF_t-{LE}_t}{\left({LF}_t+{LA}_t\right)}\right]}^{-1}+{d}_2\;\frac{PCON_{t-1}-{PCON}_{t-2}}{PCON_{t-2}} $$
PL corresponds to wage rate, LF to labor force, LE to employment, LA to armed force, PCON to price of consumption; see de Menil and Enzler (Reference de Menil and Enzler1972, p. 280).Footnote 15
Changes in the money wage rate were transmitted to prices through an oligopolistic setting price model, in which labor costs played a major role; de Menil and Enzler mentioned further refinements of the model in the direction of additional disaggregation of the labor market into unionized and non-unionized, thus focusing on the bargaining aspects of wage determination.Footnote 16 To them, in the case of separation between union and non-union labor markets, using two different wage equations, the implications for the long-run Phillips curve depends on the nature of the spillover from one sector to another (for a similar reasoning, also see Modigliani Reference Modigliani1968, Box RW 15, MP).Footnote 17 Behind their analysis, there seems to be the “lubrication” argument that, according to Forder, provided a rationalization of the persistence (long-run) trade-off, put forward before Friedman’s and Phelps’s challenge (see Forder Reference Forder2010, pp. 335–336; and Reference Forder2014, pp. 76–80). As Forder explains, by considering labor markets covering different employment situations, some in excess and some in short labor supply, and the different speeds by which money wages fall and rise, it was demonstrated that the average wages tend to rise and there can be inflation even if there is unemployment, independently from the expectations argument. Inflation “lubricates” the labor market and transforms downward nominal wage rigidity into downward real wage rigidity.
However, de Menil and Enzler did not discuss in depth the rationale behind the Phillips curve trade-off. They simply acknowledged that the effects of price change on the full model multipliers depended upon the level of unemployment, with the Phillips curve becoming steeper for low rates, and stressed that in highly inflationary periods, such as 1966 to 1968, the feedback from prices to real variables leads to a marked increase of traditional multipliers with the economy becoming highly unstable.Footnote 18 They were also aware that the trade-off was not “easy to apply” or “infallible,” arguing that the analysis of the Phillips curve’s dynamics showed the difficulties a policy maker faced if he wanted “to avail himself of the choices the curve offers” since “there is a long tail to the lag distribution for the effect of the unemployment rate on the rate of inflation,” and because “other variables besides unemployment … influence the rate of inflation” (de Menil and Enzler Reference de Menil and Enzler1972, p. 301). They concluded that they were “pessimistic”: keeping inflation within moderate bounds would have required a relatively large amount of unemployment unless the Phillips curve was shifted to the left.
Even if Friedman’s accelerationist hypothesis was on the Board of Governors’ table since the mid-1960s, model builders did not ascribe any role to a change in expectations. They concentrated on the structure of the labor market also for a possible explanation of the trade-off and its worsening.
The Accelerationist Argument at the Federal Reserve Board
During the 1965 and 1966 Academic Consultant meetings, Friedman anticipated most of the contents of his Presidential Address, criticizing the Federal Reserve discretionary policy through his emphasis on the inflationary role of price expectations.Footnote 19 At the October 1965 meeting Friedman presented a memorandum entitled “The Lessons of U.S. Monetary History and Their Bearing with Current Monetary Policy” (Friedman Reference Friedman1965, Box PS 1, MP), outlining the empirical results of his study with Anna Schwartz (Friedman and Schwartz Reference Friedman and Schwartz1963). His departure point was the high correlation between the rate of change of money and nominal income, arguing that the actual rate of money growth was not indefinitely sustainable without rises in price. The beginning of inflation would have changed expectations with nominal interest rates that would also start to rise, overcoming the so-called liquidity effect of rapid money growth, leading the economy to a recession. As in his Presidential Address, Friedman pointed out that the expansive effects of monetary policy were possible only with an unexpected inflation. Therefore, to maintain the actual rate of economic growth required a higher and higher rate of money increases, which intensified the pressure on prices.
A year later, at the June 1966 meeting, Friedman went back to this reasoning, discussing a paper on “Current Monetary Policy” (commented on by Paul Samuelson) in which he explicitly referred to the Phillips curve. He considered the memorandum an “addendum” to his 1965 analysis, arguing that the acceleration of money growth had finally led to a change of price expectations with the rise of money supply that was now accompanied by a rise of interest rates.Footnote 20 Warning again about the risk of a recession by continuing the expansive monetary policy, he relied on the Phillips curve trade-off: “In my opinion, there is no perpetual trade-off between inflation and unemployment. The trade-off is between acceleration of inflation and unemployment, which means that the real trade-off is between unemployment now and unemployment later” (June 15, 1966, Box PS 1, p. 8, MP; also see Friedman Reference Friedman1968b).
Friedman’s claims were dismissed by a laconic comment by Samuelson, who noticed only that 1966 was a year of demand-pull inflation and that, despite the “Chicago school insistence,” the monetary policy over most of 1966 was restrictive rather than expansive. However, while suggesting an expansive monetary policy, Samuelson stressed that it must be accompanied by a restrictive fiscal policy to keep inflation under control.Footnote 21
Friedman did not return to the instability of the inflation and unemployment trade-off and the expectational argument during the subsequent Academic Consultant meetings. In his 1971 (June) memorandum, he concentrated instead on the relationships between money growth and interest rates to emphasize that his previous analysis was confirmed by data. His focal point was that interest rates could not be a guide of monetary policy since the Federal Reserve did not control either the real interest rates or the nominal one in the long-run. They were a by-product of monetary policy and of other forces, and could be “neither an ultimate end nor an effective means” (Friedman Reference Friedman1971, Box PS 1, MP).Footnote 22 As it appears, the verticality issue was not central at the Board meetings, also later all over the 1970s, as it was not for model builders. Among the few exceptions were Robert Gordon’s memorandum “The Inflation and Unemployment Dilemma” he discussed at the December 1972 meeting, where he still concentrated on the modification of labor force composition as a cause for the worsening of the trade-off; and Modigliani’s 1975 memorandum in which he introduced the non-inflationary rate of unemployment concept (NIRU), which he further elaborated in a Brookings paper with Lucas Papademos (Reference Modigliani and Papademos1975).
IV. DID THE LONG-RUN VERTICAL PHILLIPS CURVE MATTER?
In a paper discussed at the Federal Reserve Financial Analysis Committee on the supposed Keynesian bias of large-scale structural models, Edward Gramlich and James Pierce (Reference Gramlich and Pierce1970) analyzed extensively the long-run behavior of the FMP-MPS model, in particular the stability of the unemployment and inflation trade-off they considered to be the most important point of contention between monetarists and Keynesians—a contention, however, they minimized. As did de Menil and Enzler (Reference de Menil and Enzler1972), Gramlich and Pierce approached the long-run shape of the Phillips curve as an empirical issue to be solved by estimating the price coefficient. They also emphasized that whatever the empirical results, they would have not undermined the model theoretical structure or its policy implications. Indeed, the FMP-MPS model combined the Keynesian and the long-run neoclassical theories, with the speed by which initial movements in real income were eliminated depending on how close the economy was to full employment. According to them, the main difference with monetarists was not the vertical Phillips curve but that for monetarists inflation was unstable at any but the natural rate, while Keynesians “would say that the economy can trade-off controllable inflation and unemployment along a stable curve” (Gramlich and Pierce Reference Gramlich and Pierce1970, p. 10). They associated the vertical Phillips curve with hyperinflationary economies, and the non-vertical Phillips curve with economies that have remained in a stable price range over most of the estimated period or that have not experienced prolonged deviations from the “natural rate of unemployment,” with no further discussion of this point (Reference Gramlich and Pierce1970, p. 12).Footnote 23 They concluded quite ambiguously that it was not “appropriate” to constrain the long-run value of the price coefficient to unity since their aim was to analyze the adjustment process: “Even if one agrees in principle that in the very long-run [the coefficient] is unity, we still may not want to impose this condition to study the questions being asked of the large scale models.” Nonetheless, they also added that “the theoretical likelihood of at least a ‘natural range’ of unemployment rate should be used in evaluating policy prescriptions. Attempts to reduce unemployment perpetually below its equilibrium value may be self-defeating in the long-run unless attempts are made to structurally reduce the natural unemployment rate” (Reference Gramlich and Pierce1970, pp. 12–13).Footnote 24 A similar argument had been already put forward by Perry (Reference Perry1966), who considered the “degree to which living costs directly affect wages” among the most interesting questions to be analyzed rather than constraining the coefficient to one (pp. 22–23; quoted from Forder Reference Forder2014, p. 61).
Ando, as well, in his notes for the monograph on the macroeconometric model, explained that the notion of a natural rate of unemployment was rejected since it was “unconvincing,” and “mainly because the purpose of formulating the long-run model” was to “establish a reference point for later short-run dynamic analysis, and the natural rate hypothesis appears unsuitable for this purpose on theoretical and empirical ground” (Ando Reference Ando1972, Box RW 14, p. 16, MP). According to Ando, the rate of change of prices could be accepted as an argument of the wage equation as long as the coefficient was substantially less than unity. He briefly mentioned the dispute in the profession on the validity of the Phillips curve trade-off, arguing that its adequacy to describe the money wage behavior was not an issue. The Phillips curve, Ando pointed out, was not “interwoven with other parts of the MPS model” and, therefore, it could be replaced with “almost any theory of the determination of money wage rates other than the marginal productivity considerations, and the remaining parts of the MPS model will accommodate the change quite easily” (Ando Reference Ando1972, Box RW 14, p. 36, MP). According to him, the most important point about the Phillips curve trade-off debate was to establish the only indirect mechanism through which changes in money supply influenced the level of nominal wages and prices: through changes in interest rates, the cost of capital, output, and unemployment; that is, by relying on the excess demand rather than changes in price expectations (Reference Ando1972, Box RW 14, p. 36, MP).Footnote 25
At the 1972 symposium “Macroeconometric Models Comparison,” Ando still referred to the long-run Phillips curve as an open, purely empirical question (Ando Reference Ando1974, p. 544). Modigliani, as well, in a paper presented at a 1972 SSRC conference, discussed de Menil and Enzler’s model based on the “conventional Phillips mechanism” because the coefficient was “a good deal less than unity” (Modigliani Reference Modigliani1975b, p. 254). And, at the February 1975 Academic Consultant meeting and in his Brookings paper with Papademos, Modigliani explicitly rejected Friedman’s accelerationist hypothesis. Modigliani and Papademos (Reference Modigliani and Papademos1975) suggested the alternative concept of NIRU, defined as the rate at which price change is stable, to support real versus monetary targets and discretionary versus fixed rules policy, arguing that expansionary policies did not accelerate inflation. Both in the paper with Papademos as in the meeting with the Board staff, Modigliani explained that the NIRU was consistent with “the vertical and nonvertical Phillips curve schools,” thus avoiding the conceptual question of the shape of the Phillips curve at extreme values. According to Modigliani and Papademos, the issue at stake was not theoretical but practical, relative to the estimate of the value and stability of the non-inflationary rate that would justify a “heavily fiscal stimuli and an accommodating monetary policy” (Modigliani Reference Modigliani1975a, Box PS2, MP).Footnote 26 Modigliani also pointed out that from 1957 to 1970, the acceleration of money stock and the acceleration of inflation went against each other, and that until 1974 the correlation between changes in inflation and change in money supply, current and lagged, was approximately zero. He finally rejected the key role of changing expectations, arguing that the monetarists’ expectations view “makes no sense” unless “one presumes that manufacturers and merchants all over the country follow avidly the monetary statistics coming out of St. Louis … and immediately change their prices,” concluding that the only influence of money on prices was indirect, through aggregate demand and employment (Reference Modigliani and Papademos1975a, Box PS2, MP).
While, in the beginning of the 1970s, modelers of the Board staff retained a negative-slope Phillips curve because of empirical estimates, and Ando and Modigliani still referred to the vertical Phillips curve as an open issue, then rejecting in 1975 Friedman’s natural rate hypothesis (Modigliani and Papademos Reference Modigliani and Papademos1975), at the 1973 Brookings conference Pierce and Enzler acknowledged that in the long run it became vertical, and discussed the role of monetary policy in the context of stagflation (Pierce and Enzler Reference Pierce and Enzler1974).
The Vertical Phillips Curve in the FMP-MPS Model Context
Commenting on the state of macroeconomics in the mid-1970s, Robert Hall argued that there was not a major defense of traditional Keynesian thinking against the fresh-water (monetarists) attack, particularly with reference to the Phillips curve (Hall Reference Hall1976, p. 6). Similarly Robert Gordon (Reference Gordon and Stein1976, pp. 54–55) explained that most of the movement of professional opinion in the non-monetarist camp was forced to accept Friedman’s natural rate hypothesis because, by 1971 data, the acceleration of inflation in the late 1960s and 1970s no longer allowed the statistical rejection of the natural rate hypothesis. Gordon was among the first to introduce the natural rate hypothesis in the Keynesian framework. In his 1978 macroeconomics textbook, he acknowledge the long-run vertical Phillips curve and accepted an explanation of stagflation as due to the change in price expectations. However, as mentioned, at the 1972 meeting with the Board of Governors, Gordon still analyzed the worsening of the Phillips curve trade-off on the basis of demographic reasons, such as change in the age–sex–color composition of the labor force, along with inflationary expectations resulting from the rapid price increases of the years from 1966 to 1969. Gordon’s speech concentrated on the construction of a proper measure of the dispersion of unemployment rates, which, for him, was the most problematic issue to be solved (Gordon Reference Gordon1972, Box PS 2, MP).Footnote 27
In 1973 Pierce and Enzler introduced the long-run vertical Phillips curve in the wage and price sector of the FMP-MPS model. However, that did not lead to a recognition of Friedman’s and Phelps’s argument, and it did not reflect the supposed consensus the accelerationist hypothesis was gaining within academia. Their vertical Phillips curve did not imply a change of the structural relationships underlying the wage and price sector, focusing on a supply shock explanation. A unit price coefficient was simply imposed on de Menil and Enzler’s (Reference de Menil and Enzler1972) wage equation because its re-estimates by adding the period from 1969 to 1971 showed that for a given unemployment rate, prices rose almost proportionally to wages. They acknowledged that any attempt to reduce unemployment below its “natural rate” (whose meaning they did not clarify) would ultimately “fuel an explosive increase in the rate of inflation” (Pierce and Enzler Reference Pierce and Enzler1974, pp. 19–20).
Pierce and Enzler did not refer to the academic debate, and explained the inflation process by the wage and price spiral that began with the exogenous oil shock. They claimed the difficulties faced in modeling the wage change process, since it was consistent with a number of alternative hypotheses about the structure of the sector, all fit the data equally well but with different implications about the impact of external shocks on the model as a whole (Pierce and Enzler Reference Pierce and Enzler1974, p. 22). The alternative specifications they had in mind were not Friedman’s or Phelps’s but Perry’s (Reference Perry1970) hypothesis on the composition of labor force, and Gordon’s (Reference Gordon1970) flexible price coefficient, which led to a kink in the Phillips curve. Despite their best fitting, and although by considering Perry’s hypothesis the price coefficient was still lower than unity, the search for a compromise between the model’s ability to explain individual sectors and the ability to track data of the entire model led to neglect of Gordon’s and Perry’s specification (Pierce and Enzler Reference Pierce and Enzler1974, p. 20).Footnote 28
The meaning of past inflation among the explanatory variables, or the role played by change in expectations for an understanding of the acceleration of prices and the wage and price spiral, was not addressed. Pierce and Enzler mentioned only the difficulty in assessing the role of adaptive expectations in the wage equation, arguing that the reading of the price change coefficient in terms of “wage earners’ expectations of future inflation” would lead to a lesser influence of price rise on wage:
One might question whether an increase in prices due to something like the formation of a cartel by the oil-producing countries would be extrapolated in the manner the price term indicates. After all, the industry cannot be cartelized more than once. To the extent wage earners are aware of the source of the initial rise in price, they may not react as strongly as the wage equation implies. (Pierce and Enzler Reference Pierce and Enzler1974, p. 29)Footnote 29
Pierce and Enzler carried out policy simulations for the period from 1969 to 1973 on the effects of an external once-and-for-all oil shock on prices and unemployment under the hypothesis of an accommodating or non-accommodating monetary policy. They acknowledged that with non-accommodating policy, the rate of inflation and unemployment returned to its original level but after a very long run, pointing out that the adjustment process required that the system must be stable, and there were no guarantees that the cyclical mechanism was stable, with the “the MPS model [that was] barely so” (Reference Pierce and Enzler1974, pp. 28, 39). Under an accommodative monetary policy, the real output declined much less, but the wage and price spiral would continue to a greater extent. For the inflation rate to return to its value in the long run, it was required that the average unemployment rate be sufficiently high in the transition period. Pierce and Enzler did not discuss further any policy conclusions. As they explained, they could have derived a monetary policy that maximized an objective function based on the unemployment and inflation rate. However, they did not want “to enter in the debate about the parameters of the objective functions,” concluding that “whether or not the outcome of [an] accommodating monetary policy is regarded as superior to the outcome with unchanged policy [depended] on the weights assigned to the unemployment and inflation objectives” (Pierce and Enzler Reference Pierce and Enzler1974, p. 46).Footnote 30
The 1974 version of the wage and price block remained almost unchanged throughout until the FMP-MPS model was dismissed in the mid-1990s. As Brayton and Eileen Mauskopf pointed out in their reports on the model updates, wage changes were still explained by the conventional Phillips curve with the distributed lag on price inflation that captures the effects of expected inflation and lagged adjustments to past price changes (Reference Brayton and Mauskopf1985, pp. 200–201; also see Brayton and Mauskopf Reference Brayton and Mauskopf1987).
In a couple of papers coauthored with Ando (Reference Ando, Brayton and Cross1995) and with Ando and Arthur Kennichell (from the Research and Statistics Division), Brayton still defended the Phillips curve performance, providing also a theoretical explanation (Ando, Brayton, and Kennichell Reference Ando, Brayton and Kennichell1991). To Ando, Brayton, and Kinnichell, stagflation implied nothing definitive about its reliability since it was not merely the result of a correlation between the rate of inflation and the rate of unemployment. Both variables were endogenous to the whole macroeconometric model. On this ground, they claimed that the traditional Phillips curve was dismissed mainly for ideological reasons:
In discussions of macroeconomic theory and policy since the mid-1970s, it is often taken for granted that conventional macroeconometric models have failed in some fundamental way to account for the actual development of the US economy and that the prime culprit in this failure is the Phillips curve (Lucas and Sargent Reference Lucas and Sargent1978, Reference Lucas and Sargent1981). For those of us who have followed the performance of the modified Phillips curve and associated equations through the 1970s and most of the 1980s, the widespread perception of the complete failure of these equations to “explain” movements of prices and wages during this period has been a complete puzzle, since these equations have been more stable and more reliable over time than most other empirical macroeconomic relationships during the period in question, at least in the US. (Ando, Brayton, and Kennichell Reference Ando, Brayton and Kennichell1991, p. 201; also see Ando and Brayton Reference Ando, Brayton and Cross1995, p. 277; Ando Reference Ando, Kmenta and Ramsey1981, p. 360)Footnote 31
Ando, Brayton, and Kennichell’s (Reference Ando, Brayton and Kennichell1991) essay was devoted to providing evidence on how the wage and price block of the FMP-MPS model accounts for the dynamic pattern of unemployment and inflation from the 1960s to the 1980s, and to understanding whether changes in money supply were inflationary regardless of other conditions. Their point of departure was a reading of the Phillips curve in terms of a structural relation; that is, the inflation and unemployment relationship was conditional on a number of external factors, such as supply conditions (Reference Ando, Brayton and Kennichell1991, pp. 202, 206; see also Ando and Brayton Reference Ando, Brayton and Cross1995). They carried out static and dynamic simulations of de Menil and Enzler’s (Reference de Menil and Enzler1972) original model and Brayton and Mauskopf’s (Reference Brayton and Mauskopf1985) most recently modified version, which embedded the vertical Phillips curve, to establish the stability of the wage and price relationship and its ability to predict over a long run (the period under consideration was 1961 to 1989). According to their estimates, de Menil and Enzler’s (Reference de Menil and Enzler1972) price and wage relationship was stable as long as the relevant external conditions were stable (i.e., over the 1960s), while in the 1970s and 1980s, the Phillips curve underestimated the rate of inflation because of the low value of the price coefficient. Nonetheless, they remarked that the model caught the “basic time pattern of the historical inflation.” although not the level of inflation, “reasonably well.” By contrast, the simplest relation between inflation and unemployment, independent from other factors, became meaningless (Ando, Brayton, and Kennichell Reference Ando, Brayton and Kennichell1991, p. 210). They emphasized that as soon as the price coefficient was recognized to be close to unity, the performance was improved by fitting the data closely, as Brayton and Musakopf’s modified version showed.
They also read the Phillips curve performance as a test against Lucas’s critique (1976) since the sample period they considered included two important breaks: the oil crisis and dollar devaluation of 1973 to 1975, and the 1979–80 shift in the monetary policy regime. According to them, because their dynamic simulations showed no evidence of modifications of agents’ behavior (i.e., in the way they formed their expectations) in response to those events, they concluded that even if one could agree in principle with the Lucas critique, what mattered was how serious the agents’ shifts were under a practical perspective (also see Ando Reference Ando, Kmenta and Ramsey1981, pp. 208–210).Footnote 32
Their empirical tests were published just when the FMP-MPS was dismissed and replaced by the FRB/US macroeconometric model in 1995. According to its builders, its structure was not very far from that of the FMP-MPS model, especially regarding the long-run neoclassical conditions for equilibrium, with short-run sticky price disequilibria and monetary policy having significant effects. Nonetheless, the introduction of rational expectations as the most important transmission channel of monetary policy represented a definite break in the approach to the inflation and unemployment relationship, now explained in terms of unexpected inflation.Footnote 33 However, more than fifteen years has passed since the earlier academic dispute on the Phillips curve unstable trade-off. The FRB/US model is described as close to the new Keynesian models, representing a new consensus within the discipline. What happened to the macroeconomics discipline during the years of the 1970s and 1980s was, instead, not captured by model builders at the Research and Statistical Division of the Federal Reserve Board.
V. CONCLUSIONS
Through policy simulations and forecasts, the FMP-MPS model analyzed the system’s response to exogenous shocks and the channels and the length of time over which monetary and fiscal actions influenced aggregate demand. This paper reconstructs the difficulties model builders faced in the specification and estimation of the wage and price sector and its endogenization with the FMP-MPS macroeconometric model in the late 1960s. Although contemporary to the debate on the long-run Phillips curve, most of the attention was devoted to proper specification of the unemployment rate and the understanding of the labor market behavior. The underlying model of wage rate determination was mainly based on bargaining, with past inflation that entered among the explanatory variables of wage changes as a measure of the cost of living.
The model builders from the Federal Reserve Board and academia did not address Friedman’s and Phelps’s accelerationist argument, even if Friedman’s critique of the long-run Phillips curve and his advocacy of a monetary rule were known since the mid-1960s, and the Federal Reserve Board economists, along with the academic economists working on the model, were well aware of rising inflation.
The dispute over the stable or unstable trade-off in the long run appears marginal; it was faced as an empirical issue that did not undermine the model theoretical structure. Indeed, the introduction of a vertical Phillips curve in the mid-1970s was not the result of academic debates. As soon as empirical estimates showed that wages adjusted proportionally to price change, a unit price coefficient was simply imposed on the previous wage equation. Since then the wage and price sector remained almost unchanged until the dismissal of the FMP-MPS model in 1995, and was regarded as one of its more stable blocks (Brayton et al. Reference Brayton, Levin, Tryon and Williams1997, p. 3; see also Ando and Brayton Reference Ando, Brayton and Cross1995).
Under this perspective, model builders who worked on the Phillips curve in the context of the FMP-MPS model showed the “disconnect” between the research line pursued within academia and the macroeconomics applied by institutions since the 1970s onwards. While it is true that Keynesian analysis appeared at that time inadequate to answer the problems associated with stagflation, it also appears that the evolution of macroeconomics from Friedman and Phelps to the new classical economics is only part of the narrative, with the discipline that moved in different directions and with different timing.
